Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

An improved hydromechanical understanding of seepage-induced instability phenomena in soil Crawford-Flett, Kaley A. 2014

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2014_september_crawfordflett_kaley.pdf [ 34.54MB ]
Metadata
JSON: 24-1.0167453.json
JSON-LD: 24-1.0167453-ld.json
RDF/XML (Pretty): 24-1.0167453-rdf.xml
RDF/JSON: 24-1.0167453-rdf.json
Turtle: 24-1.0167453-turtle.txt
N-Triples: 24-1.0167453-rdf-ntriples.txt
Original Record: 24-1.0167453-source.json
Full Text
24-1.0167453-fulltext.txt
Citation
24-1.0167453.ris

Full Text

An improved hydromechanical understanding of seepage-inducedinstability phenomena in soilbyKaley A. Crawford-FlettB.E. (First Class Honors), University of Canterbury, 2008A THESIS SUBMITTED IN PARTIAL FULFILLMENTOF THE REQUIREMENTS FOR THE DEGREE OFDoctor of PhilosophyinTHE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES(Civil Engineering)The University Of British Columbia(Vancouver)April 2014c© Kaley A. Crawford-Flett, 2014AbstractInternal instability describes phenomena that occur when a soil cannot prevent the loss of its ownsmall particles in the presence of forces induced by seepage flow. A significant proportion of incidentsand failures in water-retaining structures and their foundations are attributed to the consequences ofinternal instability. There is a concern that existing dam and canal infrastructure may be vulnerableto internal instability as a consequence of deficiencies attributed to the state-of-practice at the timeof design. In order to manage infrastructure risk, there is a need for an improved science-basedunderstanding of the state-of-art ‘hydromechanical framework’ that describes the interacting material,stress and hydraulic factors understood to govern internal instability.A series of seepage tests were undertaken on five gradations in a large permeameter. The ob-jectives were to observe critical seepage-induced phenomena at the ‘upper bound’, and to verify thepresence of an extreme ‘lower bound’ to the hydromechanical space. A modified slurry depositiontechnique was used to reconstitute saturated and homogeneous specimens, and a multi-stage seep-age regime was found satisfactory to identify the critical hydromechanical condition. Phenomenaof fluidization and hydraulic uplift were found to characterize internally stable behaviour at the ‘up-per bound’ of hydromechanical instability. A stress-independent ‘lower bound’ was experimentallydefined by tests on two very unstable suffusive soils.Existing geometric methods were evaluated and found to inadequately characterize material be-haviour in widely-graded till materials: rather, the presence of plasticity was found to inhibit internalinstability. The present study quantified three necessary conditions for internal instability in gap-gradations: (1) a theoretical porosity-based microstructure framework was adapted to identify the‘α ≈ 0’ ‘particle detachment’ condition, (2) the critical seepage condition at the hydromechanical‘lower bound’ was verified in terms of a critical seepage velocity, and (3) a novel constriction sizecriterion was proposed to assess the ‘transportation potential’ for a particle in a porous medium. Itis concluded that the hydromechanical space is not fully characterized by the stress-reduction factor‘α’ alone: the present study characterizes two distinct and necessary components of material suscep-tibility for gap-gradations: (1) particle detachment, and (2) transportation potential.iiPrefaceThis dissertation is original, unpublished, independent work by the author, Kaley Crawford-Flett.iiiTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiiList of Symbols and Acronyms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxAcknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxiii1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Seepage-induced internal instability . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Objectives of the present study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.3 Thesis outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 Literature and related work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62.2 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.2.1 Preamble . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.2.2 Internal instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.2.3 Synthesis of terminology: phenomena . . . . . . . . . . . . . . . . . . . . . 92.3 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102.3.1 Experimental apparatus and seepage conditions . . . . . . . . . . . . . . . . 112.3.2 Measured and derived parameters . . . . . . . . . . . . . . . . . . . . . . . 142.3.3 Observations on the emerging body of experimental work . . . . . . . . . . 162.4 Material susceptibility: gradation shape and microstructural geometrics . . . . . . . 162.4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162.4.2 Particle size geometrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172.4.3 Soil microstructure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20iv2.4.4 Pore and constriction size . . . . . . . . . . . . . . . . . . . . . . . . . . . 252.5 Hydraulic factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292.5.1 Quantification of hydraulic condition . . . . . . . . . . . . . . . . . . . . . 292.5.2 Experimental and theoretical thresholds . . . . . . . . . . . . . . . . . . . . 302.5.3 Scale considerations and a velocity-based framework . . . . . . . . . . . . . 322.6 The influence of stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342.7 Field manifestation of instability phenomena: Coursier Dam . . . . . . . . . . . . . 352.8 Summary of literature and role of present research . . . . . . . . . . . . . . . . . . . 373 Apparatus, materials and experimental program . . . . . . . . . . . . . . . . . . . . . 823.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 823.2 Apparatus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 823.2.1 Large permeameter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 823.2.2 Water supply and seepage configuration . . . . . . . . . . . . . . . . . . . . 843.3 Instrumentation and data acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . 853.3.1 Axial load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 863.3.2 Axial deformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 863.3.3 Pore-water pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 863.3.4 Volumetric flow rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 873.4 Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 893.4.1 Material and specimen constraints . . . . . . . . . . . . . . . . . . . . . . . 893.4.2 Widely-graded till materials . . . . . . . . . . . . . . . . . . . . . . . . . . 903.4.3 Glass bead materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 903.5 Specimen preparation, reconstitution and consolidation . . . . . . . . . . . . . . . . 913.5.1 Widely-graded till materials . . . . . . . . . . . . . . . . . . . . . . . . . . 913.5.2 Glass bead materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 933.6 Test procedure and experimental program . . . . . . . . . . . . . . . . . . . . . . . 953.6.1 Widely-graded till specimens . . . . . . . . . . . . . . . . . . . . . . . . . . 953.6.2 Glass bead specimens . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 963.6.3 Experimental program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 963.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 974 Experiments at the hydromechanical ‘upper bound’ . . . . . . . . . . . . . . . . . . . 1124.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1124.2 MC test series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1134.2.1 Pre-critical response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1134.2.2 Critical response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1144.3 SC test series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1144.3.1 Pre-critical response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1154.3.2 Critical response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115v4.4 53GB22 test series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1164.4.1 Pre-critical response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1174.4.2 Critical response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1184.5 Interpretation of results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1194.5.1 Hydraulic onset of distress . . . . . . . . . . . . . . . . . . . . . . . . . . . 1194.5.2 Critical distress mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . 1204.5.3 Hydromechanical analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 1204.5.4 Hydromechanical comparison . . . . . . . . . . . . . . . . . . . . . . . . . 1214.6 Summary of experiments at the hydromechanical ‘upper bound’ . . . . . . . . . . . 1215 Experiments at the hydromechanical ‘lower bound’ . . . . . . . . . . . . . . . . . . . 1395.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1395.2 Commissioning tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1405.2.1 Experimental and theoretical verification: ‘piping by heave’ . . . . . . . . . 1405.2.2 C-72GB22 series (no rectifying filter) . . . . . . . . . . . . . . . . . . . . . 1415.3 66GB22 test series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1445.3.1 Pre-critical response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1455.3.2 Critical response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1455.4 72GB22 test series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1475.4.1 Pre-critical response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1475.4.2 Critical response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1485.5 Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1495.5.1 General observations and data trends . . . . . . . . . . . . . . . . . . . . . . 1495.5.2 Hydraulic onset of distress . . . . . . . . . . . . . . . . . . . . . . . . . . . 1505.5.3 Critical distress phenomena . . . . . . . . . . . . . . . . . . . . . . . . . . 1535.5.4 Hydromechanical analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 1535.5.5 Hydromechanical comparison . . . . . . . . . . . . . . . . . . . . . . . . . 1545.6 Summary of experiments at the hydromechanical ‘lower bound’ . . . . . . . . . . . 1556 Controlling factors in the hydromechanical space . . . . . . . . . . . . . . . . . . . . . 1706.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1706.2 Material susceptibility of widely-graded tills at the ‘upper bound’ . . . . . . . . . . 1726.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1726.2.2 Laboratory seepage testing: widely-graded tills containing ≈ 20-30% fines . 1726.2.3 State-of-practice: material susceptibility analyses . . . . . . . . . . . . . . . 1736.2.4 State-of-art material susceptibility techniques . . . . . . . . . . . . . . . . . 1766.2.5 Plasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1776.2.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1776.3 Internal instability of gap-graded ballotini . . . . . . . . . . . . . . . . . . . . . . . 1786.3.1 Particle detachment: microstructure and α ≈ 0 potential . . . . . . . . . . . 178vi6.3.2 Hydraulic conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1846.3.3 Transportation potential for potentially mobile particles . . . . . . . . . . . . 1906.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2017 Conclusions and recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2287.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2287.1.1 Coursier Dam case study . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2287.1.2 Seepage-induced phenomena at the ‘upper bound’ . . . . . . . . . . . . . . 2297.1.3 Seepage-induced phenomena at the ‘lower bound’ . . . . . . . . . . . . . . 2307.1.4 Material susceptibility and particle transportation in gap-graded glass beadmaterials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2317.1.5 Novel contributions of the present study . . . . . . . . . . . . . . . . . . . . 2327.2 Recommendations for future studies . . . . . . . . . . . . . . . . . . . . . . . . . . 233Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235A Coursier Dam: Forensic case study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245A.1 History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245A.1.1 Site geology and dam construction . . . . . . . . . . . . . . . . . . . . . . . 245A.1.2 Service life and performance overview . . . . . . . . . . . . . . . . . . . . . 246A.1.3 Site characterization and instrumentation . . . . . . . . . . . . . . . . . . . 247A.2 The potential for internal instability . . . . . . . . . . . . . . . . . . . . . . . . . . 249A.2.1 Gradation analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249A.2.2 Interpretation of susceptibility methods . . . . . . . . . . . . . . . . . . . . 252A.2.3 The case for internal instability . . . . . . . . . . . . . . . . . . . . . . . . . 254A.3 Potential for filter incompatibility . . . . . . . . . . . . . . . . . . . . . . . . . . . 254A.3.1 Performance monitoring data . . . . . . . . . . . . . . . . . . . . . . . . . . 254A.3.2 Filtration capability at the core-foundation interface . . . . . . . . . . . . . . 256A.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259B Uncertainty in measured and derived parameters . . . . . . . . . . . . . . . . . . . . 269B.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269B.2 Instrumentation calibrations and system error . . . . . . . . . . . . . . . . . . . . . 269B.3 Uncertainty in derived parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . 271B.3.1 Uncertainty in measured quantities . . . . . . . . . . . . . . . . . . . . . . . 271B.3.2 Uncertainty in derived parameters . . . . . . . . . . . . . . . . . . . . . . . 272B.4 Uncertainty in correlation: the hydromechanical plotting position . . . . . . . . . . . 273B.5 Instrument calibrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276viiC Glass bead commissioning tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283C.1 Theoretical and experimental verification . . . . . . . . . . . . . . . . . . . . . . . 283C.2 C-72GB22 test series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285C.2.1 C-72GB22 experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285D Multi-stage permeameter tests: main test program . . . . . . . . . . . . . . . . . . . . 291D.1 ‘Upper bound’ tests: MC series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291D.1.1 MC-25 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291D.1.2 MC-100 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292D.1.3 MC-50 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294D.2 ‘Upper bound’ tests: SC series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294D.2.1 SC-25 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294D.2.2 SC-50 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295D.2.3 SC-100 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296D.3 ‘Upper bound’ tests: 53GB22 series . . . . . . . . . . . . . . . . . . . . . . . . . . 297D.3.1 53GB22-0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297D.3.2 53GB22-5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299D.3.3 53GB22-50 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300D.4 ‘Lower bound’ tests: 66GB22 series . . . . . . . . . . . . . . . . . . . . . . . . . . 302D.4.1 66GB22-0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302D.4.2 66GB22-25 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303D.4.3 66GB22-50 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305D.4.4 66GB22-100 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306D.5 ‘Lower bound’ tests: 72GB22 series . . . . . . . . . . . . . . . . . . . . . . . . . . 308D.5.1 72GB22-0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308D.5.2 72GB22-25 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309D.5.3 72GB22-100 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311D.5.4 72GB22-150 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313E Visual observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315E.1 ‘Upper bound’ tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315E.1.1 MC test series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315E.1.2 SC test series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318E.1.3 53GB22 test series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321E.2 ‘Lower bound’ tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324E.2.1 66GB22 test series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325E.2.2 72GB22 test series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329viiiF Water head distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333F.1 ‘Upper bound’ tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333F.2 ‘Lower bound’ tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338G Temporal gradient response at onset of distress . . . . . . . . . . . . . . . . . . . . . . 342G.1 ‘Upper bound’ tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342G.2 ‘Lower bound’ tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347H Controlling constriction size D∗c as percentile of CSDD curve: theoretical analysis . . . 351H.1 Constriction geometry of the transportation path: theory . . . . . . . . . . . . . . . . 351H.1.1 Verification of the probabilistic CSD . . . . . . . . . . . . . . . . . . . . . . 351H.1.2 Transportation pathway and the CSD . . . . . . . . . . . . . . . . . . . . . 352H.2 Controlling constriction size D∗c as percentile of CSDD curve . . . . . . . . . . . . . 355ixList of TablesTable 2.1 Definition of seepage-induced phenomena, in terms of temporal changes in speci-men dimension, hydraulic conductivity, and specimen mass (physically admissibleparametric combinations only), refer Figure 2.32 . . . . . . . . . . . . . . . . . . 40Table 2.2 Apparatus and experimental parameters: selected studies on internal instabilityunder unidirectional seepage (Part 1 of 2) . . . . . . . . . . . . . . . . . . . . . . 41Table 2.3 Apparatus and experimental parameters: selected studies on internal instabilityunder unidirectional seepage (Part 2 of 2) . . . . . . . . . . . . . . . . . . . . . . 42Table 2.4 Measured parameters: selected studies on internal instability under unidirectionalseepage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43Table 2.5 Lubochkov suffusion thresholds at three tolerance safety factors (Kovaˆcs, 1981) . 44Table 2.6 Microstructures in two-component mixtures (Vallejo, 2001) . . . . . . . . . . . . 45Table 2.7 Critical hydraulic thresholds in granular material . . . . . . . . . . . . . . . . . . 46Table 3.1 Uncertainty in gradients derived from pore pressure measurements . . . . . . . . 99Table 3.2 Uncertainty in discharge velocity and hydraulic conductivity parameters, as de-rived using methodology detailed in Section B.3 (where klocal is calculated for atypical local specimen region 12.5 cm in length) . . . . . . . . . . . . . . . . . . 99Table 3.3 Core gradation properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100Table 3.4 Glass bead gradation properties . . . . . . . . . . . . . . . . . . . . . . . . . . . 100Table 4.1 Initial specimen properties: ‘upper bound’ widely-graded till specimens . . . . . 123Table 4.2 Initial specimen properties: ‘upper bound’ glass bead specimens . . . . . . . . . 123Table 4.3 Distress in ‘upper bound’ materials: critical hydraulic gradients and local zone ofinitiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123Table 4.4 Identification of distress mechanisms in ‘upper bound’ materials . . . . . . . . . 124Table 5.1 Initial specimen properties: 66GB22 and 72GB22 materials . . . . . . . . . . . . 156Table 5.2 Average specimen conductivity for test duration: 66GB22 and 72GB22 materials . 156Table 5.3 Distress in ‘lower bound’ materials: critical hydraulic gradients and local zone ofinitiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156xTable 5.4 Identification of seepage-induced phenomena: change in specimen dimension atthe onset of instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157Table 6.1 Gradation properties for widely-graded glacial tills with significant fines content(multi-stage seepage testing conducted at the University of British Columbia: MCand SC gradations of the present study, C-20 and C-30 gradations from Moffat,2005) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204Table 6.2 Characteristic gradation sizes for widely-graded glacial tills with significant finescontent (multi-stage seepage testing conducted at the University of British Columbia:MC and SC gradations of the present study, C-20 and C-30 gradations from Mof-fat, 2005) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204Table 6.3 Reference curves for widely graded soils: increasing Cu and HF geometric indices,after Kenney and Lau (1985/1986) . . . . . . . . . . . . . . . . . . . . . . . . . 204Table 6.4 Gradation properties: gap-graded experimental database . . . . . . . . . . . . . . 205Table 6.5 Gradation properties: coarser and finer components of the experimental database . 206Table 6.6 Microstructural analysis: gap-graded experimental database . . . . . . . . . . . . 207Table 6.7 Proportion of free finer particle from DEM (Shire and O’Sullivan, 2013a), withmicrostructural classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208Table 6.8 Critical seepage velocity in gap-graded suffusion experiments . . . . . . . . . . . 209Table 6.9 Comparison of factors in constriction size methods . . . . . . . . . . . . . . . . . 210Table 6.10 Constriction analysis: CSD method (Indraratna, 2011), and capillary tube model(Li and Fannin, 2013) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211Table 6.11 Controlling constriction analysis, after Kenney et al. (1985) . . . . . . . . . . . . 212Table 6.12 Controlling constriction size by various methods as percentile size of ‘densest’constriction size distribution, CSDD . . . . . . . . . . . . . . . . . . . . . . . . . 213Table A.1 Results: internal erosion susceptibility analyses . . . . . . . . . . . . . . . . . . 260Table B.1 Believed uncertainty in measured quantities . . . . . . . . . . . . . . . . . . . . 272Table H.1 Controlling constriction size for an n-layer transportation passage in terms of CSDcharacteristic constriction size . . . . . . . . . . . . . . . . . . . . . . . . . . . 356xiList of FiguresFigure 2.1 Controlling factors and internal erosion mechanisms (Garner and Fannin, 2010). 47Figure 2.2 Seepage-induced phenomena according to soil microstructure (Wittmann, 1978). 48Figure 2.3 Mechanical phenomena in soils due to seepage (Kezdi, 1979). . . . . . . . . . . 49Figure 2.4 Apparatus used for investigation of ‘inherent stability’ in mixed filters (USACE,1953). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49Figure 2.5 Two rigid-walled seepage apparatus for downward flow (Kenney and Lau, 1985). 50Figure 2.6 Large permeameter schematic, Moffat (2005). . . . . . . . . . . . . . . . . . . . 51Figure 2.7 Schematic rigid-walled seepage device (Sail et al., 2011). . . . . . . . . . . . . . 51Figure 2.8 Schematic illustration of flexible walled permeameter, supplied by Soiltest Inc.,from Sun (1989). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52Figure 2.9 Schematic representation: triaxial seepage device with effluent monitoring instru-mentation (Bendahmane et al., 2008). . . . . . . . . . . . . . . . . . . . . . . . 52Figure 2.10 Schematic illustration of triaxial seepage device, from Luo et al. (2013). . . . . . 53Figure 2.11 Depiction of ‘filter box’ for a horizontal seepage condition, as used in the exper-iments of Adel et al. (1988). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53Figure 2.12 Experimental apparatus with variable angular orientation (Wittmann, 1977). . . . 54Figure 2.13 Lubochkov method to assess material susceptibility (in Kovaˆcs, 1981). . . . . . . 55Figure 2.14 Split-gradation method to assess self-filtering potential (Kezdi, 1979). . . . . . . 56Figure 2.15 Representative particle sizes: original gradation (Kezdi, 1979). . . . . . . . . . . 56Figure 2.16 Method of describing gradation curve shape (Kenney and Lau, 1985). . . . . . . 57Figure 2.17 Method of describing gradation curve shape with relation to Lubochkov limit(Kenney and Lau, 1985). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57Figure 2.18 Geometric comparison of Kezdi (1979) and Kenney and Lau (1985) methods. . . 58Figure 2.19 Comparative geometric analysis: empirical database (Li and Fannin, 2008). . . . 58Figure 2.20 Geometric assessment of stability (Burenkova, 1993). . . . . . . . . . . . . . . . 59Figure 2.21 Alternative method for assessing internal instability of broadly graded silt-sand-gravel soils (Wan and Fell, 2008). . . . . . . . . . . . . . . . . . . . . . . . . . 60Figure 2.22 Geometric assessment of 65 sand-gravel filter materials classified by field perfor-mance (Ro¨nnqvist, 2010.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60xiiFigure 2.23 Geometric assessment of 59 moraine core materials classified by field perfor-mance (Ro¨nnqvist, 2010) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61Figure 2.24 Schematic illustration of inter-granular and inter-finer void ratios in a two-componentgranular mixture. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62Figure 2.25 Hypothesized suffusion-suffosion bound for widely-graded coarse fraction (Gar-ner and Sobkowicz, 2002). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63Figure 2.26 Grain size distributions for Type A and Type B materials (A˚ berg, 1992). . . . . 63Figure 2.27 Microstructural composition, volumes, and porosities for bimodal material (Skemp-ton and Brogan, 1994). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64Figure 2.28 Bimodal mixture porosity and threshold finer fraction content (Vallejo, 2001). . . 65Figure 2.29 Characteristic microstructure schematics for bimodal granular material (Vallejo,2001). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66Figure 2.30 Intergranular matrix phase diagram for host sand with varying fines content (The-vanayagam and Mohan, 2000). . . . . . . . . . . . . . . . . . . . . . . . . . . . 67Figure 2.31 Microstructural classification (Thevanayagam et al., 2002). . . . . . . . . . . . . 68Figure 2.32 Microstructural conditions for defined phenomena (Table 2.1). . . . . . . . . . . 69Figure 2.33 Geometric arrangements of uniform spheres (Wittmann, 1979). . . . . . . . . . . 70Figure 2.34 Three-particle arrangement for minimum void size and determination of the ‘dens-est’ constriction size distribution, CSDD. . . . . . . . . . . . . . . . . . . . . . . 70Figure 2.35 Four-particle arrangement for maximum void size and determination of the ‘loos-est’ constriction size distribution, CSDL. . . . . . . . . . . . . . . . . . . . . . . 71Figure 2.36 Pore area distribution comparison, by method (Wittmann, 1979). . . . . . . . . . 71Figure 2.37 Pore area distribution with increasing filtration length (Wittmann, 1979). . . . . 72Figure 2.38 Particle and constriction size distributions for linear gradations (Kenney et al.,1985). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73Figure 2.39 Controlling constriction size, D∗c , along flowpaths D′c, for a filter (Kenney et al.,1985). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74Figure 2.40 Schematic of capillary tube model, as illustrated in Indraratna and Vafai (1997). . 75Figure 2.41 Schematic illustration of the constriction size distribution by relative density CSDRd .75Figure 2.42 Gradient thresholds and ‘typical state lines’ for suffusion (Wittmann, 1977). . . . 76Figure 2.43 Critical gradient with HF stability index (Skempton and Brogan, 1994). . . . . . . 77Figure 2.44 Critical gradient with relative density for vertical upward flow (Ahlinhan andAchmus, 2010). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77Figure 2.45 Critical gradient with HF andD′15d′85stability indices (Ahlinhan and Achmus, 2010). 78Figure 2.46 Critical gradient with seepage orientation (Ahlinhan and Achmus, 2010). . . . . 78Figure 2.47 Critical gradient in terms of geometric and hydraulic influence (Perzlmaier et al.,2007). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79xiiiFigure 2.48 Relationship between critical seepage velocity and hydraulic conductivity for lowplasticity soils (Konrad and Cote, 2013). . . . . . . . . . . . . . . . . . . . . . . 79Figure 2.49 Hydromechanical relationship for select soils, (Moffat and Fannin, 2011). . . . . 80Figure 2.50 Theoretical hydromechanical envelope for one-dimensional upward flow (Li andFannin, 2012). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80Figure 2.51 Probability density functions of normalized contact force on finer and coarser soilcomponents (Shire and O’Sullivan 2013a). . . . . . . . . . . . . . . . . . . . . 81Figure 3.1 Exploded and assembled view of large permeameter apparatus: zero top stresscondition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101Figure 3.2 Exploded and assembled view of large permeameter apparatus: surcharged con-dition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102Figure 3.3 Two-stage water filtration system, manufactured by Millipore Corporation. . . . 103Figure 3.4 Variable inlet and outlet head; gravity-driven seepage at low gradients. . . . . . . 104Figure 3.5 Dual-reservoir gravity seepage control configuration: low gradients. . . . . . . . 105Figure 3.6 Pressurized seepage control system: high gradients. . . . . . . . . . . . . . . . . 106Figure 3.7 Top load cell, Interface model 1210AF-5k. . . . . . . . . . . . . . . . . . . . . 107Figure 3.8 Submersible bottom load cell, Honeywell model 41. . . . . . . . . . . . . . . . 107Figure 3.9 Custom-made linear variable differential transformer (LVDT). . . . . . . . . . . 108Figure 3.10 Measurement of axial deformation: grid configuration for zero top stress conditions.108Figure 3.11 Particle size distributions for glass bead gradations. . . . . . . . . . . . . . . . . 109Figure 3.12 Particles size distributions for core materials, illustrating oversize treatment. . . . 109Figure 3.13 Large permeameter port configuration for core tests. . . . . . . . . . . . . . . . 110Figure 3.14 Large permeameter boundary mesh and port configuration for glass bead tests. . 111Figure 3.15 Overview of experimental program for multi-stage seepage. . . . . . . . . . . . 111Figure 4.1 MC test series: variation of discharge velocity and axial displacement with aver-age gradient. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125Figure 4.2 MC test series: normalized local conductivity ( klocalkav ) with average gradient (iav) intop (TOP), upper-mid (UM), lower-mid (LM) and bottom (BOT) specimen regions 126Figure 4.3 SC test series: variation of discharge velocity and axial displacement with averagegradient. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126Figure 4.4 SC test series: normalized local conductivity ( klocalkav ) with average gradient (iav). . 127Figure 4.5 53GB22-0: average hydraulic conductivity and axial displacement. . . . . . . . 128Figure 4.6 53GB22-0: normalized local hydraulic conductivity ( klocalkav in top, middle and bot-tom specimen regions) with average gradient, iav. . . . . . . . . . . . . . . . . . 128Figure 4.7 53GB22-5: average hydraulic conductivity and axial strain. . . . . . . . . . . . . 129Figure 4.8 53GB22-5: normalized local hydraulic conductivity ( klocalkav in top, middle and bot-tom specimen regions) with average gradient, iav. . . . . . . . . . . . . . . . . . 129Figure 4.9 53GB22-50: average hydraulic conductivity and axial strain. . . . . . . . . . . . 130xivFigure 4.10 53GB22-50: normalized local hydraulic conductivity ( klocalkav in top, middle andbottom specimen regions) with average gradient, iav. . . . . . . . . . . . . . . . 130Figure 4.11 MC tests: Stress distribution along specimen length: initial and final stages. . . . 131Figure 4.12 SC tests: Stress distribution along specimen length: initial and final stages. . . . 131Figure 4.13 53GB22 tests: Stress distribution along specimen length: initial and final stages. 132Figure 4.14 MC-25: local hydromechanical paths to pre-critical seepage stage. . . . . . . . . 133Figure 4.15 MC-100: local hydromechanical paths to pre-critical seepage stage. . . . . . . . 133Figure 4.16 All MC tests: critical hydromechanical state (including a third test previouslyconducted at UBC, ‘MC-50’). . . . . . . . . . . . . . . . . . . . . . . . . . . . 134Figure 4.17 SC-25: local hydromechanical paths to pre-critical seepage stage. . . . . . . . . 134Figure 4.18 SC-50: local hydromechanical paths to pre-critical seepage stage. . . . . . . . . 135Figure 4.19 SC-100: local hydromechanical paths to pre-critical seepage stage. . . . . . . . . 135Figure 4.20 All SC tests: critical hydromechanical state. . . . . . . . . . . . . . . . . . . . . 136Figure 4.21 53GB22-0: local hydromechanical paths for all pre-critical seepage stages. . . . 136Figure 4.22 53GB22-5: local hydromechanical paths for all pre-critical seepage stages. . . . 137Figure 4.23 53GB22-50: local hydromechanical paths for all seepage stages (no critical con-dition observed). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137Figure 4.24 All 53GB22 tests: critical hydromechanical state in 53GB22-0 and 53GB22-5tests; non-critical condition in 53GB22-50 test. . . . . . . . . . . . . . . . . . . 138Figure 5.1 66GB22-0: average hydraulic conductivity and axial displacement. . . . . . . . 158Figure 5.2 66GB22-25: average hydraulic conductivity and axial displacement. . . . . . . . 158Figure 5.3 66GB22-50: average hydraulic conductivity and axial displacement. . . . . . . . 159Figure 5.4 66GB22-100: average hydraulic conductivity and axial displacement. . . . . . . 159Figure 5.5 66GB22 test series: normalized local conductivity with average stage gradient. . 160Figure 5.6 72GB22-0: average hydraulic conductivity and axial displacement. . . . . . . . 161Figure 5.7 72GB22-25: average hydraulic conductivity and axial displacement. . . . . . . . 161Figure 5.8 72GB22-100: average hydraulic conductivity and axial displacement. . . . . . . 162Figure 5.9 72GB22-150: average hydraulic conductivity and axial displacement. . . . . . . 162Figure 5.10 72GB22 test series: normalized local conductivity with average stage gradient. . 163Figure 5.11 66GB22 tests: Stress distribution along specimen length: initial and final stages. 164Figure 5.12 72GB22 tests: Stress distribution along specimen length: initial and final stages. 164Figure 5.13 66GB22-0: local hydromechanical paths to final seepage stage. . . . . . . . . . . 165Figure 5.14 66GB22-25: local hydromechanical paths to final seepage stage. . . . . . . . . . 165Figure 5.15 66GB22-50: local hydromechanical paths to final seepage stage. . . . . . . . . . 166Figure 5.16 66GB22-100: local hydromechanical paths to final seepage stage. . . . . . . . . 166Figure 5.17 All 66GB22 tests: critical hydromechanical state prior to onset of suffusion. . . . 167Figure 5.18 72GB22-0: local hydromechanical paths to final seepage stage. . . . . . . . . . . 167Figure 5.19 72GB22-25: local hydromechanical paths to final seepage stage. . . . . . . . . . 168Figure 5.20 72GB22-100: local hydromechanical paths to final seepage stage. . . . . . . . . 168xvFigure 5.21 72GB22-150: local hydromechanical paths to final seepage stage. . . . . . . . . 169Figure 5.22 All 72GB22 tests: critical hydromechanical state prior to onset of suffusion. . . . 169Figure 6.1 Chapter outline. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214Figure 6.2 Widely-graded till materials testing in a large permeameter apparatus at UBC:MC and SC (present study), C-20 and C-30 from Moffat (2005). . . . . . . . . . 215Figure 6.3 Continuously graded particle size distributions: (1) Fuller curve, Fuller exponentb = 0.3,0.5,0.7 (Cu = 390, 36 and 13, respectively), and (2) linear-logarithmicgradation, Cu = 1.1,3.2,10 and 32. . . . . . . . . . . . . . . . . . . . . . . . . . 