Multi-scale modeling of cyclic

shearing and liquefaction

response of granular materials
by

Andres Ricardo Barrero

B.Sc., Universidad de Los Andes, 2011

M.A.Sc., Universidad de Los Andes, 2012

A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF

THE REQUIREMENTS FOR THE DEGREE OF

DOCTOR OF PHILOSOPHY

in

The Faculty of Graduate and Postdoctoral Studies

(Civil Engineering)

THE UNIVERSITY OF BRITISH COLUMBIA

(Vancouver)
December 2019

c Andres Ricardo Barrero, 2019
The following individuals certify that they have read, and recommend to the Faculty of
Graduate and Postdoctoral Studies for acceptance, a dissertation entitled:
Multi-scale modeling of cyclic shearing and liquefaction response of granular materials

submitted by Andres Ricardo Barrero in partial fulfillment of the requirements for

the degree of Doctor of Philosophy

in Civil Engineering

Examining Committee:
Dr. Mahdi Taiebat, Civil Engineering at University of British Columbia
Supervisor
Dr. José E. Andrade, Civil Engineering at California Institute of Technology
External Examiner
Dr. Dharma Wijewickreme, Civil Engineering at University of British Columbia
University Examiner
Dr. Scott Dunbar, Mining Engineering at University of British Columbia
University Examiner
Dr. Arcesio Lizcano, Industry Expert (SRK Consulting)
Supervisory committee member

Additional Supervisory Committee Members:

Dr. John Howie, Civil Engineering at University of British Columbia
Supervisory committee member
Dr. Davide Elmo, Mining Engineering at University of British Columbia
Supervisory committee member

ii
Abstract

Cyclic shearing of granular materials under undrained conditions can induce a reduction of
mean effective stress and increase of the pore water pressure. In extreme cases, the mean
effective stress can temporarily vanish and lead to a “semifluidized state”, in which large
shear strains are developed and accumulated. Predicting the level of deformations developed
during liquefaction and especially in the post-liquefaction stage using constitutive models is a
changeling task, and yet important to evaluate the safety of geotechnical structures. A sand
plasticity model, which is the precursor of the SANISAND family of models, was considered
as the reference model in this study. The model has proven success in the simulation of
monotonic and cyclic response of sand in the pre-liquefaction state. A series of modifications
were introduced out to improve the predictability of the model for the post-liquefaction
cyclic shear strain. The modifications were motivated by carrying out a number of constant-
volume cyclic shear triaxial simulations using the discrete element method (DEM). The DEM
simulation results revealed that a high number of floating particles with zero contact in a
semifluidized state, which explained the vanishing of load-bearing structures and large shear
strain accumulations. Thus, linking discrete and continuum modelings via the semifluidized
state, inspired introducing a new state internal variable named strain liquefaction factor
(SLF) to model the degradation of stiffness. The SLF evolves within the semifluidized range;
its constitutive role is to reduce the values of parameters controlling the plastic modulus and
dilatancy, maintaining the same plastic volumetric strain rate, in the semifluidized range.
The evolution rate equation of the SLF includes a back-to-zero recovery term under drained
loading. The extended model was validated against a series of undrained cyclic simple
shear tests at the element level. Then this model was implemented in a finite difference
platform and used in the benchmark study LEAP for simulating centrifuge experiments of
a submerged slope subjected to dynamic excitations. Comparisons between experiments

iii
and simulations were satisfactory, and especially the simulated horizontal displacement was
improved using the SLF. This work is expected to extensively benefit the numerical modeling
of liquefaction-related problems in the future.

iv
Lay Summary

In saturated granular deposits, earthquake-induced cyclic shearing can lead to the

suspension of grains in the water medium, technically referred to as “liquefaction”. For
successful modeling of response of soil deposits in such events, it is essential to understand
and accurately simulate the response of soil using a constitutive model for the relation of
the stresses and strains. A fundamental link between discrete and continuum modeling is
provided through a semifluidized state of the granular material. This concept inspired
introducing a new constitutive ingredient named strain liquefaction factor for reducing the
degradation of stiffness while only inside the semifluidized state. The developed modeling
approach improves the predictability of the post-liquefaction response of geotechnical
structures and allows a more reliable evaluation of their performance when subjected to
seismic excitations.

v
Preface

This study was undertaken under the direction and supervision of my research supervisor
Dr. Mahdi Taiebat, and my supervisory committee members Dr. Arcesio Lizcano, Dr. John
Howie, and Dr. Davide Elmo. Dr. Mahdi Taiebat is a Professor in the Civil Engineering
department at The University of British Columbia. Dr. Arcesio Lizcano is a principal
geotechnical engineer at SRK consulting. Dr. John Howie and Dr. Davide Elmo are
Associate Professors in the Civil Engineering department and in the School of Mining
Engineering, respectively at The University of British Columbia.

Three original stand-alone research chapters (2-4) are included in this thesis and are
intended for peer-reviewed journal publications. I, Andres Ricardo Barrero, am the
primary contributor to all six chapters of this work. In all of the research chapters, my
contributions included: 1) Identification of the research question, 2) planning the studies,
3) conducting all the research activities (formulation, implementation, verification and
validation), and 4) preparing the initial and subsequent drafts of all chapters.
My Ph.D. supervisor, Dr. Mahdi Taiebat, advised me in developing the research direction,
carrying out and verifying the analyses, interpreting the results, and overcoming a number
of research pitfalls. He also developed the main idea of Chapter 3 in our close collaboration
with Dr. Yannis F. Dafalias. The industry expert Dr. Arcesio Lizcano worked closely with
me in verifying the results and revising the manuscript presented in chapter 2. He also
provided me with valuable comments during our regular group meetings. Dr. John Howie
and Dr. Davide Elmo have approved my doctoral research proposal and provided me with
valuable comments throughout the development of the work.
The main findings of this thesis have been submitted to a peer-reviewed journal or are
under preparation for submission to peer-reviewed journals. Equally important several

vi
conference papers have been published with relevant content to the research herein.

Chapter 2: Insight from discrete element modeling of cyclic shearing of granular assemblies

The numerical results presented in this chapter were obtained using a revised version
of the numerical code of Dr. William Fernando Oquendo. He helped in adapting
the numerical tool for simulating cyclic shearing under constant volume conditions
on granular assemblies. He also provided me with valuable comments for revision of
several drafts. The findings of this chapter have lead to:

• Barrero, A. R., Oquendo. W. F., Taiebat, M., and Lizcano A. (2018) “Cyclic
shearing response of granular material in the semi-fluidized regime” in Proceedings
of fifth geotechnical earthquake engineering and soil dynamics conference. Austin,
TX, US.

• Barrero, A. R., Oquendo. W. F., Taiebat, M., and Lizcano A. (In preparation)
“Insight from discrete element modeling of cyclic shearing of granular
assemblies” to be submitted to a peer-reviewed journal.

Chapter 3: Stress-strain constitutive modeling of cyclic shearing response in the

semi-fluidized state

Chapter 3 is based on a manuscript submitted for publication in a peer-review journal:

• Barrero, A. R., Taiebat, M., and Dafalias, Y. F. (2020) “Modeling cyclic

shearing of sand in the semifluidized state” in International Journal for
Numerical and Analytical Methods in Geomechanics.

Chapter 4: Application of the constitutive model in continuum modeling of a saturated slope

subjected to seismic loading

This chapter presents selected results of my contribution to the liquefaction

experimental and analysis project (LEAP) in 2017-2019. My peer graduate students,

vii
Ming Yang and Andres Reyes, were also involved in the development of the numerical
framework for the analyses of the centrifuge experiments and preparing the findings
in form of research documents. More extensive outcomes of my contribution in this
project have been published in:

• Yang M., Barrero A. R., and Taiebat, M. (2020) “Application of a SANISAND

model for numerical simulations of the LEAP 2017 experiments.” In Kutter, B.,
Manzari, M., and Zeghal, M., editors, in Model tests and numerical simulations of
liquefaction and lateral spreading: LEAP-UCD-2017, pages 563-578. Davis, CA,
US.

• Kutter, B.L., Manzari, M.T., Zeghal, M., Arduino, P., Barrero, A. R., Carey,
T.J., Chen, L., Elgamal, A., Ghofrani, A., Montgomery, J., Ozutsumi, O., Qiu, Z.,
Taiebat M., Tobita, T., Travasarou, T., Tsiaousi, D., Ueda, K., Ugalde, J., Wada,
T., Yang, M., Zheng, B. L. and Ziotopoulou, K. (2020) “Numerical sensitivity
study compared to trend of experiments for LEAP-UCD-2017” In Kutter, B.,
Manzari, M., and Zeghal, M., editors, in Model tests and numerical simulations
of liquefaction and lateral spreading: LEAP-UCD-2017, pages 209-225. Davis,
CA, US.