215Figure 6.4 HF with increasing Cu for (1) ‘Fuller’ curve, and (2) linear-logarithmic particlesize distributions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216Figure 6.5 Geometric assessment method of Burenkova (1993), extrapolated for widely-graded till soils. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216Figure 6.6 Geometric assessment method of Wan and Fell (2008) for widely-graded till soils:MC∗ and SC (present study), C-20 and C-30 from Moffat (2005). . . . . . . . . 217Figure 6.7 Domains of the intergranular matrix phase diagram. . . . . . . . . . . . . . . . . 217Figure 6.8 Intergranular matrix phase plot for experimental gradations: glass bead gap-gradations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218Figure 6.9 Intergranular matrix phase plot for experimental gradations: natural soil gap-gradations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218Figure 6.10 Gap-graded experimental gradations (glass bead particles). . . . . . . . . . . . . 219Figure 6.11 Gap-graded experimental gradations (soil particles). . . . . . . . . . . . . . . . 219Figure 6.12 Percent of finer component in ‘free’ stress state, (Shire and O’Sullivan, 2013a)with intergranular matrix phase classification. . . . . . . . . . . . . . . . . . . . 220Figure 6.13 Influence of microstructure on α-based plotting position in the hydromechanicalspace (as postulated by Li, 2008). . . . . . . . . . . . . . . . . . . . . . . . . . 221Figure 6.14 Experimental mean pore velocity and theoretical transportation thresholds, afterPerzlmaier et al. (2007). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222Figure 6.15 Capillary tube analysis for transportation potential, after Kovaˆcs (1981) and Liand Fannin (2013). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223Figure 6.16 Controlling constriction size analysis for internal instability (Indraratna, 2011). . 224Figure 6.17 Controlling constriction size for filtration, after Kenney et al. (1985). . . . . . . . 225Figure 6.18 Semi-empirical threshold for internal instability: 95th percentile of the densestconstriction size distribution, CSDD, against characteristic finer fraction size (d′85). 226Figure 6.19 Clast-supported microstructures: influence of controlling constriction size onplotting position in hydromechanical space. . . . . . . . . . . . . . . . . . . . . 227Figure A.1 Coursier Dam: interpreted plan view. . . . . . . . . . . . . . . . . . . . . . . . 261Figure A.2 Coursier Dam, decommissioned. Aerial photograph indicating sinkhole locationsrelated to Unit 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262xviFigure A.3 Interpreted section A-A’. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263Figure A.4 Sinkhole A (1992) surface expression (Garner et al., 2004. . . . . . . . . . . . . 263Figure A.5 Sinkholes B and C: exposure of subsurface pipes. . . . . . . . . . . . . . . . . . 264Figure A.6 2011 sampling locations at elevation near section B-B’. . . . . . . . . . . . . . . 265Figure A.7 Unit 4 and Unit 5: foundation samples and historic gradation envelopes. . . . . . 265Figure A.8 Foundation samples: unified stability chart (after Li and Fannin, 2008). . . . . . 266Figure A.9 Foundation samples: Burenkova (1993) ‘suffosion’ chart (after Burenkova, 1993). 266Figure A.10 Foundation samples: stability plot, after Wan and Fell (2008)). . . . . . . . . . . 267Figure A.11 1992 sinkhole A incident: downstream piezometric response. . . . . . . . . . . . 267Figure A.12 Lower core and sinkhole gradations; Unit 4 gradation envelope and design filterenvelope. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268Figure B.1 MC material: Uncertainty in ‘upper bound’ hydromechanical plotting position. . 274Figure B.2 SC material: Uncertainty in ‘upper bound’ hydromechanical plotting position. . . 274Figure B.3 53GB22 material: Uncertainty in ‘upper bound’ hydromechanical plotting position.275Figure B.4 DPT2: calibration regression curve and standard error, s(y). . . . . . . . . . . . 276Figure B.5 DPT3: calibration regression curve and standard error, s(y). . . . . . . . . . . . 277Figure B.6 DPT4: calibration regression curve and standard error, s(y). . . . . . . . . . . . 277Figure B.7 DPT5: calibration regression curve and standard error, s(y). . . . . . . . . . . . 278Figure B.8 DPT6: calibration regression curve and standard error, s(y). . . . . . . . . . . . 278Figure B.9 TPT3: calibration regression curve and standard error, s(y). . . . . . . . . . . . . 279Figure B.10 TPT4: calibration regression curve and standard error, s(y). . . . . . . . . . . . . 279Figure B.11 TPT5: calibration regression curve and standard error, s(y). . . . . . . . . . . . . 280Figure B.12 TPT6: calibration regression curve and standard error, s(y). . . . . . . . . . . . . 280Figure B.13 TPT7: calibration regression curve and standard error, s(y). . . . . . . . . . . . . 281Figure B.14 Bottom load cell: calibration regression curve and standard error, s(y). . . . . . . 281Figure B.15 LCT: calibration regression curve and standard error, s(y). . . . . . . . . . . . . 282Figure C.1 C-GB100-0: Test summary: variation in discharge velocity with average hy-draulic gradient. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283Figure C.2 C-GB100-0: water head distribution along specimen length for all pre-criticalseepage stages. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284Figure C.3 C-72GB22-0: Test summary: variation in discharge velocity with average hy-draulic gradient. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286Figure C.4 C-72GB22-0: Significant accumulations of finer particles on specimen top sur-face at gradients exceeding iav = 0.7. . . . . . . . . . . . . . . . . . . . . . . . 287Figure C.5 C-72GB22-50: Test summary: variation in discharge velocity with average hy-draulic gradient. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288Figure C.6 C-72GB22-50: Local vertical region deficient of finer fraction particles. . . . . . 289xviiFigure C.7 C-72GB22-100: Test summary: variation in discharge velocity with average hy-draulic gradient. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290Figure C.8 C-72GB22-HME: Tentative hydromechanical effective stress-gradient relationfor C-72GB22 test series. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290Figure E.1 MC-25 test: visual observations. . . . . . . . . . . . . . . . . . . . . . . . . . . 316Figure E.2 MC-100 test: visual observations. . . . . . . . . . . . . . . . . . . . . . . . . . 317Figure E.3 SC-25 test: visual observations. . . . . . . . . . . . . . . . . . . . . . . . . . . 318Figure E.4 SC-50 test: visual observations. . . . . . . . . . . . . . . . . . . . . . . . . . . 319Figure E.5 SC-100 test: visual observations. . . . . . . . . . . . . . . . . . . . . . . . . . . 320Figure E.6 53GB22-0 test: visual observations. . . . . . . . . . . . . . . . . . . . . . . . . 321Figure E.7 53GB22-5 test: visual observations. . . . . . . . . . . . . . . . . . . . . . . . . 322Figure E.8 53GB22-50 test: visual observations. . . . . . . . . . . . . . . . . . . . . . . . 323Figure E.9 Initial specimen condition; pre- and post- consolidation. . . . . . . . . . . . . . 324Figure E.10 66GB22-0 test: visual observations. . . . . . . . . . . . . . . . . . . . . . . . . 325Figure E.11 66GB22-25 test: visual observations. . . . . . . . . . . . . . . . . . . . . . . . 326Figure E.12 66GB22-50 test: visual observations. . . . . . . . . . . . . . . . . . . . . . . . 327Figure E.13 66GB22-100 test: visual observations. . . . . . . . . . . . . . . . . . . . . . . . 328Figure E.14 72GB22-0 test: visual observations. . . . . . . . . . . . . . . . . . . . . . . . . 329Figure E.15 72GB22-25 test: visual observations. . . . . . . . . . . . . . . . . . . . . . . . 330Figure E.16 72GB22-100 test: visual observations. . . . . . . . . . . . . . . . . . . . . . . . 331Figure E.17 72GB22-150 test: visual observations. . . . . . . . . . . . . . . . . . . . . . . . 332Figure F.1 MC-25: Water head distribution along specimen length. . . . . . . . . . . . . . 333Figure F.2 MC-100: Water head distribution along specimen length. . . . . . . . . . . . . . 334Figure F.3 SC-25: Water head distribution along specimen length. . . . . . . . . . . . . . . 334Figure F.4 SC-50: Water head distribution along specimen length. . . . . . . . . . . . . . . 335Figure F.5 SC-100: Water head distribution along specimen length. . . . . . . . . . . . . . 335Figure F.6 53GB22-0: Water head distribution along specimen length. . . . . . . . . . . . . 336Figure F.7 53GB22-5: Water head distribution along specimen length. . . . . . . . . . . . . 336Figure F.8 53GB22-50: Water head distribution along specimen length. . . . . . . . . . . . 337Figure F.9 66GB22-0: Water head distribution along specimen length. . . . . . . . . . . . . 338Figure F.10 66GB22-25: Water head distribution along specimen length. . . . . . . . . . . . 338Figure F.11 66GB22-50: Water head distribution along specimen length. . . . . . . . . . . . 339Figure F.12 66GB22-100: Water head distribution along specimen length. . . . . . . . . . . 339Figure F.13 72GB22-0: Water head distribution along specimen length. . . . . . . . . . . . . 340Figure F.14 72GB22-25: Water head distribution along specimen length. . . . . . . . . . . . 340Figure F.15 72GB22-100: Water head distribution along specimen length. . . . . . . . . . . 341Figure F.16 72GB22-150: Water head distribution along specimen length. . . . . . . . . . . 341xviiiFigure G.1 MC-25: Local variation in hydraulic gradient at critical seepage condition. . . . 342Figure G.2 MC-100: Local variation in hydraulic gradient at critical seepage condition. . . . 343Figure G.3 SC-25: Local variation in hydraulic gradient at critical seepage condition. . . . . 343Figure G.4 SC-50: Local variation in hydraulic gradient at critical seepage condition. . . . . 344Figure G.5 SC-100: Local variation in hydraulic gradient at critical seepage condition. . . . 344Figure G.6 53GB22-0: Local variation in hydraulic gradient at critical seepage condition. . . 345Figure G.7 53GB22-5: Local variation in hydraulic gradient at critical seepage condition. . . 345Figure G.8 53GB22-50: Local variation in hydraulic gradient; no observation of distress. . . 346Figure G.9 66GB22-0: Local variation in hydraulic gradient at critical seepage condition. . . 347Figure G.10 66GB22-25: Local variation in hydraulic gradient at critical seepage condition. . 347Figure G.11 66GB22-50: Local variation in hydraulic gradient at critical seepage condition. . 348Figure G.12 66GB22-100: Local variation in hydraulic gradient at critical seepage condition. 348Figure G.13 72GB22-0: Local variation in hydraulic gradient at critical seepage condition. . . 349Figure G.14 72GB22-25: Local variation in hydraulic gradient at critical seepage condition. . 349Figure G.15 72GB22-100: Local variation in hydraulic gradient at critical seepage condition. 350Figure G.16 72GB22-150: Local variation in hydraulic gradient at critical seepage condition. 350Figure H.1 Constriction size distributions by DEM/Delaunay and the Locke et al. (2001)probabilistic method. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357Figure H.2 Pore network, illustrating multiple pore exits. . . . . . . . . . . . . . . . . . . . 358Figure H.3 Particle transport passage with ‘n’ inter-pore movements: controlling transporta-tion pathway Dc,passage considering ‘n’ encounters with the CSD. . . . . . . . . . 359Figure H.4 Effect of increasing inter-pore movements on constriction size of the transporta-tion passage, CSDpassage for ‘n’ layers. . . . . . . . . . . . . . . . . . . . . . . . 360Figure H.5 Illustration of controlling constriction size by various methods. . . . . . . . . . . 361Figure H.6 Semi-empirical threshold for internal instability: 75th percentile of the densestconstriction size distribution, CSDD, against representative finer fraction size (d85). 362Figure H.7 Semi-empirical threshold for internal instability: 80th percentile of the densestconstriction size distribution, CSDD, against representative finer fraction size (d85). 363Figure H.8 Semi-empirical threshold for internal instability: 85th percentile of the densestconstriction size distribution, CSDD, against representative finer fraction size (d85). 364Figure H.9 Semi-empirical threshold for internal instability: 90th percentile of the densestconstriction size distribution, CSDD, against representative finer fraction size (d85). 365Figure H.10 Semi-empirical threshold for internal instability: 100th percentile of the densestconstriction size distribution, CSDD, against representative finer fraction size (d85). 366xixList of Symbols and Acronymsα — proportion of applied stress transferred to finer particles within a homogeneoussoil (after Skempton and Brogan, 1994)∆H — difference in total head∆iav — increment of hydraulic gradient applied in successive seepage stagesγ ′ — submerged unit weightγd — dry unit weightγw — unit weight of waterσ ′vb — vertical effective stress at the bottom of the specimenσ ′vt — vertical effective stress at the top of the specimenσ ′vm — mean vertical effective stress in the defined zone¯σ ′vm — normalized mean vertical effective stressA — cross-sectional area in direction of seepage flowBCHydro — British Columbia Hydro and Power Authoritycc — cubic centimetresCu — Coefficient of Uniformity, =D60D10CSD — constriction size distributionCSDD — ‘densest’ constriction size distribution, corresponding to probabilistic occur-rence of three-particle arrangements based on a soil particle size distribution,after Silveira (1965)CSDDEM — constriction size distribution generated by discrete element modellingCSDL — ‘loosest’ constriction size distribution, corresponding to probabilistic occur-rence of four-particle arrangements based on a soil particle size distribution,after Silveira et al. (1975)CSDRd — constriction size distribution as a function of CSDD, CSDL and Rd , by methodof Locke et al. (2001)d′xx — particle diameter corresponding to ‘XX%’ mass passing for the finer compo-nentd0 — average pore diameter, after Li and Fannin (2013)D′xx — particle diameter corresponding to ‘XX%’ mass passing for the coarser com-ponent(D′15d′85)gap— ratio of the 15th percentile size of the coarser fraction divided by the 85th per-centile size of the finer fraction in a gap-graded materal, where the gradationsplit occurs at the size gapDxx — particle diameter corresponding to ‘XX%’ mass passing (complete gradation)D∗c — controlling constriction size, after Kenney et al. (1985)xxDc,35 — 35th percentile size of the constriction size distribution by the probabilisticmethod of Locke et al. (2001) and Indraratna (2011)Dc,DXX — XX th percentile size of the ‘densest’ constriction size distribution (after Sil-veira, 1965)DEM — discrete element modellingDPT — differential pore-water pressure transducere — void ratioec — intergranular void ratio (refer Figure 2.24)ec,max — maximum void ratio, emax, of the coarser soil fractionec,min — minimum void ratio, emin, of the coarser soil fractione f — interfiner void ratio (refer Figure 2.24)e f ,max — maximum void ratio, emax, of the finer soil fractione f ,min — minimum void ratio, emin, of the finer soil fractionemax — maximum void ratio, as determined by ASTM-D4253 (2006b) and ASTM-D4254 (2006c) or similaremin — minimum void ratio, as determined by ASTM-D4253 (2006b) and ASTM-D4254 (2006c) or similaremix,max — maximum void ratio of a two-component mixtureF — mass fraction passingFd or FD — mass fraction passing particle diameter d, or D (respectively)g — gramsh’ — = D90D60 , conditional factor of uniformity (Burenkova, 1993)h” — = D90D15 , conditional factor of uniformity (Burenkova, 1993)H — mass increment = F4d−Fd , after Kenney and Lau (1985)HF — mass increment divided by mass fraction passing, after Kenney and Lau (1985)i — hydraulic gradient = ∆H∆Liav — average hydraulic gradient across the specimen length (differential total headdivided by specimen length)ich — ‘threshold’ hydraulic gradient for movement of particles independent of grav-ity (Skempton and Brogan, 1994)icr — critical hydraulic gradientilocal — average hydraulic gradient across the local zone of measurement (differentialtotal head divided by local length)ICOLD — International Commission on Large Damsk — hydraulic conductivitykav — average hydraulic conductivity across the specimenkav,i — initial average hydraulic conductivity across the specimenkav, f — final average hydraulic conductivity across the specimenkbot — hydraulic conductivity in the bottom zone of the specimen (refer Figure 3.13and Figure 3.14)klocal — average hydraulic conductivity across the local zone of measurementkLM — hydraulic conductivity in the lower-mid zone of the specimen (refer Fig-ure 3.13)kmid — hydraulic conductivity in the middle zone of the specimen (refer Figure 3.14)ktop — hydraulic conductivity in the top zone of the specimen (refer Figure 3.14)kUM — hydraulic conductivity in the upper-mid zone of the specimen (refer Fig-ure 3.13)xxikPa — kiloPascalL — specimen lengthL — litreL0 — initial specimen lengthLD — ratio of specimen length to diameterLVDT — linear variable displacement transformer‘Lower bound’ — lower bound to the onset of seepage-induced internal instability in the hy-dromechanical spaceM — mass of soil particlesM0 — initial mass of soil particlesn — porositync — intergranular porosity (refer Figure 2.24)nc,max — maximum porosity, nmax, of the coarser soil fractionnc,min — minimum porosity, nmin, of the coarser soil fractionn f — interfiner porosity (refer Figure 2.24)n f ,max — maximum porosity, nmax, of the finer soil fractionn f ,min — minimum porosity, nmin, of the finer soil fractionnmax — maximum porosity, as determined by ASTM-D4253 (2006b) and ASTM-D4254 (2006c) or similarnmin — minimum porosity, as determined by ASTM-D4253 (2006b) and ASTM-D4254 (2006c) or similarnmix,min — minimum porosity of the two-component mixturePI — plasticity indexPSD — Particle size distributionPsi — pounds per square inchQ — volumetric flow rateRd — relative densityRd,C — relative density of the coarser componentRd,F — relative density of the finer componentS — (selected studies in the literature) mass fraction passing, equivalent to F in thepresent studyS f — finer fraction content in a gap-graded specimenS∗ — critical finer fraction contentTs — Stage durationTPT — total pore-water pressure transducer‘Upper bound’ — upper bound to the onset of seepage-induced internal instability in the hy-dromechanical spacevcr — critical velocity at the onset of instabilityvd — discharge velocityvs — seepage velocity =vdnvs,cr — critical seepage velocityvp,av — mean pore velocity =vsT , where tortuosity, T =pi2Vv — volume of voids in a unit volumeVv,C — volume of voids separating coarser fraction particles in a unit volumeVp — volume of solid particles in a unit volumeVp,C — volume of coarser fraction particles in a unit volumeVp,F — volume of finer fraction particles in a unit volumexxiiAcknowledgmentsThis body of work would not have been possible without incredible support from a number of sources.Much credit is owed to Professor R. Jonathan Fannin, whose focus and enthusiasm kept the Ph.D.program ‘honest’. I sincerely appreciate the constructive feedback that I have received over the courseof this study. I would also like to thank the wider geotechnical research group at UBC, particularlyDrs. John Howie and Mahdi Taiebat who, along with Dr. Roger Beckie, have asked a number offruitful questions throughout the program of study.I owe special thanks to Steve Garner of BC Hydro for his unbridled enthusiasm and thoughtfulinsights on the wider university-industry collaboration. This industry contribution has made my timeat UBC a very meaningful experience.I would like to express my gratitude to those who assisted in my experimental efforts. Within theUBC Civil Engineering group, I am particularly grateful to Harald Schrempp, Scott Jackson, MarkRigolo and Bill Leung. Bill brightened the mood in the laboratory on a near-daily basis and I amgrateful for the perspective he provided in times of experimental difficulty. The comradeship of myfellow graduate students is also greatly appreciated.This research was funded by Natural Sciences and Engineering Research Council of Canada(NSERC) in conjunction with the British Columbia Hydroelectric and Power Authority (BC Hydro). Iam also extremely grateful for the support provided by the Hume Fellowship (courtesy of the Instituteof Professional Engineers of New Zealand, IPENZ) at the outset of my graduate education.I began this PhD program as a result of the enthusiasm and curiosity instilled in me by a numberof people over many years. To my family: thank you for a most practical and robust upbringing.Thankfully I survived to tell the tale. To the engineers that led me head-first into this field: gratefulthanks for your mentorship and trust.Finally, sincere thanks to my friends who have served as unwavering navigational stars duringthis period. I’m truly blessed to know a remarkable collection of people who have persisted in timesof cloud cover and short-sightedness (Aimee, Alana, Antonio, Mark H., Matt D., Matt K., Matthias,Shona). Thanks for keeping me honest and true, and largely on the charted course. In particular:Brent, who has always told me what I needed to hear, not what I wanted to hear. Your faith has meantthe world to me at various times over the past decade. Jess/Cabbage, who has been along for the ride- sharing drafting duties at various times - over the past 25 years. To Gail, Michelle and Bertie, trueand honest friends in every sense of the word. Lastly, most heartfelt thanks to Mark ‘too-legit-to-quit’Styler, for his enduring patience.xxiiiChapter 1Introduction1.1 Seepage-induced internal instabilitySeepage-induced internal instability describes the downstream transportation of soil particles fromwithin a soil unit. Comprehensive analyses of international dam failures (i.e. Foster et al., 2000 andXu et al., 2009) suggest that around half of all dam incidents and failures are attributed to internal‘seepage erosion’ or piping. Excluding non-structural causes of failure (i.e. over-topping; equipmentor appurtenant failures), seepage-induced erosion or piping is believed to account for approximately90% of structural failures in embankment dams.The present ‘state-of-practice’ for internal instability is based on a body of work characterizingthe geometric shape of the soil particle size distribution (e.g. Kezdi, 1979; Sherard, 1979; Kenneyand Lau, 1985; Burenkova, 1993) and believed verified by decades of industry experience. In reality,most embankment dams constructed for the purposes of hydro-electricity generation were designedto a now-superseded state-of-practice. As of 2003, the median age of hydro dams in the Canadianprovince of British Columbia (B.C.) was 40 years, while 90% of dams were more than 20 yearsold (Prowse et al., 2004). This ageing infrastructure is critical to B.C. society as hydro-power ac-counts for approximately 80% of electricity generation in the province. Similar statistics exist inother parts of the world: for example, as of 2013, 60% of electricity in New Zealand is generatedby hydro-electric means, via infrastructure with a median age of 57 years (New Zealand Ministryof Economic Development, 2012, and NZEA, 2013). Further compounding the risk managementassociated with ageing infrastructure is a tendency for internal instability to manifest itself in a time-dependent fashion - i.e. long after first filling (Foster et al., 2000). Ongoing differential settlements,changes in hydro-geology due to climate or water allocation changes, and the cumulative impactof cyclical reservoir loading may also contribute to episodic events of internal instability long be-fore macro-scale symptoms are observed (ICOLD, 2013a). Accordingly, soil structures may perform‘incident-free’ for decades prior to distress, a likelihood exemplified by sinkhole incidents at CoursierDam (Section 2.7), Bennett Dam, BC (Stewart, 2000) and Tekapo Canal, New Zealand (Amos et al.,2010). Engineers today therefore act as custodians of ageing dam and canal infrastructure that may1be vulnerable as a consequence of deficiencies attributed to the state-of-practice at the time of design.Indeed, the need to explicitly address internal instability in engineering design has only re-cently become evident, with the first publications on geometric ‘criteria’ emerging around 1970 (e.g.Lubochkov, 1969 (in Kovaˆcs, 1981); De Mello, 1975; Kezdi, 1979; Sherard, 1979). Such geometricconcepts were subject to years of critique, refinement (e.g. Kenney and Lau, 1985,1986) and industryapplication before becoming generally accepted as the ‘state-of-practice’.Furthermore, today’s ‘state-of-practice’ does not yet encompass more recent ‘state-of-art’ contri-butions by researchers in the past 10 to 20 years. Acknowledging the need for continued advancementin internal instability research, the International Commission on Large Dams (ICOLD) has attemptedto capture the present state-of-art in a Bulletin on Internal Erosion (ICOLD, 2013a,b). The presentstate-of-art encompasses factors beyond the gradation shape that is the focus of the present state-of-practice. Specifically, the initiation of internal instability is attributed to the interaction of threefactors: (1) material susceptibility, (2) stress conditions, and (3) hydraulic loading (Garner and Fan-nin, 2010).The driver for contributions of the present thesis is established in the recent research of Moffat(2005) and Li (2008). Moffat (2005) proposed a relation between stress on a potentially unstable soiland the critical hydraulic gradient required to initiate internal instability. Li (2008) further expandedthe empirical database of Moffat (2005) and theoretically unified the proposed ‘hydromechanical’framework (Li and Fannin, 2012). Specifically, Li (2008) postulated a linear relation between criti-cal gradient and mean vertical stress at the onset of instability in a number of materials. Li (2008)found that the slope of the stress-gradient (i.e. hydromechanical) threshold for a given material isa function of material susceptibility. It follows that an infinite number of material-dependent hy-dromechanical thresholds may exist as a spectrum in stress-gradient space, given that each gradationwill possess a unique degree of susceptibility to internal instability. The spectrum of material sus-ceptibility will range from extremely unstable materials, to stable materials deemed unsusceptible toseepage-induced internal instability.This thesis follows the work of Moffat (2005) and Li (2008), and seeks to clarify the upper andlower limits to the spectrum of material susceptibility in the hydromechanical stress:gradient space.‘Stable’ soils plot at - or close to - the theoretical hydromechanical ‘upper bound’ to the onset ofinternal instability 1. The material characteristics of stable soils vary greatly: from uniform grada-tions, to coarse gap-gradations, to very widely graded soils containing fines. To-date, no geometricor hydromechanical framework can reliably predict stable behaviour for the diverse group of soilsin which stability is observed. At the hydromechanical ‘upper bound’, the thesis has pragmatic in-tent: specifically, to observe and characterize ‘stable’ mechanisms of fluidization or uplift in the verywidely graded soils commonly used in dam construction, yet rarely tested in the laboratory due toparticle size constraints. At the ‘upper bound’, the intent of the thesis is to: (1) explore the validity of1In this thesis, ‘upper bound’ hereafter refers to the upper bound to the onset of seepage-induced internal instabilityin the hydromechanical space. By definition, the theoretical ‘upper bound’ represents equilibrium between seepage andweight forces on a soil element. Accordingly, at the ‘upper bound’, a granular material will experience seepage force (perunit volume) sufficient to induce a state of zero effective stress, yielding fluidization.2the current ‘state-of-practice’ for typical construction materials, and (2) advance the state-of-art forthe hydromechanical framework of Li (2008).Conversely, an extreme ‘lower bound’ to the onset of seepage-induced internal instability is pro-posed by Li and Fannin (2012) 2. This hydromechanical bound is postulated to represent the highlyunstable behaviour of ‘borderline segregable’ materials. At the ‘lower bound’, this thesis aims tocontribute to the present state-of-art through testing of ‘ideal’ glass bead gradations that can be mod-elled theoretically. Specifically, the present study explores shortcomings in the current understandingof instability in gap-graded soils. Seepage experiments on glass bead gradations will provide data toinform companion work by other researchers in the field of micromechanical modelling.In summary, the intent of the thesis is to advance our understanding of internal instability bydefining ‘where’ - in stress:gradient space - concerns should lie for the onset of the phenomenon. Thethesis goal is a contribution to the state-of-art understanding that will inform decision-support toolsfor prioritization of remedial earthworks, and aid in future monitoring of those remedial works. Theengineering need is exemplified in recent seepage-related remediation activities at Bennett Dam, BC(Stewart, 2000), and Tekapo Canal, New Zealand (Amos et al., 2010).1.2 Objectives of the present studyIn professional practice, there exists a clear need for an improved science-based understanding ofthe factors governing internal instability mechanisms in soil (ICOLD, 2013a,b). The thesis aims toaddress some present ‘knowledge-gaps’ in the science-based understanding of internal instability.Objectives of the study are as follows:• Conduct a forensic investigation at the decommissioned Coursier Dam site, in order to:– Better understand the ‘signature’ of instability phenomena in the field;– Gain an understanding of the state-of-practice in material susceptibility and compare es-tablished geometric techniques with present state-of-art analysis frameworks;– Gain insights into spatial and temporal factors influencing the manifestation of seepagephenomena; and,– Inform experimental foci and potential limitations of the laboratory research program.• Investigate seepage phenomena at the ‘upper bound’ to the onset of hydromechanical insta-bility with seepage testing on two widely-graded soils typically encountered in practice. Thewidely-graded till soils (Cu >> 100) were provided by the industry partner, British ColumbiaHydro and Power Authority (BC Hydro) and are representative of materials commonly used inconstruction yet rarely tested in laboratory seepage conditions. Objectives at the upper-boundare three-fold:2In this thesis, ‘lower bound’ hereafter refers to the lower bound to the onset of seepage-induced internal instability inthe hydromechanical space. In the effective stress:gradient domain below the ‘lower bound’, seepage force is not requisitefor internal particle movement within a soil. Granular materials plotting below the ‘lower bound’ will segregate due togravity.3– Observe the manifestation of a critical seepage-induced phenomenon or phenomena -anticipated as either fluidization or uplift - in two internally stable soils;– Examine the adequacy of state-of-practice material susceptibility techniques, and considerexperimental insights in the context of an emerging state-of-art analysis framework; and,– Assess the hydromechanical plotting position of these poorly-understood materials in re-lation to the hydromechanical upper bound proposed by Li (2008).• Experimentally verify the proposed hydromechanical ‘lower bound’ through testing of gap-graded materials comprising glass beads whose properties lend themselves to theoretical anal-yses. Again, three specific objectives follow:– Observe the phenomenon of suffusion in very unstable soils possessing a clast-supportedmicrostructure;– Assess material susceptibility of gap-graded soils according to the current state-of-practice,considering results and implications for the current state-of-art; and,– Examine the correlation of an experimental hydromechanical threshold in terms of theproposed ‘α’-based lower bound plotting position, after Li (2008).• Advance the understanding of instability phenomena by re-visiting material susceptibility andparticle transportation, with reference to ‘ideal’ glass bead materials, giving consideration to:– The role of effective stress on a particle in terms of particle detachment and availabilityfor transportation;– The necessary hydraulic force required to impart motion to an available particle; and,– Geometric constraints of a porous medium that may prevent passage of a particle that isotherwise able to be transported by seepage flow.1.3 Thesis outlineThesis objectives are addressed in seven chapters, detailed as follows:• Chapter One introduces the topic of internal instability and highlights the challenges facingengineers in professional practice. The study is tailored to address unresolved hydromechanicalinteractions at the upper and lower bounds of a proposed hydromechanical instability spectrum.• Chapter Two summarizes key literature in the field and introduces relevant concepts exploredin other geotechnical studies.• Chapter Three introduces the experimental component of the study. Apparatus, test proceduresand material properties are detailed for multi-stage permeameter tests undertaken on five uniquegradations.4• Chapter Four presents results of seepage tests exhibiting stable behaviour at the upper boundof hydromechanical instability.• Chapter Five presents results of seepage tests demonstrating very unstable behaviour at thelower bound of hydromechanical instability.• Chapter Six provides a theoretical exploration of controlling material, hydraulic and stressfactors at extreme bounds of the proposed hydromechanical space.• Chapter Seven summarizes important findings of the experimental and theoretical thesis com-ponents. Recommendations for future work are presented.5Chapter 2Literature and related work2.1 IntroductionSystematic investigations on the phenomenon of seepage-induced internal instability in soils began inthe 1950’s (i.e. USACE, 1953, Istomina 1957, see Kovaˆcs, 1981). Gradation shape parameters werethe main focus of early research efforts (USACE, 1953; Istomina, 1957 (in Kovaˆcs, 1981); Kenneyand Lau, 1985; Adel et al., 1988; and Burenkova, 1993), while the influence of effective stress andhydraulic gradient were investigated later (i.e. Skempton and Brogan, 1994; Moffat, 2005; Li, 2008).The impact of soil microstructure and fabric in a soil was acknowledged in a conceptual sensein many early studies. However, only recently have computational advances allowed successful im-plementation of theoretical models. The interaction of three broad factors are commonly understoodto influence the onset and occurrence of internal erosion phenomena: (1) material susceptibility, (2)hydraulic conditions, and (3) stress conditions (see Figure 2.1, after Garner and Fannin, 2010).This chapter summarizes the state of knowledge preceding the present study on internal erosion.Section 2.2 has relevant definitions of internal erosion phenomena and associated terminology, neces-sary to the understanding of related work presented herein. The scope and limitations of past researchare then summarized with reference to four thematic contributions:1. Laboratory experiments (Section 2.3);2. Geometric criteria, geometric analyses and material susceptibility (Section 2.4);3. Hydraulic factors (Section 2.5); and,4. The influence of stress (Section 2.6).Thereafter, the findings of a field study at the now-decommissioned Courser Dam in BritishColumbia are used to identify knowledge gaps in considerations of internal instability in engineer-ing practice (Section 2.7). Finally, the role and intended novel contributions of the present study aredescribed in Section 2.8.62.2 Definitions2.2.1 PreambleHistorically, internal erosion phenomena have been described in terms of measured parameters andvisual observations. However, many terms have been used in the literature without complete defini-tion: rarely have phenomena, mechanisms and processes been defined in terms of standard parametersthat can be universally-identified, either quantitatively or qualitatively. Section 2.2.2 clarifies use ofterminology in a subset of internal erosion studies, specifically concerning internal instability. Arationale for selection of terminology adopted in the present study is summarized in Section 2.2.3.2.2.2 Internal instabilityIn general, ‘internal instability’ describes the preferential loss of particles from within a single homo-geneous soil unit. USACE (1953) first investigated the general phenomenon of internal stability inmixed sand and gravel filter materials, and used the term ‘inherent stability’ to describe the resistanceof the material to segregation and piping within itself. Kenney and Lau (1985) provide a generallyaccepted definition of internal stability as ‘the ability of a granular material to prevent loss of its ownsmall particles due to disturbing agents’.Internal instability can result from a number of distinct phenomena. Wittmann (1977 and 1978)describes a number of distinct internal instability phenomena for a sand and gravel mixture basedon microstructure (Figure 2.2). In soils with a gravel skeleton (< 25− 30% sand) transportationphenomena consist of: (1) suffosion, ‘transportation out of the skeleton’ or (2) colmatation or sluicing,‘transportation into the skeleton’. In cases (1) and (2), the gravel skeleton of the soil remains intact.For a soil microstructure where no gravel skeleton exists, ‘the (seepage-induced) deformation of thesemixtures can only act on the whole mixture’. For a sand-dominant microstructure with a ‘higheramount of coarser particles’ (up to 70% sand, Figure 2.2), the relevant phenomenon is describedas (3) ‘erosion’ or ‘piping’. Erosion ‘leads to collapse of the mixture, with finer particles drawnaway by seeping water’. Finally, for pure sand, or sand with a small gravel fraction, (4) ‘piping byheave’ (as defined by Terzaghi and Peck, 1948), will occur under sufficient hydraulic load. Wittmann(1977) suggests that ‘erosion’ (phenomenon 3) may follow extensive ‘suffosion’ (phenomenon 2) ifhydraulic conditions permit.Kezdi (1979) terms suffusion ‘a phenomenon where water, while seeping through the pores, car-ries along the fine particles without destroying the soil structure’. Based on the location of particlemovement, (a) internal, (b) external and (c) contact suffusion phenomena are defined (subfigures 159,160 and 161 of the author, respectively, Figure 2.3).Kovaˆcs (1981) considers ‘hydrodynamic parameters’ as well as the stability of the layer (i.e. mi-crostructure) to classify particle movement phenomena. The phenomenon in which a coarse-grainedskeleton remains unaffected by finer particle loss is termed suffusion, ‘the motion of fine grains’. LikeKezdi (1979), Kovaˆcs (1981) describes internal suffusion as the redistribution of particles within asoil unit, resulting in local changes to the hydraulic conductivity profile, but no loss of finer particles.7Similarly, external suffusion is described as the ‘scouring of fines’, or transportation of particles outof the layer. External suffusion results in a reduction of layer mass and a net increase in conductiv-ity. Kovaˆcs (1981) terms phenomena involving ‘destruction of the soil skeleton’ as either subsidence(a decrease in total volume resulting from movement of soil skeleton particles) or piping or boiling(‘considerable’ movement of grains, localized along a flowline and ‘boiling’ at the channel exit sur-face). An additional phenomenon described by Kovaˆcs (1981) is that of liquidization, where an entirehorizontal surface loses bearing capacity. Soil grains ‘float’ due the fact that hydrodynamic upliftforces equal the weight of the grains (i.e. hydrodynamic force equals Terzaghi’s theoretical criticalhydraulic gradient (see Terzaghi and Peck, 1948)).Burenkova (1993) defines suffosion in terms equivalent to Kovaˆcs’ 1981 ‘suffusion’; with internaland external methods characterized by local and global conductivity changes, respectively. Someconfusion arises, however, where Burenkova (1993) describes ‘outer suffosion’ to potentially causeinstability of the whole soil structure, leading to ‘sinking of the ground surface or landslides’. Thisstatement appears to be at odds with a later remark: ‘it is... important... to evaluate the dimensions oferoded particles (soil skeleton should not be disturbed).’.Skempton and Brogan (1994) use the term ‘segregation piping’ to describe the piping of fineparticles where gravel particles remain practically undisturbed. Chapuis (1996) defines the ‘migrationof fine particles of a soil within its own pore space’ as ‘suffossion’.More recently, terminology has begun to converge somewhat. ‘Suffusion’ is the term used toexplicitly describe movement of fine particles within an unaffected coarse-grained skeleton in manyrecent studies (e.g. Bendahmane et al., 2008; Benamar, 2010; Ahlinhan and Achmus, 2010; Maknoonand Mahdi, 2010; Sadaghiani and Witt, 2012; and Ke and Takahashi, 2012). Like Kezdi (1979) andKovaˆcs (1981), many studies (e.g. Sun, 1989, and Semar et al., 2010) distinguish between internal,external and/or contact suffusion. In all cases the coarse-grained skeleton remains intact, assuring nosettlement or volumetric deformation.However, it is important to note that a number of recent papers (e.g. Sail et al., 2011, and Chang,2012) use the term ‘suffusion’ to describe the instability phenomenon in gap-graded specimens con-taining 35−40% finer fraction, with corresponding ‘settlement’ of the specimen. Wan and Fell (2008)also associate the suffusion phenomenon with settlement.Recognizing the need for clarity in the literature, Garner and Sobkowicz (2002) and Moffat et al.(2011) distinguish between suffusion - where the coarse-grained skeleton remains intact, with zerototal volume change - and suffosion, where loss of particles results in collapse of the soil structureand net reduction in total volume. This distinction is adopted in the present study. The phenomenadescribed by Wittmann as ‘suffosion’ and ‘colmatation’ (Figure 2.2) should thus be termed ‘suffusion’(external and internal, respectively). Wittmann’s (1978) ‘erosion’ or ‘piping’ and Kovaˆcs’ (1981)‘subsidence’ correspond exactly to suffosion, in the proposed terminology. Given the confusion inBurenkova’s (1993) definitions of ‘suffosion’ it is possible that the work of Burenkova involves bothsuffusion and suffosion. Moffat and Fannin (2011) present experimental evidence of widespreadsuffosion preceded by minor suffusion, a process earlier postulated by Wittmann (1977).8Numerous other terms may be used to describe the progression of soil distress phenomena (in-cluding suffusion and suffosion) under specific conditions. Piping should be considered a macro-scaleterm, used to describe the local and confined occurrence of any form of internal erosion. The termsegregation piping in the work of Skempton and Brogan (1994), is interpreted as suffusion withina pipe-like local zone, given that the gravel skeleton of the soil remains undisturbed. Accordingly,Ke and Takahashi (2012) consider suffusion and piping ‘coupled phenomena’. Similarly, backwarderosion describes the macro-scale occurrence of any particle transport phenomenon - specifically, aphenomenon initiating at an exit boundary and retrogressing in the upstream direction.Moreover, some studies describe ‘piping’ and ‘backward erosion’ as modes of seepage failurein internally stable soils (i.e. van Beek et al., 2011; Wilhelm, 2000). Wilhelm (2000) describesseepage failure as a ‘loss of stability due to the seepage force counterbalancing the stabilizing effect ofgravity’. The experiments of Wilhelm (2000) involve uniformly-graded (internally stable) sands, thusseepage phenomena occur at, or close to, the theoretical critical hydraulic gradient (after Terzaghi,1939). For comparative purposes in the present study, hydraulic distress phenomena are defined forinternally stable soils (i.e. soils not susceptible to suffusion or suffosion) as follows:• Fluidization occurs when the weight of the total soil mass is overcome by opposing seepageforce per unit volume (defined by the theoretical critical hydraulic gradient, icr =γ ′γw ). A stateof zero effective stress occurs, resulting in total loss of soil strength. The term ‘boiling’ refersto a specific particulate motion that may occur during fluidization. ‘Heave’ is considered aspecific case of fluidization, where a state of zero effective stress is reached and the soil surfaceexpands in the direction of flow.• Uplift or hydraulic uplift occurs when an intact soil mass is translated in the direction of flow.No change occurs in soil properties, and no inter-particle movement results. In practice, upliftis commonly preceded by hydraulic fracture, where displacement of the unmodified soil massoccurs from the fracture surface.2.2.3 Synthesis of terminology: phenomenaIn the present study, seepage-induced phenomena are defined in terms of quantifiable soil parame-ters. There is a clear need to define phenomena in terms of their effect on soil properties. In mostcases, internal erosion experiments involve qualitative or quantitative observations of three importantparameters (detailed further in Section 2.3.2 and Table 2.4):1. Specimen length (in 1-dimensional space), or volume (in 3-dimensional space);2. Hydraulic conductivity (global and/or local); and,3. Mass loss.9When subject to seepage flow, the total length (or volume) of a soil element may increase, de-crease, or remain unchanged. A change in length signifies a change in the fabric of the soil unit, whileno change indicates that microstructural skeleton remains intact.Similarly, hydraulic conductivity within a test specimen or unit volume may increase, decrease,or remain constant. Measurements of hydraulic conductivity may be local or global (average acrossthe specimen), changes of which could be associated with: (1) loss of specimen mass (i.e. fineparticle transport), or (2) increase in specimen dimensions (fluidization or heave), (3) accumulationof particles in pore space (i.e. clogging), or (4) collapse of the soil structure (accompanied by decreasein specimen length). No change in hydraulic conductivity suggests that the average cross-sectionalflow area remains constant in the zone of measurement.Assuming particles are not added to the porous system at the inlet boundary, total specimen mass(mass of solid grains in the specimen or unit volume) can either decrease or remain constant. Locally,an increase in specimen mass is possible due to an accumulation of transported particles. Other thanforensic gradation analysis, local mass increase is rarely measured. However, particle accumulationis inferred from local pore pressure response where the accumulation of particles may result in a localregion of low hydraulic conductivity. An observed decrease in specimen mass occurs when particlesare lost from the specimen, however the ‘constant mass’ condition may arise in a number of scenarios(1) particles are not ‘free’ or ‘detached’ and are therefore not available for transport (internally stablematerial), or (2) the hydraulic load is insufficiently large to move the particles (internally unstablematerial), or (3) particles are able to be transported a short distance before becoming ‘trapped’ withinthe soil fabric.Considering all possible combinations of changes in: (1) unit length (or volume in a three-dimensional sense), (2) hydraulic conductivity, and (3) unit mass, Table 2.1 summarizes phenomenaof suffusion (external and internal), suffosion, fluidization, uplift and consolidation. Any of the de-fined phenomena could cause internal erosion in a soil structure (see Section 2.2.1); however, internalinstability of a soil mass is characterized by the phenomena of suffusion (internal and external) andsuffosion. Suffusion is characterized by no change in soil volume or length: external suffusion resultsin an increase in hydraulic conductivity and a decrease in mass, while internal suffusion results in alocal decrease in conductivity (in the zone of particle accumulation) with no (or minor) mass loss.Suffosion is characterized by mass loss accompanied by volumetric deformation of the soil element.2.3 ExperimentsAn early appreciation of internal erosion first emerged from observations on base soil-filter compat-ibility. Terzaghi first began a formal synthesis of filtration ‘rules’ in the 1920’s (see Fannin, 2008).Subsequently, Bertram (1940), USACE (1953), and Karpoff (1955) conducted systematic filtrationstudies to verify proposed empirical stability rules, based largely on grain size ratios. While generallyserving as a good ‘screening tool’ for base-filter compatibility, the applicability of geometric filtrationcriteria continues to be challenged (e.g. Tomlinson and Vaid, 2000).First to note the importance of internal stability within the filter component of the base soil-10filter system, USACE (1953) conducted downward-flow seepage tests on homogeneous specimensof mixed sand and gravel. Numerous systematic experimental investigations regarding internallyunstable soils have since explored the early premises of ‘self-filtration’ in coarse-grained cohesionlesssoils.The importance of field observations and construction experience must be recognized in theemerging science-based understanding of internal instability. Many early laboratory studies in the1970’s and 80’s were prompted by observations related to problematic soils in the field, particularlygap-graded and widely-graded materials (e.g. De Mello, 1975; Wittmann, 1977; Sherard, 1979; andKenney and Lau, 1985). Some later research has focused on the systematic investigation of materialsusceptibility factors (e.g. Li, 2008), yet many recent laboratory studies are still intended to addressspecific issues, problems or performance questions in the dam, dyke, and levee industry (e.g. Cividiniet al., 2009; Wan and Fell, 2008; Moffat and Fannin, 2011; Moffat et al., 2011).Across seven decades of experimental research, the nature and motivation of individual studiesvaries greatly. In the absence of standard internal instability test methods, direct comparisons betweenstudies can be problematic: experimental focii, boundary conditions and measured parameters varygreatly.2.3.1 Experimental apparatus and seepage conditionsA summary is given of apparatus characteristics, seepage conditions, and material type for 25 ex-perimental studies concerning internal instability of soils subject to unidirectional seepage flow (seeTable 2.2 and Table 2.3).2.3.1.1 ApparatusRigid-walled permeameter devices were most commonly used for internal stability studies prior toabout 2005. Rigid-walled devices simplify the specimen reconstitution process and permit numerousoptions for instrumentation, radial volumetric control, simple operation and maintenance. The inter-nal diameter of rigid-walled apparatus varies from approximately 70 mm (Cividini et al., 2009; Sterpi,2003), to approximately 275 mm or more (e.g. Kenney and Lau, 1985; Honjo et al., 1996; Moffat,2005; Li, 2008; Wan and Fell, 2008; Ahlinhan and Achmus, 2010; and Sail et al., 2011). Earlytest devices such as those of USACE (1953) and Kenney and Lau (1985) imposed a flow regimewith limited control of seepage (Figure 2.4 and Figure 2.5). Hydraulic conditions were monitoredby simple standpipe piezometers (Figure 2.4), and/or through measurement of volumetric discharge(Figure 2.5). Axial loading can be controlled in some rigid-walled devices; however, stress inho-mogeneities may occur under high axial loads due to the effects of side-wall friction. While mostexperiments on coarse-grained soils have been conducted under zero or nominal axial load, Moffat(2005) and Li (2008) present repeated tests at different axial stress conditions. With advances ininstrumentation technology, researchers have implemented additional measurements of specimen re-sponse due to seepage, as discussed below. Recent devices of Moffat (2005) and Sail et al. (2011)include more comprehensive instrumentation arrangements to measure parameters such as axial load11at both top and bottom of the specimen, and specimen density (as shown in Figure 2.6 and Figure 2.7,respectively).Over the past decade, more complex stress conditions have been invoked in internal instabilitytesting through use of flexible-walled apparatus. While some flexible-walled devices have been man-ufactured commercially (e.g. Sun, 1989, Figure 2.8), most apparatus used in permeameter testingare custom-designed, often created from a modified triaxial apparatus. Flexible-walled apparatus aresometimes termed ‘stress-controlled’ devices, emphasizing the ability to control three-dimensionalstresses on the specimen. Most tests on fines-rich soils are conducted in the smaller triaxial stresscontrolled apparatus, due to scale considerations discussed below. Only Chang and Zhang (2011)and Luo et al. (2013) have addressed internal instability in coarse-grained soil under triaxial stresscontrol. Marot et al. (2011, 2009) used slightly different configurations of the triaxial device of Ben-dahmane et al. (2008), equipped with effluent monitoring instrumentation (Figure 2.9). However, adownside of the flexible-walled device is the difficulty in quantifying volumetric change. Specimensize is usually limited to that of a standard triaxial specimen with diameter ≈ 50 to 65 mm. Recentexceptions are that of Chang and Zhang (2011) and Luo et al. (2013), who developed flexible-walleddevices accommodating a specimen 100 mm in diameter. The device of Luo et al. (2013) additionallyallows for volumetric quantification via hoop strain instrumentation (Figure 2.10).To a lesser degree, horizontal flow conditions have been studied using flume tests. Flume teststypically involve a larger-scale rectangular specimen, without stress control (e.g. Adel et al., 1988;Ahlinhan and Achmus, 2010, Figure 2.11). The impact of specimen geometry and scale has beenstudied under horizontal flow conditions (i.e. Richards and Reddy, 2012 and van Beek et al., 2011);however, this research details the initiation of backward erosion in internally stable materials at anunfiltered exit, with non-unidirectional seepage. The erosion phenomena described by Richards andReddy (2012) and van Beek et al. (2011) therefore lie beyond the scope of the present study, whichstrictly considers suffusion and suffosion under unidirectional seepage.2.3.1.2 Specimen characteristicsAcross studies in all devices, specimen length varies significantly in relation to apparatus diameter.Length-to-diameter ratios range from LD < 0.7 (Sun, 1989; A˚ berg, 1993; and Honjo et al., 1996) toLD ≥ 2 (i.e. Wittmann, 1977; Sterpi, 2003; Moffat, 2005; and Cividini et al., 2009). In the past decade,studies have most commonly employed specimen dimensions of LD ≈ 1 (Bendahmane et al., 2008; Li,2008; Wan and Fell, 2008; Marot et al., 2009; Ahlinhan and Achmus, 2010; Marot et al., 2011; andLuo et al., 2013, Table 2.2).Specimen dimensions dictate the maximum particle size for materials tested. Most permeameterstudies adhere to a ratio of specimen diameter to maximum particle size of greater than 8 to 12, inaccordance with ASTM D-2434 (2006a) and/or ASTM D-5101 (2012). Hence, a larger apparatuspermits testing of widely-graded sands and gravels (e.g. Kenney and Lau, 1985; Honjo et al., 1996;Moffat, 2005; and Sadaghiani and Witt, 2012); whereas a smaller device is typically used in the test-ing of sands and finer materials. Materials possessing high fines contents often require significant12consolidation periods and can withstand very high gradients (iav > 100) prior to the onset of particlemovement (Bendahmane et al., 2008; Marot et al., 2009). Higher gradients can be achieved in speci-mens of smaller dimensions where a differential water head is applied over a shorter specimen length.For these reasons, materials possessing a high fines content tend to be tested in a smaller apparatus,with maximum particle size, D100 < 3 mm (Sun, 1989; Bendahmane et al., 2008; Marot et al., 2009;and Marot et al., 2011).Independent of maximum particle size constraints, the types of soil employed in internal insta-bility investigations vary in terms of: (1) grain size range, (2) grain properties, (3) gradation ‘shape’or slope, and (4) plasticity of any fines content. Coarse-grained soils were the focus of most earlyexperimental studies (with the exception of Sun, 1989). However, some recent studies have employedglass bead gradations as an ‘ideal’ representation of internally unstable porous media (e.g. Moffat andFannin, 2006; Li and Fannin, 2008; and Sail et al., 2011). A tendency toward instability in gap-gradedand widely-graded gradations has led to a number of studies on gradations of these shapes (Table 2.2and Table 2.3). A number of experiments on soils with significant fines content have emerged in thepast decade. Wan and Fell (2004) conducted tests on 20 widely-graded soils of predominantly coarseparticles with varying, but limited, fines content and of variable plasticity (< 10% clay-size fraction;plasticity, PI < 12%). The tests of Wan and Fell (2004) are otherwise of similar experimental designto many early studies in the field of internal instability (Table 2.2). In contrast, the small-scale triaxialprograms of Sun (1989), Bendahmane et al. (2008), and Marot et al. (2011, 2009) test fines-dominantgradations containing significant clay-size fractions and notable plasticity (≤ 30% clay-sized fractionand PI ≤ 33%), with a maximum particle size of D100 ≈ 3 mm.2.3.1.3 Seepage conditionsSeepage conditions vary among reported experimental studies, typically in terms of: (1) flow direc-tion; (2) head vs. flow control; (3) the addition of other influences (e.g. vibration); and (4) seepagemagnitude and duration.Vertical seepage is most commonly employed in permeameter tests (Table 2.2 and Table 2.3). Inthe test configuration, any mesh bounding the downstream extent of the specimen (the outlet mesh)needs to be coarse enough to allow passage of transported particles. The inlet boundary of the spec-imen can typically accommodate a tighter mesh, as the boundary lies upstream of the specimen andtherefore does not restrict particle transport. Downward flow allows easy collection of eroded parti-cles below the specimen; however, specimen integrity may suffer as some fine particles tend to be lostat the bottom (outlet) boundary during specimen reconstitution (Moffat, 2005, and Li, 2008). In theupward flow configuration, the inlet boundary mesh is at the base of the specimen and material recon-stitution can therefore proceed with very little particle loss. While the upward seepage configurationfacilitates improved specimen homogeneity at the base of the specimen, quantification of materialtransport during testing becomes problematic. Many apparatus permit either upward or downwardflow configuration and testing programs may involve flow in either direction. Sun (1989) reversedflow direction during individual tests, while Burenkova (1993), Wan and Fell (2004), Moffat (2005),13and Li (2008) conducted separate tests in upward and downward unidirectional seepage conditions.Ahlinhan and Achmus (2010) conducted tests in two separate apparatus under upward and horizontalseepage conditions. Adel et al. (1988) conducted tests with horizontal seepage flow, while the per-meameter of Wittmann (1977) was fixed with a central pivot to permit variable angular orientation ofthe specimen under longitudinal flow conditions (Figure 2.12).Seepage regimes may vary based on the focus of the individual study. Generally, seepage isimposed under conditions of flow control or head control. Some tests involve constant imposedseepage conditions for the duration of the test; many such tests impose vibration and/or very largegradients to test for an ultimate susceptibility to instability (Kenney and Lau, 1985; A˚ berg, 1993;Honjo et al., 1996; and the downward flow test series of Wan and Fell, 2004). In identifying a criticalhydraulic condition, seepage force is typically increased incrementally in multiple stages. Gradient-controlled multi-stage seepage flow was employed in 15 of the 25 reported studies in Table 2.2 andTable 2.3.The magnitude of seepage force applied in experimental testing is typically governed by materialcharacteristics. Stable soils with high fines contents may withstand very high hydraulic gradients, insome cases in excess of iav = 100 under adequate confining stress conditions (e.g. Bendahmane et al.,2008; and Marot et al., 2009). In contrast, very internally unstable sands and gravels may demonstrateinstability at gradients much less than one, as described in Table 2.2 and Table 2.3. The duration ofimposed seepage typically ranges from minutes to hours and, in some cases, weeks (e.g. Cividiniet al., 2009). The magnitude of gradients imposed in the testing of some materials may be greaterthan that typically expected in the field, given that macro-scale gradients in large dams are oftenmodelled at magnitudes of less than one. Conversely, experimental seepage is inevitably imposed fordurations much shorter than those experienced by dams and canals with an intended design life of50 years. Most comprehensive experimental studies in the literature are undertaken not to simulatean assumed field condition, rather, with the intent to further an understanding of material behaviourthrough a controlled parametric study of phenomena.In the present study, seepage is imposed in the upward direction, under head control, with novibration. Gradients of less than 50 are applied in testing of internally stable soils, while internallyunstable soils are tested at gradients typically less than one. The present experimental program isfully described in Chapter 3.2.3.2 Measured and derived parametersResults of internal instability experiments are reported in terms of measured and calculated param-eters. Reported parameters differ among experimental studies, based on apparatus constraints andexperimental objectives. Table 2.4 presents measured parameters for 25 experimental studies. Asin Table 2.2 and Table 2.3, selected studies concern internal instability under unidirectional seepageconditions.The focus of an individual study will determine the parameters of concern and resolution ofmeasurement. Strict monitoring of hydraulic conditions is necessary in studies focused primarily14on the onset of particle movement (e.g. Skempton and Brogan, 1994, and Moffat, 2005), while theprecise measurement of eroded particles is necessary in studies focusing specifically on the rate andprogression of mass loss or the fraction of eroded (vs erodible) materials (Bendahmane et al., 2008;Marot et al., 2009; Cividini et al., 2009; Luo et al., 2013; Ke and Takahashi, 2012).The initial state of the specimen (prior to seepage) is generally detailed in terms of: (1) materialtype and gradation properties, (2) initial void ratio or relative density, and (3) initial specimen dimen-sions (i.e. length or volume). Seepage-induced changes in specimen properties (typically hydraulicconductivity, specimen length and mass) characterize the specimen response.Calculation of local or global hydraulic gradients (ilocal and iav, respectively) is common, givena known axial dimension and pore-water pressure measured in either local or global terms duringseepage. Volumetric flow rate throughout the specimen is usually measured, either by collectionof outflow in a given time period, monitoring of influent and/or effluent tanks, or via an automatedflowmeter. Where volume changes occur in the specimen, volumetric discharge should be adjustedaccordingly to give the rate of flow through the specimen. Knowing the cross-section area of the spec-imen, average discharge velocity, vd , is calculated for unidirectional flow. Given hydraulic gradient,i, and volumetric flow rate through the specimen, Q, hydraulic conductivity, k, is typically derived forlaminar flow conditions assuming Darcy’s Law governs:Q = k · i ·A (2.1)Visual observations of mass loss typically concern either: (1) particle movement within the spec-imen in a transparent-walled device, or (2) the colour and/or turbidity of effluent exiting the testspecimen. Quantitative deformation measurements are common in at least one dimension; howeversome studies rely solely on a qualitative (visual) assessment of changes in specimen volume.Eroded particles are typically collected during discrete time periods in downward flow configura-tions, permitting calculation of mass loss and measurement of the size of eroded particles. Pre- andpost- test gradation analyses may permit the derivation of maximum eroded particle size via the curvematching technique (Kenney and Lau, 1985; Wan and Fell, 2004; and Ke and Takahashi, 2012).Monitoring of stress conditions is possible only where apparatus design permits. Moffat (2005)and Li (2008) monitored axial stress at the top and bottom of the specimen in a large rigid-walledpermeameter (Figure 2.6). Like the smaller-scale tests of Li (2008), Sail et al. (2011) also conductedseepage tests in a rigid-walled permeameter where axial stress was monitored at the top surface only.In all triaxial devices, confining pressure is measured for the duration of the test.Recent experimental studies have introduced novel measurements. Bendahmane et al. (2008) andMarot et al. (2009) employed an optical sensor to measure entrained fine particle concentration intest effluent (Figure 2.9), with improvements of the instrumentation reported by Marot et al. (2011).Sail et al. (2011) quantified average density of the specimen in a non-destructive manner using agamma-ray scintillometer (Figure 2.7), while Ke and Takahashi (2012) quantified changes in speci-men strength due to internal instability via miniature cone penetration test profiling. Luo et al. (2013)added hoop strain gauges to a triaxial apparatus to quantify volumetric change in the specimen.152.3.3 Observations on the emerging body of experimental workOver the past six decades, a number of experimental studies have investigated internal instability dueto unidirectional seepage. However, the nature of these studies differ immensely. The validationof unique experimental procedures has been achieved in some cases by re-testing common materialgradations. For example, the G4-C gradation of Honjo et al. (1996) has been subsequently tested byMoffat and Fannin (2006) and Sail et al. (2011), while gradation ‘A’ of Skempton and Brogan (1994)has been tested independently by Li (2008) and Luo et al. (2013). An emerging body of experimentalwork concerning fine-grained soils (Sun, 1989; Bendahmane et al., 2008; Marot et al., 2009 andMarot et al., 2011) provides much-needed empirical guidance in the use of fine-grained materials inpractice, given the historical focus on coarse-grained soils in the literature.Furthermore, the effects of experimental scale remain unresolved. Li (2008) noted significantdifferences in critical hydraulic condition for the same material when tested in different-sized rigid-walled seepage apparatus. Recent studies by Marot et al. (2012) and van Beek et al. (2011) attemptto resolve scale issues through comparative multi-scale seepage tests and centrifuge experiments,respectively. Conclusions are not consistent.Whilst many individual studies have investigated the different factors associated with internalinstability, cross-study comparisons can be challenging when results are not reported using the samemeasured and derived parameters.2.4 Material susceptibility: gradation shape and microstructuralgeometrics2.4.1 IntroductionSoils whose material characteristics do not preclude preferential movement of particles are termed‘susceptible’ to internal instability. Along with effective stress and hydraulic conditions (addressed inSection 2.5 and Section 2.6, respectively), material susceptibility is identified by Garner and Fannin(2010) as one of the three necessary factors controlling the expression of internal instability in practice(Figure 2.1).Historically, the susceptibility of granular soils to internal instability has been assessed by the‘shape’ of the grain-size distribution, often termed geometric criteria. Geometric criteria have beenempirically developed (e.g. Kezdi, 1979; Kenney and Lau, 1985; Burenkova, 1993; Wan and Fell,2008; and Li and Fannin, 2012) and applied with reasonable success (e.g. Ro¨nnqvist, 2010). However,the applicability of methods is limited to predominantly coarse-grained soil types and engineeringjudgement is requisite in the interpretation of results. Strictly, geometric criteria do not address otherpotential material susceptibility factors that may be implicated based on limited empirical experience.Additionally, factors influencing material susceptibility may include plasticity, the in-situ state ofthe soil mass (i.e. compaction density, defects created by local conditions, excess void space), soilmicrostructure, or inter-granular cementation (e.g. Wittmann, 1979; Garner and Fannin, 2010, and16ICOLD, 2013a).Recently, capillary and constriction-based criteria have been developed by Li and Fannin (2013)and Indraratna (2011), to assess the likelihood for particle transport in suffusion. Like the fundamentalstudies of Wittmann (1979) and Kezdi (1979), these criteria refer to microstructural limits (relatingto the relative proportions of finer and coarser soil components) in identifying potentially mobileparticles.Geometric criteria are described in Section 2.4.2, while microstructure geometrics are discussedin Section 2.4.3 with regard to internal erosion phenomena. Pore geometry analyses are detailed inSection 2.4.4.2.4.2 Particle size geometricsEarly studies estimated the potential for instability as a function of grain-size ratio. Kovaˆcs (1981)cites Istomina’s (1957) guidance on an upper limit of coefficient of uniformity, Cu, recommended forinternally stable filter design:No suffusion if:Cu ≤ 10 (2.2)Transition condition if:10≤Cu ≤ 20 (2.3)Suffusion liable if:Cu ≥ 20 (2.4)In addition, Kovaˆcs (1981) refers to three publications preceding the reported work of Istomina,where Cu ≤ 6 to 20 is recommended for internally stable behaviour in filters.Kezdi (1979) published an application of the Terzaghi (1939) base soil-filter retention criterionas a test for ‘self-filtration’ in a single soil unit (Figure 2.14). Geometrically, the soil gradation issplit into a coarser fraction and finer fraction at an arbitrary point on the gradation curve (d0, S0),as illustrated in Figure 2.14. The representative filter size, D′15, corresponds to the diameter of thecoarser curve ( 1©) at 15% mass passing (S = 15%). The representative size of the finer component,d′85, is that of the finer curve ( 2©) at 85% mass passing (S = 85%). A material is deemed susceptibleif:D′15d′85> 4 (2.5)Kezdi (1979) advocated that the gradation be assessed across its entire particle size range bycalculating the filter ratio (Equation 2.5) at numerous arbitrary split points (d0,n+1, S0,n+1) for theentire gradation (d0 < d0,n+1 < d100). Sherard (1979) discusses the same split-gradation analysis of17coarse broadly-graded soils, following an application by De Mello (1975). Sherard (1979) observesthat typical core materials give D′15d′85ratios of 2 to 4, while the noted ‘problematic soils’ associated withsinkholes in dams have D′15d′85ratios varying from 5 to 20, or greater. Notably, Sherard (1979) advocatesthat the gradation be divided in two parts at any size greater than approximately 0.1 or 0.2 mm.Kovaˆcs (1981) reports the early shape-analysis of Lubochkov (1962 and 1965). The premise ofthe Lubochkov method is that ‘the (soil) layer is not susceptible to suffusion when the slope of thedistribution curve is equal to, or smaller than a given limit in each diameter-interval’. The methodis depicted in Figure 2.13. Given a point (Dn,SDn) in (D, S) space, for a standard gradation curvewith particle diameter, D, and mass percent passing represented as S%, a forward diameter-interval(Dn to Dn+1) and backward diameter interval (Dn to Dn−1) are considered. Corresponding ‘massincrements’, ∆S1 and ∆S2 are defined as the mass fraction of particles in the size range Dn≤D≤Dn+1and Dn ≤ D ≤ Dn−1, respectively. Forward and backward diameter intervals satisfy the followingequality:Dn−1Dn=DnDn+1= K (2.6)where K = 2.5, 5 or 10, based on the ‘tolerance factor of safety’ required. The upper limits tostability are defined by the ratio of forward to backward mass increments, given for the three levelsof confidence in Table 2.5.Kenney and Lau (1985) conducted an extensive suite of seepage tests under large hydraulic loadsand delineated ‘stable’ and ‘unstable’ materials based on the shape of the particle size distributioncurve in the finest 20% or 30% of the gradation curve (for widely graded and narrowly gradedsoils, respectively). Conceptually, the method of Kenney and Lau (1985) is very similar to that ofLubochkov (refer Figure 2.13, Kovaˆcs, 1981), but considering a forward ‘diameter-interval’ only. Forany point on the gradation curve in (D,F) space, a ratio of mass increment, H, to mass percent pass-ing, F , is obtained. Mass increment, H (comparable to the mass increment ∆S1 of Lubochkov), is thepercentage of mass between particle size, D, and four times the particle size (4D):H = F(4D)−F(D) (2.7)as illustrated in Figure 2.16. Based on a geometric assessment of tested soils (Figure 2.17), theminimum value of HF indicates the most susceptible region of the gradation curve. The gradation isdetermined to be susceptible to internal instability if:(HF)min≤ 1.0 (2.8)Originally, Kenney and Lau (1985) recommend a threshold ratio of (HF )min ≤ 1.3, in accordancewith the previous gradation-shape technique of Lubochkov (1969) (Figure 2.17). Subsequent discus-sion in the literature (Milligan, 1986, and Sherard and Dunnigan, 1986) resulted in a revision of thegeometric stability threshold to (HF )min ≤ 1.0 (Kenney and Lau, 1986).18Chapuis (1992) documented geometric similarities in the gradation shape assessments of Lubochkov,Kezdi (1979) (attributed to Sherard, 1979) and Kenney and Lau (1985). Specifically, all methods de-fine a stability bound based on a limiting value for the gradation curve secant slope.Li (2008) explored the observations of Chapuis (1992), illustrating the graphical similarity unitingthe Kezdi (1979) and Kenney and Lau (1985) methods in Figure 2.18. From various laboratorystudies, Li and Fannin (2008) assemble a database of 57 stable and unstable sand-gravel gradations.The experimental database shows the secant slope limit of Kenney and Lau (1986) to best predictstability in the mass fraction range 0% ≤ F ≤ 15%, while the gradation slope limit of Kezdi (1979)proves most accurate in the range 15%≤ F ≤ 30%. The resulting ‘unified’ method of Li and Fannin(2008) suggests a two-part geometric threshold for internal instability (Figure 2.19):H =F if 0≤ F ≤ 15%15 if F ≥ 15%(2.9)Burenkova (1993) delineates geometric stability via three discrete points on the gradation curve.