• Manzari, M.T., Ghoraiby, M.E., Zeghal, M., Kutter, B.L., Arduino, P., Barrero,
A. R., Bilotta, E., Chen, L., Chen, R., Chiaradonna, A., Elgamal, A., Fasano,
G., Fukutake, K., Fuentes, W., Ghofrani, A., Ichii, K., Kim, D.S., Kiriyama, T.,
Lascarro, C., Mercado, V., Montgomery, J., Ozutsumi, O., Qiu, Z., Taiebat, M.,
Travasarou, T., Tsiaousi, D., Ueda, K., Ugalde, J., Wada, T., Wang, R., Yang,
M., Zhang, J.-M., and Ziotopoulou, K. (2020) “LEAP-2017 Simulation Exercise:
Calibration of Constitutive Models and Simulation of the Element Tests” In
Kutter, B., Manzari, M., and Zeghal, M., editors, in Model tests and numerical
simulations of liquefaction and lateral spreading: LEAP-UCD-2017, pages
153-173. Davis, CA, US.

• Manzari, M. T., Ghoraiby, M. E., Zeghal, M., Kutter, B. L., Arduino, P., Barrero,
A. R., Bilotta, E., Chen, L., Chen, R., Chiaradonna, A., Elgamal, A., Fasano,

viii
 G., Fukutake, K., Fuentes, W., Ghofrani, A., Haigh, S., Hung, W.-Y., Ichii, K.,
Kim, D. S., Kiriyama, T., Lascarro, C., Madabhushi, M. S. P. G., Mercado, V.,
Montgomery, J., Okamura, M., Ozutsumi, O., Qiu, Z., Taiebat, M., Tobita, T.,
Travasarou, T., Tsiaousi, D., Ueda, K., Ugalde, J., Wada, T., Wang, R., Yang, M.,
Zhang, J.-M., Zhou, Y.-G., and Ziotopoulou, K. (2020) “LEAP-2017: Comparison
of the Type-B Numerical Simulations with Centrifuge Test Results” In Kutter, B.,
Manzari, M., and Zeghal, M., editors, in Model tests and numerical simulations of
liquefaction and lateral spreading: LEAP-UCD-2017, pages 175-207. Davis, CA,
US.

• Reyes, A., Barrero A. R., and Taiebat, M. (Submitted in August 2019)

“Application of a SANISAND model for numerical simulations of the
LEAP-ASIA 2019 experiments” In Kutter, B., Manzari, M., and Zeghal, M.,
editors, in Proceedings of LEAP-ASIA 2019, pages 1-30. Japan.

• Reyes, A., Yang, M., Barrero A. R., and Taiebat, M. (Submitted in December
2019) “Numerical modeling of soil liquefaction using the SANISAND-Sf model
in the leap experiments” Soil Dynamics and Earthquake Engineering.

Appendix A: Constitutive model finite difference implementation

The constitutive model implementation in the finite difference platforms F LAC and
F LAC 3D has been presented, verified and evaluated in this chapter. The
implemented model has been used in peer-reviewed journal publication and several
conference papers:

• Barrero, A. R., Taiebat, M., and Lizcano A. (2015) “Application of an advanced

constitutive model in nonlinear dynamic analysis of tailings dam” in Proceedings
of the Sixty Eighth Canadian Geotechnical Conference. Quebec, BC, Canada.

• Ramirez, J., Badanagki, M., Abkenar, M. R., ElGhoraiby, M., Manzari, M. T.,
Dashti, S., Barrero, A. R., Taiebat, M., Ziotopoulou, K., and Liel, A. (2017)
“Seismic Performance of a Layered Liquefiable Site: Validation of Numerical

ix
 Simulations Using Centrifuge Modeling” in Geotechnical Frontiers 2017.
Orlando, FL, US.

• Ramirez, J., Barrero A. R., Chen, L., Dashti, S., Ghofrani, A., Taiebat, M., and
Arduino, P. (2018) “Site Response in a Layered Liquefiable Deposit: Evaluation
of Different Numerical Tools and Methodologies with Centrifuge Experimental
Results” in Journal of Geotechnical and Geoenvironmental Engineering.

• Ramirez, J., Barrero A. R., Chen, L., Dashti, S., Ghofrani, A., Taiebat, M., and
Arduino, P. (2018) “Capabilities and Limitations of Different Numerical Tools
in Capturing Seismic Site Performance in a Layered Liquefiable Site” in Fifth
geotechnical earthquake engineering and soil dynamics conference. Austin, TX,
US.

• Adinata, J., Reyes, A., Barrero A. R., and Taiebat, M., (2018) “An Investigation
on the Effect of Bidirectional Seismic Loading on Volumetric Strain and Surface
Settlement of Sand Deposits” in Fifth geotechnical earthquake engineering and
soil dynamics conference. Austin, TX, US.

x
Table of Contents

Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii

Lay Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi

Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi

List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv

List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvi

List of Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxii

Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxviii

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Motivation of research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Objectives and methodology . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2 Insight from discrete element modeling of cyclic shearing of granular

assemblies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Observations from cyclic undrained triaxial laboratory tests on sands . . . . 12
2.3 An overview of the adopted DEM framework . . . . . . . . . . . . . . . . . 14
2.3.1 Basic assumptions and contact definition . . . . . . . . . . . . . . . 15

xi
 2.3.2 Normal contact law . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3.3 Sliding, rolling, and torsion contact laws . . . . . . . . . . . . . . . . 18
2.3.4 Summary of the contact parameters . . . . . . . . . . . . . . . . . . 20
2.3.5 Description of control variables . . . . . . . . . . . . . . . . . . . . . 21
2.4 Description of triaxial DEM test . . . . . . . . . . . . . . . . . . . . . . . . 24
2.4.1 Sample generation and compression stage . . . . . . . . . . . . . . . 25
2.4.2 Constant volume cyclic shearing stage . . . . . . . . . . . . . . . . . 26
2.5 Analysis of results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.5.1 Macromechanical response: ru , stress-path and stress-strain . . . . . 28
2.5.2 Micromechanical response: anisotropy and coordination evolution . . 31
2.5.3 Transitional state: semifluidized state . . . . . . . . . . . . . . . . . 35
2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

3 Stress-strain constitutive modeling of cyclic shearing response in the

semifluidized state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.2 Synthesis of observations from cyclic undrained simple shear tests . . . . . . 46
3.3 Constitutive model formulation within the semifluidized range . . . . . . . . 49
3.3.1 Reference constitutive model . . . . . . . . . . . . . . . . . . . . . . 49
3.3.2 Proposed modification: the SANISAND-Sf model . . . . . . . . . . . 50
3.4 Performance of the proposed modified model . . . . . . . . . . . . . . . . . 55
3.4.1 Effects of the new model parameters . . . . . . . . . . . . . . . . . . 58
3.4.2 Simulation of undrained cyclic response on Ottawa F-65 and Toyoura
sand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

4 Application of the constitutive model in dynamic continuum modeling of

slope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.2 Liquefaction Experiments and Analysis project: LEAP . . . . . . . . . . . . 73
4.2.1 Model geometry and material characterization . . . . . . . . . . . . 74

xii
 4.2.2 Model instrumentation and recorded data . . . . . . . . . . . . . . . 75
4.3 Description of the numerical model . . . . . . . . . . . . . . . . . . . . . . . 77
4.3.1 General numerical platform . . . . . . . . . . . . . . . . . . . . . . . 77
4.3.2 Model geometry, initial and boundary conditions, and input motion . 78
4.3.3 Sequence of analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.4 Comparison between measured and simulated results . . . . . . . . . . . . . 82
4.5 Exploring the strain liquefaction factor . . . . . . . . . . . . . . . . . . . . . 88
4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

5 Summary, Conclusions and Future work . . . . . . . . . . . . . . . . . . . . 96

5.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.2 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
5.3 Recommendations for future research . . . . . . . . . . . . . . . . . . . . . . 100

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

A Constitutive model implementation . . . . . . . . . . . . . . . . . . . . . . 116

A.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
A.2 Model overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
A.2.1 Elastic and plastic strains . . . . . . . . . . . . . . . . . . . . . . . . 117
A.2.2 Critical state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
A.2.3 Yield surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
A.2.4 Critical, dilatancy and bounding surfaces . . . . . . . . . . . . . . . 118
A.2.5 Plastic flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
A.2.6 Evolution laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
A.2.7 Treatment for overshooting response . . . . . . . . . . . . . . . . . . 121
A.3 Numerical implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
A.3.1 Pressure dependent sub-stepping . . . . . . . . . . . . . . . . . . . . 122
A.3.2 Stress integration scheme . . . . . . . . . . . . . . . . . . . . . . . . 124
A.3.3 Numerical treatment for low effective stress state . . . . . . . . . . . 127
A.3.4 Numerical treatment for mixed-discretazation . . . . . . . . . . . . . 127

xiii
A.4 Verification of the model implementation . . . . . . . . . . . . . . . . . . . 130
A.4.1 Verification of reference model implementation . . . . . . . . . . . . 130
A.4.2 Verification of the treatment for overshooting response . . . . . . . . 132
A.4.3 Verification of the treatment for pressure dependent sub-stepping . . 134
A.4.4 Accuracy of the model implementation . . . . . . . . . . . . . . . . . 134
A.4.5 Verification of the treatment for low effective stress state . . . . . . . 135
A.5 Summary of the constitutive model implementation . . . . . . . . . . . . . . 137

xiv
List of Tables

2.1 Initial conditions (initial void ratio e0 , relative density Dr and initial confining
pressure p0 ) and limit of qamp for the triaxial tests . . . . . . . . . . . . . . . 12
2.2 Suggested correlations used for the contact parameters in the DEM
simulations (Oquendo, 2013; Luding, 2008) . . . . . . . . . . . . . . . . . . . 21
2.3 Alternative definitions for coordination number z according to the different
measurements of effective number of particles, N eff . . . . . . . . . . . . . . 23
2.4 DEM model simulations and contact parameters used for the triaxial simulation 26
2.5 Initial conditions (initial void ratio e0 , initial confining pressure p0 , initial
coordination number z0 , and initial contact anisotropy ac0 ) for the triaxial tests 28