‘Conditional factors of uniformity’, h′ and h′′, are defined for a gradation curve based on the D90, D60and D15 particle sizes, where:h′ =D90D60(2.10)h′′ =D90D15(2.11)Four stability regions are defined in (h′,h′′) space (Figure 2.20) 1 for a database of seepage testsfor approximately 110 widely-graded non-cohesive granular materials. Gradation Zones I and III ofFigure 2.20 are deemed susceptible; while Zone IV comprises ‘artificial’, geometrically implausiblesoils (requiring D60 ≤ D15). Hence, a domain for internally stable soils is defined:0.76 · log(h′′)+1 < h′ < 1.86 · log(h′′)+1 (2.12)Modifying the approach of Burenkova (1993), Wan and Fell (2008) examine the D90, D60, D20and D5 particle sizes for a database of 58 selected silt-sand-gravel soils. Log functions ofD90D60and D20D5are plotted to define three zones of material susceptibility: (1) unstable, (2) transition, and (3) stable(specifically, ‘low likelihood’ of instability) (Figure 2.21). The method of Wan and Fell (2008) is theonly assessment method that explicitly applies to soils with silt content; all other methods are basedstrictly on empirical experience with coarse-grained soils.With the exception of Kezdi (1979) (likely also Lubochkov, as presented by Kovaˆcs, 1981), em-pirical data used in the development of analysis methods originate entirely from laboratory perme-ameter testing (i.e. Kenney and Lau, 1985; Burenkova, 1993; Li and Fannin, 2008; and Wan and Fell,2008). Acknowledging a disconnect from field experience of internal instability, Ro¨nnqvist (2010)1Burenkova (1993) defines susceptibility in terms of ‘suffosion’; the interpretation of which is discussed in Section 2.2.19assembled a database of moraine-core dams categorized by field performance. Figure 2.22 and Fig-ure 2.23 present geometric analyses corresponding to observed field performance in 65 moraine filtermaterials and 59 core gradations, respectively. For the coarse-grained filter material dataset (Fig-ure 2.22), both the geometric method of Kenney and Lau (1986) and the unified method of Li andFannin (2008) prove similarly accurate in identifying dams with historic evidence of internal insta-bility. For the database of core materials, for which coarse-grained material susceptibility techniquesdo not strictly apply (Figure 2.23), Ro¨nnqvist (2010) found that the unified method of Li and Fannin(2008) improved the accuracy and reduced the degree of conservatism when compared to the methodof Kenney and Lau (1986).2.4.3 Soil microstructureSoil microstructure is a critical, yet complex, factor controlling material susceptibility. Early studieson internal instability (e.g. Wittmann, 1979, and Kovaˆcs, 1981), rely on the conceptual geometricsof soil microstructure as a basis for understanding particle detachment and transportation. However,microstructure is not explicitly addressed in most common geometric techniques, due to dependenceon a number of stochastic variables, such as:• Porosity• Particle shape• Layer or unit homogeneityIn the present study, the following terminology is used:Sf = finer fraction content; expressed as a fraction (0≤ S f ≤ 1) or percent (0%≤ S f %≤ 100%) of thetotal mass of solids (and volume, where specific weight of finer and coarse fractions is equal).S∗ = critical finer fraction content at which microstructure of a two-component mixture transitionsfrom coarse-skeleton to matrix-supported. The minimum porosity of the mixture occurs at S∗.n and e = overall porosity and void ratio, respectively, of the mixture.nc and ec = inter-granular porosity and inter-granular void ratio, respectively (Vaid, 1994).nf and ef = inter-finer porosity and inter-finer void ratio, respectively (Thevanayagam, 1998).Given the overall porosity of the gap-graded soil and the finer fraction content by mass, the indi-vidual porosities of the coarser and finer components can be calculated. Specifically, inter-granularand inter-finer void ratios are derived based on fundamental volume and porosity definitions, as shownin Figure 2.24, where: VV = total volume of voids, VP,C = volume of coarse fraction particles, VP,F =volume of finer fraction particles, and VV,C = volume not occupied by coarse particles (=VP,F +VV ):ec =VV,CVP,C=e+S f1−S f(2.13)20e f =VVVP,F=eS f(2.14)Two general concepts are presented in the literature (e.g. Wittmann, 1978; Kezdi, 1979; Kovaˆcs,1981; Kenney and Lau, 1985; and Skempton and Brogan, 1994). Firstly, a primary coarse-grainedskeleton exists when:ec < ec,max (2.15)Where ec,max is the maximum void ratio of the coarse particles alone (as determined by ASTM-D4253 (2006b) and ASTM-D4254 (2006c) or similar). In this case, a stable network of inter-granularcontacts exists within the coarse particles alone, and the coarse skeleton exists at relative density0 < Rd < 1, fully occupying the volume of the porous medium. Specifically, for a clast-supportedmicrostructure to exist, the mass fraction of finer particles (with inter-finer void ratio e f > e f ,min)must be geometrically capable of fitting within the void spaces of the (geometrically permissible)clast-supported skeleton.Secondly, a critical finer fraction content, S∗, can be determined for a coarse primary skeleton,such that the finer fraction fully occupies the void space of the coarse primary skeleton when S f = S∗.S∗ is a function of ec. The critical finer fraction content delineates the boundary between a coarse-grained clast-supported microstructure and a finer fraction matrix-supported structure.Four equivalent expressions are presented in the literature, each describing the two-componentmixture at the critical fines content, S∗:1. Wittmann (1978) defines S∗ in terms of the unit weight of solid particles, γd,S; dry unit weightof the gravel skeleton, γd,G; and dry unit weight of sand component, γs:S∗ =γd,S(γs− γd,G)γs · γd,G + γs · γd,S− γdG · γd,S(2.16)2. Kezdi (1979) defines βmax ≡ S∗1−S∗ in terms of the porosity of the coarse skeleton, n1 ≡ nc, andthe inter-finer porosity, n2 ≡ n f :βmax = n11−n21−n1(2.17)3. Kenney and Lau (1985) define the critical coarse fraction content, fp ≡ 1−S∗, in terms of theintergranular void ratio of the primary coarse skeleton, ep ≡ ec, and the inter-finer porosity,nl ≡ n f :fp =11+ ep(1−nl)(2.18)214. Skempton and Brogan (1994) define the critical finer fraction fraction content, S∗, in terms ofthe intergranular void ratio of the primary coarse skeleton, ec, and the inter-finer porosity, n f :S∗ =A1+A(2.19)and:A = nc1−n f1−nc(2.20)Three microstructural types are defined in the present study, based on the comparison of S f to S∗:• Sf << S∗: Clast-supported microstructure (Type C): A ‘skeletal’ load-bearing microstructurecomprising the coarser fraction of a gap-graded material. Specifically, the clast-supported mi-crostructure possesses a stable network of inter-granular contacts within the coarse particlesalone. The coarse skeleton exists at relative density 0 < Rd,C < 1, and is capable of fully occu-pying the volume of the porous medium.• Sf >> S∗: Matrix-supported microstructure (Type M): A microstructure in which the finer com-ponent of the gap-gradation constitutes the primary fabric. Coarser particles are dispersedbetween the bulk of finer particles which exist at a relative density 0 < Rd,F < 1.• Sf ≈ S∗: Transitional microstructure (Type T): A transitional microstructure will arise at, ornear, the critical finer fraction content: both coarse and finer particles will form part of theload-bearing microstructure. Void spaces of a clast-supported skeleton may be fully infilledwith finer particles; alternatively, both coarser and finer components may exist at void ratiosclose to - or exceeding - maximum void ratios (ec,max and e f ,max, respectively).Wittmann (1978) observes that the ‘structural classification’ of the sand-gravel mixture deter-mines the observed phenomenon at a critical seepage condition. Figure 2.2 shows the proposedmechanistic framework, based on coarse-to-finer fraction composition. For the clast-supported mi-crostructure, (S f < S∗), particle transportation can occur while the soil structure remains intact. ‘Suf-fosion’ (sic) (equated to suffusion in the present study, see Section 2.2) or colmatation (internal suf-fusion or clogging) phenomena result. In contrast, two-component mixtures with a matrix-supportedmicrostructure (S f > S∗) are not susceptible to these ‘filtration effects’ due to the absence of a coarseskeleton: ‘the deformation of these mixtures can only act on the whole mixture’. ‘Collapse’ occursand the finer particles are transported, resulting in ‘erosion’ - the equivalent of suffosion in the presentstudy. When composed primarily of the finer component (S >> S∗), the material is susceptible toclassic ‘heave’, i.e. fluidization (Figure 2.2).In a similar vein, Kovaˆcs (1981) critiques the method of Lubochkov, pointing out that ‘the possi-bility of suffusion can be investigated only along the lower part of the distribution curve’ representingthe finer, potentially mobile, mass fraction that exists independent of the fixed soil skeleton. Any crit-ical geometric value does not imply high susceptibility to suffusion when generated for the portion ofthe gradation curve constituting the clast-supported skeleton. Kovaˆcs (1981) reiterates the proposal22of Wittmann (1978), stating that a matrix-supported microstructure will be susceptible to phenomenaother than suffusion (for example, ‘subsidence’).Likewise, Kenney and Lau (1985) state that an unstable grading requires: (1) a primary fabric ofparticles which supports imposed stress and remains fixed in position (the coarse skeleton), and (2)loose particles within the pores of this primary fabric. Clearly, Kenney and Lau (1985) are referringto the case where S < S∗ implies susceptibility to suffusion in the framework of Wittmann (1978).Kenney and Lau (1985) proceed to define the range of applicability for their method by mi-crostructural geometry. The gradation curve is assessed for mass percent passing F ≤ 0.3 in nar-rowly graded materials and F ≤ 0.2 in widely graded materials based on an approximation of ‘howmany loose particles the soil can contain’ within the primary fabric (the fixed coarse skeleton). NoteF = 0.2− 0.3 corresponds closely to the critical fines content S∗ as defined by Wittmann (1978) inEquation 2.16, shown for a specific application in Figure 2.2.For F > 0.2 in a widely-graded primary fabric (as per Kenney and Lau, 1985), Garner andSobkowicz (2002) suggest in Figure 2.25 that suffosion will occur. Specifically, particles lost in theportion of the gradation curve where F > 0.2 form part of the load-bearing microstructure. No fixedclast-supported skeleton can exist, and loss of particles will result in collapse of the soil structure.In an analysis of void sizes in soils, A˚ berg (1992) distinguishes between ‘Type A’ matrix-supported soils - whereby all grains transfer inter-particle stress and ‘have influence on the totalvolume of the material’ - and ‘Type B’ clast-supported microstructures comprising a skeleton ofcoarse grains in a fixed position, with loose interstitial finer particles that do not contribute to globalvolume or force transfer (shown in Figure 2.26). For ‘Type B’ materials, A˚ berg (1992) proposes amethod to determine the point on the grain size curve, (xa,ya), that delineates the fixed coarse skeletonfrom the finer fraction of loose particles. Using an analysis of grain and void chords, the global voidratio, e, is defined for a ‘Type B’ material based on an empirical grain shape coefficient and integralsof the volumetric particle size distribution.Skempton and Brogan (1994) similarly consider microstructure as a basis for interpretation ofinternal instability experiments. The theoretical critical finer fraction content for a gap-graded soil, S∗,is defined given porosities of the coarse and finer components, nc and n f , respectively (Equation 2.19and Equation 2.20, above). Figure 2.27 illustrates the relative composition of coarse particles, finerparticles and void space for gap-graded mixture proportions. The critical fines content is reportedto fall within a narrow range: 24% < S∗ < 29%, with lower critical fines content at greater density.Minimum global porosity, n, occurs at the critical fines content whereby all void spaces within thecoarse skeleton are fully occupied by finer particles.Vallejo (2001) compares theoretical and experimental porosities for gap-graded glass bead mix-tures. Figure 2.28 shows the variation in mixture porosity with coarse-to-finer fraction compositionat low normal stress. Like Wittmann (1978) and Skempton and Brogan (1994), Vallejo (2001) reportstheoretical critical finer-fraction contents (corresponding to the minimum porosity condition in Fig-ure 2.28) between 23% < S∗ < 25%. The experimental data of Figure 2.28 suggest S∗ could be ashigh as 30% in practice. Vallejo (2001) distinguishes four mixture microstructures based on porosity23and strength characteristics, defined in terms of coarse- or fine- grain support (see Figure 2.29 andTable 2.6).Thevanayagam and Mohan (2000) and Thevanayagam et al. (2002) further investigate the effectof microstructural geometry on soil strength characteristics. Microstructure is classified in terms ofconstituent void ratios; with the intergranular (ec or es) and interfine (e f ) void ratios describing thecoarse and finer components of the soil mixture, respectively (Equation 2.13 and Equation 2.14).The concept of threshold fines content, FCth, is introduced. FCth is similar, in principle, to thecritical or threshold finer-fraction content, S∗, defined above; however, it is important to note thatFCth is defined by the contribution of finer fraction to soil strength characteristics in the work ofThevanayagam and Mohan (2000), who propose a matrix phase diagram (Figure 2.30) to determinevoid ratios of constituent coarser and finer components (ec and e f ) at any given finer fraction andglobal void ratio. Schematic microstructural diagrams and associated void-geometry conditions arerefined by Thevanayagam et al. (2002), reproduced in Figure 2.31.For a coarse grain soil mixture (low finer fraction content, FC < FCth), three microstructuralcases are defined in terms of increasing finer fraction, FC,% (Figure 2.30 and Figure 2.31):• Case I: for a mixture with little finer fraction, finer particles are confined within the void spacesof the coarse skeleton.• Case II: finer particles are generally confined within the void spaces of the coarse skeleton, yetmay partially support the coarse skeleton.• Case III: finer particles may partially separate the coarse skeleton.In Cases I through III, the coarse grain contacts play a primary role in the shear strength response,while the finer fraction may offer a minor contribution.For a fine grain soil mixture (high finer fraction content, FC > FCth), the coarse grains are fullydispersed in the finer fraction matrix. Fine grain contacts dictate the strength charactersitics of thesoil (Figure 2.30 and Figure 2.31):• Case IV-1: above some limiting finer fraction content, FCL (where FC > FCL > FCth), thesoil behaviour is determined by the finer fraction characteristics. Coarse particles are fullydispersed and do not contribute to soil behaviour.• Case IV-2: the finer particle matrix is fully established, however interactions between separatecoarse-grain particles may form a minor contribution to soil microstructure (FCth < FC <FCL).A ‘finer fraction contribution parameter’, b, is proposed, where b = ‘portion of the fine(r) grainsthat contribute to the active inter-grain contacts’. Conceptually, when b = 0 none of the fine grainsactively support the fixed coarse skeleton. For b = 1, all of the fines actively distribute load andsupport the coarse skeleton. Subsequent studies (Rahman and Lo, 2007; Ni et al., 2004; Rahman24et al., 2008; and Rahman et al., 2011 have provided quantitative evaluation of ‘b’. It is commonlyaccepted that b depends on the finer fraction content, relative density (i.e. inter-granular and inter-finer void ratios), and a measure of relative particle size in the coarse and finer mixture components.Recent studies investigate the effects of changes in soil microstructure arising from seepage-induced particle loss. An emerging mechanics-based framework for modelling temporal changes ingradation and density is presented by Muir Wood (2007). The need for an improved understandingof the consequences of internal erosion was prompted by deficiencies identified in existing analyticalmodels during the interpretation of the 1996 sinkhole incidents at W.A.C. Bennett Dam (Muir Wood,2007, and Muir Wood and Maeda, 2007). Accordingly, the effects of gradation change (i.e. finerparticle loss) on soil strength and critical state have been modelled, using discrete element techniques,for idealized internally unstable soils (Muir Wood and Maeda, 2007, and Maeda et al., 2010). Thiswork, considering the microstructural effects of particles loss, is viewed as complimentary to thepresent study which focuses on the potential susceptibility of soils based on the initial soil gradationprior to any onset of instability.In summary, soil microstructure is a critical factor in determining the susceptibility of a soil tointernal instability. To date, microstructural geometrics have been considered primarily in conceptualterms. Considering the microstructural categories introduced in the work of Thevanayagam and Mo-han (2000), Vallejo (2001), and Thevanayagam et al. (2002) (Figure 2.29 and Figure 2.31), the internalinstability phenomena defined parametrically in the present study (see Table 2.1), can be assigned to asimilar microstructure-based framework (supported by the mechanistic concepts of Wittmann, 1978).Figure 2.32 presents general microstructural conditions for each of the instability phenomena definedin Section 2.2.3 for the present study.2.4.4 Pore and constriction sizeUnder sufficient hydraulic load, unstressed finer particles may be transported through the soil skele-ton, provided there is a sufficiently large continuous pore passage. The third premise for gradationsusceptibility provided by Kenney and Lau (1985) requires that the grading possess a viable trans-portation pathway such that loose particles may be transported within, and out of, the void spaces ofthe soil structure. Thus, pore size and pore constriction geometrics are an additional factor determin-ing susceptibility to suffusion. In the present study, the terms pore and void refer to any space in agranular medium not occupied by soil particles. A constriction is a narrow opening between largerpore spaces.Kovaˆcs (1981) states that, in theory, it is very simple to determine the potential for suffusionwithin a soil: particle transport is geometrically impossible when:Dc < d (2.21)where Dc = constriction diameter and d = diameter of potentially mobile particle. In practice,the assessment of this condition is difficult. For a single particle to be transported at the macro-scale, many inter-pore movements are required; thus the range of constriction diameters within a soil,25Dc,i...i+1, must be evaluated with respect to the range of particle sizes in the potentially mobile frac-tion. Soil microstructure will determine: (1) the size and number of fixed particles forming the poreconstrictions, and (2) the degree of freedom available to any potentially mobile finer particles. Therange of pore and constriction sizes further depend on particle shape, soil grading and density. Fur-thermore, soil fabric is stochastic in nature: particle arrangements will vary even within a gradationat a given density. Simplified, yet accurate, methods of pore size analysis are therefore sought.For uniform spheres, the densest (three-particle) and ‘loosest’ (four-particle) arrangements aredepicted in Figure 2.33. Minimum and maximum theoretical packing porosities are typically givenas nmax = 0.48 and nmin = 0.26 based on the uniform sphere arrangements.Silveira (1965) discretizes the original soil gradation curve into five particle sizes, each with agiven probability of occurrence. The basic geometrics of a three-particle arrangement are analysed todefine a minimum constriction size distribution curve (Figure 2.34). All possible particle size com-binations are considered in the three-particle arrangement, where triangulation of the three particlediameters, di, d j and dk, determines the constriction diameter, d¯. The corresponding probability ofoccurrence, p¯, of a constriction of size d¯, can be calculated based on the individual probabilities ofindividual particles, pi, p j and pk:p¯ =3!ri! · r j! · rk!· pi · p j · pk (2.22)where ri, r j and rk are integers (0 ≤ r ≤ 3) representing the number of times particle i, j, and koccur in the three-particle arrangement. Accordingly, ri + r j + rk = 3.Similarly, a four-particle arrangement (Figure 2.35) is considered to define a ‘maximum’ poresize distribution curve. Silveira et al. (1975) define the maximum constriction size of the four particlearrangement by solving for the interior angles of the quadrilateral ABCD such that the shaded voidarea Sv is a maximum. Constriction diameter, d¯, is then calculated assuming the pore is an equivalentcircle of area Sv. The four particle arrangement for paticles di, d j, dk and dm possesses a corresponding‘percent of occurrence’ based on combination theory:p¯ =4!ri! · r j! · rk! · rm!· pi · p j · pk · pm (2.23)In considering the likelihood of combinations of particles forming a particular void size, a particlesize distribution by mass may over-represent the occurrence of large particles. Wittmann (1979)conducted analysis of epoxy-impregnated gravel slices, and advocated a particle size distributionby number (whereby the gradation is plotted based on the number of particles smaller than a givendiameter). A method of calculation for average pore diameter, d¯p, is presented as a function of: (1)void ratio, (2) particles sizes and their corresponding probability of occurrence by number, and (3) the‘α parameter’, correcting pore dimensions to accommodate passage of a standard sphere. Figure 2.36compares the calculated values of average pore diameter, d¯p, to experimentally measured void areasizes (dp) and the theoretical method of Silveira (1965) (with reference to the coarse particle gradationcurve, D): the constriction size distribution of Silveira (1965) is finer than the empirical pore-size26curves of Wittmann (1979), which is to be expected given that Wittmann (1979) quantifies all poreareas in the cross-section.Wittmann (1979) further expands his void size investigation to consider the transportation ofa small particle through numerous interconnected pores: i.e. a ‘pore channel’. The likelihood ofpassage along a pore channel is dictated by the finest pore size encountered in the void spaces. Resultsare presented in Figure 2.37: as filtration length increases the controlling size for the pore channeltends to the minimum range of the pore sizes measured experimentally.Kenney et al. (1985) define a controlling constriction size, D∗c , based on analytical premises andverify findings through base-filter seepage experiments. Beginning with the basic geometric three-particle arrangement (Figure 2.33) and the general approach of Silveira (1965), theoretical constric-tion size distributions are obtained for soils of different gradings with a common minimum particlesize, D0 (Figure 2.38). As in the method of Wittmann (1979), the probability of occurrence of particlesizes is determined based on the relative number of particles of each particular size in a gradation. Theconstriction size distribution is then assumed to represent individual constrictions along a pore pas-sage, by means of a filter composed of discrete layers with independent flow channels (Figure 2.39).Controlling constriction size, D∗c , is found to be generally independent of the gradation shape (specifi-cally, Cu) and filter thickness (where most real filters comprise more than 10 unit layers, approximatedas 10 times the size of a particle in the primary microstructure). Analysis suggests D∗c is controlledby the size of the finer ‘fixed’ particles, where the fixed particle is taken as that which forms theclast-supported microstructure.Base-filter seepage tests were conducted in order to verify the analytical findings of Figure 2.38,given that D∗c would, in practice, depend on factors not considered in the analytical process (e.g.representative grain size, grading characteristics, particle shape, filter porosity, and filter length).Experiments confirmed that the proposed controlling constriction size, D∗c , is closely related to thesize of small particles in the filter gradation. Upper limits for successful filtration are given by tworelations:D∗cD5≤ 0.25 (2.24)or:D∗cD15≤ 0.20 (2.25)where D5 and D15 are the 5th and 15th percentile sizes of the (coarse) filter gradation. Kenneyet al. (1985) take care to emphasize the fact that pore constriction size, not pore area (as measured inWittmann, 1979), controls particle transportation potential.A further advance in the implementation of the pore constriction geometry was provided in theanalytical work of Humes (1996), who showed that the probability of void sizes corresponding togiven particle arrangements was better predicted using a particle size distribution associated withlateral surface area. This alternative was verified using a three-dimensional void size model, and27found to produce better estimates of the void size distribution than the typical grain size curve bymass (Silveira, 1965) or the grain size distribution by particle number (Wittmann, 1979, and Kenneyet al., 1985).Kovaˆcs (1981) approaches the issue of pore geometry by analysing the characteristics of intercon-nected pore spaces. The ‘capillary tube model’ considers the fixed soil skeleton gradation as a seriesof parallel tubes, with average, largest, and smallest pore diameters (d0, d2, and d1, respectively; Fig-ure 2.40). Largest and smallest diameters are calculated as multiples of the average pore diameter,d0:do =4n1−nDhα (2.26)where n = porosity of fixed soil skeleton, α = shape coefficient based on hydrodynamic equality(specifically, surface-to-volume ratio), and Dh is Kozeny’s effective diameter:Dh =1∑ ∆Fi∆Di(2.27)where ∆Fi = weight of particles in the ith interval of the fixed coarse gradation and ∆Di = corre-sponding mean diameter of particles in the ith interval. Kovaˆcs (1981) assesses the potential passageof a finer particle within the capillary tubes based on the smallest or average constriction withinthe tube in relation to d85 size of the finer fraction (for which 85 percent of particles, by mass, arefiner). Based on empirical findings, Li and Fannin (2013) later adapt the capillary tube model for theassessment of particle transportation potential in potentially suffusive materials.In recent years, the analysis of pore geometry has accelerated with advances in computationalcapacity. Many recent studies (e.g. Indraratna and Vafai, 1997; Locke et al., 2001; Raut, 2006;Indraratna and Raut, 2006; Indraratna et al., 2007; Raut and Indraratna, 2008; Reboul et al., 2008;and Reboul et al., 2010) are based on key geometric concepts proposed by Silveira (1965), Silveiraet al. (1975) and Wittmann (1979).The analytical body of work on filtration with respect to constriction size was expanded to addressthe problem of internal instability in Indraratna (2011). The gradation curve is split into ‘coarser’and ‘finer’ components, based on the method of A˚ berg (1992). The coarser component is thenconsidered the ‘filter gradation’, from which a Constriction Size Distribution (CSD) is derived -based on relative density - within the constriction size limits as determined by the three and fourparticle arrangements (as per Silveira et al., 1975). The CSD by relative density, CSDRd is shownschematically in Figure 2.41, relative to the densest and loosest probabilistic distributions of Silveira(1965 and 1975) (CSDD and CSDL, respectively).Reboul et al. (2008) use discrete element modelling (DEM) to characterize void microstructure.Later, Reboul et al. (2010) provide a comprehensive review of CSD methods and limitations associ-ated with density assumptions and propose an improved DEM method to derive the constriction sizedistribution considering density inhomogeneities. The proposed method is verified through compar-ison with the experimental pore constriction findings of Soria et al. (1993). Later, Cavarretta et al.28(2012) undertake advanced particle-scale characterization of select glass bead materials used in ex-perimental tests to validate DEM simulations. Cavarretta et al. (2012) aim to verify the suitabilityof the parameters used in DEM simulations when modelling specific experimental scenarios under-taken on idealized glass bead ‘soils’. Recently, Shire et al. (2012) further develop the DEM algorithmproposed by Reboul et al. (2008) to quantify void constriction anisotropy based on stress orientation.2.5 Hydraulic factorsSusceptible soils will exhibit internal instability only when subject to sufficient hydraulic load (e.g.Wittmann, 1979; Kovaˆcs, 1981; Skempton and Brogan, 1994; and Ahlinhan and Achmus, 2010).Early studies used extreme hydraulic loads as a means of investigating controlling geometric param-eters (i.e. Kenney and Lau, 1985; Sun, 1989; and A˚ berg, 1993). Specifically, early experimentalprograms applied extreme hydraulic force such that the hydraulic criterion for particle transport wasfulfilled at all times; it was thus assumed that material susceptibility parameters (soil geometrics, inparticular) were controlling any observed threshold between unstable and stable behaviour. More re-cently, the magnitude of the critical hydraulic load has been studied in a more systematic manner viaincremental seepage loading (Table 2.2). Adel et al. (1988) first emphasized the need for a hydraulicapproach in assessing the potential susceptibility of a soil, suggesting that pure geometric methodsare overly conservative given that the hydraulic and dynamic forces used in their development (suchat those imposed in the experiments of Kenney and Lau, 1985), are unlikely to be encountered in thefield.2.5.1 Quantification of hydraulic conditionSeepage force (per unit volume) is quantified using a number of metrics in the literature. The hy-draulic condition is most commonly reported in terms of hydraulic gradient, i, also in terms of veloc-ity, v, and less commonly classified in terms of hydrodynamic parameters (e.g. ‘seepage resistance’,R, the hydrodynamic number, R′; or with reference to Reynolds number).Hydraulic gradient is defined as a change in total head over unit distance in the direction of flow:i =∆H∆L(2.28)Gradient is a macro-scale parameter, defined as an average value between the two points ofmeasurement. No consideration is given to the actual cross-sectional area available for flow in themedium, or the velocity at which water moves in one - or any - direction within the medium.Velocity is a stochastic variable due to the irregular shape of the pore skeleton (Adel et al., 1988).Accordingly, any calculation of seepage velocity will be an average representation of the true rangeof flow speeds and directions present in the porous system.Discharge velocity, vd , (also termed Darcian flux or ‘superficial’ velocity) is commonly reportedin the literature for unidirectional seepage flow. vd is an artificial average velocity, derived frommeasurement of volumetric discharge, Q, assuming flow is one-dimensional and water is the only29phase present in the total cross-section area, A, of the porous medium. Seepage, or pore, velocity,vs, represents the average one-dimensional velocity in the cross-sectional area of flow, i.e. the porearea only. In some applications (i.e. flume tests of Gaucher et al., 2010), a ‘dimensionless’ velocityis discussed, where measured velocity is divided by the sink velocity of the average-sized particle, ω .This approach assumes that the sink velocity represents the threshold for particle motion, thereforemotion can be imparted when ω > 1.Less commonly, the hydraulic condition is quantified using specific parameters calculated fromflow measurements and hydraulic conditions. Kovaˆcs and Ujfaludi (1983) report that ‘seepage resis-tance’, R, is the ‘most suitable parameter to evaluate the process of suffosion’ (equated to ‘suffusion’in the present study):R =∆HQ(2.29)Where ∆H is the pressure head applied across the sample, and Q the measured discharge. Kon-rad and Cote (2013) note that Equation 2.29 can be expressed as R = LkA , a function of hydraulicconductivity, k, and specimen dimensions (length, L, and area, A). Seepage resistance is then usedin empirical definition of a ‘lower boundary’ to the critical flow velocity for internal instability