3.1 Calibrated parameters for reference model . . . . . . . . . . . . . . . . . . . 56

3.2 Calibrated additional parameters of the new model . . . . . . . . . . . . . . 56
3.3 Default additional parameters of the new model . . . . . . . . . . . . . . . . 57

4.1 Specific gravity (Kutter et al., 2018) and grain size parameters for Ottawa
F-65 sand obtained from (Carey et al., 2019) . . . . . . . . . . . . . . . . . . 75

A.1 Analysis of computational cost for different m . . . . . . . . . . . . . . . . . 137

xv
List of Figures

2.1 Experimental results of an undrained cyclic triaxial test on Karlsruhe fine

sand at Dr = 78, 61, and 25%: (a)(d)(g) excess pore pressure ratio vs.
number of cycles (b)(e)(h) stress-path and (c)(f)(i) stress-strain loops. Data
from Wichtmann and Triantafyllidis (2016a) . . . . . . . . . . . . . . . . . . 13
2.2 Schematic representation of the overlapping distance, hij , between two
spherical particles i and j . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Schematic illustration for the force-displacement response of the normal
contact model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4 Schematic illustration for the force-displacement response for tangential
contact model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.5 Illustration of the DEM sample at the end of an isotropic triaxial compression
using six rigid walls. In this case Lx , Ly , Lz are the initial walls lengths prior
to the cyclic shearing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.6 Variation of void ratio and contact anisotropy at the end of the compression
stage with sliding friction coefficient . . . . . . . . . . . . . . . . . . . . . . . 27
2.7 Results of DEM cyclic triaxial tests under constant volume for different initial
configurations: (a),(d),(g),(j) pore-pressure ratio history, (b),(e),(h),(k) stress
path and (c),(f),(i),(l) stress-strain curve . . . . . . . . . . . . . . . . . . . . 30
2.8 Illustration of a blue spherical particle with different average number of
contacts (z): (a)-(d) scenarios associated with unstable states and (e)
associated with stable state . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

xvi
2.9 Domain of the semifluidized state (0< p ≤ pth ) via coordination number:
(a),(d),(g),(j) coordination number evolution excluding particles with 0,1 and
2 contacts, (b),(e),(h),(k) coordination number evolution with mean stress,
and (c),(f),(i),(l) coordination number evolution with emphasis in low mean
stress values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.10 Domain of the semifluidized state (0< p ≤ pth ) via floaters ratio: (a),(d),(g),(j)
floaters ratio excluding particles with 0 and 1 contacts, (b),(e),(h),(k) floaters
ratio evolution with mean stress, and (c),(f),(i),(l) floaters ratio evolution with
emphasis in low mean stress values . . . . . . . . . . . . . . . . . . . . . . . 34
2.11 Triax-04 test mean stress evolution and shear strength correlation with
anisotropy sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.12 Results of DEM constant cyclic triaxial test for different initial densities:
(a),(d),(g),(j)contact anisotropy, (b),(e),(h),(k) normal anisotropy and
(c),(f),(i),(l) tangential anisotropy . . . . . . . . . . . . . . . . . . . . . . . . 37
2.13 Post-liquefaction response for Triax-01 test: (a) stress-path and (b) stress-
strain. Red markers 1 and 4 are non-liquefied states. Red markers 2 and 5
correspond to the peak deviatoric stresses. Red markers 3 and 6 are liquefied
states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.14 Tangent shear modulus during the post-liquefaction response for Triax-01 test 38
2.15 Post-liquefaction response for Triax-01 test: (a) coordination number
excluding particles with 0, 1 and 2 contacts and (b) floating ratio
considering particles with 0 and 1 . . . . . . . . . . . . . . . . . . . . . . . . 39
2.16 Post-liquefaction anisotropy evolution for test Triax-01: (a) contact
anisotropy, (b) normal anisotropy and (c) tangential anisotropy . . . . . . . 40

3.1 Experimental results of an undrained cyclic simple shear test on Ottawa F-65
sand at Dr = 77%: (a) stress-path and (b) stress-strain loops. The red dashed
and red continuous lines correspond to the 3rd and 23rd cycle, respectively,
encompassing all cycles in the post-liquefaction stage. (Data from Bastidas
(2016)) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

xvii
3.2 Details of shear strain accumulation in the semifluidized state: (a,b)
stress-path and the corresponding stress-strain loops in the 23rd loading
cycle Fig. 3.1, illustrating the shear strains generated within the
semifluidized state, γsf , while increasing and decreasing the shear stress
between points A and B, and points C and D, respectively; and (c) absolute
values of the γsf in the half cycles of all loading cycles . . . . . . . . . . . . 48
3.3 Variations of the h0 /h00 and A0 /A00 ratios with pr at different levels of l,
illustrating the upper and lower bounds of these ratios in the semifluidized
state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.4 Illustration of the effect of x on the maximum amplitude of post-liquefaction
shear strain, in simulation of undrained cyclic simple shear loading: (a-c)
stress-strain loops, (d-f) evolution of γsf with the strain liquefaction factor up
to l = 0.6, and (g-i) evolution of the γsf with the number of loading cycles . . 60
3.5 Illustration of the effect of cl on the pace of evolution of post-liquefaction shear
strain amplitude toward its saturation level, in simulation of undrained cyclic
simple shear loading: (a-c) stress-strain loops, (d-f) evolution of γsf with the
strain liquefaction factor up to l = 0.6, and (g-i) evolution of the γsf with the
number of loading cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.6 Illustration of the effect of cr on the post-liquefaction response after a drained
re-consolidation stage, where the first undrained cyclic shearing is shown in
grey and the second undrained cyclic shearing is shown in red: (a-c) stress-
strain loops, (d-f) evolution of γsf with the strain liquefaction factor up to
l = 0.6, and (g-i) evolution of the γsf with the number of loading cycles . . . 62
3.7 Summary of the effects of x, cl , and cr on different aspects of the
post-liquefaction response from Figure3.4, 3.5 and 3.6: (a) effect of x on the
attained values of |γsf | at different levels of l, (b) effect of cl on the number
of cycles to reach different levels of l, and (c) effect of cr on the number of
cycles to reach different levels of |γsf | in the second liquefaction round . . . . 63

xviii
3.8 Illustration of the effect of a on the number of cycles to reach 3% of single
amplitude shear strain (NγSA =3% ) in simulation of undrained cyclic simple
shear loading with different level of cyclic stress ratio CSR . . . . . . . . . . 64
3.9 Stress-path and stress-strain loops of an undrained cyclic simple shear test
on Ottawa F-65 sand at Dr = 77%: (a-b) Experimental results
from Bastidas (2016), (c-d) reference model simulation results, (e-f)
extended model accounting for semifluidized strains . . . . . . . . . . . . . . 65
3.10 Stress-path for different CSR levels on undrained cyclic simple shear test on
Ottawa F-65 sand at Dr = 77%: Experimental and extended model results,
data from Bastidas (2016) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.11 Stress-strain loops for different CSR levels on undrained cyclic simple shear
test on Ottawa F-65 sand at Dr = 77%: Experimental and extended model
results, data from Bastidas (2016) . . . . . . . . . . . . . . . . . . . . . . . . 67
3.12 Comparing the experimental results and model simulations in terms of cyclic
stress ratio versus number of loading cycles to reach 3% single-amplitude of
shear strain in undrained simple shear tests on Ottawa sand F-65 sand at Dr
=77% . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.13 Stress-path and stress-strain loops of an undrained cyclic torsional shear test
on Toyoura sand at Dr = 70%: (a) Experimental results from Zhang (1997),
and (b) extended model accounting for semifluidized strains . . . . . . . . . 68