inlow-plasticity soils, based on soil conductivity.Kenney and Lau (1985) and A˚ berg (1993) report experimental seepage conditions in terms of‘hydrodynamic number’, R′. Based on the pore channel experiments of Kenney et al. (1985), hydro-dynamic number is defined:R′ =qD5nν (2.30)Where q = Darcian flux, equivalent to vd in the present study; D5 = particle diameter correspond-ing to 5% mass fraction passing, n = porosity and ν = kinematic viscosity of water. R′ is hencecomparable to seepage velocity multiplied by a measure of pore channel, and divided by the viscosityof water.2.5.2 Experimental and theoretical thresholdsAn incremental hydraulic regime has been implemented in a number of studies in order to identifythe critical hydraulic condition at the onset of internal instability. The hydraulic criterion (i.e. hy-draulic threshold) has been reported in a number of ways, presented in Table 2.7 with reference tothe Terzaghi (1939) theoretical critical hydraulic gradient for ‘piping by heave’ of an internally stablesoil.Expressions for the hydraulic threshold for internally unstable soils have been postulated in theliterature based on a number of parameters. Perzlmaier et al. (2007) cite the hydraulic threshold ofBusch (1993), with critical gradient defined in terms of material susceptibility parameters (specif-ically: dry density, gradation shape, porosity, particle size, hydraulic conductivity, Table 2.7). Theconceptual ‘stress reduction factor’, α , was conceived by Skempton and Brogan (1994) who proposed30that movement of the finer fraction of a soil would occur at a fraction of Terzaghi’s critical hydraulicgradient. Li and Fannin (2012) further consider the ‘α’ stress reduction factor and propose a criticalhydraulic gradient derived from a vertical force balance on a soil unit for upward flow conditions.A discharge velocity-gradient relation is proposed by Wittmann (1977) in Figure 2.42 based on theconcentration of finer fraction in a suffusive soil. The hydraulic gradient comprises three ‘dissipationmechanisms’ or sub-critical gradient components: (1) hydraulic friction, ia; (2) interaction effectof particles, ib; and, (3) sedimentation effect, ic. The interaction of particles is overcome at a pre-critical ‘break-point’ gradient ib; corresponding to the point where some sand particles at ‘a fewlocations’ will ‘begin to move in place’. The critical hydraulic gradient, ic, where ic > ib, occurs when‘sedimentation effects’ are overcome - i.e. the flow velocity exceeds the particle sink velocity forsedimentation. The hydraulic gradient is large enough to overcome any settling of particles loosenedat the pre-critical gradient, and loose particles are thus free to be transported from the void spacewithin the coarse soil skeleton.Similarly, Skempton and Brogan (1994) hypothesize that some ‘threshold gradient’, ich, is re-quired to move particles independently of gravity, regardless of their presence in any soil microstruc-ture. As a first estimate based on the authors’ experiments, Skempton and Brogan (1994) suggestich ≈ 0.16, from horizontal flow tests on material with a similar geometric stability index.Difficulties arise in the determination of a threshold gradient in practice. The ‘critical gradient’depends on identification of the perceived ‘critical’ condition. In practice, there is no universal defini-tion of a critical distress condition for the onset of internal instability. The onset of internal instabilityhas been defined based on different parameters, either alone or collectively:1. Visual observations of mass loss: i.e. first observation of particle washout (e.g. Bendahmaneet al., 2008; Wan and Fell, 2008; Luo et al., 2013) or other visual threshold (i.e. Skemptonand Brogan, 1994 report the critical hydraulic condition when ‘strong general piping of fines’is first observed)2. Rate of change of mass loss (e.g. Adel et al., 1988; Honjo et al., 1996)3. Change in pore pressure (or hydraulic gradient) within the specimen (e.g. Burenkova, 1993;Moffat and Fannin, 2006)4. Change in hydraulic conductivity within the specimen (e.g. Chapuis, 1996)5. Rate of change of hydraulic conductivity within the specimen (e.g. Sun, 1989)6. Change in net flow of water out of the specimen (at constant gradient), or change in discharge-gradient relationship (e.g. Wan and Fell, 2008; Ahlinhan and Achmus, 2010)7. Change in particle size distribution of soil, based on forensic analysis (e.g. Kenney and Lau,1985; A˚ berg, 1993)8. Change in specimen volume or length (e.g. A˚ berg, 1993; Sail et al., 2011)31Some authors observe changes in multiple test parameters at the onset of instability. Wan andFell (2004), Moffat (2005), Li (2008), and Ke and Takahashi (2012) use a suite of the parameterslisted above to determine the ‘critical’ condition. Ultimately, it is necessary to understand the termsby which any experimental hydraulic ‘threshold’ is defined.Skempton and Brogan (1994) report a tentative relation between the geometric stability index, HF(of Kenney and Lau, 1985), and critical hydraulic gradient, as illustrated in Figure 2.43. Ahlinhanand Achmus (2010) further explore the influence of material susceptibility on critical gradient throughtesting of stable, narrowly-graded sands (series A) and gap-graded sands susceptible to internal insta-bility (Series E). The dependence of critical hydraulic gradient was investigated with respect to twomaterial susceptibility factors: (1) relative density of soil (Figure 2.44), and (2) geometric suscepti-bility indices, D′15d′85and HF , after Kezdi (1979) and Kenney and Lau (1985), respectively (Figure 2.45).Finally, the effect of seepage orientation of critical hydraulic gradient was considered, with resultspresented in Figure 2.46.In stable gradations, increasing density results in a higher material unit weight. Therefore, thetheoretical critical hydraulic gradient is slightly greater at higher density. The experimental resultsof Ahlinhan and Achmus (2010) in Figure 2.44 show values of icr in agreement with the theoreticalcritical hydraulic gradient (Terzaghi, 1939) for stable soils A1 and A2 at various densities. For soilsthat are very susceptible to internal instability (gap-graded gradations E2 and E3; for which HF << 1and D′15d′85>> 4), density has very little influence on critical gradient. For these very unstable gradations,icr ≈ 0.2. In contrast, the E1 gradation had geometric indices close to proposed stability thresholds( HF = 1.1 andD15d85= 3.3) and the critical gradient varied with relative density. The critical gradientfor E1 material at a very dense state (icr ≈ 0.6) was found to be double that for the same soil at amedium-dense state (icr ≈ 0.3).The results of Ahlinhan and Achmus (2010) support the tentative geometric-hydraulic relationshipproposed by Skempton and Brogan (1994). Figure 2.45 shows critical gradient as a function of HFand D′15d′85stability indices, with the results of Skempton and Brogan (1994) included for comparison.Critical gradients much smaller than the theoretical hydraulic gradient (≈ 1) are reported for HF < 1,while icr decreases with increasingD′15d′85in an approximate second-order fashion.The critical gradient is tentatively found to be smaller for horizontal flow; however, the rela-tion appears more valid for ‘better’ soils with higher critical gradients (Figure 2.46). For highlyunstable soils (with correspondingly low critical gradients), the effect of flow orientation is not well-understood.Perzlmaier et al. (2007) interpret the geometric-hydraulic influences on critical hydraulic gradi-ent in Figure 2.47, indicating that icr is governed by geometric factors at lowD′15d′85(stable soils withcorrespondingly high icr) and hydraulic factors at highD′15d′85(highly susceptible soils).2.5.3 Scale considerations and a velocity-based frameworkPerzlmaier et al. (2007) consider the quantification of controlling hydraulic criteria, arguing:32‘Looking at the soil on a macro scale as a uniform homogeneous porous medium, hy-draulic gradients causing zero stress condition... can be derived. It is known that internalerosion and suffusion may initiate at gradients lower than this. This can be explainedby assessing the process on a micro scale where the hydraulic load on single particles incross flow exceeds their drag force.’ (Perzlmaier et al., 2007).The force balance on a soil particle is considered theoretically by Perzlmaier et al. (2007), with re-spect to: (1) particle sink velocity, (2) drag force on a resting particle, and (3) sediment transportationtheory. The critical velocity calculated using the three methods converge to a similar value. Theo-retically, critical hydraulic velocity remains almost constant for particles smaller than 1 mm; whiletransport is unlikely for a particle of any size for velocities ≤ 10−3 ms .Factoring seepage velocity, vs, to account for porosity and tortuosity properties, Perzlmaier et al.(2007) estimate that the actual micro-scale (inter-pore) velocity experienced by potentially mobileparticles is four to eight times the macro-scale discharge velocity, vd . The authors further suggestthat an additional factor be considered to account for stochastic variation in soil fabric and, therefore,conductivity within the soil.Recent experimental studies on backward erosion and piping have found that seepage velocity, vs,more accurately correlates to particle movement (e.g Reddi et al., 2000; Richards and Reddy, 2007;Bendahmane et al., 2008; and Marot et al., 2009). Most recently, Konrad and Cote (2013) adapted thework of Kovaˆcs and Ujfaludi (1983) to observe an empirical ‘lower boundary’ to the critical seepagevelocity across 34 tests in the literature (Figure 2.48). The minimum critical seepage velocity forthe initiation of particle movement, vs,cr, is defined as a function of hydraulic conductivity, k, wherevs,cr = 0.22 · k0.68 cms . While all reported data plot above the minimum threshold defined by Konradand Cote (2013), the velocities observed experimentally ranged from one to 10 times the proposedvs,cr threshold. However, the empirical velocity threshold must be further examined to provide aphysical explanation for the observed threshold and inform a science-based understanding of particlemovement. Additionally, the lower bound does not consider microstructure type and the potentialdetachment of finer particles in the load-bearing microstructure, i.e. soils possessing ‘transitional’microstructures, which may occur at 25-35% finer fraction content, are included along with suffusiveclast-supported soils in the empirical study of Konrad and Cote (2013).In summary, a state of zero effective stress must be achieved for particle movement to occur.However, large differences may exist in inter-particle force distribution within a soil at the micro-scale, particularly for highly susceptible finer particles present in the voids of a fixed skeleton (low‘α’, after Skempton and Brogan, 1994). Hydraulic gradient is a very coarse macro-scale measurementof energy differential for an entire soil volume between two points. This suggests that future attemptsto define hydraulic thresholds for particle motion should focus on seepage quantification at the micro-scale (Perzlmaier et al., 2007). Within the tortuous pore network of a soil, the magnitude of localvelocities will vary significantly. There exists no practical way to measure the range of velocitiesat the local scale. However, an average seepage velocity can be deduced for a soil element, whichmay provide insight into the critical hydraulic condition. In the present research, a velocity-based33framework is further explored with regard to experiments on internally unstable materials.2.6 The influence of stressIn Section 2.5, apparent hydraulic thresholds were summarized. Proposed hydraulic thresholds arefunctions of either: (1) material susceptibility factors (i.e. Busch, 1993, in Perzlmaier et al., 2007;and Ahlinhan and Achmus, 2010), or (2) stress conditions (or force-balance) within the soil (e.g.Terzaghi, 1939, the α concept of Skempton and Brogan, 1994, and Li and Fannin, 2012). Recently,the influence of stress conditions on seepage-induced distress has been studied in greater detail.Wittmann (1978) clearly notes the significance of soil microfabric and stress, noting that seepage-induced phenomena are stress-independent in the case of suffusion (termed ‘suffosion’ and ‘colmata-tion’ in the original publication, see Section 2.4.3, Figure 2.2) and stress-dependent in all other cases.In terms of soil microstructure, suffusion requires a clast-supported ‘skeleton’ with finer particleslocated in void spaces. The finer, potentially mobile, particles are conceptually exempt from inter-particle force chains no matter what the applied stress condition. Where the microstructure becomestransitional or matrix-supported, all particles are subject to stress transfer and therefore the resultingphenomena are stress-dependent.The interaction of hydraulic and mechanical (stress) factors was addressed in the doctoral thesesof Moffat (2005) and Li (2008), to produce the ‘theoretical hydromechanical envelope’ from whichequations for critical gradient are derived (Li and Fannin, 2012, Table 2.7). The hydromechanicalenvelope is shown in Figure 2.49 and Figure 2.50. For the soils considered, increasing mean verticaleffective stress, σ ′vm, results in a higher critical hydraulic gradient, icr. The linear σ¯ ′vm - icr relationshipis proposed by Li and Fannin (2012) as a function of α (after Skempton and Brogan, 1994).The experimental programs of Moffat (2005) and Li (2008) focus on the testing of real and ar-tificial soils representing various degrees of geometric susceptibility (after Kezdi, 1979 and Kenneyand Lau, 1985). It is important to note, however, that none of the gradations used in definition of ahydromechanical threshold possess a pure clast-supported microstructure according to the analysesin the literature (e.g. Wittmann (1979), A˚ berg (1992), Skempton and Brogan (1994), Vallejo (2001),or Thevanayagam et al. (2002), Section 2.4.3).Given the interdependence of effective stress and relative density in normally consolidated spec-imens, care must be taken to decouple the effects in published findings. At higher effective stress,higher relative densities will be achieved. As shown in Figure 2.45, the onset of internal instabilitymay depend significantly on relative density, especially for borderline-unstable materials.Shire and O’Sullivan (2013a) model particle connectivity and stress distribution in soils of variousmicrostructures using Discrete Element Modelling (DEM) techniques. Modelling confirms that, for acoarse-grained microstructure (i.e. a gap-gradation with 20% finer fraction), a high proportion of finerparticles are subject to negligible (normalized) contact force, as shown in Figure 2.51. Specifically,when small volumes of finer fraction particles are present, many are unstressed within the coarserskeletal microstructure. As postulated in earlier studies (e.g. Wittmann, 1978; Skempton and Brogan,1994, and Vallejo, 2001), an increasing proportion of finer particles results in a tendency toward a34matrix-supported structure, with greater stresses on a larger portion of the finer particles.2.7 Field manifestation of instability phenomena: Coursier DamAs noted in the draft ICOLD Bulletin on internal erosion (ICOLD, 2013a,b), case histories exist inthe published literature for many mechanisms of internal erosion. However, well-documented casestudies are sparse, and critical information (e.g. performance monitoring data; loading and materialcharacteristics) is presented in varying degrees of detail. In particular, there exists a pivotal lack ofdocumented evidence of suffusion in the field (ICOLD, 2013b).A practical understanding of the research focus is important to the relevance of any laboratorystudy; indeed, many, if not most, of the present topics in geotechnical research are supported by largevolumes of field evidence or trials (e.g. studies concerning liquefaction, piling, soil-structure inter-action, slope stability). Due, in part, to corporate risk-aversion and publicity concerns surroundingin-service dams, dikes, levees, and canals, many field occurrences of seepage phenomenon remainunpublished. There is a distinct need for documented case histories on the manifestation of suffusion,and other seepage phenomena, to provide a basis for applicable and relevant research on the topic.As a component of the present research, a forensic field study was undertaken at Coursier Dam(B.C., Canada). The case study is presented in Appendix A. The intent of this study was to appreciatethe likely manifestation of suffusion, and subsequent filter incompatibility, in the field. Goals of thecase study were to assess:• Spatial and temporal ‘evidence’ of seepage-induced distress at the field scale, and its relevanceto laboratory measurements and observations.• The suitability of existing material susceptibility assessment techniques, given the types ofmaterial used in construction and the availability of characterization data in practice.• Gaps in the state-of-practice understanding of phenomena that prioritized research should ad-dress.Coursier Dam was a 19 m high earth dam constructed in 1963 from widely-graded soils of glacialorigin. The dam experienced a number of seepage events during its service life, prior to decommis-sioning in 2003. Decommissioning was undertaken in recognition of the challenging risk manage-ment associated with potential failure mechanisms. Full details of the case study are presented inAppendix A, with findings most pertinent to the current research presented herein.Performance monitoring records at Coursier Dam included piezometric readings, weir flow dataand visual observations. Piezometric response within the dam reliably characterized an anomaloushydraulic profile associated with sinkhole observations. Evidence suggests that the sinkhole andseepage phenomena manifest in a manner that is highly variable, both spatially and temporally. Atime-series of unusual performance monitoring data spans more than 30 years prior to observationsof daylighting sinkholes during the 1990’s, suggesting that seepage-induced phenomena may developfor many years prior to observation of major distress. The location of sinkholes proves spatially35dependent: gradation analyses indicate that sinkholes likely developed due to incompatibility of theunderlying site geology in relation to engineered dam fill. Furthermore, the incompatible foundationsoil takes the form of a ‘buried channel’ within the underlying site geology. The formation of thisburied channel is attributed to suffusion.Analysis of gradation data at Coursier Dam provides insight as to the analytical challenges facedin engineering practice. At the outset, there exists a lack of historic gradation data for the site. Of thefew gradation analyses undertaken at the time of construction, very few hydrometer analyses wereincorporated. The limited gradation data demonstrates large variability in the gradation ‘envelopes’that characterize Zones of material used in construction. Due to the lack of complete gradation data,forensic sampling was undertaken as a part of the present study in 2011, based on locations identifiedin historic construction documents. A small number of high-quality gradation curves were obtainedfrom field samples. However, even with the benefit of high-quality data, there are few establishedmaterial susceptibility criteria for internal instability in widely-graded soils.With regard to the present study, the Coursier Dam field investigation yielded the following in-sights:• Spatial and temporal ‘evidence’ of seepage-induced distress is observed at the field scale overperiods spanning days, months, and years, in the form of piezometric response, visual reportsof reservoir disturbance, and sinkholes and pipes. It is recognized that the spatial and tempo-ral nature of instability phenomena may not be fully replicable in the laboratory; however, thetesting of large-scale specimens is important to better understand spatial progression of phe-nomena in these materials. Pore pressure transducers can serve as the laboratory counterpart topiezometer instrumentation in the field; both instruments are capable of recording characteris-tic pore pressure ‘signatures’ associated with instability phenomena. In addition, the relativelylow overburden pressures at Coursier Dam suggest that comparable effective stress levels andappropriate gradients can be imposed in laboratory seepage testing.• There exist few established material susceptibility criteria for evaluating the susceptibility ofwidely-graded soils to internal instability. State-of-practice geometric methods (i.e. Kezdi(1979) and Kenney and Lau, 1985 and 1986) may suggest marginal to considerable susceptibil-ity in the foundation materials at Coursier Dam; however, these techniques are not intended forwidely-graded soils. Current state-of-art techniques for widely-graded soils (i.e. Burenkova,1993 and Wan and Fell, 2008) remain largely unverified in practice. Overall, results of suscep-tibility analyses concerning the widely-graded soils at Coursier Dam are inconclusive.• Gaps in the state-of-practice understanding of seepage-induced instability phenomena include:– A present lack of experimental familiarity with the type of widely-graded till materialsoften used in construction.– An incomplete understanding of the factors controlling material susceptibility in widely-graded soils.36– The applicability of existing state-of-practice ‘gradation shape’ techniques as ‘screeningtools’ for widely-graded soils remains unverified.Guided by insights gained from field observations at Coursier Dam, the present research seeks tocontribute to an experimental and analytical understanding of seepage-induced behaviour of widelygraded till materials. Specifically, the thesis seeks to address some deficiencies in the current state-of-practice understanding, listed above and presented in terms of the thesis structure in Section 2.8.2.8 Summary of literature and role of present researchThe research described in this thesis is intended to contribute to a science-based understanding ofseepage-induced internal instability phenomena, by expanding on principles and concepts presentedin the literature to date.In the current chapter, a summary of terminology concerning internal instability is first presented.It is found that many terms have been used in the literature without adequate definition. Furthermore,terms such as suffusion, suffosion, and piping have been used interchangeably to describe fundamen-tally different phenomena. At the outset of this study, terminology in the literature is unified, and aunique set of definitions is presented to define seepage-induced phenomena (Section 2.2.3). Termsare defined based on changes in quantifiable parameters; namely, soil volume (or length), hydraulicconductivity, and mass loss.The material presented in the current chapter defines the focus of the novel seepage tests under-taken in the present study (Chapter 4 and Chapter 5). Experimental literature presented in the currentchapter Section 2.3, in conjunction with the case study at Coursier Dam site (Section 2.7), shows alack of experimental testing in the widely-graded till materials commonly used in practice. To helpaddress this knowledge deficiency, the present study augments the limited experimental database oftests on widely-graded till materials, with tests on materials with fines of varying plasticity. Testresults for widely-graded tills comprise part of Chapter 4.With further regard to the knowledge-gaps highlighted by the Coursier Dam case study (Sec-tion 2.7), the present research seeks to contribute to an understanding of material susceptibility inwidely-graded soils. The present study examines the validity of state-of-practice internal instabilitycriteria with respect to widely-graded soils in Section 6.2, through: (1) a comparison of experimen-tal observations with outcomes of state-of-practice assessment criteria; and (2) parametric analysisof soil gradation shape with varying Cu, to assess the influence of grading on the applicability ofthresholds for ‘gradation-shape’ criteria.In the current chapter, glass bead particles are found to enable theoretical analysis of microstruc-ture and pore constriction sizes (such as Silveira, 1965; Silveira et al., 1975 and Locke et al., 2001),and permit comparison with recent DEM modelling on virtual soils (e.g. Shire and O’Sullivan,2013a). For this reason, glass bead gradations are selected for use in the present experimental study,as detailed in Chapter 4 and Chapter 5.Uniting material susceptibility, hydraulic, and stress factors (Section 2.4, to Section 2.6), a ‘lower37bound’ to the hydromechanical instability spectrum is proposed by Li and Fannin (2012), correspond-ing to very unstable soils with finer particles in the α = 0 condition (after Skempton and Brogan,1994). However, none of the gradations used to define a hydromechanical threshold in the literaturepossess the pure clast-supported microstructure that is theoretically requisite for the α = 0 conditionand suffusion (after Wittmann, 1978, and Skempton and Brogan, 1994). In the present study, glassbead gap-gradations are therefore carefully manufactured to ensure a clast-supported microstructurefor seepage testing (Chapter 3).The literature review identifies three necessary factors required for internal instability, summa-rized by Kenney and Lau (1985). The present study aims to better quantify these factors, namely:(1) particle detachment, (2) a threshold hydraulic criterion, and (3) ‘transportation potential’. Thesefactors form the framework for analyses on glass bead materials tested in the present study:Particle detachment forms the conceptual basis of the ‘α’ stress-reduction framework of Skemp-ton and Brogan (1994) and Li and Fannin (2012), presented in this chapter. In qualitativeterms, Wittmann (1978) details the relation between soil microstructure and instability phe-nomena. More recently, microstructural frameworks have been proposed for interpretation ofsoil strength in gap-gradations (e.g. Vallejo, 2001, Thevanayagam and Mohan, 2000 and The-vanayagam et al., 2002). Based on novel experiments detailed in Chapter 4 and Chapter 5,Chapter 6 of this thesis presents an adaptation of the porosity-based microstructural analysisof Thevanayagam (2000 and 2002) as a framework to assess for particle detachment and the‘α = 0’ condition for internal instability.Threshold hydraulic criterion is typically characterized in terms of hydraulic gradient at the onsetof instability (Section 2.5). The hydromechanical framework of Li and Fannin (2012) doesnot explicitly address the critical hydraulic gradient for ‘α = 0’, while Skempton and Brogan(1994) suggest that some ‘threshold gradient’ will be required to move unstressed particlesindependent of gravity. Synthesis of the literature suggests that the sub-critical gradient thresh-olds proposed by Wittmann (1977) as ‘break point’ and ‘critical gradient’ correspond to: (1)a threshold for particle detachment (as a function of gradient), and (2) a threshold requiredto overcome the ‘self-weight’ of the particle. The latter resonates with the ‘threshold gradi-ent’ concept of Skempton and Brogan (1994). The present study therefore seeks to clarifythe hydraulic threshold required to impart movement to an unstressed particle (Section 6.3.2).Specifically, the present study aims to experimentally verify a tentative velocity-based frame-work proposed by Perzlmaier et al. (2007) for suffusion in upward flow conditions, based onsink velocity as a function of particle self-weight.Transportation potential has only recently been addressed in the literature. Constriction geometryanalyses proposed by Indraratna (2011) and Li and Fannin (2013) are further examined in thisthesis against a database of gap-graded suffusive (clast-supported) soils, augmented by exper-imental data presented in Chapter 4 and Chapter 5. Given recent computational advances,the early principles of constriction and pore passage geometry (e.g. Silveira, 1965 and 1975;38Wittmann, 1979, and Kenney et al., 1985) are revisited and verified with the aid of perme-ameter tests undertaken in the present study using glass bead materials. Based on principlespresented in the literature herein, a novel criterion for ‘transportation potential’ is then devel-oped in Chapter 6.39Table 2.1: Definition of seepage-induced phenomena, in terms of temporal changes in specimen dimension, hydraulic conductivity, and speci-men mass (physically admissible parametric combinations only), refer Figure 2.32Measured parameterPhenomenon DescriptionLength, L Hydraulic con-ductivity, kSpecimenmass, ML< L0 k < k0 M =M0 Consolidation Decrease in specimen volume at constant massL> L0 k > k0 M =M0 Fluidization State of zero effective stress in microstructural particlesL= L0 k = k0 M =M0 Uplift Pure translation of intact specimen; no relative inter-particlemovementL= L0 k ≤ k0 M <M0 Internal suffusion Selective mass transportation within the specimen; no corre-sponding volumetric deformationL= L0 k > k0 M <M0 External suffusion Selective mass loss with no corresponding volumetric defor-mationL< L0 k ≤ k0 M <M0 Suffosion Selective mass loss accompanied by volumetric deformation40Table 2.2: Apparatus and experimental parameters: selected studies on internal instability under unidirectional seepage (Part 1 of 2)Publication Apparatus AxialorientationSpecimendiameter(mm)Length-to-diameterratio ( LD )D100(mm)Seepage regime SeepagemagnitudeDegreeof stresscontrol ∗Vibration Material typeUSACE (1953)Rigid-walledVertical 127 N/S 23 Step-wise, increas-ing, DFiav = 0.5 to161 P sand and gravelmixWittmann(1979)Rigid-walledVariable 150 2.3 60 Variable orientation N/S 1 N gap-graded allu-vium: sand andgravel mixKenney andLau (1985)Rigid-walledVertical 245/580 0.7 to 1.0 38/100 Constant, DF R′ ≥ 10 2 (10kPa) Y sand and gravelAdel et al.(1988)Rigid-walledflumeHorizontal N/S N/A 100 Step-wise increas-ing and decreasing,HFiav = 0.18to 11 N minestone, arti-ficial gradations:sand and gravelSun (1989)ModifiedtriaxialVertical 64 0.4 4.75 Constant, reversedUF/DFiav ≈ 20 4 N sand and clayeysilt mixA˚ berg (1993)Rigid-walledVertical 190 0.7 16 Constant, DF; iav ≈ 22 2 P silt-sand-gravelBurenkova(1993)Rigid-walledVertical N/S N/S 60 to100Various tests: UF,DF, HFiav ≤ 2.5 N/S N/S predominantlysand and gravel;some siltSkempton andBrogan (1994)Rigid-walledVertical 139 1.1 10 Step-wise, increas-ing, UFiav ≤ 1 1 N sand and gravelmixChapuis (1996)Rigid-walledVertical 103 N/S 20 Saturated specimensubject to cyclic-load induced gradi-entsN/S 3 N crushed gravel;crushed limestoneHonjo et al.(1996)Rigid-walledVertical 300 0.3 3 or 20 DF iav ≤ 36 or522 (8.5 gcm2 ) P sand and gravelmix; glass spheresSterpi (2003)Rigid-walledVertical 70 2 2 Constant, UF iav = 0.18to ≈ 11 Y sand and gravelWan and Fell(2004)Rigid-walledVertical 300 0.83 to 1 75 DF: constant. UF:step-wise, increas-ingiav,DF ≈ 8 1 (UF); 2(DF)N clay-silt-sand-gravel∗ Degree of stress control: 1 - zero applied stress; 2 - nominal applied stress; 3 - applied axial stress; 4 - triaxial stress control41Table 2.3: Apparatus and experimental parameters: selected studies on internal instability under unidirectional seepage (Part 2 of 2)Publication Apparatus AxialorientationSpecimendiameter(mm)Length-to-diameterratio ( LD )D100(mm)Seepage regime SeepagemagnitudeDegreeof stresscontrol ∗Vibration Material typeMoffat (2005)Rigid-walledVertical 279 1.2 to 2 75 UF, DF; generallystep-wise, increas-ing≈ 1≤ iav ≤603 N silt-sand-gravelMoffat andFannin (2006)Rigid-walledVertical 279 1.6 3.3 Step-wise, increas-ing, DFiav ≤ 8.3 3 N glass spheresBendahmaneet al. (2008)ModifiedtriaxialVertical 50 1 0.7 Constant, DF iav = 5 to1404 N sand and kaolinmixLi (2008)Rigid-walledVertical 102/279 0.9 to 1.2 19 UF, DF; generallystep-wise, increas-ingiav ≤ 30 3 N sand and gravel;glass spheresCividini et al.(2009)Rigid-walledVertical 74 2.7 9.5 Step-wise, increas-ing, UFiav ≈ 0.2 to11 N silt-sand-gravelMarot et al.(2009)ModifiedtriaxialVertical 50 1 2 Constant, DF iav = 2 to1684 N sand and kaolinmixAhlinhan andAchmus (2010)(1) Rigid-walled, (2)Flume(1) Ver-tical, (2)Horizontal(1) 285 (1) 1.05 3 ‘Gradually’ in-creasingiav ≈ 0.2 to11 N sandChang andZhang (2011)ModifiedtriaxialVertical 100 1 5 Step-wise, increas-ing, DFiav = 0.15to 84 N sand and gravelMarot et al.(2011)ModifiedtriaxialVertical 50 1 1 Constant, DF iav = 0.6 to204 N sand and kaolinmixSail et al.(2011)Rigid-walledVertical 280 1.6 3.3 Step-wise, increas-ing, DFiav = 1 to4.93 N glass spheresKe and Taka-hashi (2012)Rigid-walledVertical 100 1.7 3 Step-wise, increas-ing, UFiav ≤ 0.51 1 N finer and coarsersand mixLuo et al.(2013)ModifiedtriaxialVertical 100 1 10 (1) Short-term,step-wise, increas-ing; (2) long-term,constant.iav ≈ 0 to4.54 N sand and gravelSadaghiani andWitt (2012)Rigid-walledVertical 300 1.7 30 Step-wise, increas-ing, DFiav = 0.1 to12(< 30kPa)N silt-sand-gravel∗ Degree of stress control: 1 - zero applied stress; 2 - nominal applied stress; 3 - applied axial stress; 4 - triaxial stress control42Table 2.4: Measured parameters: selected studies on internal instability under unidirectional seepageHydr. grad. Visual obs. Mass lossPublication Av. Local Spec. Effluent ErodedmassPSD Flow rate Spec. di-mensionsStress(location∗)OtherUSACE (1953) X X X X X XWittmann (1979) X X XKenney and Lau (1985) X(indirect) X X X X(implied) X(T)Adel et al. (1988) X X XSun (1989) X X X XA˚ berg (1993) X(indirect) X XBurenkova (1993) X X XSkempton and Brogan (1994) X X X XChapuis (1996) X X X(implied)Honjo et al. (1996) X X X X XSterpi (2003) X X X X(implied)Wan and Fell (2004) X X X X(DFseries)X(DFseries)XMoffat (2005) X X X X X X X(T, B)Moffat and Fannin (2006) X X X X X X X(T, B)Bendahmane et al. (2008) X X(opticalsensor)X X(volume) X(Tx)Li (2008) X X X X X X X(T, B)Cividini et al. (2009) X X X XMarot et al. (2009) X X(opticalsensor)X X(volume) X(Tx)Ahlinhan and Achmus (2010) X X X(HFonly)XChang and Zhang (2011) X X X X X X X(Tx) Shear strengthMarot et al. (2011) X X(opticalsensor)X X(volume) X(Tx)Sail et al. (2011) X X X X X X X(T) Density (gamma-densiometer)Ke and Takahashi (2012) X X X X X X X Cone resistance profileLuo et al. (2013) X X X X X X(Tx) Volumetric change (axialdisplacement and hoopstrain gauge)Sadaghiani and Witt (2012) X X X X X∗ T = top; B = bottom, Tx = Triaxial43Table 2.5: Lubochkov suffusion thresholds at three tolerance safety factors (Kovaˆcs, 1981)Tolerance safety factor K = Dn−1Dn =DnDn+1Mass increment thresholdlimit, ∆S1∆S21.0 10.0 ≤ 4.01.5 5.0 ≤ 2.62.3 2.5 ≤ 1.744Table 2.6: Microstructures in two-component mixtures (Vallejo, 2001)Finer frac-tion (%)Microstructure Description Reference schematic (Figure 2.29)0 - 20 Coarse grain supported Zero or very small amount offine grains in the voidsA, B20 - 30 Transitional coarse grain supported Mostly supported by coarsegrainsC30 - 60 Transitional fine grain supported Mostly supported by fine grains D60 - 100 Fine grain supported Zero or very small amount ofcoarse grains. Coarse grainsfloat in matrix of fine grainsE, F45Table 2.7: Critical hydraulic thresholds in granular materialStudy Critical threshold ConditionsTerzaghi icr =(1−n)(γs−γw)γw =γ ′γw =Gs+11+e Theoretical critical hydraulic gradient for uniform, internallystable material, subject to no external force under upward ver-tical flow conditionsBusch (1993) (in Perzlmaieret al., 2007)icr = 0.6(ρdρw − 1)(0.82 − 1.8 · n +0.0062 · (Cu − 5)) · sin(30◦ − α8 ) ·√n·g·d2sν ·kFor suffusion in widely-graded soils subject to downward flowinclined α to the vertical. Cu = coefficient of uniformity;n = porosity; ρd and ρw = dry density of soil and density ofwater, respectively; k = Darcian hydraulic conductivity; andν = kinematic viscosity of the fluid. ds = the largest parti-cle subject to suffusion, estimated ds = 0.27 6√Cu n1−nd17, afterWittmann (1980, see Perzlmaier et al., 2007 )Skempton and Brogan (1994)icr = α γ′γw For suffusion, where α is a stress reduction factor accountingfor reduced stress on the finer mobile particlesLi and Fannin (2012)icr = 2(σ¯ ′vm+0.5γ ′γw ) Uniform stress distribution for all particles in the soil (α = 1)icr = α1−0.5α (σ¯′vm+0.5γ ′γw ) Reduced stress on finer soil fraction 0≤ α < 146Figure 2.1: Controlling factors and internal erosion mechanisms (Garner and Fannin, 2010).With permission from the authors; reproduced in Garner and Fannin, 2010.)47Figure 2.2: Seepage-induced phenomena according to soil microstructure (Wittmann, 1978).Reprinted from Symposium on the Effect of Flow through Porous Media, Copyright 1978, with permis-sion from the International Association for Hydro-Environment Engineering and Research.48Figure 2.3: Mechanical phenomena in soils due to seepage (Kezdi, 1979). Reprinted from SoilPhysics, Developments in Geotechnical Engineering, Volume 25, Arpad Kezdi, Cases of phase move-ment, Pages No. 96-153, Copyright 1979, with permission from Elsevier.Figure 2.4: Apparatus used for investigation of ‘inherent stability’ in mixed filters (USACE,1953). Reprinted without restriction under U.S. copyright laws.49Figure 2.5: Two rigid-walled seepage apparatus for downward flow (Kenney and Lau, 1985).Republished with permission of NRC Research Press; permission conveyed through Copyright Clear-ance Center, Inc.50Figure 2.6: Large permeameter schematic, Moffat (2005). Reprinted with permission of the author.Figure 2.7: Schematic rigid-walled seepage device (Sail et al., 2011). Reprinted from EuropeanJournal of Environmental and Civil Engineering, with permission from Taylor and Francis.51Figure 2.8: Schematic illustration of flexible walled permeameter, supplied by Soiltest Inc.,from Sun (1989). Reprinted with permission of the author.Figure 2.9: Schematic representation: triaxial seepage device with effluent monitoring instru-mentation (Bendahmane et al., 2008). With permission from ASCE.52Figure 2.10: Schematic illustration of triaxial seepage device, from Luo et al. (2013). Re-published with permission of Kluwer Academic Publishers; permission conveyed through CopyrightClearance Center, Inc.Figure 2.11: Depiction of ‘filter box’ for a horizontal seepage condition, as used in the experi-ments of Adel et al. (1988).53Figure 2.12: Experimental apparatus with variable angular orientation (Wittmann, 1977).Reprinted from Proceedings of the 9th Australasian Conference on Hydraulics and Fluid Mechan-ics, with permission from Engineers Australia.54Figure 2.13: Lubochkov method to assess material susceptibility (in Kovaˆcs, 1981). Reprintedfrom Seepage Hydraulics, Developments in Water Science, Volume 10, Gyorgy Kovacs, Page No 354,Copyright 1981, with permission from Elsevier.55Figure 2.14: Split-gradation method to assess self-filtering potential (Kezdi, 1979). Reprintedfrom Soil Physics, Developments in Geotechnical Engineering, Volume 25, Arpad Kezdi, Cases ofphase movement, Pages No. 96-153, Copyright 1979, with permission from Elsevier.Figure 2.15: Representative particle sizes: original gradation (Kezdi, 1979). Reprinted from SoilPhysics, Developments in Geotechnical Engineering, Volume 25, Arpad Kezdi, Cases of phase move-ment, Pages No. 96-153, Copyright 1979, with permission from Elsevier.56Figure 2.16: Method of describing gradation curve shape (Kenney and Lau, 1985). Republishedwith permission of NRC Research Press; permission conveyed through Copyright Clearance Center,Inc.Figure 2.17: Method of describing gradation curve shape with relation to Lubochkov limit. Leg-end:‘WG’ - soils widely graded in the range F = 0.2−1.0; ‘NG’ - soils narrowly gradedin the range F = 0.3− 1.0 (Kenney and Lau, 1985). Republished with permission of NRCResearch Press; permission conveyed through Copyright Clearance Center, Inc.57Figure 2.18: Geometric comparison of Kezdi (1979) and Kenney and Lau (1985) methods (fromLi and Fannin, 2008). Republished with permission of NRC Research Press; permission conveyedthrough Copyright Clearance Center, Inc.Figure 2.19: Comparative geometric analysis: empirical database (Li and Fannin, 2008). Repub-lished with permission of NRC Research Press; permission conveyed through Copyright ClearanceCenter, Inc.58Figure 2.20: Geometric assessment of stability. Zones 1 and III - potentially susceptible; Zone 2- unsusceptible; Zone 4 - artificial soils (Burenkova, 1993). Reprinted from Proceedings of the1st International Conference on Filters in Geotechnical and Hydraulic Engineering, with permissionfrom Taylor and Francis.59Figure 2.21: Alternative method for assessing internal instability of broadly graded silt-sand-gravel soils (Wan and Fell, 2008). With permission from ASCE.Figure 2.22: Geometric assessment of 65 sand-gravel filter materials classified by field per-formance, using the unified method of Li and Fannin (2008) (from Ro¨nnqvist, 2010).Reprinted with permission of the author.60Figure 2.23: Geometric assessment of 59 moraine core materials classified by field perfor-mance, using the unified method of Li and Fannin (2008) (from Ro¨nnqvist, 2010).Reprinted with permission of the author.61Figure 2.24: Schematic illustration of inter-granular and inter-finer void ratios in a two-component granular mixture.62Figure 2.25: Hypothesized suffusion-suffosion bound for widely-graded coarse fraction, shownwith reference to the internal instability plot of Kenney and Lau, (1985) (Garner andSobkowicz, 2002). Reprinted with permission of the author.Figure 2.26: Grain size distributions for Type A and Type B materials, with delineation ofcoarser and finer fractions at point [xa,ya] (A˚ berg, 1992). With permission from ASCE.63Figure 2.27: Microstructural composition, volumes, and porosities for bimodal material(Skempton and Brogan, 1994). Republished with permission of ICE Publishing; permissionconveyed through Copyright Clearance Center Inc.64Figure 2.28: Bimodal mixture porosity and threshold finer fraction content: typical result fromVallejo (2001), for laboratory sample at 13.9kPa normal stress. Republished with permissionof NRC Research Press; permission conveyed through Copyright Clearance Center, Inc.65Figure 2.29: Characteristic microstructure schematics for bimodal granular material (refer Ta-ble 2.6): (A), (B) - coarse grain support; (C) - transitional coarse grain support; (D) -transitional fine grain support; (E), (F) - fine grain support (Vallejo, 2001). Republishedwith permission of NRC Research Press; permission conveyed through Copyright Clearance Center,Inc.66Figure 2.30: Intergranular matrix phase diagram for Host Sand ‘A’ with varying fines con-tent: (a) microstructure cases 1-4; (b) effect of fines on soil matrix at constant e (The-vanayagam and Mohan, 2000). Republished with permission of ICE Publishing; permissionconveyed through Copyright Clearance Center Inc.67Figure 2.31: Microstructural classification (Thevanayagam et al., 2002). With permission fromASCE.68Figure 2.32: Microstructural conditions for defined phenomena (Table 2.1).69Figure 2.33: Geometric arrangements of uniform spheres (Wittmann, 1979). Reprinted from Pro-ceedings of the 7th European Conference on Soil Mechanics and Foundation Engineering, with per-mission from the British Geotechnical Association.Figure 2.34: Three-particle arrangement for minimum void size and determination of the ‘dens-est’ constriction size distribution, CSDD (after Silveira et al., 1975).70Figure 2.35: Four-particle arrangement for maximum void size and determination of the ‘loos-est’ constriction size distribution, CSDL (after Silveira et al., 1975).Figure 2.36: Pore area distribution comparison, by method (Wittmann, 1979). Reprinted fromProceedings of the 7th European Conference on Soil Mechanics and Foundation Engineering, withpermission from the British Geotechnical Association.71Figure 2.37: Pore area distribution with increasing filtration length, in terms of ‘m’ filter layers(Wittmann, 1979). Reprinted from Proceedings of the 7th European Conference on Soil Mechanicsand Foundation Engineering, with permission from the British Geotechnical Association.72Figure 2.38: Particle and constriction size distributions for linear gradations (Kenney et al.,1985). Republished with permission of NRC Research Press; permission conveyed through Copy-right Clearance Center, Inc.73Figure 2.39: Controlling constriction size, D∗c , along flowpaths D′c for a filter comprising ‘m=3’layers (Kenney et al., 1985). Republished with permission of NRC Research Press; permissionconveyed through Copyright Clearance Center, Inc.74Figure 2.40: Schematic of capillary tube model, as illustrated in Indraratna and Vafai (1997).With permission from ASCE.Figure 2.41: Schematic illustration of the constriction size distribution by relative densityCSDRd (by the method of Locke et al., 2001 and Nguyen, 2012); and the bounding 3-and 4- particle arrangements, CSDD and CSDL (as defined by Silveira et al., 1975).75Figure 2.42: Gradient thresholds and ‘typical state lines’ for suffusion (Wittmann, 1977).Reprinted from Proceedings of the 9th Australasian Conference on Hydraulics and Fluid Mechan-ics, with permission from Engineers Australia.76Figure 2.43: Critical gradient with HF stability index (Skempton and Brogan, 1994). Republishedwith permission of ICE Publishing; permission conveyed through Copyright Clearance Center Inc.Figure 2.44: Critical gradient with relative density for vertical upward flow (Ahlinhan andAchmus, 2010). With permission from ASCE.77Figure 2.45: Critical gradient with HF andD′15d′85stability indices (Ahlinhan and Achmus, 2010).With permission from ASCE.Figure 2.46: Critical gradient with seepage orientation (Ahlinhan and Achmus, 2010). Withpermission from ASCE.78Figure 2.47: Critical gradient in terms of geometric and hydraulic influence (Perzlmaier et al.,2007). Reprinted from Assessment of the Risk of Internal Erosion of Water Retaining Structures:Dams, Dykes and Levees: Intermediate Report of the European Working Group of ICOLD, withpermission from ICOLD.Figure 2.48: Relationship between critical seepage velocity and hydraulic conductivity for lowplasticity soils (Konrad and Cote, 2013). Reprinted with permission of the author.79Figure 2.49: Hydromechanical relationship for select soils, (Moffat and Fannin, 2011). Repub-lished with permission of NRC Research Press; permission conveyed through Copyright ClearanceCenter, Inc.Figure 2.50: Theoretical hydromechanical envelope for one-dimensional upward flow (Li andFannin, 2012). Republished with permission of ICE Publishing; permission conveyed throughCopyright Clearance Center Inc.80Figure 2.51: Probability density functions of normalized contact force on finer and coarser soilcomponents for a bimodal gradation at 20% finer fraction (Shire and O’Sullivan 2013a).Republished with permission of Springer Verlag; permission conveyed through Copyright ClearanceCenter, Inc.81Chapter 3Apparatus, materials and experimentalprogram3.1 IntroductionThe research objectives were investigated by means of multi-stage seepage tests in a large rigid-walled permeameter. This chapter addresses experimental preliminaries, including a description ofthe apparatus, instrumentation and materials used in testing. Details of the material preparation andspecimen reconstitution processes are provided. The testing program investigated two material types:1. Widely-graded tills2. Gap-graded glass bead mixturesA total of five different gradations were subject to upward unidirectional seepage flow undervarious levels of axial stress. A summary of the testing program is presented.3.2 Apparatus3.2.1 Large permeameterThe experiments were conducted in a large rigid-walled permeameter, with all specimens subjectto upward seepage flow. The large permeameter was originally commissioned for an experimentalstudy on the internal instability in the soils of the WAC Bennett Dam (Moffat, 2005; Moffat andFannin, 2011, and Moffat et al., 2011). The device was later used in an experimental study concerningseepage-induced instability in a range of select gradations of soil and glass beads (Li, 2008).The permeameter cell is made of transparent acrylic, with wall thickness of 13 mm and a totallength of 998 mm. The inside diameter of the cylindrical cell is 279 mm. The cell (A) is fixedto the permeameter base plate (B) via an aluminium top ring (C) and a series of three longitudinalstainless steel tie rods (D) (Figure 3.1 and Figure 3.2). The base plate disperses flow from an inlet82port, permitting uniform flow conditions. The bottom load cell (E) mounts centrally on the base plate,on top of which the base pedestal (F) is positioned.The specimen is confined within the rigid permeameter cell (A). It is supported on a perforatedbase pedestal (F), permitting uniform distribution of unidirectional seepage flow in the longitudinaldirection. A select series of wire meshes are positioned sequentially at the bottom of the specimen,forming an interface with the perforated bottom pedestal (shown later in detail, Figure 3.14). Thebottom wire meshes perform two functions:1. Provide a permeable base platform fine enough to retain particles during specimen reconstitu-tion; and,2. Encourage flow uniformity via lateral dispersion of flow between base plate perforations andthe specimen.Configuration of the upper half of the apparatus (downstream of specimen, for the case of upwardflow examined herein) is dependent on experimental stress conditions. Two configurations are usedfor: (1) σ ′vt = 0: zero vertical effective stress condition, where no vertical effective stress is appliedat the top of the specimen, and (2) σ ′vt > 0: surcharged test condition, where a constant vertical loadis applied to the top surface of the specimen for the duration of the test. These two configurations aredescribed in following paragraphs.3.2.1.1 Tests with σ ′vt = 0In the zero top stress condition, the top surface of the specimen is unconfined (Figure 3.1). The rigidpermeameter cell (A) is fixed via the aluminium top ring (C), with an additional fabricated perspextop ring (P) used to direct discharge flow into a collection hose (Q). Thus, under zero top stress thepermeameter is configured as a constant-head overflow system.3.2.1.2 Tests with σ ′vt > 0In the surcharged top stress condition (Figure 3.2), the top (downstream) surface of the specimen isbounded by a further series of wire meshes, which perform three functions:1. Prevent expulsion of material through top plate (G) perforations during loading;2. Encourage flow uniformity via lateral dispersion of flow between the specimen and top plate(G) perforations; and3. Permit free passage of any eroded particles at the specimen top surface (the flow exit surface inthe upward flow configuration).For surcharged test conditions, a loading rod (H) is fixed to a perforated top plate (G) positionedat the top surface of the specimen. A sealed bearing assembly (I) guides the loading rod through thetop cap (J), permitting frictionless displacement of the top plate while maintaining a water-tight seal.83The top cap is bolted in tight compliance with the aluminium top-ring assembly. Both top ring andtop cap are secured by the longitudinal rods (D) that anchor the permeameter cylinder (A) to the baseplate (B). A lubricated O-ring prevents leakage at the interface between top-ring and top-cap, whilean outlet port (K) within the top cap allows discharge of water to the outlet reservoir.The top load cell (L) and LVDT (M) are mounted at the top of the loading rod. A pneumaticloading ram (N) applies stress to the top of the specimen via the upper load cell (L), loading rod (H)and perforated top plate (G). The entire permeameter assembly is aligned in a reaction frame (O) thathouses the pneumatic loading ram.3.2.2 Water supply and seepage configurationPermeameter testing was conducted with filtered de-aired water, prepared in two stages from thepublic mains water supply (Vancouver, BC) prior to use. Tap water was passed through a two-stagefilter system, manufactured by Millipore Corporation (Figure 3.3). Particles larger than 10 µm werefirst removed from flow passing through the sand filter, with remaining particulate matter larger than3 µm captured in the carbon membrane. Filtered water was held in a modified water cylinder ofapproximately 284 L capacity (John Wood, model JW80SDE) and a vacuum of approximately 80kPa applied for a minimum period of 24 hours to de-air the water.Effectiveness of the de-airing procedure was monitored periodically during the testing programwith a dissolved oxygen meter manufactured by Extech Instruments (Model: 407510). Concentra-tions < 6 mgL were measured for prepared water, thus satisfying the criterion typically used to ensurethat seepage results are not adversely affecting by the presence of air bubbles in geotextile testing(ASTM D-5101 and ASTM D-4491, in Aydilek and Kutay, 2004).The large permeameter permits seepage in either upward or downward flow directions. The find-ings of Moffat (2005) suggest that the hydromechanical onset of internal instability is not governedby flow direction. Li (2008) subsequently conducted experiments on the same material in both up-ward and downward seepage conditions and found that the onset of instability proved independentof flow direction. As detailed in Section 2.3.1.3, the main benefit of an upward flow configuration isimproved specimen homogeneity, resulting from the ability to install a finer mesh at the lower (in-let) boundary of the specimen. A finer basal mesh results in less particle loss during reconstitution,yielding a specimen more uniform along its length. For this reason, seepage flow was imposed in anupward direction in all tests.Seepage flow was imposed via one of two systems, depending on the required magnitude ofdifferential water head to be applied across the specimen. Accordingly, filtered de-aired water wastransferred to one of two supply tanks (at atmospheric pressure for the manual configuration, Sec-tion 3.2.2.1, or fitted with an air-water interface for automated pressurized seepage configuration,Section 3.2.2.2) in preparation for seepage testing, as described in the following paragraphs:843.2.2.1 Manual configuration (differential elevation head)For testing at relatively low differential head (typical average gradients less than 1.5), a manually-adjusted gravity-driven reservoir system was employed to achieve greater precision in gradient con-trol. Two small constant-head devices were wall-mounted on stainless steel rods (Figure 3.4): eachis fitted with an overflow outlet to impose a condition of constant total head. The inlet reservoirreceived de-aired water from a closed supply tank via a Cole Parmer Masterflex variable-speed peri-staltic pump (model 7549-30, with Masterflex Easy Load pump head, model 7529-20), schematicallyillustrated in Figure 3.5. The difference in elevation between the inlet and outlet reservoirs establishesthe total head applied to the system.3.2.2.2 Automated configuration (pressurized seepage head)For testing of materials tested at hydraulic gradients significantly greater than unity, inlet flow wascontrolled via pressurization of a diaphragm tank (WellXTrol WX302, manufactured by Amtrol),illustrated schematically in Figure 3.6. The inlet reservoir contains an internal diaphragm at the air-water interface, allowing pressurization of the water supply while maintaining the de-aired seepagecondition. Pressure in the diaphragm tank is controlled by Labview computer software, via a Field-point TB-10 data acquisition module and an electronic pressure regulator, manufactured by MarshBellofram (model T2000, 0-100 PSI output). Outlet flow exits to a floor-mounted constant headoverflow reservoir.3.3 Instrumentation and data acquisitionTest data are automatically acquired by computer-based data acquisition software, via a number oftransducers wired to A/D (analog-to-digital) hardware. The data acquisition components are manufac-tured by National Instruments (NI): NI LabVIEW software supports the NI Fieldpoint A/D hardwareconfiguration.LabVIEW (Laboratory Virtual Instrumentation Engineering Workbench) 8.2.1 software uses vi-sual programming language incorporating dataflow and graphical programming components. ‘Virtualinstruments’ were created to: (1) monitor and record transducer output, and (2) control output voltageto the electronic pressure regulator in the pressurized seepage configuration. The read/write frequencyfor data acquisition is specified in LabVIEW programming. Data were typically recorded at a fre-quency of 1 to 2 Hz. The software was installed on an AMD Athlon TMX2 desktop computer runningthe Windows XP (Home Edition) platform.The NI Fieldpoint hardware configuration consists of an FP-1600 ethernet network interface sup-porting four I/O (input-output) modules: two FP-AI-110 analog input modules (eight channel, 16 bit),assigned ‘A’ and ‘B’; one FP-TC-120 thermocouple input module (eight channel, utilized for differ-ential inputs in the millivolt range); and one FP-AO-V5 analog voltage output module coupled withthe necessary FP-TB-10 terminal base. Use of specific modules is described along with transducerand load cell specifications, in the following sub-sections.853.3.1 Axial loadAxial load is monitored by two load cells, each of 2250 kgf capacity. The two load cells produceelectrical responses (in millivolts per volt excitation) that are converted to digital output via two chan-nels of the Fieldpoint TC-120 module. The top load cell is positioned below the loading piston atthe top of the specimen, while the bottom load cell is submerged in the permeameter cell, below thespecimen base pedestal (Figure 3.2). The top load cell (Figure 3.7) was manufactured by Interface,model 1210AF-5k, with system calibration indicating a standard uncertainty of±0.0538 kN. The sub-mersible bottom load cell (Figure 3.8) was sourced from the the Model 41 product line manufacturedby Honeywell, with standard measurement uncertainty of ±0.0135 kN. Accordingly, the deducedvalues of axial stress at the top and bottom of the specimen (279 mm cross-sectional diameter) arerated accurate to ±0.8 kPa and ±0.2 kPa, respectively.3.3.2 Axial deformationIn all tests with σ ′vt > 0, axial deformation of the specimen (specifically, change in specimen length)is monitored by continuous measurement of displacement of the top plate. The base elevation of thespecimen is fixed. A linear variable differential transformer (LVDT, Figure 3.9) is mounted on theloading rod (to which the perforated top plate is fixed, Figure 3.2). The custom-manufactured LVDTmeasures displacement of the top plate relative to the fixed top cap, to an accuracy of ±0.05 mmacross the usual 25 mm displacement range. A single channel of the Fieldpoint FP-AI-110 A moduleis assigned to process the electrical LVDT signalFor tests undertaken in the σ ′vt = 0 configuration, the top surface of the specimen was not loadedby the perforated top plate (Section 3.2.1, Figure 3.1). Accordingly, axial deformation could notbe quantified using the LVDT arrangement. Instead, specimen length was monitored graphically,with reference to a series of grid-references affixed to the outside of the permeameter (Figure 3.10).Visual records of top surface elevation were obtained throughout the test, using two methods: (1)photographs taken around the entire perimeter of the specimen at the beginning and end of eachseepage stage (horizontal line of sight, Figure 3.10); (2) continuous time-lapse video recording ata single location on the permeameter wall (using a wide view angle to capture approximately onequarter to one third of the specimen perimeter). Change in the average length of the specimen wastherefore obtained at the specimen perimeter throughout the test.3.3.3 Pore-water pressureThe permeameter side-wall is equipped with 16 ports, each with a porous filter, through which pore-water pressure at the specimen perimeter can be monitored. Saturated semi-rigid nylon tubing (≈ 1.8mm inner diameter) transmits pore-water pressure from the port to pressure transducer. Two arraysof transducers are installed on opposite sides of the permeameter to monitor the spatial distributionof water pressure across the specimen.In any rigid-walled seepage device, there exists concern for variation in hydraulic conditionsbetween the boundary and interior of the specimen. However, experience in large permeameter testing86suggests that generalized flow conditions are achieved in the device. First, the differential and totalpressure transducer arrays, located on opposite sides of the device, provide near-identical responsesin experimental testing, as observed by Moffat (2005), Li (2008), and in the present experimentalprogram. Second, when testing internally unstable material with the aid of perimeter O-rings and arectifying filter (see Section 3.5.2), particle accumulations are observed across the top surface, bothmid-surface and near the side-wall (e.g. Figure C.4 and Figure E.14c), providing visual evidence ofa generalized specimen response. Finally, localized seepage paths observed at the side-wall in thelatter stages of testing on internally unstable materials (e.g. Figure E.14c and Figure E.15b) haveno discernible impact on measured pore pressures (recall that two pressure transducer arrays areinstalled at opposite side-wall locations, providing two points of measurement at a single elevationalong the specimen length). Based on these observations it is reasonable to believe that a generalizedspecimen response is achieved in testing, and that pore water pressures measured at the side-wallprovide satisfactory characterization of the specimen response at that elevation.Differential pressure transducers (DPT) are manufactured by Sensotec/Honeywell (model PDW,part 060-E972-05-01). The operating range of the differential transducers is -70 to +70 kPa. Cali-bration of the sensor and data acquisition system indicates measurement uncertainty of ±0.06 kPa inDPT data (based on conditional standard deviations for calibration data, Appendix B). Five channelsof the Fieldpoint FP-AI-110 ‘B’ input module are assigned to the conversion of DPT signals. Eachdifferential pressure transducer is connected to two instrumentation ports and monitors the differencein pore-water pressure between the two port locations: five DPTs typically span six port locations onthe permeameter side-wall (Figure 3.13 and Figure 3.14).Total pressure transducers are manufactured by Schaevitz, model PS10000, with a range of 0to 350 kPa with measurement uncertainty of ±0.2 kPa when calibrated using the experimental dataacquisition system (Appendix B). Six total pressure transducers are assigned to the Fieldpoint FP-AI-110 ‘A’ input module. The total pressure transducers monitor pore-water pressures at individual portlocations.Local zones within the specimen are delineated based on the location of these pressure transduc-ers, illustrated in Figure 3.13 and Figure 3.14 for till and glass bead materials, respectively. Table 3.1summarizes the anticipated error in values of hydraulic gradient derived, in part, from pressure trans-ducer readings. It is important to note that the measurement of specimen length introduces an ad-ditional source of error. Details of transducer calibrations and uncertainty analyses are presented inAppendix B.3.3.4 Volumetric flow rateDischarge velocity and hydraulic conductivity were calculated from measured volumetric flow rate.Accuracy in the calculation of discharge velocity depends primarily on the uncertainty in measure-ment of: (1) weight, and (2) time. The uncertainty in deduced hydraulic conductivity parameters(average conductivity, kav, and local conductivity, klocal) further depends on the uncertainty in hy-draulic gradient calculations (Table 3.1). Table 3.2 summarizes the uncertainty in parameters derived87from volumetric discharge measurements. The values of Table 3.2 are calculated as the propagationof contributing uncertainties using the methodology and analyses presented in Section B.3.Typical test configurations involve two modes of volumetric flow measurement, described inforthcoming paragraphs.3.3.4.1 Automated measurementAutomated measurement was typically used in quantifying volumetric flow for materials tested underthe pressurized seepage configuration. The pressurized supply tank is mounted on a platform weighscale manufactured by Mars (MSG series) and possesses a 900 kg capacity, accurate to ±0.1 kg. Theload cell of the scale is read continuously via a single channel of the Fieldpoint FP-TC-120 module,providing the change in mass of the inlet supply tank and hence the rate at which water enters thepermeameter.3.3.4.2 Manual measurementManual measurement of volumetric flow was required in two test situations:1. At early stages of testing (average gradients < 5) with automated measurement on materials ofrelatively low hydraulic conductivity, for which the cumulative volumetric flow during a teststage is much less than the resolution of the weigh scales; and,2. Where the manual gravity-driven seepage configuration of Section 3.2.2.1 is used.In the first case of a low-conductivity specimen, near-constant data are obtained from the inletscale. For the second case, a small quantity of water will be spent at the constant-head overflow aswater travels from the supply tank to the permeameter via the elevated constant-head device (Fig-ure 3.4 and Figure 3.5), yielding an overestimation of seepage flow from the weigh-scale data. Inboth cases, water is collected downstream of the permeameter at the overflow of the constant-headoutlet device to verify volumetric flow rate.For (1) above, (low-conductivity materials), increased accuracy was achieved by extending theduration of the seepage stage (see Section 3.6) and collecting cumulative water discharge for theentire stage duration. At low gradients, total volumetric discharge for a 12 hour seepage stage couldbe as little as 5 mL, therefore evaporation rates in the laboratory were measured and the values forseepage discharge adjusted accordingly.For case (2) above, (manual seepage configuration, Figure 3.5, Section 3.2.2.1), three measure-ments of volumetric discharge were taken during each seepage stage. Water was collected over peri-ods ranging from 5 minutes at low gradients, to 30 seconds as gradients approached iav ≈ 1. Collecteddischarge was weighed at the end of each time period, yielding approximate volumes between 500and 1,500 mL.883.4 MaterialsFive gradations were tested in the main experimental program: two widely-graded soils, and threegap-gradations composed of spherical glass beads. Materials processing is influenced by specimenand apparatus constraints detailed in Section 3.4.1. The widely graded soils are described in Sec-tion 3.4.2, and the glass bead gradations in Section 3.4.3.3.4.1 Material and specimen constraintsSpecimens tested in the large permeameter must conform to a number of requirements, related to dataquality and inherent limitations of the apparatus. Specifically, consideration is given to matters ofspecimen length, maximum particle size and hydraulic conductivity.3.4.1.1 Specimen lengthThe ratio of specimen length to cross-sectional diameter has been studied in both consolidation andpermeability applications. Valdes and Caban (2006) considered a range of conductivity studies withlength-to-diameter ratios of 0.5 to 2.0, noting that a specimen requires dimensions ‘sufficiently largeto evade boundary conditions and properly represent the tested soil assemblage’. The experiments ofValdes and Caban (2006) subsequently employed a length-to-diameter ratio of 1.0. Similarly, 11 of24 test programs reported in Table 2.2 possess length-to-diameter ratios in the range 0.7 ≤ LD ≤ 1.2.In accordance with the reasoning of Valdes and Caban (2006) and past experimental studies (Kenneyand Lau, 1985; Moffat, 2005; and Li and Fannin, 2008), the present study adopts a minimum length-to-diameter ratio of approximately 1.0.3.4.1.2 Maximum particle sizeThe maximum size of individual particles within the specimen is limited to facilitate uniformity ofcross-sectional flow. In addition, the particle size limit further promotes specimen uniformity duringthe reconstitution process by reducing the likelihood that individual particles will protrude above thesurface of a batch placement of prepared soil slurry. Furthermore, the presence of large particles nearthe top boundary of the specimen may cause irregular distribution of applied stress.The limiting maximum particle size in seepage testing is generally developed from guidance inseepage testing standards ASTM D-5101 (2012) or ASTM D-2434 (2006a). ASTM D-5101 speci-fies a maximum particle size 110thof the apparatus diameter, while ASTM D-2434 requires that soilparticles be less than 18thto 112ththe inner diameter of the rigid-walled seepage apparatus. Maximumparticle size in the present study is 37.5 mm (1 12 ” size sieve opening), corresponding to the standardsieve size closest to 18ththe 279 mm inner diameter of the large permeameter.3.4.1.3 Hydraulic conductivity: constraintsThe seepage regime for relatively high-conductivity materials may be constrained depending on thetest configuration, namely, the outflow configuration arising from zero or surcharged top-stress con-89figurations (Figure 3.1 and Figure 3.2). Under zero top-stress conditions, the discharge capacityof the permeameter is approximately 4.0 L/min, corresponding to a maximum specimen dischargevelocity of approx. 