4.1 Schematic geometry for LEAP-UCD-2017 experiment for shaking in the

circumferential direction of the centrifuge. Taken from Kutter et al. (2019a) 76
4.2 Acceleration time history and response spectrum (5% damped) of the input
harmonic motion at the base of the numerical model . . . . . . . . . . . . . 76
4.3 2-dimensional view of the grid with the boundary conditions, and the
recordings locations for the pore pressure (P), horizontal acceleration (AH),
and horizontal displacement (D) . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.4 Contour of effective horizontal stress before the dynamic shaking . . . . . . . 80
4.5 Contour of effective vertical stress before the dynamic shaking . . . . . . . . 81

xix
4.6 Experimentally measured and numerically computed acceleration time
histories from accelerometers located at different depths of the centre midline 84
4.7 Experimentally measured and numerically computed response spectra (5%
damped) time histories from accelerometers located at different depths of the
centre midline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.8 Experimentally measured and numerically computed excess pore water
pressure (EPWP) time histories from pressuremeters located at different
depths of the centre midline . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.9 Reference model horizontal displacements and displacement vectors at the end
of the dynamic shaking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.10 Extended model horizontal displacements and displacement vectors at the end
of the dynamic shaking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.11 Experimentally measured and numerically computed horizontal displacements
at centre midline displacement marker . . . . . . . . . . . . . . . . . . . . . 89
4.12 Numerically computed mean effective stress p at different depths from the
centre midline: (a)(d) bottom, (b)(e)mid-height and (c)(f) surface. The grey
shaded region denotes areas in which the semifluidized state is experienced.
The black and red dashed lines are the values for the pth and pinr , respectively 90
4.13 Numerically computed strain liquefaction factor, l, at different depths from
the centre midline. The grey shaded region denotes areas in which the
semifluidized state is active . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.14 Numerically computed shear stress τ at different depths from the centre
midline. The grey shaded region denotes areas in which the semifluidized
state is active . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.15 Numerically computed shear strain γ at different depths from the centre
midline. The grey shaded region denotes areas in which the semifluidized
state is active . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

xx
A.1 Comparisons between simulations of monotonic drained triaxial compression
tests with initial confinements of 100 kPa (black color), 200 kPa (red color)
and 300 kPa (blue color) and initial void ratios of 0.555, 0.64 and 0.698, using
different numerical platforms . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
A.2 Comparisons between simulations monotonic constant volume triaxial
compression with initial confinements of 50 kPa (green color), 100 kPa
(black color) and 200 kPa (red color) and initial void ratios of 0.555, 0.64
and 0.698, using different numerical platforms . . . . . . . . . . . . . . . . . 131
A.3 Comparisons of results from different numerical platforms simulating a cyclic
constant volume triaxial test . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
A.4 Overshooting performance for a drained triaxial compression test on
isotropically consolidated samples subjected to (a) small unloading, (b)
medium level of unloading, and (c) moderate level of unloading . . . . . . . 133
A.5 Sensitivity analyses using different s on a drained triaxial compression test . 134
A.6 ISO-error maps for the initial stress state at p0 = 100 kPa and e0 = 0.555
showing the reduction of the realative error with the sub-stepping tolerance s 136
A.7 Sensitivity analyses on m simulating a undrained cyclic triaxial test . . . . . 136

xxi
List of Symbols

⊗ Dyadic product operator

× Cross product operator
· Single tensorial contraction or vector contraction
: Double contraction
|| Norm of a tensor or vector
tr( ) Trace of a tensor
hi Macauley brackets
σ Stress tensor
σv Vertical stress
σh,0 Initial horizontal stress
σxx xx-component of the stress tensor
σyy yy-component of the stress tensor
σzz zz-component of the stress tensor
σw Wall confinement stress
δij Kronecker symbol
ε Strain tensor
εa Axial strain
εxx xx-component of the strain tensor
εyy yy-component of the strain tensor
εzz zz-component of the strain tensor
εv Volumetric strain
ε̇v Volumetric strain rate
ε̇pv Plastic volumetric strain rate

xxii
εq Triaxial deviatoric strain
ρs Density of solids
ε̇a Axial strain rate
τ Shear stress
~τi Torque vector for particle i
~τsliding Sliding torque vector
~τrolling Rolling torque vector
~τtorsion Torsion torque vector
γ Shear strain
γ0 Rolling torsion coefficient
γn Normal damping coefficient
γr Rolling damping coefficient
γsf Semifluidized strain
γt Tangential damping coefficient
γS.A Single amplitude shear strain
ξ Critical void ratio power parameter
ξ~t Tangential spring vector
ξ~ Deformation spring vector
∆ε Increment of strain
∆q Increment of deviatoric stress
∆t Integration time step
∆lx difference between current and initial length for the x direction
∆ly difference between current and initial length for the y direction
∆lz difference between current and initial length for the z direction
ω
~i Angular velocity vector for particle i
µs Particle to particle coefficient of sliding friction
µr Particle to particle coefficient of rolling friction
µ0 Particle to particle coefficient of torsional friction
χ Tangential force anisotropy tensor
χn Normal force anisotropy tensor

xxiii
χ0 Initial semifluidized second parameter
κ Contact stiffness parameter
ρ Particle density
ψ State parameter
α Back stress-ratio tensor
αin Initial back stress-ratio of a new loading process
αbθ Image back stress ratio tensor on the bounding surface
αdθ Image back stress ratio tensor on the dilatancy surface
ν Poisson’s ratio
λc Slope of critical void ratio line
a Cyclic ratio semifluidized parameter
ac Magnitude of the contact anisotropy
a0i Corrected radius for particle i
a0ij Effective corrected radius for particle i and j
aij Reduced radius without correction for particle i and j
an Magnitude of the normal force anisotropy
at Magnitude of the tangential force anisotropy
A0 Dilatancy parameter
A00 Semifluidized dilatancy parameter
c Compression and extension critical ratio
ch Plastic modulus second parameter
cf Rate semifluidized parameter
cl Strain liquefaction factor rate parameter
cr Strain liquefaction healing rate parameter
cz Fabric-dilatancy rate parameter
Cu Uniformity coefficient
d¯ Mean particle diameter
D Dilatancy
D50 Mean grain size passing at 50%
Dr Relative density

xxiv
e Deviatoric strain tensor
e Void ratio
en Normal coefficient of restitution
emin Minimum void ratio
emax Maximum void ratio
eref Critical void ratio reference
ēpeq First overshooting correction parameter
F Fabric tensor
FC Fine content
F~nda Normal viscous force vector
F~tda Tangential viscous force vector
F~nel Normal elastic force vector
F~tel Tangential elastic force vector
F~i Force vector of particle i
F~n Normal force vector
F~t Tangential force vector
fr Floaters ratio
fl Semifluidized limit parameter
G Shear modulus
G0 Shear reference parameter
Gs Specific gravity
h0 Plastic modulus first parameter
h00 Semifluidized plastic modulus first parameter
hij Overlap between particles i and j
I Second order identity tensor
Ii Moment of Inertia of particle i
I Inertial parameter
k0 Torsion stiffness
kn Normal stiffness
kr Rolling stiffness

xxv
kt Tangential stiffness
Kp Plastic modulus
l Strain liquefaction factor
~li Branch vector joining the center of particle i to the point of contact
Lx Initial wall length in the x direction
Ly Initial wall length in the y direction
Lz Initial wall length in the z direction
hLi Plastic multiplier
m Yield surface size parameter
mi Mass of particle i
mij Reduced mass for particles i and j
M Critical stress ratio
n̄ Second overshooting correction parameter
n Loading direction tensor
nb Bounding surface parameter
nd Dilatancy surface parameter
nl Semifluidized parameter
N Total number of particles
Nc Total number of contacts
NγSA=3% Number of cycles to reach 3% of single amplitude shear strain
n̂ij Normal unit vector from particle i to particle j
p Mean effective stress
pat Atmospheric pressure
pcyc Cyclic ratio
pin Mean stress at the initiation of a new loading process
pinr Mean stress reference semifluidized parameter
pr Pressure ratio
pth Mean efective stress threshold
qamp Deviatoric stress amplitude
q Triaxial deviatoric stress

xxvi
R Particle radius
~ri Position vector of particle i
~rij Position vector from particle i to particle j
ru Excess of pore pressure ratio
s Deviatoric stress tensor
t̂ij Tangential unit vector from particle i to particle j
~vij Relative velocity vector for particles i and j
~vij,0 Torsion relative velocity vector for particles i and j
~vij,n Normal relative velocity vector for particles i and j
~vij,r Rolling relative velocity vector for particles i and j
~vij,s Sliding relative velocity vector for particles i and j
vw,x Wall speed in the x direction
vw,y Wall speed in the y direction
vw,z Wall speed in the z direction
Vs Volume occupied by the solids
x Semifluidized second parameter
z Fabric dilatancy tensor
z Coordination number
zmax Maximum fabric-dilatancy parameter

xxvii
Acknowledgements

I am grateful to my Ph.D. supervisor, Dr. Mahdi Taiebat, for the support, direction,
guidance, and encouragement that he provided to me throughout my PhD journey. I am
also grateful to Dr. Arcesio Lizcano for our discussions and his encouragements. I would
also like to thank my additional supervisory committee members, Drs. John Howie and
Davide Elmo, for their support and mentorship throughout these years.
I extend my deepest gratitude to the past and present members of the theoretical and
geomechanics applied (TAG) research group, for our many fruitful technical discussions
and for their friendship. I also thank the faculty and graduate students of the UBC
geotechnical engineering group for their valuable comments at the geotechnical seminars,
and also the UBC Civil Engineering department staff for the support over the last years.
I am also grateful for the research partnership with SRK consulting, and in particular to
Dr. Maritz Rykaart. Specially, I want to thank Aaron Fultz for his valuable comments in
revising this manuscript and Dr. Frederic Dubois for several hours of discussion in different
chapters.
I would also like to extend my deepest gratitude to my Persian friends, soccer friends and
many other friends that I have meet along this journey.