0.11 cm/s. For the σ ′vt > 0 configuration, experience indicates that excess pore-water pressures develop in the permeameter cell downstream of the specimen at flow rates exceedingapproximately 8.0 L/min. This is due to the limited capacity of the 3/4” diameter discharge valve.Specimens of relatively low hydraulic conductivity are subject to fewer limitations in testing. Thetime taken for one-dimensional consolidation may be significant given the length of a drainage pathin a large permeameter specimen. In terms of multi-stage seepage, any change in differential waterhead (i.e. application of an inter-stage gradient increment) will take a period of time to be reflectedin pore-water pressure measurements within the specimen at distances ‘downstream’ of the pressurechange, given that response time is directly related to specimen conductivity. Stage duration will bedictated by this response time.3.4.2 Widely-graded till materialsTwo soils were sampled from seepage applications at locations in British Columbia (B.C.), Canada,and supplied by the industry partner. The two soils are distinguished by gradation codes ‘MC’ and‘SC’. MC and SC gradations are representative samples of low-conductivity materials of glacial ori-gin; they are similar to many glacial till materials used in construction of dam core zones in B.C. andcountries such as New Zealand and Sweden.MC material is a widely-graded glacial till, comprising 31.4% fines (26.4% silt), 39.6% sand and29% gravel (Figure 3.12). Gradation properties are shown in Table 3.3. The soil has a plasticity index(PI) of 4 to 5% and a coefficient of uniformity, Cu ≈ 170.SC material is a widely-graded glacial till of moderate plasticity (PI = 11% to 16%). SC ma-terial comprises 30.5% fines (19.5% silt), 25.5% sand and 44% gravel (Figure 3.12), with gradationproperties summarized in Table 3.3.3.4.3 Glass bead materialsGlass beads were chosen as an ‘idealized’ medium to investigate geometric gradation and pore prop-erties. The beads are highly uniform rigid spheres, meaning valid comparisons can be drawn betweenexperiments and discrete element soil models.All glass beads were sourced from Potters Industries Inc.. Beads comprised soda-lime silica glasswith a density of 2.5 g/cc and a coefficient of static friction of 0.9 - 1.0. The glass beads possesseda minimum crushing resistance of 96 MPa and a Moh scale hardness of 5 to 6. Particle roundnessranged from 80% for the A-series beads comprising the gap-gradations’ coarser fraction (A-100, A-120 and A-130 beads) to 90% for P-series beads used as the rectifying filter and the finer fraction ofall gap-gradations (P-0170 and P-0070 beads).Three glass bead material gradations were reconstituted from manufactured glass beads of varioussizes (Figure 3.11). Test gradations were bi-modal mixtures of glass beads, with(HF)min = 0 (as perthe method of Kenney and Lau, 1985) and D′15d′85> 5, as per Kezdi (1979). The first two digits of the9053GB22, 66GB22 and 72GB22 gradation ‘codes’ correspond to(D′15d′85)gapratios of 5.3, 6.6 and 7.2,respectively, while ‘GB’ signifies the glass bead material type. The finer fraction for all specimenswas a P-0070 (0.12-0.18 mm) glass bead comprising 22% of the total gradation mass (the final twodigits of the gradation code). The three glass bead gap-gradations were characterized by differentsized coarse components, with A-100, A-120 and A-130 beads constituting the coarser fraction of53GB22, 66GB22 and 72GB22 gradations, respectively. Table 3.4 and Figure 3.11 summarize prop-erties of the glass bead gradations, as well as properties of the uniform ‘rectifying filter’ employed toimprove flow uniformity (detailed in Section 3.5).3.5 Specimen preparation, reconstitution and consolidationThe aim of the specimen reconstitution process was to produce a homogeneous and saturated testspecimen. Two methodologies were employed, based on specific characteristics of the test materials.3.5.1 Widely-graded till materialsWidely-graded till core materials were reconstituted to form a uniform and saturated specimen ina manner adapted from Moffat (2005) and Li (2008). Specimen preparation techniques requiredspecific modification due to the extremely widely-graded nature of the SC material (Cu = 5380) andthe particularly low conductivity of both materials, in the order of k = 10−6 to 10−7 cm/s.Core materials were supplied by the industry partner, BC Hydro. Following pre-screening anddrying, materials were typically prepared and reconstituted in four or five batches over a three toseven day period . A single batch of material comprised six bowls, each containing approximately1.5 kg of soil to a total dry specimen mass of approximately 45 kg.3.5.1.1 Material preparationCore materials were dried and processed in three particle size fractions: D < 19 mm, 19 ≤ D < 38mm and oversize (D > 38 mm based on the permeameter D100 constraint, Section 3.4.1). Oversizeparticles in the as-received gradation were removed and replaced with an identical mass of additionalparticles in the next largest size fraction ( 34 ” to 112 ”). Required mass fractions were weighed separatelyfor each≈ 1.5 kg bowl of prepared material to achieve the ‘as-tested’ gradation of Figure 3.12, whichvaries from the ‘as-received’ gradation in the medium-to-coarse gravel size range only. Each bowlwas then filled with de-aired water and boiled on a hot plate for a minimum of 30 minutes whilesubject to frequent stirring. Core material was then removed from the heat, cooled slightly, andplaced under ≈-70 kPa vacuum to cool completely for a minimum of 12 hours. The cooling materialcontinued to boil for up to 30 minutes due to the low air pressure within the vacuum desiccator.Following vacuum treatment, each bowl of material was then syphoned of excess standing waterand stirred into a slurry. MC materials formed a relatively thick and uniform slurry; however, SCmaterials were much more widely-graded and formed an inhomogeneous mixture of settled gravelswithin a very thin fluid with a relatively large amount of suspended solids.91An additional de-watering phase was therefore introduced in the preparation of SC materials.Each batch of ‘boney’ de-aired slurry was gently heated on a hotplate for a period of three to sixhours to facilitate evaporation of excess water. Frequent stirring of the material ensured uniformevaporation and maintained saturation. The fluid component thickened considerably during the de-watering period as excess water was removed, yielding a mixture of high viscosity such that largeparticles were entrained in the uniform, saturated slurry. Each batch of thickened slurry was thencooled for a second time under vacuum in preparation for placement.3.5.1.2 Specimen reconstitutionThree rigid basal wire meshes were first placed on the perforated base pedestal (Section 3.2), asdescribed above. Two ‘spacer meshes’ of 6.0 mm and 2.0 mm opening size were first placed, servingto disperse flow laterally between base pedestal perforations and the specimen base. The uppermostwire mesh of the base mesh series was of 35 µm opening size, corresponding to the approximate D20to D25 size of MC and SC gradations. The 35 µm mesh was adequate to retain the overlying corematerial. A rigid plastic O-ring was installed to secure the meshes and prevent preferential flow alongthe permeameter side-wall.Each batch of prepared core slurry was then sequentially placed in the large permeameter cylinderusing the modified discrete slurry deposition method, after Moffat (2005). The permeameter cell wasfilled to an elevation ≈ 5 mm above the uppermost basal mesh. Large spoonfuls of the saturated tillslurry were deposited in gentle horizontal motion under the 5 mm deep standing film of water. Pre-pared batches of the test material progressively placed, with excess standing water syphoned prior toplacement of the next batch of material to prevent segregation during placement. When the requiredspecimen length was achieved, the top surface of the specimen was levelled with a flat spoon. Threeboundary meshes were then installed at the top surface of the specimen. The exit meshes served asspacers to encourage lateral flow between the specimen surface and the perforated top plate, whilepermitting free passage of any potentially mobile fine particles. The meshes were placed in a pro-gressively coarser arrangement with the finest mesh of 1.1 mm opening size followed by ‘spacer’meshes of 2.8 mm and 6.0 mm opening size. Meshes secured at the perimeter by an O-ring beforethe perforated top plate, top cap, LVDT and top load cell were sequentially installed.The top cap of the permeameter was installed, followed by the LVDT and top load cell. Theremaining permeameter cylinder volume was then carefully filled with filtered de-aired water. A slowstream of prepared water was introduced via a thin hose placed at the solid mid-section of top plateto disperse flow and prevent disturbance of the specimen. Finally, water was back-flowed from theconstant-head outlet reservoir through the outflow hose to fully saturate the apparatus (Figure 3.5 andFigure 3.6). The outlet valve (installed in the large permeameter top cap) was filled with water, and awet-to-wet hose connection made. A bleed valve permitted back-flow of water from the constant-headoutlet device to purge any residual air in the hose connection.Global homogeneity of the reconstituted till specimens is inferred in measurement of water headdistribution along the ≈ 30 cm specimen length in the initial seepage stages. While all core speci-92mens are reconstituted from a material slurry prepared and placed in five or six discrete batches, thedistribution of water head along the specimen length is highly linear in the initial stage of testing(hydraulic gradients of approximately 1 to 2, Section F.1), indicating that batch placement yields auniform specimen. All intact core specimens appear visually homogeneous (Section E.1).3.5.2 Glass bead materialsGlass bead materials were similarly prepared, reconstituted and consolidated using a process mod-elled on that of Moffat (2005) and Li (2008). In order to satisfy uniformity and saturation require-ments, glass bead specimens were typically prepared and reconstituted in approximately five batchesover a three to four day period. A single batch of material comprised six bowls, each containingapproximately 1-1.5 kg of glass beads, for a total specimen mass of approximately 35 kg.3.5.2.1 Material preparationDry glass beads of the required particle size ranges were measured into each of the stainless steelbowls in the appropriate mass ratio for the given gradation. Each bowl was then filled with de-airedwater and boiled for a minimum of 30 minutes. Frequently stirring encouraged the release of entrainedair. Each bowl of glass bead material was then removed from the heat and vacuum treated accordingto the method described above (Section 3.5.1).Following the vacuum period, each bowl of cooled material was then syphoned of excess standingwater and stirred into a uniform slurry. Care was taken to minimise the entrainment of air whilestirring.3.5.2.2 Specimen reconstitutionThe large permeameter required specific preparation to ensure that the reconstitution method yieldeda saturated, uniform specimen. A series of commissioning tests were first undertaken on gap-gradedglass bead specimens to optimize the apparatus configuration and experimental program for testing ofvery unstable specimens (described in Section 5.2 and Appendix C). A homogeneous rectifying filterwas subsequently introduced at the base of all glass bead specimens in the main experimental programto satisfy flow uniformity requirements within the lower specimen (Figure 3.14). The rectifying filteradhered to the permeability requirements specified by Tanaka and Toyokuni (1991) and adopted inthe experimental work of Wilhelm (2000).First, a series of three rigid wire meshes was placed on the perforated base pedestal: two spacermeshes of 6 mm and 2 mm opening size were placed below a ‘retention’ mesh of 0.23 mm openingsize (refer detail, Figure 3.1). A rigid plastic O-ring was installed in tight compliance with the innerperimeter of the permeameter cylinder to secure the meshes and prevent preferential flow along thepermeameter side-wall during seepage testing.Prepared material was placed in the large permeameter cylinder using the modified discrete slurrydeposition method, detailed above. A single-sized glass bead material (Table 3.4) was first placed onthe basal meshes to form the rectifying filter 8 to 9 cm in length (Appendix C); such that the boundary93between rectifying filter and specimen lay between the ports of DPT 5 (Figure 3.14). The top surfaceof the rectifying filter was levelled using a custom-built slotted aluminium plate. Two wire mesheswere placed on top of the rectifying filter layer. The first boundary mesh (0.17 mm opening size)served as a rigid horizontal delineation between the rectifying filter and specimen. The overlyingwire fabric mesh (35µm opening size) retained the finer fraction of the test specimen and preventedparticle loss into the rectifying filter during reconstitution of the test specimen. A second O-ring wasplaced on top of the boundary meshes at the inner perimeter of the permeameter cylinder.Prepared batches of the test material were then placed atop the rectifying filter using the modifieddiscrete slurry deposition technique until an adequate specimen length was achieved. The top surfaceof the specimen was levelled by carefully removing excess material with a large flat spoon. For testsconducted under σ ′vt = 0 conditions, the top surface of the specimen required no further treatment.For tests subject to applied axial stress (σ ′vt > 0), a series of three wire meshes was installed at thetop (flow exit) surface of the specimen. The exit meshes served as spacers to encourage diffuse flowbetween the specimen surface and the perforated top plate. The meshes were placed in a progressivelycoarser arrangement (0.6 mm, 2.8 mm, and 6.0 mm opening sizes, respectively), with the finest meshin contact with the top surface of the specimen and secured at the perimeter by an O-ring. The openingsize of the finest mesh was selected such that any potentially mobile particles within the specimencould pass freely: all GB22 gradations had a common finer fraction size and, therefore, the finestmesh at the exit interface was common to all tests (0.6 mm opening size).The perforated top plate was then carefully fitted so as to avoid specimen disturbance. A schematicof the rectifying filter, specimen and mesh arrangements is shown in Figure 3.14.On completion of the specimen reconstitution process, the top of the permeameter cell was fullyfilled with water. In the σ ′vt = 0 apparatus configuration, a 5 mm film of standing water remainedat the top of the specimen following reconstitution. The large flat spoon was carefully placed atthe height of the standing fluid, and a gentle stream of de-aired water fed via a small hose onto thespoon to aid dissipation of the flow such that the introduction of water in the cell did not disturbthe specimen surface. As the water level in the cell rose, de-aired water was then streamed downthe cell wall to minimize specimen disturbance. For glass bead specimens tested in the σ ′vt > 0apparatus configuration, the saturation process was identical to that used in testing of widely-gradedtill materials, detailed in Section 3.5.1.2.Due to their refractive properties, glass bead materials provide good visual evidence of any inho-mogeneities in a specimen (e.g. Figure E.9a). Local inhomogeneities are an inevitable consequencewhen reconstituting very internally unstable materials that have a tendency toward segregation. How-ever, as shown in Figure E.9b, the consolidation process appears to reduce minor fabric inhomo-geneities in some ‘lower bound’ specimens. Despite the appearance of local inhomogeneities (asso-ciated with placement of a discrete spoonful of prepared slurry), global homogeneity of the specimenis inferred through a highly-linear distribution of water head measured along the ≈ 30 cm specimenlength on imposition of seepage flow (hydraulic gradients of approximately 0.05 to 0.10, Section F.2).943.6 Test procedure and experimental programFollowing the completion of material preparation, specimen reconstitution and apparatus saturation,the two-part test procedure commenced. Reconstitution by slurry deposition results in a loosely-packed specimen. For tests undertaken in the surcharged test configuration (tests with σ ′vt > 0), spec-imens were consolidated to the target top stress prior to the main multi-stage seepage program.Details of the test procedure are provided in terms of: (1) consolidation, and (2) multi-stageseepage for widely-graded till materials and glass bead materials in Section 3.6.1 and Section 3.6.2,respectively.3.6.1 Widely-graded till specimens3.6.1.1 ConsolidationConsolidation was carried out under double-drained conditions, via incremental application of loadto the top-plate through the loading ram (Figure 3.2). Axial load was applied gradually, in loading in-crements of approximately 2 to 5 kPa, such that excessive gradients (greater than the initial hydraulicgradient imposed in multi-stage seepage) were not induced within the specimen prior to seepage test-ing. Due to the low conductivity of core materials and the relatively long drainage length in the largepermeameter configuration ( L2 ≈ 150 mm), the excess pore-water pressures generated by each loadincrement required significant dissipation periods. Thus, loading increments were typically appliedat 24 hour intervals. The time-frame for consolidation of a till specimen was dependent on the hy-draulic conductivity of the material and the target top stress for the test: typical consolidation periodsspanned one to two weeks. Seepage testing commenced when an ≈ 95% degree of consolidation wasachieved for the target top stress.3.6.1.2 Multi-stage seepageMulti-stage seepage was imposed in an upward-flow configuration under either: (1) zero top stress,σ ′vt = 0, or (2) constant applied top stress, σ ′vt > 0, conditions. The ‘gradient increment’ (∆iav, increasein average gradient between sequential stages) was typically ∆iav ≈ 1, between 3 and 8% of theexpected critical gradient for the specimen, depending on the applied top stress. In early (non-critical)seepage stages, inter-stage gradient increments were often increased slightly, but limited to ∆iav ≤ 2.Stage duration was typically ≈ 90 minutes, else, at very low gradients, the time taken for theincrease in gradient to be observed as a change in pore pressure response along the length of thespecimen. Periodic long-duration stages of seepage flow were imposed overnight and in the initialseepage stages (i.e. very low applied hydraulic gradients). Long-duration stages were typically ≈ 8to 12 hours, during which the collection of outflow allowed improved quantification of volumetricflow at low gradients. Sequential seepage stages of increasing hydraulic gradient were applied until acritical seepage condition was reached.953.6.2 Glass bead specimens3.6.2.1 ConsolidationFor testing in the σ ′vt > 0 configuration, consolidation was carried out under single-drainage con-ditions via incremental application of load to the top-plate through the loading ram (Figure 3.2).Loading increments of approximately 5 kPa were applied at intervals of sufficient duration to allowfull dissipation of excess pore-water pressure. Due to the relatively high hydraulic conductivity of theglass bead specimens, pressure dissipation within the specimen was near-instantaneous, thus the fullconsolidation load was applied in a matter of minutes.3.6.2.2 Multi-stage seepageMulti-stage seepage was imposed in an upward-flow configuration under constant top stress. Eachstage entailed seepage flow at a constant gradient for a standard duration. To permit adequate resolu-tion in determining the critical hydraulic gradient, the increase in gradient between sequential stages(‘gradient increment’, ∆iav) was less than 20% of the expected critical gradient for the specimen, or∆iav ≈ 0.05, whichever was smaller. Stage duration was typically 20±2 minutes or, for the first stageof seepage at very low gradient, the time taken for one full pore volume exchange. Stage durationcommenced when the target hydraulic gradient for the stage was reached.Incremental seepage proceeded until either: (1) apparatus outflow limits were reached, or (2)a constant gradient condition could not be achieved due to rapidly changing specimen properties(corresponding to well-established instability in latter stages of testing).3.6.3 Experimental programSeepage was imposed in a one-dimensional upward flow configuration for all tests. The test codes ofthe main experimental program consist of two components:1. A soil descriptor (two or six characters)• Widely-graded till core materials are represented by a two-character descriptor: the firstletter representing the material source (‘M’ or ‘S’) and the second letter representing thesoil type (‘C’ for core)• Glass bead gradations are described using a three-part, six-character code:(a) Stability index is represented by the first two characters. Specifically, the(D′15d′85)gapratio of the bimodal gradation (after Kezdi, 1979) is presented to two significantfigures, with no decimal point.(b) Glass bead material type is labelled ‘GB’.(c) Finer fraction content (%) is detailed in the final two characters.Thus, a bimodal glass bead gradation with a geometric stability index of (D′15d′85)gap = 6.6and 22% finer fraction is allocated the ‘66GB22’ material descriptor.962. Effective stress on the top of the specimen (numerical value, in kPa), held constant throughoutthe test. Effective stresses ranged from 0≤ σ ′vt ≤ 150 kPa.A summary of the testing program is presented in Figure 3.15. Two tests were undertaken on MCmaterial to complement a single unpublished test undertaken by another researcher prior the presentstudy. Three tests were undertaken on SC material at various top stresses. The MC and SC materialshave plasticity indices of 4-5% and 11-16%, respectively, and provide insight as to the influence ofplasticity on material susceptibility.Commissioning tests were undertaken prior to commencement of the main glass bead testing pro-gram. The commissioning test program is presented in detail in Section 5.2 and Appendix C. A singletest was undertaken on the uniformly-graded P-0070 glass bead material used as the finer componentof gap-gradations in the main testing program. The aim of this test was to observe the theoreticalcritical gradient for an internally stable granular material and verify the quality of data obtained in theexperimental configuration. A second series of commissioning tests was undertaken on the very un-stable 72GB22 gradation in the experimental configuration use by previous researchers (Moffat, 2005and Li, 2008). Specifically, no rectifying filter was employed in the second set of commissioning tests.Extremely unstable gradations had not previously been tested in the large permeameter device. There-fore, the commissioning tests served three functions: (1) verify data quality, (2) define experimentalparameters (e.g. expected critical gradient and appropriate inter-stage gradient increments, head con-trol resolution, and volumetric outflow configuration as determined by material conductivity), and (3)identify improvements to the apparatus for very unstable gradations. As a result of commissioningtests, the rectifying filter was introduced for the main testing program to improve flow uniformity anddata quality, particularly in very unstable materials after the onset of instability.Two tests on the 72GB22 gradation undertaken in the commissioning series at 0 and 100 kPa topstress were replicated in the main testing program (with the addition of a rectifying filter). Within thebounds of data resolution, the onset of instability was observed at comparable gradients (Appendix C),and initial hydraulic conductivity was replicated within 30% for tests at both 0 and 100 kPa.Three glass bead materials were examined in the main testing program, with(D′15d′85)gapgeometricindices of 5.3, 6.6 and 7.2. Four tests were undertaken on each of the 66GB22 and 72GB22 gradationsat applied effective stresses ranging from 0 to 150 kPa. Three tests were undertaken on the 53GB22gradation at applied effective stresses between 0 and 50 kPa.Results of the main testing program are presented according to hydromechanical plotting position(Figure 3.15). Chapter 4 presents results for internally stable specimens plotting at the ‘upper bound’of hydromechanical instability (specifically MC, SC and 53GB22 gradations). Results for very un-stable specimens at the ‘lower bound’ of hydromechanical instability are presented in Chapter 5.3.7 SummaryA large permeameter device was used to conduct large-scale seepage tests. Data were acquired au-tomatically via a series of load, pressure and displacement transducers, and processed via Fieldpoint97I/O hardware and Labview computer software. Manual and automated seepage configurations wereused to ensure adequate experimental control and data resolution.Two types of material were tested: (1) widely-graded till soils (MC and SC gradations) are similarto many well-graded glacial soils used in dam engineering practice, and the results provide insightas to behaviour of soils that lie outside the bounds of traditional geometric susceptibility indices (i.e.Kezdi, 1979 and Kenney and Lau, 1985), in terms of both particle size and plasticity; and (2) glassbeads (GB gradations) provide an ‘ideal’ medium to which geometric analyses can be readily applied.Material processing, reconstitution and consolidation techniques were modified based on unique soilcharacteristics and apparatus limitations. Seepage was imposed using either: (1) the manual seepageconfiguration, employing constant-head devices for higher precision at lower seepage gradients, or (2)the automated seepage configuration, with which high gradients could be imposed via a pressurizedsupply tank at the permeameter inlet and a floor-mounted constant-head outlet device.The main experimental program consisted of 16 large-scale multi-stage seepage tests, conductedunder unidirectional upward flow conditions at various levels of applied axial stress (σ ′vt = 0 to 150kPa). Separate apparatus configurations were employed for the σ ′vt = 0 and σ ′vt > 0 test conditions.Test results are grouped in terms of relative hydromechanical plotting position. Results of tests onstable ‘upper bound’ materials are presented in Chapter 4 and results for very unstable ’lower bound’materials presented in Chapter 5.98Table 3.1: Uncertainty in gradients derived from pore pressure measurementsAverage hydraulic gradient, iav0.20 1.0 10.0 20.0Specimen type DPT TPT DPT TPT DPT TPT DPT TPTWidely-graded tilliav( cms ) - - ±< 0.1 ±0.1 ±0.2 ±0.2 ±0.3 ±0.4ilocal1 ( cms ) - - ±0.1 ±0.2 ±0.4 ±0.6 ±0.8 ±1.0ilocal2 ( cms ) - - ±0.6 ±1.3 ±2.8 ±3.5 ±5.3 ±6.0Glass beadiav( cms ) ±0.02 ±0.07 ±< 0.1 ±0.1 - - - -ilocal1 ( cms ) ±0.05 ±0.17 ±0.1 ±0.2 - - - -1 - calculated for a typical local specimen region 12.5 cm in length (Figure 3.13, Figure 3.14)2 - calculated for a small local measurement zone ≈ 2 cm in length, (see ‘bottom’ region, Figure 3.13)Table 3.2: Uncertainty in discharge velocity and hydraulic conductivity parameters, as derivedusing methodology detailed in Section B.3 (where klocal is calculated for a typical localspecimen region 12.5 cm in length)Average hydraulic gradient, iavSpecimen type 0.2 1.0 10.0 20.0Widely-graded tillvd( cms ) - ±5% ±3% ±2%kav( cms ) - ±9% ±5% ±4%klocal( cms ) - ±14% ±8% ±6%Glass beadvd( cms ) ±1% ±3% - -kav( cms ) ±13% ±7% - -klocal( cms ) ±26% ±12% - -Note: relative uncertainties decrease with increasing iav, as contributing absolute uncertainties (i.e.uncertainties in pore water pressure and volumetric discharge) become smaller relative to theincreasing absolute magnitude of the measurement99Table 3.3: Core gradation propertiesCharacteristic particle size (mm) Fines< 74µm(%)Gradation PI % D100 D90 D60 D50 D20 D10 Cu(D′15d′85)gap(HF)minMC 4−5 38 21 1.6 0.5 0.031 0.008 31.4 173 22@F = 4.3% 0.8@F = 4.3%SC 11−16 38 29 7.0 2.4 0.014 0.0013 30.5 5384 25@F = 12% 0.4@F = 10%Table 3.4: Glass bead gradation propertiesCoarse component Finer componentGradation BeadtypeDiameter(mm)BeadtypeDiameter(mm)Cu(D′15d′85)gap(HF)min72GB22 A-130 1.2 - 1.7 P-0070 0.12 - 0.18 9.2 7.2 066GB22 A-120 1.0 - 1.4 P-0070 0.12 - 0.18 8.2 6.6 053GB22 A-100 0.8 - 1.2 P-0070 0.12 - 0.18 6.6 5.3 0RectifyingfilterP-170 0.3 - 0.43 N/A N/A100Figure 3.1: Exploded and assembled view of large permeameter apparatus: zero top stress con-dition.101Figure 3.2: Exploded and assembled view of large permeameter apparatus: surcharged condi-tion.102Figure 3.3: Two-stage water filtration system, manufactured by Millipore Corporation.103Figure 3.4: Variable inlet and outlet head; gravity-driven seepage at low gradients.104Figure 3.5: Dual-reservoir gravity seepage control configuration: low gradients.105Figure 3.6: Pressurized seepage control system: high gradients.106Figure 3.7: Top load cell, Interface model 1210AF-5k.Figure 3.8: Submersible bottom load cell, Honeywell model 41.107Figure 3.9: Custom-made linear variable differential transformer (LVDT).Figure 3.10: Measurement of axial deformation: grid configuration for zero top stress condi-tions.108Figure 3.11: Particle size distributions for glass bead gradations.Figure 3.12: Particles size distributions for core materials, illustrating oversize treatment.109Figure 3.13: Large permeameter port configuration for core tests.110Figure 3.14: Large permeameter boundary mesh and port configuration for glass bead tests.Figure 3.15: Overview of experimental program for multi-stage seepage.111Chapter 4Experiments at the hydromechanical‘upper bound’4.1 IntroductionThe present study serves to explore the limits of the ‘hydromechanical space’ proposed by Li (2008),after Moffat (2005). Specifically, a unique linear relation between effective stress and critical hy-draulic gradient is proposed by Li (2008), and Li and Fannin (2012), with internally stable materialsforming an ‘upper bound’ to the onset of internal instability in the hydromechanical space.This study seeks to define factors controlling material susceptibility at the hydromechanical ‘up-per bound’, specifically investigating the influence of plasticity in fines, and the potential for particletransportation in gap-graded materials. In order to characterize the ‘upper bound’ to material suscep-tibility, multi-stage seepage tests were conducted on three materials postulated to be internally stable.The three ‘upper bound’ materials comprise: (1) two widely-graded till soils with ≈ 20−30% finesand varying degrees of plasticity (MC and SC material codes, described in Section 3.4.2), and (2) onegap-graded glass bead gradation (53GB22 material code, described in Section 3.4.3) with a geometricsusceptibility index of(D′15d′85)gap= 5.3.The aim of the experimental program was to characterize the onset of a critical seepage conditionin effective stress:gradient space. A multi-stage seepage regime identified the critical seepage gra-dient, while a series of tests was undertaken on each material at a range of applied axial stresses tocharacterize the relation between critical gradient and effective stress.Summarized results are presented by material type in Section 4.2, Section 4.3 and Section 4.4.Detailed results are presented on a test-by-test basis in Appendix D. The interpretation of materialbehaviour is discussed in Section 4.5. Consolidation periods and properties of consolidated specimensat the onset of seepage testing are presented in Table 4.1 and Table 4.2 for widely-graded till and glassbead specimens, respectively.1124.2 MC test seriesTwo tests were performed on the MC material to supplement a single prior test (undertaken outsideof the current research program) that suggested the MC material was likely internally stable (Sec-tion D.1.3). The main objective of the tests on MC material was to determine the hydromechanicalplotting position of a widely-graded soil of low plasticity (PI = 4−5%) and observe the critical fail-ure mechanism. The MC test series serves as a companion to the SC test series which is undertakenat comparable stresses on soils with higher plasticity. Full details of MC-25 and MC-100 tests arepresented in Appendix D.Upon completion of consolidation, multi-stage seepage flow was imposed in an uninterruptedsequence until a critical mechanism was observed. The rationale dictating the seepage stage durationfor each test is described in detail in Section 3.6. Low-gradient seepage stages were maintainedfor 10 to 24 hours to permit adequate quantification of discharge flow, otherwise the standard 90-minute stage duration was used in testing. Initial seepage stages were progressed at a slightly largerthan typical gradient increment of ∆iav ≈ 1.5 to 3. A standard gradient increment of ∆iav ≈ 1.0 wasmaintained for the latter stages of both tests (iav > 6.0 for MC-25, and iav > 32.1 for MC-100, tests).Measurements obtained across the relatively thin local basal region of the MC specimens (‘bot-tom’ zone of Figure 3.13) were subject to a large degree of measurement uncertainty due to the shortgauge length over which differential pore-water pressure readings were obtained. The nature and re-liability of conductivity measurements in the lowermost zone of the widely-graded till experimentalset-up are discussed in Section 3.3 and Appendix B. Given the uncertainty of ≈ 20−60% in gradientand conductivity measurements (Table 3.1 and Table 3.2), less emphasis is placed on data concerningthe lowermost ≈2 cm of the specimen (‘bottom’ zone, Figure 3.13).MC-25 and MC-100 specimens had LD ratios of 0.97 and 1.2, respectively (Table 4.1). Seep-age was imposed in stages of increasing gradient over uninterrupted periods of 132 hours and 307hours for MC-25 and MC-100 specimens, respectively. The tests both terminated with uplift of thespecimen.4.2.1 Pre-critical responseSeepage commenced at gradients of iav = 1.5 and iav = 0.5 for MC-25 and MC-100 tests, respec-tively. At the onset of seepage, the distribution of water head along both specimens was highly linear(Figure F.1 and Figure F.2), indicating that the specimens were homogeneous.No axial displacement occurred in the initial stages of seepage. LVDT instrumentation first in-dicated minor upward movement of ≤ 0.1 mm at gradients of iav = 9.0 (MC-25) and iav = 37.1(MC-100). Very minor upward displacement of < 1.0 mm occurred during remaining pre-criticalseepage stages (Figure 4.1).No change in specimen condition was observed for pre-critical seepage stages at gradients iav <12.0 in the MC-25 specimen and iav < 33.0 in the MC-100 specimen. No evidence of particle loss wasobserved. Fine horizontal cracks were found close to the base of the specimen at iav = 12 (Figure E.1a,MC-25 test) and iav = 33.0 (MC-100 test). Cracks elongated and widened very slightly in subsequent113seepage stages, remaining . 