Agradezco a todas las personas que estuvieron involucradas de alguna u otra forma
durante el transcurso de la elaboración de este documento. Claramente este manuscrito va
dedicado en especial al apoyo espiritual e incondicional de mis padres, Arcesio y Vicky. Por
fin la logramos. También quiero mencionar a otros miembros de mi familia, a mis hermanas
Angela y Paola, a mis cuñados Rene y Camilo, a mis sobrinos Minichu y Juanse, mi tı́a
Clara y a todos mis tı́os y tı́as, primos y primas por su constante apoyo. También quiero
agradecer a Dios por guiarme en este camino.

xxviii
 También quiero aprovechar para agradecer a todos esas personas que son fundamentales
y me apoyaron para hacer este trabajo un camino agradable y muy alegre.
Lastly I have to say thank you to that special person, Neshat, for being besides me and for
being my fundamental cornerstone throughout the development of this complex work.

Financial support provided by the Natural Sciences and Engineering Research Council of
Canada (NSERC) and Mitacs is gratefully acknowledged.

xxix
1. Introduction

1.1 Motivation of research

Cyclic loading has many practical applications in problems associated with geotechnical
structures. In the field, cyclic loading may be caused by traffic (heavy trucks, trains, and
machinery), wind turbines, waves, earthquakes, and other types of forces that follow
loading and unloading scenarios. In many cases, cyclic loading is applied in different
directions. For example, earthquake loading is usually described with: two perpendicular
horizontal and one vertical component. Additionally, the irregular nature of earthquake
loading usually covers a wide range of frequencies that may induce strong soil-structure
interaction effects. Understanding the complexity of cycling loading is a current challenge
for geotechnical engineers wanting to predict how soils respond when subjected to
aforementioned types of loadings.
Depending on the boundary and drainage conditions of soil elements, the cyclic loading
in soils might lead to a high strain accumulation, a reduction in the effective stress, or
both. The level of strain accumulation or stress reduction ultimately depends on the cyclic
amplitude, while taking into account other parameters from the soil initial conditions such
as density, stress, or fabric. In highly engineered structures like tailings dams, low tolerance
during earthquake loading concerning differential settlements of the dam crest is required.
Thus, geotechnical engineers need a reliable numerical tool that allows obtaining reliable
and accurate results of the boundary value problem, as well as predict the expected level of
the soil deformation. Specifically, predicting the level of deformation in saturated
cohesionless soils subjected to cyclic loading under partly drained or undrained conditions
is highly complex. During this process, the soil effective stress is reduced and subsequently
its stiffness decrease while the excess of pore pressure increases.

1
In extreme scenarios, the pore pressure might rise to the total stress leading to large
deformations as the shear stiffness and soil dilatancy capability are reduced temporarily. In
this point, the soil behaves like a fluid, often called “liquefied”. After reaching liquefaction
and continuing the cyclic loading, the soil develops large strain accumulation during
post-liquefaction response. The consequences of liquefaction were first seen after the event
of the 1964 Niigata earthquake in Japan. Infrastructure near the Shinano river (Japan)
suffered bearing capacity failures and tilted severely. After this event, engineers saw for the
first time the potential damage of cyclic loading on saturated cohesionless soils and the risk
that liquefaction poses on civil structures. Catastrophic consequences of soil liquefaction
led to the need to deepen scientific understanding of liquefaction, and specially to develop
a systematical approach to predict shear deformations accumulated during
post-liquefaction response. As a result, multiple theoretical, numerical, and experimental
advances regarding soil liquefaction have been developed.
Soil liquefaction experimental testing on cohesionless soils has extensively improved our
collective understanding of the mechanism behind the phenomenon (Seed and Lee, 1966;
Ishihara et al., 1975; Castro, 1975; Casagrande, 1976; Alarcon-Guzman et al., 1988; Vaid
et al., 1990; Bardet et al., 1999). To this extent, Lade (1992) made valuable contributions to
understand the stress state in which a soil sample under monotonic compression is not able to
recover its shear strength during a strain controlled undrained triaxial test. After undrained
monotonic tests on samples with different confinement pressures meanwhile sharing the same
initial void ratio, Lade proposed the instability line connecting the maximum deviators stress
points representing the onset of liquefaction. This type of liquefaction usually is termed as
flow liquefaction, commonly occurs to loose, saturated, and cohesionless soil deposits. Flow
liquefaction is when the applied deviator stress triggers the failure at the peak strength and
goes beyond the steady state strength. Flow liquefaction is directly linked to the loss of shear
strength which leads to a rapid increase of strains and pore pressure (Casagrande, 1976).
This type of liquefaction can be also observed in laboratory tests with cyclic loading under
undrained conditions (Yamamuro and Covert, 2001; Andrade et al., 2013). When the soil
sample is sheared with a cyclic stress amplitude inside the steady state strength a second
type of liquefaction, termed as cyclic mobility, is identified.

2
 Cyclic mobility is associated with saturated cohesionless soil during cyclic loading when
the applied cyclic stress amplitude is lower than the steady state strength and is directly
linked to the progressive reduction of the mean effective stress due to the buildup of pore
pressure that may lead to vast strain accumulation if there is a complete loss of the mean
effective stress, also called initial liquefaction Seed and Lee (1966). Beyond the first
instance of liquefaction, the material dilates and is able to recover its shear strength
(Kramer, 1996). In situ, this type of liquefaction has been associated with lateral spreading
(Bardet et al., 1999) and it represents a risk on geotechnical structures due to its potential
of accumulating high levels of strain during the post-liquefaction response. The inclusion of
the post-liquefaction response into constitutive models offers a promising approach for
predicting the maximum magnitude of lateral spread. Developing advanced constitutive
models for simulating post-liquefaction response remains a major challenge among
geotechnical engineers. Indeed, many advanced models have been formulated
mathematically, and few of them have been implemented on numerical platforms.
Therefore, their usage among industry practitioners and academic researchers remains
limited. Constitutive models such as, the pressure-dependent multi yield surface model
developed by Elgamal et al. (2002), the bounding surface plasticity model developed by
Dafalias and Manzari (2004), the UBCSAND model developed by Beaty and Byrne (2011),
the PM4SAND model developed by Boulanger and Ziotopoulou (2013), the unified version
of the post-liquefaction model developed by Wang et al. (2014), have contributed to the
modeling of post-liquefaction response in sands. However, they share a shortcoming
common to similar models. They incorporate accumulative terms to permanently reduce
(i.e., soil stiffness or soil dilation) and there is no constitutive mechanism in their model
formulation to eliminate the permanent reduction when it is imperative. For example, in
simulating multiple liquefaction test requires several changes from cyclic undrained to
monotonic drained loading and from monotonic drained to cyclic undrained loading. Using
accumulative terms to permanently reduce either soil stiffness or soil dilation, or both,
leads to unrealistic response. Hence, the current state of post-liquefaction response during
cyclic mobility modeling using advanced constitutive models for saturated cohesionless soils
requires further research.

3
 Focusing in cyclic mobility, this dissertation aims to expand the knowledge of shear
strain generation during liquefied states from a grain-to-grain point of view and develop
a novel constitutive mechanism to reproduce soil post-liquefaction response, that can be
used in future research and applications related to geotechnical engineering in liquefiable soil
deposits.

1.2 Objectives and methodology

This study includes three main research objectives: (1) using fundamental discrete element
simulations at the particle scale to improve the understanding of soil cyclic mobility and its
accumulation of strains when saturated granular materials are subjected to cyclic loading;
(2) to develop a constitutive mechanism that can be used to predict saturated granular
material post-liquefaction response when subjected to cyclic loading; (3) to evaluate the
model performance and its potential usage in continuum modeling for future applications on
more realistic geotechnical boundary value problems. The aforementioned research objectives
resulted in contributions at the discrete and continuum modeling scales in the form of three
interrelated chapters that comprise the main body of this dissertation.

1. The first objective was achieved by computer driven simulations of element behavior
at the particle scale to reproduce cyclic mobility. Discrete element simulations were
performed using granular packings of spherical particles. The packings were
isotropically consolidated and cyclically sheared in a triaxial configuration under
constant volume conditions. The qualitative results obtained from the numerical
simulations provided deeper insights from a particle-to-particle point of view into
shear strain generation during liquefaction. The results were a vital source of
inspiration for the development of a constitutive model to reproduce stress-strain
response of saturated granular materials (sand) subjected to cyclic loading under
undrained conditions at the element level.

2. Inspired by the discrete element simulation results, the second objective was
accomplished by developing a constitutive mechanism to reduce the plastic stiffness

4
 and soil dilation below a stress threshold. The constitutive mechanism was
incorporated in the formulation of the existing Simple ANIsotropic SAND
(SANISAND) plasticity model to evaluate quantitatively the performance of the
extended model in simulating direct simple shear and torsional tests subjected to
cyclic loading under constant volume conditions. Achieving agreement between the
number of cycles to reach liquefaction in both model and experiment was an
important step in validating the proposed mechanism.

3. Continuing with the model validation, the third objective was fulfilled by simulating
a centrifuge experiment of a submerged slope subjected to cyclic loading. To this
end, the model was first implemented in a finite difference platform using a C++
programming environment. The implementation was performed using a semi-explicit
integration method that was complemented with a pressure dependent sub-stepping
technique. The verification of the model implementation was undertaken by simulating
triaxial tests subjected to monotonic and cyclic loading under drained and undrained
conditions. The stress-path and stress-strain results obtained with both numerical
implementation and available results from the literature were compared to verify the
solution of the implemented model. The validation of the continuum model was carried
out quantitatively by comparing the results obtained in the numerical model and with
those from the actual centrifuge experiment. The comparison was performed with
the data obtained from the sensor measurements of the horizontal acceleration, pore
pressure and horizontal displacements.

Beyond the common current approach found in the literature to include accumulative
plastic volumetric or shear strains (Elgamal et al., 2002; Boulanger and Ziotopoulou, 2013;
Wang et al., 2014) to permanently affect the degradation of stiffness. This dissertation
provides a unique mechanism to reduce the degradation of stiffness only at semifluidized
state, i.e. at low effective confinement stress and to eliminate its influence in subsequent
loadings. The mechanism its inspired after discrete element simulations that provided key
physical description of the semifluidized state. The plausible and logical semifluidized
concept presented here allows future researchers and engineers to model sand cyclic

5
post-liquefaction behavior under undrained soil conditions.

1.3 Outline
The research undertaken in this dissertation is described in five chapters. The term ‘multi-
scale modeling’ is adopted in the title of this work to best describe the simulations performed
at different scales. More specifically, Chapter 2 used a discrete modeling approach to simulate
cyclic triaxial element tests, and Chapters 3 and 4 used a continuum modeling approach to
simulate laboratory element tests and a centrifuge experiment, respectively. As oppose to
the traditional approach to have an independent literature review chapter, it is important
to mention the literature review is performed independently in each chapter.
Chapter 2 described the contributions of previous laboratory element tests on granular
materials cyclic shearing under undrained conditions, focusing primarily on soil cyclic
mobility. The explored laboratory element tests are based on available triaxial cyclic
experiments. The limited laboratory results are presented by showcasing deviatoric and
mean stress, axial strain and pore pressure ratio plots. The results consistently suggested
that any saturated granular deposit is susceptible to cyclic mobility regardless of its initial
conditions. Later, virtual computer simulations are carried out by simulating triaxial cyclic
shearing under constant-volume conditions, with the aim of exploring the origin of
liquefaction and particle-to-particle interactions. Beyond the limited quantities, such as
deviator and mean stress and axial strain, the computer simulations offered a valuable
insight to the behavior of the particle network and fabric. Numerical simulations shown
with average contact of particles and average of particles without contacts (also known as
floaters) suggest liquefaction occurs when particles enter a suspended state at low effective
stress levels. Other fabric measurements are also used to complement the description and
explain the origin of the suspended state. Numerical results suggested a stress threshold at
low stress levels in which the soil behaves as a fluid, called the “semifluidized state”.
Chapter 3 presents the formulation of a constitutive model developed for describing
sand response when subjected to cyclic loading under undrained conditions. Inspired by
the observations from the previous chapter, the development of the model starts by

6
quantifying the accumulation of strains at low stress levels during cyclic simple shear tests
carried out on Ottawa F-65 sand under undrained conditions. The novel contribution of
the model is the newly added internal variable, the Strain Liquefaction Factor, which is
responsible for reducing the plastic stiffness and soil dilation inside the semifluidized state.
This variable represented a key component for simulating large strain accumulation beyond
the first instance of liquefaction. Furthermore, the generic structure of the newly proposed
equations allows other constitutive models to be incorporated in their formulation. The
model validation is performed in comparison with cyclic simple shear tests on Ottawa F-65
sand under undrained conditions and cyclic torsional test on Toyoura sand under constant
volume conditions.
In Chapter 4, the performance of the extended model implementation is evaluated by
simulating a centrifuge experiment on a saturated sloping sand deposit subjected to dynamic
loading. For this purpose, a numerical model is developed in a finite difference platform to
represent the soil conditions during the static and dynamic analysis. The dynamic analysis
is carried out using two constitutive models, the extended model presented in Chapter 3 and
its former version prior to the modifications. The numerical results are compared with the
recordings of acceleration sensors, pressure transducers and displacement measurements from
the centrifuge experiment. The results and comparisons validated the model implementation
and highlighted the capability of the extended model to improve the reproduction of the
sand response during the dynamic analysis. Additionally, the comparison is used to point
out shortcomings of the constitutive models.
Chapter 5 presents a summary of the research performed on each chapter. Also, it is
presented the important contributions found on each chapter and suggestions for future
work research.
Appendix A described the model implementation carried out using a C++
programming language and incorporated into a finite difference platform for solution of
boundary value problems. To this end, the appendix presents the formulation and
implementation of the constitutive model developed by Dafalias and Manzari (2004) with
the constitutive mechanism described in Chapter 3 and an overshooting scheme proposed
by Dafalias and Taiebat (2016). The method chosen for the integration of the rate

7
equations is cutting-plane algorithm. A pressure dependent sub-stepping technique is
added to the model implementation to ensure that the stress-strain response is integrated
correctly. Details of the steps undertaken in all the algorithms used for the development of
the model implementation are presented. The code verification of the implementation is
carried out by simulating triaxial tests subjected to monotonic and cyclic loading under
drained and undrained conditions. The model implementation results are compared with
the results obtained using a verified model implementation to ensure that all numerical
algorithms are correctly implemented in the code.

1.4 Notation
Within this dissertation, scalar variables are denoted by characters with normal letters (e.g.
e), vectors are noted with an arrow symbol (e.g. ~a), second order-order tensors are formatted
bold (e.g. σ). Fourth-order tensors are noted by open face symbols (C). A dyadic product
is denoted by a ⊗ b (i.e. aij bkl in index notation), a vector cross-product is denoted by
~a × ~b = εijk aj bk , a single tensorial contraction a · b = aik bkj , a vector contraction ~a · ~b = ai bi
√
and a double contraction by a : b = akl bkl . The norm of a tensor is defined as |a| = a : a,
√
similarly the norm of a vector is defined as |~a| = ~a · ~a, the trace of a tensor as tr(a) = aii .
Throughout the thesis the stress tensor, σ is decomposed very often into its hydrostatic
p = tr(σ)/3 and deviator s = σ − pI components, with I = δij the second-order identity
tensor. The Kronecker symbol δij is equal to 1 for i = j and 0 for i 6= j. Similarly, the strain
tensor ε is decomposed as well into its volumetric εv = tr(ε) and deviator  = ε − εv I/3
components.

8
2. Insight from discrete element
modeling of cyclic shearing of
granular assemblies

2.1 Introduction
Understanding the fundamental mechanism of soil liquefaction is one of the most challenging
tasks for the geotechnical community. Liquefaction may occur in saturated granular soils
when they are subjected to monotonic or cyclic shearing, and may lead to a temporary or
permanent loss of shear stiffness and strength in the soil matrix. Usually, the temporary loss
of strength is due to the excessive generation of pore pressure. Soil liquefaction often leads
to large displacement in the soil, causing considerable damage of structures frequently seen
after earthquakes. Extensive research on soil liquefaction has improved the understanding of
the mechanism behind this phenomenon (Seed and Lee, 1966; Ishihara et al., 1975; Castro,
1975; Casagrande, 1976; Alarcon-Guzman et al., 1988; Vaid et al., 1990; Bardet et al., 1999).
As previously mentioned in the Chapter 1, the current practice distinguishes two types
of liquefaction: 1) flow liquefaction, and, 2) cyclic mobility. Flow liquefaction commonly
observed when loose sands are subjected to monotonic or cyclic loading, is linked to the
permanent loss of shear strength which leads to a rapid increase of strains and pore pressure.
During cyclic loading, cyclic mobility is associated with the temporary loss of strength when
the pore pressure in each cycle rises momentarily to the confining pressure. Point in which
the material liquefies under cyclic loading, the effective stress goes to zero and the material
experiences large shear strains. Cyclic mobility can be observed in almost any type of
granular material, even dense sands.