2 mm in width (Figure E.2a).Discharge velocity, vd , increased linearly with average hydraulic gradient, iav, throughout thepre-critical stages of both MC-25 and MC-100 tests. In accordance with Darcian flow conditions,constant average specimen conductivities of kav = 3.0×10−6 cms and kav = 2.4×10−6 are establishedfor MC-25 and MC-100 tests, respectively, in Figure 4.1.Local hydraulic conductivities in the upper-mid, lower-mid, and bottom zones were highly uni-form and constant throughout the pre-uplift seepage regime. Specifically, local conductivities werewithin 20% of the specimen average ( klocalkav ≈ 1.0±0.2) for the MC-25 specimen (Figure 4.2a), whilethe top, upper-mid and lower-mid zones exhibited conductivity ratios of klocalkav ≈ 0.75, 0.95 and 1.3,respectively, during the pre-uplift seepage regime of the MC-100 test (Figure 4.2b). Specifically, nospatial changes in conductivity developed as a result of the imposed seepage regime.4.2.2 Critical responseIn both tests, the application of a successive gradient increment resulted in rapid crack propagationand upward displacement of the specimens. The MC-25 specimen uplifted as an intact unit whenthe gradient was increased to a target value of iav = 14 (Figure E.1b). Displacement of the MC-25specimen occurred over a 50 second period. A similar uplift of the MC-100 specimen occurred over aperiod of approximately three minutes when a target gradient of iav = 42 was imposed (Figure E.2b).Movement was arrested by closing the inlet valve to the permeameter, when the LVDT travel limithad been reached (axial displacement of ≈ 35 to 40 mm, Figure 4.1). With seepage flow terminated,the specimen length (above and below the displacement void) was measured. No change in specimenlength was inferred in either test. Preferential flow paths developed along the premeameter side-wall,over a period of 10 min, as uplift ceased and the specimens moved downward under gravity and axialstresses (Figure E.1c and Figure E.2c). No disturbance of fines on the top of the specimen (the flowexit surface) was observed before, during, or after specimen uplift (e.g. Figure E.2d).4.3 SC test seriesThree tests were conducted on SC material. The pressurized seepage control configuration (detailedin Section 3.2) was used to impose seepage gradient for all tests. The objective of the SC materialtest series was to determine the hydromechanical plotting position and observe the critical failuremechanism of a widely-graded soil of moderate plasticity (PI = 11−16%). The SC test series servesas a companion to the MC test series, permitting a comparison of two widely-graded till soils withvarying degrees of plasticity.For reasons previously discussed (Section 4.2), measurements obtained across the thin local ‘bot-tom zone’ of the SC specimens were subject to a larger degree of measurement uncertainty. As for theMC test series, less emphasis is therefore placed on data for the lowermost ≈2 cm of the specimen(‘bottom’ zone, Figure 3.13).The consolidated SC specimens yielded LD ratios of 1.3 to 1.4 (Table 4.1). Seepage was appliedfor periods ranging from 44 hours (SC-25) to 152 hours (SC-100). All three tests concluded with114uplift of the specimen mass on application of a successive gradient increment, at target gradients ofiav = 12.0, 20.0, and 40.0 for SC-25, SC-50, and SC-100 tests, respectively.Seepage was imposed at a gradient of iav ≈ 1.2 to 1.5. The duration of both the first and secondseepage stages was extended in excess of nine hours to permit volumetric flow quantification. Fromthe second seepage stage, a gradient increment of ∆iav ≈ 1.0 was maintained. At the onset of seep-age, the distribution of water head along both specimens was highly linear (Figure F.3, Figure F.4and Figure F.5), indicating that the specimens were homogeneous. Stage duration was typically 90minutes; however, stages were periodically extended for up to 12 hours overnight in the pre-criticalseepage regime to allow continuous seepage.4.3.1 Pre-critical responseNo visual changes were observed, and no axial displacements recorded, for gradients iav < 10.0 in anyof the three SC tests. Small, discontinuous hairline cracks first developed at the base of each specimenat gradients of iav = 10.0 (Figure E.3a), iav = 19.0 (Figure E.4a), and iav = 37.6 (Figure E.5a), inSC-25, SC-50, and SC-100 specimens, respectively. The small basal crack in the SC-25 specimenwidened to approximately 2mm, propagating around the entire basal perimeter of the specimen, at agradient of iav = 10.9 (Figure E.3b). Measured axial displacements were < 2 mm for all specimensduring respective pre-critical seepage stages (Figure 4.3).Average hydraulic conductivity remained constant throughout the pre-critical seepage stages inall tests, as indicated in Figure 4.3. A linear relation between discharge velocity and average gradientshows an average hydraulic conductivity of between kav = 5.0×10−7 cm/s and kav = 2.1×10−7 cm/sin each of the three SC tests.Spatially, hydraulic conductivity varied little between local zones of the specimens (bounded bythe port locations of each pressure transducer). In SC-25, SC-50 and SC-100 tests, the three localspecimen zones exhibited constant hydraulic conductivity for the pre-critical seepage regime prior tothe onset of hydraulic displacement. In the SC-25 specimen, relative conductivities of ktopkav ≈ 0.80,kUMkav≈ 1.0 and kLMkav ≈ 1.3, are shown for top, upper-mid and lower-mid specimen zones, respectively(Figure 4.4a). Local conductivities in the top, upper-mid and lower-mid specimen zones were sim-ilarly highly uniform in the SC-100 specimen, remaining within 20% of the specimen average inall zones for the duration of the test (0.8 < klocalkav < 1.2, Figure 4.4c). Local conductivities wereslightly non-uniform in the SC-50 specimen; moreso that for other tests on SC material, as dis-cussed in Section 4.5. Regardless, local hydraulic conductivities were constant throughout the test, atktopkav≈ 0.60; kUMkav ≈ 0.90, andkLMkav≈ 2.4, in top, upper-mid and lower-mid specimen zones. Across allSC specimens, spatial changes in conductivity did not occur as a function of imposed gradient.4.3.2 Critical responseUplift of each specimen occurred from the base of the specimen - at the location of previously ob-served cracks - on application of a successive gradient increment. Specifically, bulk uplift occurredon imposition of target gradients iav = 12, 20, and 40, for SC-25, SC-50, and SC-100 specimens, re-115spectively (Figure E.3c, Figure E.4b and Figure E.5b). No preferential loss of particles was observedduring any stage of the test and measurement of the displaced specimen indicated no axial deforma-tion of its intact bulk. Mass uplift signalled the end of the test, and upward movement was arrested ataxial displacements of between 24 and 28 mm (Figure 4.3).Uplifted specimens remained displaced, and intact, following inlet closure (i.e. still subject toapplied top stress, with no applied seepage force). Over a period of 24 hours, particles at the base ofthe specimen fell through the saturated displacement void and settled on the basal mesh arrangement(e.g. Figure E.4c and Figure E.5c). The specimen periphery and upper zones remained undisturbed.A forensic assessment of void ratio was undertaken on the displaced intact portion of the SC-25and SC-100 specimens. Each specimen was carefully extracted and loose material removed fromthe surface of the hydraulic fracture. The specimen was weighed and the specimen volume derivedvia submersion technique. An identical void ratio was derived in both the pre-test and post-upliftcondition for both SC-25 and SC-100 specimens (e = 0.37, and e = 0.31, respectively), confirmingthat no volumetric change occurred at the onset of uplift.4.4 53GB22 test seriesThree tests were performed on 53GB22 material as a comparison to the 66GB22 and 72GB22 testseries. The tests were undertaken at σ ′vt = 0 kPa, 5 kPa, and 50 kPa (53GB22-0, 53GB22-5, and53GB22-50, respectively). The main objective of the tests on 53GB22 material was to characterizethe failure mechanism and determine the hydromechanical plotting position of a gap-graded glassbead gradation with(D′15d′85)gap= 5.3. In contrast to the two widely-graded till materials plotting atthe ‘upper bound’, the 53GB22 glass bead gradation is gap-graded and contains no fines. Like othertests on glass bead material (those plotting at the ‘lower bound’, Chapter 5), the 53GB22 materialwas tested with a rectifying filter and the instrumentation port locations shown in Figure 3.14.The consolidated 53GB22 specimens yielded LD ratios of 1.1 to 1.2 (Table 4.2). Seepage wasapplied for a cumulative period of between 295 and 570 minutes for each 53GB22 test, with periodicinterruption of seepage flow to permit recharge of the water supply. Full details of the seepage regimesare given in Section D.3.1, Section D.3.2 and Section D.3.3, for 53GB22-0, 53GB22-5, and 53GB22-50 tests, respectively.Multi-stage upward seepage flow was imposed in a sequence of increasing gradient until a criticalmechanism was observed, or the limits of the seepage configuration reached. The hydraulic con-ductivity of the 53GB22 glass bead gradation was relatively high in comparison to MC and SC tillmaterials, and experienced showed that it became difficult to apply relatively high gradients (at higherstresses) due to the limited discharge capacity of the apparatus. For these reasons, it was not possi-ble to impose a gradient large enough to trigger a critical failure mechanism in the 53GB22 materialwhen subject to σ ′vt = 50 kPa . The 53GB22-50 test was observed to be stable within the limits of theapparatus (Section D.3.3).1164.4.1 Pre-critical responseThe initial water head distributions for 53GB22 specimens were linear at the onset of seepage, in-dicating that specimens were homogeneous in their reconstituted state (Figure F.6, Figure F.7 andFigure F.8).Negligible axial strain occurred during the first eight to nine seepage stages in 53GB22-0 and53GB22-5 specimens (0< iav < 0.85, Figure 4.5, and 0< iav ≤ 3.1, Figure 4.7, respectively). Slightaxial compression of 1 mm (axial strain εa = 0.3%) was recorded at iav = 0.85) in the 53GB22-0specimen. Both 53GB22-0 and 53GB22-5 specimens underwent minor net axial expansion during thelatter stages of the pre-critical seepage regime. The 53GB22-0 specimen underwent axial expansionat gradients of iav = 0.93 and 0.99, resulting in a net axial strain of εa = -0.3%. Axial expansion ofthe 53GB22-5 specimen occurred at gradients of iav = 3.88 and iav = 4.3, resulting in minor net axialstrain εa = -0.18%. Negligible axial strain (εa = 0.02%) accrued over the duration of the 53GB22-50test.Initial specimens are shown prior to seepage testing in Figure E.6a, Figure E.7a, and Figure E.8a.No visual changes were observed for gradients 0.06≤ iav ≤ 0.81 in the 53GB22-0 test (Figure E.6b).Very small, localized finer-fraction-deficient channels developed during the 10th seepage stage (Fig-ure E.6c). Discontinuous regions of apparent particle transport were observed at the specimen side-wall localized within the top quarter of the specimen. Minor accumulations of finer particles becamevisible at the specimen top surface in subsequent stages (iav = 0.93, 0.99, and 1.12, Figure E.6d).Visual observations at the permeameter side-wall indicated that particle transportation occurred in alocal manner within the upper 10 cm of the specimen only.Similarly, the 53GB22-5 specimen maintained its original appearance throughout the majority ofthe test. A slight ‘marbled’ appearance was observed locally at the specimen side-wall, near the topsurface, at gradients of iav = 2.6 and 2.8. No further changes were observed in following pre-criticalseepage stages, and no particle transportation was observed. The subtle ‘marbled’ effect is shown inFigure E.7b, at a gradient of iav = 3.9.The same slight ‘mottling’ or ‘marbling’ of initial horizontal striations appeared in the 53GB22-50 specimen at gradients in excess of iav = 3.33 (seen in comparison of Figure E.8a and Figure E.8b);yet the effect was very minor. No particle transportation was observed and no significant change wasobserved in the condition of the specimen throughout the 53GB22-50 test. The specimen is shown inpre-test condition in Figure E.8a, and during the final seepage stage, iav = 6.12, in Figure E.8b.A near-linear relation was observed between average gradient and discharge velocity across thepre-critical seepage stages for gradients 0 < iav < 1.5, in all three 53GB22 specimens (Figure 4.5,Figure 4.7 and Figure 4.9). Average hydraulic conductivities between kav = 0.062 and 0.073 cmsare demonstrated for Darcian flow conditions. At gradients exceeding approximately iav = 1.5, theslope of the iav-vd curve flattens and becomes non-linear (Figure 4.7 and Figure 4.9). The non-lineartransition is attributed a transition in flow regime from laminar to non-laminar.The observation of a curvilinear vd − iav relation at gradients exceeding ≈ 1 to 2 in 53GB22materials (Figure 4.7 and Figure 4.9) is consistent with the assessment of laminar conditions in Taylor117(1948). Taking the effective particle diameter as the mean diameter of the 53GB22 gradation, theapproach of Taylor (1948) yields Re≈ 5 at a gradient of iav = 1.5, which exceeds the proposed limitof Re ≈ 1 for laminar flow. The apparent decrease in hydraulic conductivity observed for iav > 1 in53GB22 materials therefore does not represent a true decrease in conductivity; rather, at high seepagevelocities, flow conditions become semi-turbulent or turbulent and Darcian proportionality in the iav-vd relation ceases.Local hydraulic conductivities at the top, mid, and bottom zones of 53GB22-0, 53GB22-5 and53GB22-50 specimens were generally tightly grouped at the klocalkav ≈ 1 bound throughout the tests(0.92 < klocalkav ≤ 1.10, Figure 4.6; 0.85 <klocalkav≤ 1.16, Figure 4.8; and 0.95 < klocalkav < 1.05, Figure 4.10;respectively). The close agreement and constant nature of the local conductivities indicates: (1)a near-uniform spatial conductivity distribution, and (2) no time-dependent change in conductivity.The local conductivities are independent of applied seepage gradient.A deviation in local conductivities was observed in the latter pre-critical stages of the 53GB22-0test, associated with the localized transportation of particles and minor dilation in the uppermost 10cm of the unconstrained specimen. This effect is considered secondary to the predominant failuremechanism and is described further in Section D.3.1.4.4.2 Critical responseFluidization occurred in 53GB22-0 and 53GB22-5 specimens on imposition of a successive gradientincrement. Fluidization resulted in an increase in discharge velocity that was accompanied by adecrease in average gradient across the specimen and visual evidence of widespread, non-preferential,particle motion. The maximum (critical) gradients for the 53GB22-0 and 53GB22-5 tests, icr = 1.14and icr = 5.1, respectively, are annotated by dashed vertical lines in Figure 4.6 and Figure 4.8.Visually, fluidization manifested itself in a wave-like fashion, with particle motion involving largezones of both coarser and finer particles, progressing upward from the base of the specimen. The phe-nomenon resulted in local size-segregated zones within both specimens (Figure E.6e and Figure E.7c)and expansion of the specimen skeleton. For the unconstrained 53GB22-0 test in the σ ′vt = 0 condi-tion, final axial displacement of the top surface was 8.6 mm, corresponding to an axial strain of εa= -2.7% (Figure 4.5). Final axial strain in the 53GB22-5 specimen (constrained by the top plate at 5kPa vertical stress) was εa = -0.96% (Figure 4.7).No critical response was observed in the 53GB22-50 specimen. The test was terminated whenthe outflow limit of the permeameter in the pressurized seepage configuration (Section 3.2.2.2) wasreached, at a discharge velocity of ≈ 0.26 cms .In order to confirm that the 53GB22-50 specimen remained internally stable, the initial four seep-age stages of the test were repeated. Discharge characteristics and hydraulic conductivity (both localand average) remained identical to those observed at the start of the test, confirming that the specimenremained stable.1184.5 Interpretation of results4.5.1 Hydraulic onset of distressThe onset of a distress phenomenon is defined in terms of change in critical specimen parametersdefined in Section 2.2.3. More specifically, in the large permeameter device, the onset of a criticalseepage-induced phenomenon is characterized by a combination of the following indicators:1. change in average hydraulic conductivity2. spatial change in local hydraulic conductivity3. temporal changes in local gradient4. axial displacement5. visual observations of particle transportNo change was found to occur in the average conductivity of MC, SC, and 53GB22 specimens,for the portion of the iav-vd curve corresponding to Darcian flow. No significant visual changes wereobserved in the specimens during the pre-critical seepage regime. Across all tests on MC, SC, and53GB22 materials, visual observations, specimen length and pore-water pressure measurements, andconductivity derivations indicate that specimen properties remained unchanged.In MC and SC material, a temporal change in hydraulic response corresponds to significant axialdisplacement and visual observations of mass uplift. Thus, the onset of distress corresponds to thesimultaneous occurrence of: (1) temporal hydraulic response (Section G.1, Appendix G), (2) axialdisplacement measurement via LVDT, and (3) visual observation of specimen displacement. Themechanism originates within the lowermost zone of the specimen, identified by inspection of the localgradient time-series (Appendix G). Specifically, large changes in porewater pressure response occurthroughout the specimen as uplift occurs, with the pressure response originating in the lowermostzone of pore-pressure measurement (Figure G.1 to Figure G.5).Similarly, in 53GB22-0 and 53GB22-5 test specimens, a sudden decrease in hydraulic resistance(gradient time-series, Figure G.6 and Figure G.7) coincides with significant axial expansion and visualobservations of specimen fluidization. The onset of instability is localized within the specimen basedon local stress and hydraulic conditions, identified through inspection of the temporal local hydraulicresponse (Appendix G). In both 53GB22 and 53GB22-0 specimens, a drop in gradient associatedwith loss of effective stress (i.e. fluidization) first occurs in the lowermost region of the specimen(Figure G.6 and Figure G.7).None of the aforementioned distress indicators (items 1 to 5, above) were observed in the 53GB22-50 specimen, thus the specimen is deemed stable for the range of applied gradients (0≤ iav ≤ 6.12).Correspondingly, no disproportionate change in gradient response is seen in Figure G.8.The locations and hydraulic conditions corresponding to the onset of distress are summarized for‘upper bound’ materials in Table 4.3.1194.5.2 Critical distress mechanismsPost-onset changes in specimen (1) dimensions (axial strain, εa), (2) hydraulic conductivity (kav),and (3) mass (visual observations of particle transportation), are summarized in Table 4.4. Given theparametric definitions of seepage-induced distress mechanisms (Section 2.2.3), results indicate:• MC materials experienced failure due to hydraulic uplift• SC materials similarly experienced uplift at the critical hydraulic condition• 53GB22 materials succumbed to fluidization (provided an adequate seepage force per unit vol-ume could be applied)4.5.3 Hydromechanical analysisIn the present study, stress calculations are based on an assumption of linear total stress distributionalong the specimen length (after Moffat, 2005 and Li, 2008). The vertical effective stress distributionis calculated from load cell readings and measured pore pressures within the specimen. The uncer-tainty of stress derivations is addressed in Appendix B. Initial effective stress measurements (‘initial’,i.e. pre-seepage, series, Figure 4.11 to Figure 4.13) indicate that as little as 40% of the applied verticaleffective stress is transferred to the base of the specimen. Specifically, up to 60% of applied top stressis resisted by side-wall friction at the specimen periphery.As shown in initial and pre-critical stress distributions (Figure 4.11, Figure 4.12 and Figure 4.13),effective stress within the specimen decreases as seepage-induced pore pressures increase. Appliedtop stress is held constant throughout the test, thus minimum effective stresses are generally experi-enced at the base of the specimen.In contrast, the 53GB22-0 specimen in the σ ′vt = 0 condition had the highest effective stress at thebottom of the specimen (Figure 4.13). The base load cell measures only the buoyant self-weight ofthe material, minus any effects of side-wall friction.The onset of seepage-induced instability has been interpreted in a proposed hydromechanicalspace (Moffat, 2005, Li, 2008, and Li and Fannin, 2012). Accordingly, local hydraulic gradient(ilocal) and mean vertical effective stress (σ ′vm) conditions are plotted for the incremental seepageregimes described for each MC, SC, and 53GB22 test in Section 4.2 to Section 4.4. Based on thelocal zone of distress initiation (Table 4.3), the hydromechanical progression of the test is presentedfor MC, SC, and 53GB22 materials in Figure 4.16, Figure 4.20, and Figure 4.24, respectively.A common hydromechanical stability bound is observed for tests in each material: distress occurswhen a local zone of the material approaches the linear regression annotated in σ ′v,m - ilocal space.Experimentally, the linear regression represents a tentative hydromechanical threshold, based on thetheoretical equation proposed by Li and Fannin (2012):icr,local =α1−0.5 ·α(σ¯ ′v,m +0.5γ ′γw)(4.1)120where icr,local is the local critical hydraulic gradient at the onset of instability, α is a parameterpostulated to represent the proportion of stress on the finer fraction of the soil, σ¯ ′v,m is the normalizedlocal mean vertical effective stress, γ ′ is the submerged unit weight of the soil, and γw is the unitweight of water.Based on the hydromechanical plotting positions of Figure 4.16, Figure 4.20, and Figure 4.24, the‘α’ material susceptibility parameter is defined for ‘upper bound’ materials:MC material:α = 0.95±0.05 (4.2)SC material:α = 1.00±0.07 (4.3)53GB22 material:α = 0.95±0.15 (4.4)The stated uncertainties in ‘α’ value indicate the maximum and minimum values of ‘α’ for whichEquation 4.1 will adequately describe the σ ′v,m-icr,local correlation, accounting for the experimentalmargins of uncertainty for the data of the present study. Using individual uncertainties derived forlocal stress and gradient parameters, a margin of expected uncertainty is defined in the hydromechan-ical space for the critical condition of each test (detailed in Section B.4). The stated uncertaintiesdescribe the limits of ‘α’ that produce a linear regression plotting within the margins of expecteduncertainty for the test series.4.5.4 Hydromechanical comparisonThe position of apparent hydromechnical thresholds in MC, SC, and 53GB22 materials correspondswell with the proposed hydromechanical threshold characterized by α = 1 for stable (‘upper bound’)materials (Li and Fannin, 2012).4.6 Summary of experiments at the hydromechanical ‘upper bound’This chapter presents experiments on internally stable materials at the ‘upper bound’ to the onset ofinternal instability in the proposed hydromechanical space. A multi-stage seepage regime identifiedthe critical seepage gradient, while a series of tests was undertaken on each material at a range ofapplied axial stresses to characterize the relation between critical gradient and effective stress.Two widely-graded till materials of varying plasticity were tested. The tests on MC and SC mate-rials enhance the current understanding of seepage-induced behaviour by providing a unique insightinto the behaviour of widely-graded till materials that are vastly prevalent in industry practice, yetrarely tested in controlled conditions. Further tests on the 53GB22 gap-graded glass bead materialprovide insight regarding the apparent stability of some gap-gradations deemed potentially suscepti-ble by classic material susceptibility analyses (Kezdi, 1979, and Kenney and Lau (1985,1986)), andpossessing a relatively low proportion of finer particles.121Internal stability was experimentally established for the MC, SC, and 53GB22 gradations in thischapter. No significant preferential movement of finer particles was observed. Two distinct phenom-ena were observed:• Fluidization was observed in gap-graded 53GB22 glass bead specimens, as characterized bysudden axial dilation and visual observations of fluid-like particulate movement, accompaniedby a decrease in hydraulic gradient.• Uplift was observed in the widely-graded MC and SC till soil gradations, characterized bya constant hydraulic gradient and discharge velocity profile in all pre-uplift seepage stages,followed by axial displacement at the critical condition. No change in specimen length or massoccurred due to uplift.All three materials subject to fluidization or uplift were found to plot at the ‘upper bound’ ofthe proposed hydromechanical space (after Li and Fannin, 2012), with each soil demonstrating a hy-dromechanical threshold of ‘α ≈ 1’ within the bounds of experimental uncertainty. The experimentalbody of data presented in this chapter thus verifies the presence of a hydromechanical ‘upper bound’to the onset of internal instability, and demonstrates that the proposed ‘upper bound’ characterizesinternal stability in soils subject to both fluidization and uplift phenomena.122Table 4.1: Initial specimen properties: ‘upper bound’ widely-graded till specimensTest Code Spec. length (cm) Tconsol (days) e n σ ′t σ ′bMC-25 27.2 8 0.30 0.23 24.3 16.2MC-100 32.5 5 0.29 0.22 99.0 39.4SC-25 38.0 6 0.37 0.27 24.8 11.4SC-50 37.0 13 0.35 0.26 50.2 21.8SC-100 35.1 17 0.31 0.23 98.8 42.9Table 4.2: Initial specimen properties: ‘upper bound’ glass bead specimensLength, L (cm) Effective stressTest Code Rect.Filter Specimen Tconsol (hr) e n σ ′vt σ ′vb53GB22-0 8.7 32.4 < 0.5 0.44 0.31 0 5.653GB22-5 8.9 31.4 < 0.5 0.41 0.29 4.6 5.753GB22-50 8.7 31.9 < 0.5 0.42 0.29 48.9 17.1Table 4.3: Distress in ‘upper bound’ materials: critical hydraulic gradients and local zone ofinitiationTest code Av. gradient, iav Local zone of initiation Local gradient, ilocalMC-25 13.0 Base 13.0MC-100 41.0 Base 35.1SC-25 10.9 Base 6.35SC-50 19.0 Base 8.5SC-100 38.8 Base 37.953GB22-0 1.12 Lower (DPT4) 1.3953GB22-5 5.1 Lower (DPT4) 4.053GB22-50 N/A N/A N/A123Table 4.4: Identification of distress mechanisms in ‘upper bound’ materialsAxial displacement Hydraulicconductivity,kavMass lossobserva-tionMechanismTest code MeasuredvalueSpecimenlengthMC-25 D = −39.0mmL = L0 kav = kav,0 x UpliftMC-100 D = −34.5mmL = L0 kav = kav,0 x UpliftSC-25 D = −23.8mmL = L0 kav = kav,0 x UpliftSC-50 D = −27.6mmL = L0 kav = kav,0 x UpliftSC-100 D = −26.0mmL = L0 kav = kav,0 x Uplift53GB22-0 εax =−2.7% L > L0 kav > kav,0 x Fluidization53GB22-5 εax =−0.96% L > L0 kav > kav,0 x Fluidization53GB22-50 εax = 0.02% L≈ L0 kav = kav,0 x - Stable -124Figure 4.1: MC test series: variation of discharge velocity and axial displacement with averagegradient.125(a) MC-25: Normalized local conductivity(b) MC-100: Normalized local conductivityFigure 4.2: MC test series: normalized local conductivity ( klocalkav ) with average gradient (iav) intop (TOP), upper-mid (UM), lower-mid (LM) and bottom (BOT) specimen regionsFigure 4.3: SC test series: variation of discharge velocity and axial displacement with averagegradient.126(a) SC-25: Normalized local conductivity(b) SC-50: Normalized local conductivity(c) SC-100: Normalized local conductivityFigure 4.4: SC test series: normalized local conductivity ( klocalkav ) with average gradient (iav) intop (TOP), upper-mid (UM) and lower-mid (LM) specimen regions.127Figure 4.5: 53GB22-0: average hydraulic conductivity and axial displacement.Figure 4.6: 53GB22-0: normalized local hydraulic conductivity ( klocalkav in top, middle and bottomspecimen regions) with average gradient, iav.128Figure 4.7: 53GB22-5: average hydraulic conductivity and axial strain.Figure 4.8: 53GB22-5: normalized local hydraulic conductivity ( klocalkav in top, middle and bottomspecimen regions) with average gradient, iav.129Figure 4.9: 53GB22-50: average hydraulic conductivity and axial strain.Figure 4.10: 53GB22-50: normalized local hydraulic conductivity ( klocalkav in top, middle and bot-tom specimen regions) with average gradient, iav.130Figure 4.11: MC tests: Stress distribution along specimen length: initial and final stages.Figure 4.12: SC tests: Stress distribution along specimen length: initial and final stages.131Figure 4.13: 53GB22 tests: Stress distribution along specimen length: initial and final stages.132Figure 4.14: MC-25: local hydromechanical paths to pre-critical seepage stage.Figure 4.15: MC-100: local hydromechanical paths to pre-critical seepage stage.133Figure 4.16: All MC tests: critical hydromechanical state (including a third test previously con-ducted at UBC, ‘MC-50’).Figure 4.17: SC-25: local hydromechanical paths to pre-critical seepage stage.134Figure 4.18: SC-50: local hydromechanical paths to pre-critical seepage stage.Figure 4.19: SC-100: local hydromechanical paths to pre-critical seepage stage.135Figure 4.20: All SC tests: critical hydromechanical state.Figure 4.21: 53GB22-0: local hydromechanical paths for all pre-critical seepage stages.136Figure 4.22: 53GB22-5: local hydromechanical paths for all pre-critical seepage stages.Figure 4.23: 53GB22-50: local hydromechanical paths for all seepage stages (no critical condi-tion observed).137Figure 4.24: All 53GB22 tests: critical hydromechanical state in 53GB22-0 and 53GB22-5tests; non-critical condition in 53GB22-50 test.138Chapter 5Experiments at the hydromechanical‘lower bound’5.1 IntroductionIn addition to the testing of materials at the proposed hydromechanical ‘upper bound’ (Chapter 4), thepresent study aims to verify experimentally the presence of a ‘lower bound’ to the hydromechanicalspace postulated, but not experimentally confirmed, by Li and Fannin (2012). Additionally, thisstudy seeks to define factors controlling material susceptibility and the critical hydraulic condition inunstable gap-graded materials at the hydromechanical ‘lower bound’ to the onset of seepage-inducedinternal instability. Multi-stage seepage tests were therefore conducted on two glass bead materialspostulated to be highly internally unstable. The gap-graded glass bead gradations consist of a mixtureof two size fractions - a ‘coarser fraction’ and a ‘finer fraction’, described in Section 3.4.3. The finerfraction is common to both 66BG22 and 72GB22 ‘lower-bound’ materials as well as the companion53GB22 gap-gradation which plots at the ‘upper bound’ to the hydromechanical space (Chapter 4).The 66GB22 and 72GB22 materials have geometric susceptibility indices of(D′15d′85)gap= 6.6 and 7.2,respectively (after Kezdi, 1979).As for the ‘upper bound’ experiments, the aim of seepage testing at the ‘lower bound’ is to char-acterize the onset of a critical seepage condition in effective stress - gradient space. A multi-stageupward seepage regime was therefore imposed with increasing gradient. Four tests were undertakenon each of the 66GB22 and 72GB22 gradations at axial stresses ranging from 0 to 150 kPa in orderto characterize any relation between critical gradient and effective stress at the onset of instability.Prior to the main experimental program, commissioning tests were undertaken on glass bead ma-terials (Section 5.2 and Appendix C). Extremely unstable gradations had not previously been testedin the large permeameter apparatus, and, therefore, the purpose of these tests was threefold: (1) ver-ify experimental data quality, (2) define experimental parameters for very unstable gap-gradations(e.g. expected magnitude of critical gradient, required resolution of data, inter-stage gradient incre-ments, volumetric outflow configuration as determined by material conductivity), and (3) identify139improvements to the apparatus that may aid data quality at very low gradients. As a result of thecommissioning tests, a rectifying filter was introduced for the main testing program to improve flowuniformity and data quality in very unstable materials, particularly after the onset of instability.This chapter focuses on characterization of the critical seepage response in 66BG22 and 72GB22materials (the main ‘lower bound’ testing program). Typical pre- and post- consolidation specimensare shown in Figure E.9 and initial specimen properties for the ‘lower bound’ test series are presentedin Table 5.1. Results of seepage tests are presented by material type in Section 5.3 and Section 5.4 anddescribed on a test-by-test basis in Appendix D. The interpretation of material behaviour is discussedin Section 5.5.5.2 Commissioning tests5.2.1 Experimental and theoretical verification: ‘piping by heave’A single test was undertaken on a very narrowly-graded specimen reconstituted entirely of the P-0070glass bead material used as the ‘finer fraction’ of gap-gradations in the main experimental program.The objectives of this test were as follows:1. Observe the theoretical critical gradient for ‘piping by heave’ in a uniformly-graded material(Terzaghi, 1939)2. Verify the reliability of measurements in the large permeameter apparatus at very small valuesof hydraulic gradient ≤ 1.3. Obtain a value of hydraulic conductivity for the ‘finer component’ of the 53GB22, 66GB22 and72GB22 gradations4. Examine the best means to quantify top surface displacement in the σ ′vt = 0 apparatus configu-ration5.2.1.1 Experimental summaryThe unconsolidated specimen was reconstituted by the modified discrete slurry deposition method toan initial void ratio of 0.65. The corresponding theoretical critical hydraulic gradient for the glassbead material is icr =γ ′γw = 0.91 (where γ′ is the buoyant unit weight of the specimen, and γw the unitweight of water).A very light-weight mesh platform was constructed to sit upon the specimen top surface such thatconnection to the LVDT was possible in the σ ′vt = 0 apparatus configuration. In addition, a transparentreference grid was installed around the outside perimeter of the permeameter cell in the region of thespecimen top surface. Video and photographic records were used to track changes in specimen lengthat the top surface throughout the test.140A test summary is presented in Figure C.1, along with the variation in discharge velocity withaverage gradient during the test. The specimen exhibited no visual signs of distress and no changein length (by grid-measurement or LVDT record) for all stages up to, and including, iav = 0.85. Verysmall axial expansion (< 1mm) was observed at a gradient of iav = 0.90. The iav−vd relation is linear(Figure C.1), indicating an average hydraulic conductivity of kav = 0.019 cms .Imposition of iav,target = 0.95 resulted in fluidization of the specimen (the mechanism identifiedvisually and with evidence of ≈ 15 mm average axial expansion, corresponding to an increased voidratio of 0.73). Axial expansion was measured at the top surface using grid-based visual recording (thecorresponding LVDT measurement was 7mm, discussed below). The flow rate through the specimenincreased as a result of fluidization. In response, the pump rate to the inlet supply constant headdevice was increased to maintain constant differential elevation head. The corresponding averagehydraulic gradient across the post-fluidized specimen was 0.78, with implied hydraulic conductivityof kav = 0.030 cms (represented in the final data point of Figure C.1).5.2.1.2 Summary remarksThe pre-critical average hydraulic gradient observed experimentally, iav = 0.90, agrees very wellwith the theoretical critical hydraulic gradient for the specimen icr = 0.91. The homogeneity of thespecimen was well-represented by pore-water pressure data in terms of a linear water head distribution(Figure C.2). The deduced hydraulic conductivity of the P-0070 material kav = 0.019 cms provides anestimate of the lower-most value expected for the glass bead mixtures used in the main experimentalprogram.Grid-based visual recording of the specimen top surface at the cell perimeter was deemed the bestmethod of quantifying top-surface displacement in the σ ′vt = 0 configuration. The visual grid-systemis preferable due to: (1) poor accuracy in the measurement of average top surface displacement in theLVDT arrangement due to potential skewing of the top surface contact due to small variations in topsurface height, and (2) obscuration of the specimen top surface by the mesh platform in the LVDTarrangement. One of the main benefits of the σ ′vt = 0 configuration is the opportunity for visual ob-servation of the top surface of the specimen. The visual grid-based system (detailed in Section 3.3.2)provides a better quantification of average specimen length and permits visual observations of parti-cles exiting the top surface of test specimens. For these reasons, the grid-based system is employedin all tests undertaken in the σ ′vt = 0 configuration in the the main testing program.5.2.2 C-72GB22 series (no rectifying filter)Potentially unstable specimens with a pure clast-supported microstructure had not been tested pre-viously in the large permeameter device. An initial commissioning program was undertaken on the72GB22 gradation, with the apparatus configuration based on that of Moffat (2005) and Li (2008).The intent of the commissioning program was to:• Identify appropriate testing parameters to permit adequate resolution of gradient and stress141measurements;• Ensure that the resolution of the instrumentation was adequate for very low-gradient seepageapplications;• Identify potential improvements in the configuration of the apparatus for testing of very unsta-ble materials; and,• Verify the suitability of the gap-gradation containing 22% finer fraction content for ‘lowerbound’ testing: specifically, verify that the gradation will exhibit instability by suffusionA summary is provided of three tests on the 72GB22 gradation (with no rectifying filter) in thefollowing sections. Seepage was applied via the gradient control configuration (Figure 3.5). Seepagestages were of 20−30 minutes duration in all tests.5.2.2.1 Experimental summaryThree tests were undertaken on the C-72GB22 gradation, at vertical effective stresses of σ ′vt = 0 kPa,50 kPa, and 100 kPa. The three tests are summarized in annotated discharge velocity:gradient spacein Figure C.3, Figure C.5 and Figure C.7.C-72GB22-0 Particle loss associated with internal instability was observed at iav ≈ 0.3. The initialstages of the test were undertaken at a gradient increment of ∆iav = 0.1, thus the pre-criticalseepage stage was that of iav ≈ 0.2. Local gradients in the pre-critical seepage condition wereilocal = 0.14 to 0.23.C-72GB22-50 The initial stages of the test were undertaken at a gradient increment of ∆iav = 0.05.Particle loss associated with internal instability was observed at iav ≈ 0.16 to 0.20. Localgradients in the pre-critical seepage condition ranged from ilocal = 0.06 to 0.34.C-72GB22-100 The initial stages of the test were undertaken at a gradient increment of ∆iav = 0.15.Particle loss associated with internal instability was first observed at iav ≈ 0.33 (thus the pre-critical seepage stage was that o