9
 Focusing only on cyclic mobility, extensive research has been carried out to by means of
element level laboratory experiments. Experimental results of cyclic triaxial, cyclic torsional
and cyclic simple shear tests often report that large strains continue to be generated after
initial liquefaction, i.e., in the post-liquefaction response (Kutter et al., 1994; Wichtmann,
2016; Wahyudi et al., 2016; Bastidas, 2016; Rouholamin et al., 2017). Initial liquefaction
refers to the first occurrence of soil liquefaction i.e., the first occurrence of zero effective
stress. Experimental results constitute a reliable database for the evaluation, calibration
and validation of constitutive models for the analysis of cyclic mobility. At the sample
scale, however, explanations of the mechanism underlying the origin of large shear strains
observed in cyclic mobility at a particle scale is missing in the current literature. Performing
fundamental research in the laboratory to understand the physical interactions between
particles at a grain-to-grain scale might be challenging, or even impossible with current
technology.
Due to the difficulties in measuring the particle interaction during cyclic loading using
laboratory testing procedures, virtual experiments based on numerical simulation methods
such as molecular dynamics, also known as the discrete element method (DEM) (Cundall and
Strack, 1979), or contact dynamics (Jean and Moreau, 1987), may be suitable alternatives
for this purpose. The simulation of soil behaviour using DEM is becoming a widely used
source of information for understanding macroscopic physical phenomena from the micro-
scale interaction in granular assemblies. Since the pioneering work of Cundall and Strack,
many researchers have developed models for contact forces (Walton and Braun, 1986), time
integration methods (Lee and Herrmann, 1993), algorithms for the detection of contact
between particles (Nezami et al., 2004), and modeling particles with complex geometries
(Boton et al., 2013). In this chapter, computer simulations are used to reproduce cyclic
mobility in granular material using DEM.
From a micromechanical perspective, research has been performed by means of numerical
simulations using DEM to understand granular material response when it is subjected to
cyclic loading under constant volume conditions. In particular, Wang and Wei (2016)
carried out computational undrained cyclic biaxial tests using DEM to highlight a proposed
measure called “centroid difference” for quantifying the geometrical arrangement between

10
particles and the surrounding voids. With a direct correlation between the centroid difference
and the level of shear strain amplitude γ, the study suggests that stress-strain loops reach the
maximum level of γ that corresponds to the minimum centroid difference. Using the same
configuration, Wang et al. (2016) looked at the average distance of the three closest particles
per grain during “semi-suspended particle” regime, also known as the Mean Neighbouring
Particle Distance (MNPD), to establish a direct correlation between the maximum level of
γ and MNPD. Although these 2-dimensional DEM studies have consistently shown that
shear strain amplitude increase occurs at low stresses and low coordination number, more
simulations in 3-dimensions are still missing in the literature to understand the phenomenon
from a micromechanical perspective.
The objective of this chapter is to explore the mechanism underlying the large strain
accumulation in post-liquefaction response during cyclic mobility using 3-dimensional DEM
simulations. The qualitative results shown in terms of macromechanical (e.g., stress-path,
stress-strain response and shear modulus) and micromechanical variables (e.g., number of
contacts in the particle network and anisotropy) aims to develop novel concepts to improve
the continuum constitutive modeling. For this purpose, DEM simulations of cyclic triaxial
under constant volume conditions are carried out on samples with polydisperse spherical
particles. At the particle contact level, the linear spring-dashpot model is used to describe
normal and tangential forces. The sliding between particles is imposed in the tangential
direction to take into account the sliding/sticking condition.
The chapter is organized as follows. The observed results from cyclic triaxial laboratory
tests on sands are summarized in Section 2.2. A brief overview of the numerical method
and its fundamental contact laws underlying the DEM simulations are presented in Section
2.3. The description of the sample preparation and the shearing procedure is shown in
Section 2.4. This is followed by the main macromechanical and micromechanical results
used to characterize the shear strain generation in Section 2.5. Conclusions drawn out of
this chapter are shown in Section 2.6.

11
2.2 Observations from cyclic undrained triaxial
laboratory tests on sands
This section deals with the response of sandy material subjected to cyclic triaxial loading
in the laboratory. The experimental results from tests on Karlsruhe fine sand published by
Wichtmann and Triantafyllidis (2016a,b), are used to describe the main components of the
cyclic undrained behavior of sands.
The undrained cyclic triaxial tests were conducted at the Karlsruhe Institut for
Technology. The batch of the tested Karlsruhe sand is a clean sand with a fines content
F C = 0.9%, a mean grain size D50 =0.14 mm, a uniformity coefficient Cu ≈ 1.45, maximum
and minimum void ratios emax = 1.054 and emin = 0.677, and specific gravity GS = 2.65.
The samples were prepared by dry air pluviation in a mold of 100 mm in diameter and 100
mm in height, to p0 = 100 kPa of confinement pressure at three different relative densities,
Dr : 78, 61, and 25%. The shearing stress cycles have been applied using a constant axial
strain rate of ε̇a = 0.5%/min. The strain rate direction was changed when the
pre-described shear stress limit of qamp , was reached either in compression or extension.
Table 2.2 presents all the initial conditions (densities, relative density and confinement
pressure) of the samples prior to the cyclic loading and the limit of shear stress applied
during the undrained shearing.

Table 2.1: Initial conditions (initial void ratio e0 , relative density Dr and initial confining
pressure p0 ) and limit of qamp for the triaxial tests

Test ID e0 Dr (%) p0 (kPa) qamp (kPa)

TCUI-01 0.759 78 101 30
TCUI-02 0.825 61 102 30
TCUI-03 0.961 25 99 25

For practical purposes liquefaction is considered when excess pore water pressure ratio
(ru ) reaches 99%. The definition of ru is defined with the measured excess of pore water

12
pressure ∆u as

∆u
ru = . (2.1)
p0

Figure 2.1 shows the mechanical response in three samples in terms of ru with the number
of cycles, stress-path (q-p), and stress-strain (q-εa ) plots. Several qualitative observations of
the sand response can be highlighted:

1. Figures 2.1(a),(d) and (g) shows the evolution of ru with the number of cycles. The

(a) (b) (c)

(d) (e) (f)

(g) (h) (i)

Figure 2.1: Experimental results of an undrained cyclic triaxial test on Karlsruhe fine sand
at Dr = 78, 61, and 25%: (a)(d)(g) excess pore pressure ratio vs. number of cycles (b)(e)(h)
stress-path and (c)(f)(i) stress-strain loops. Data from Wichtmann and Triantafyllidis
(2016a)

13
 results suggests the number of cycles to reach ru = 99% is increased with the level of
relative density.

2. In the dense and medium samples, the mean effective stress is reduced with cyclic
loading until it reaches the failure line i.e., the point in which the stress-path in Figures
2.1(b) and (e) starts to overlap in almost identical butterfly-shaped loops, a signature
in cyclic mobilty. In the loose sample, the stress-path shown in Figure 2.1(h) reached
the failure line reducing permanently the mean effective and deviatoric stress.

3. The post-liquefaction response in Figures 2.1(c) and (f) is characterized by entering and
exiting liquified states two times in each loading cycle, a signature in cyclic mobility.
Figure 2.1(i) is characterized by the entering once the liquefied state, a characteristic
response in flow liquefaction. The post-liquefaction response during liquefied states is
characterized by a large axial strain development.

These qualitative observations are used later to verify the DEM simulations in capturing
the intended cyclic mobility.

2.3 An overview of the adopted DEM framework

The numerical tool used in this study is an open-source DEM code developed by Oquendo
(2013). This tool provides a numerical option to study the 2 or 3-dimensional mechanical
response of disks or spheres respectively. The contact laws between particles consist of
normal collision, tangential sliding, rolling and torsion. The code has been used to study
the influence of the rotation of particles on the material shear strength when subjected to
monotonic triaxial compressions under drained conditions (Oquendo et al., 2011). The code
also has been used to derive an equation of state for granular materials during isotropic
compressions (Oquendo et al., 2016). In this study, the code has been modified to simulate
triaxial cyclic loading under constant volume conditions. In the following section a brief
overview is given on the numerical tool used to carry out the 3-dimensional simulations.

14
2.3.1 Basic assumptions and contact definition

Granular materials consist of a large number of particles, N , ranging in size between

micrometers and centimeters. The particles interact via a short-range interaction, i.e.,
interaction by physical contact. The dynamic of a particle i with mass mi and moment of
inertia Ii moving with the translational and rotational velocities ~vi and ω
~ i , respectively, at
any time t is governed by Newton-Euler equations

d~vi
mi = F~i (2.2)
dt
d~ωi
Ii = ~τi (2.3)
dt

where F~i and ~τi are the total force and total torque on the particle i. F~i is defined in terms
of normal force F~n and tangential force F~t by

N 
X 
F~i = ~ ~
Fn + Ft (2.4)
j
j=N

~τi is defined in terms of sliding ~τsliding , rolling ~τrolling , and torsion ~τtorsion torques by

N
X
~τi = (~τsliding + ~τrolling + ~τtorsion )j . (2.5)
j=N

The sum in Equations 2.4 and 2.5 takes into account all contact partners j of particle i.
The computed F~i and ~τi with Equations 2.4 and 2.5 are the key of every DEM simulation.
The non-linear Equations 2.2 and 2.3 should be solved numerically. In order to obtain
a numerical solution to these equations (e.g., the calculations of trajectories and particles
displacements) one has to 1) define the boundary conditions and initial particles positions
and trajectories, 2) calculate the total forces and torques acting on each particle, and 3)
numerically integrate Newton’s second law of motion.
Modeling deformations of particles in contact with each other is simplified to the
interaction force being related to the overlap hij of the pair of spherical particles in
contact. Following the principle of the DEM at a specific time, the two particles i and j of

15
radii Ri and Rj are allowed to have a limited overlap hij , given by

hij = Ri + Rj − |~rij | (2.6)

where ~rij = ~ri − ~rj , and ~ri and ~rj are the vector positions to the center of each particle (see
Fig 2.2). Then, two particles are in mechanical contact if the sum of their radii exceeds the
distance of their centers; i.e., hij > 0 (remember hij < 0 is not considered due to the short-
range interaction). For any other particle shape, the contact definition is more complex than
for spherical particles. The force on particle i, from particle j, at contact c, can be broken
down into its components in normal and tangential directions F~n and F~t , respectively. The
normal component is along the unit vector n̂ij joining the centres of the mass of particles i
and j,
~rij
n̂ij = (2.7)
|~rij |
and the tangential component is along the unit vector t̂ij such that n̂ij · t̂ij = 0. The unit
vector t̂ij is defined either by the relative tangential velocities, or, from the tangential forces
explained later in the following sections.

2.3.2 Normal contact law

Each one of the intergranular normal F~n and tangential F~t forces consists of an elastic F~ el
and viscous F~ da component. The simplest normal elastic force F~nel , representing the repulsive
particle force, follows the Hooke’s law (linear model) and is proportional to hij and particle

Figure 2.2: Schematic representation of the overlapping distance, hij , between two spherical
particles i and j

16
normal stiffness kn ,
F~nel = kn hij n̂ij . (2.8)

The normal viscous force F~nda , representing the dissipative term, is calculated from the
reduced mass mij = mi mj /(mi + mj ), the normal relative velocity ~vij,n , and the normal
damping coefficient γn by
F~nda = −γn mij ~vij,n (2.9)

with
~vij,n = (~vij · n̂ij )n̂ij . (2.10)

The damping coefficient γn in Equation 2.11 is calculated using the well-known concept
of the normal coefficient of restitution en . The normal coefficient of restitution is defined as
the ratio between the final and initial relative speed of two particles before and after they
collide. A perfectly elastic contact results in en = 1. In this scenario, the velocity after the
collision reaches the same value before the collision. An inelastic contact results in en < 1, in
which a portion of the total energy is dissipated during the collision. Therefore, a perfectly
inelastic contact results in en = 0. In this case, the relative velocity after the collision is
reduced to zero. This study uses the relationship between γn and en proposed by Pöschel
and Schwager (2005) and given by
s
kn mij
γn = 2 ln(en ) . (2.11)
ln (en )2 + π

The force-displacement response for the direction normal to the contact plane of two
particles is illustrated in Figure 2.3, where |F~n | represents the magnitude of the normal
force and h represents the particle overlap. The response follows a linear spring-dahspot

|Fn|

kn

-h h
Figure 2.3: Schematic illustration for the force-displacement response of the normal contact
model

17
model in the compression direction (e.g., when h > 0), and zero force in the extension
direction (e.g., when h < 0). In order to complete the description of the total force and
total torque in Equations 2.4 and 2.5 respectively, the F~t together with the resulting torques
(~τsliding , ~τrolling , ~τtorsion ) are defined in the following section.

2.3.3 Sliding, rolling, and torsion contact laws

The tangential direction includes the observed dynamic and static friction forces and torques.
There are three main processes considered; sliding, rolling and torsion. The force-torque
contact model is the same for these interactions. The main difference is the computation
of the appropriate relative tangential velocity. It is important to note that the tangential
force and sliding torque in Equations 2.4 and 2.5 are calculated from the sliding process.
The latter two cases (rolling and torsion) result from a pseudo-force, used as an auxiliary
quantity to calculate the acting friction torque included in Equation 2.5. In the present
study, all the simulations were carried out using the tangential contact model proposed by
Luding (2008). In this contact model, the set of equations describing the sliding process are
used to describe rolling and torsion processes. Each process is distinguished by using the
corresponding relative velocity (e.g., ~vij,s for sliding, ~vij,r for rolling, and ~vij,0 for torsion),
accumulated tangential displacement (e.g., ξ~t,s for sliding, ξ~t,r for rolling, and ξ~t,0 for torsion),
and model parameters involved in each process. The tangential contact model is described
based on the sliding case, rolling and torsion are identical to the sliding case.
The elastic and viscous terms of the tangential force F~t are analogous to the normal force
configuration. The tangential elastic force F~tel , acting in the opposite direction of the sliding
relative velocity, is proportional to ξ~t,s with the particle tangential stiffness kt , according to

F~tel = −kt ξ~t,s (2.12)

where
ξ~t,s = ξ~s − (n̂ij · ξ~s )n̂ij (2.13)

and the accumulated tangential displacement is updated as

ξ~t,s = ξ~t,s + ~vij,s ∆t (2.14)

18
where ∆t is the integration time-step, and ~vij,s is the sliding relative velocity at the contact
point giving by
~vij,s = ~vij − (~vij · n̂ij )n̂ij (2.15)

the relative linear velocity ~vij between particles i and j is determined from
~ i + a0j ω
~vij = ~vi − ~vj + n̂ij × (a0i ω ~ j) (2.16)

where ~vi , ~vj and ω ~ j are the translational and angular velocities respectively, and a0i is
~ i, ω
the corrected radius from the centre of mass for particle i to the contact point defined by
a0i = Ri − hij /2. The tangential viscous force F~tda , representing the dissipative term, is
computed as
F~tda = −γt mij ~vij,s (2.17)

where γt is the tangential damping coefficient. The magnitude of the normal and tangential
forces at the particle contact are given by |F~n | = |F~nel + F~nda | and |F~t | = |F~tel + F~tda |,
respectively. If a Coulomb friction law is used at the particle contact level and µs is the
particle-to-particle coefficient of sliding friction. The particles are in no-sliding condition as
long as |F~t | ≤ µs |F~n |; otherwise they slide against each other when |F~t | > µs |F~n |. During
the sliding condition, the friction tangential force is |F~t | = µs |F~n |t̂ij , and the accumulated
tangential displacement is updated as
1
ξ~t,s = − (F~t + γt mij ~vij,s ) (2.18)
kt
according to the computed F~t . The corresponding friction torque is described as
τsliding = ~li × F~t (2.19)

where ~li is the branch vector joining the particle i centre with the contact point.
The force-displacement response for the direction parallel to the contact plane of two
particles is illustrated in Figure 2.4, where |F~t | represents the magnitude of the accumulated
tangetial displacement and |ξ~t,s | represents the magnitude of the tangential spring. The
response follows a linear spring-dahspot model in the tangential direction until the Coulomb
slider µs is active when a finite normal force is achieved µs |F~n |.
To complete rolling and torsion mechanisms, the rolling relative velocity at the contact
point for the rolling process is defined as
~vij,r = −a0ij n̂ij × (~ωi + ω
~ j) (2.20)

19
 |Ft|
s|Fn|

kt

| t,s|

Figure 2.4: Schematic illustration for the force-displacement response for tangential contact
model

where a0ij is the effective reduced radius defined by a0ij = a0i a0j /(a0i + a0j ). For the torsion
process, the torsion relative velocity is the normal component of the relative angular velocity,
defined as
~vij,0 = aij {n̂ij · (~ωi + ω
~ j )} n̂ij (2.21)

calculated from the reduced radius without correction aij = Ri Rj /(Ri + Rj ) according to
Oquendo (2013).

2.3.4 Summary of the contact parameters

The DEM contact model needs 11 model parameters to define the particle-to-particle
interaction. However, it is possible to reduce the number of input parameters by defining
correlations for selected parameters. Table 2.2 shows the micromechanical parameters used
in each particle-to-particle contact. Most of these parameters are defined as ratios of four
parameters: kn , γn , µs , and the particle-to-particle coefficient of rolling friction µr . Hence,
the model parameters used in the rolling process, e.g., the rolling stiffness kr and the rolling
damping γr are defined from kn and γn , respectively. The parameters involved in the
torsion process, e.g., the torsion stiffness k0 , the torsion damping γr , and the
particle-to-particle coefficient of torsion friction µ0 are obtained similarly as in the rolling
process. These ratios are also adopted in other research studies to simulate granular
material response (Luding, 2008; Oquendo et al., 2011; Oquendo, 2013; Barrero et al.,
2018).

20
Table 2.2: Suggested correlations used for the contact parameters in the DEM simulations
(Oquendo, 2013; Luding, 2008)

Parameter name Symbol Value

Normal stiffness kn T.B.A∗
Sliding stiffness kt 0.8kn
Rolling stiffness kr 0.4kn
Torsion stiffness k0 0.4kn
Normal damping γn T.B.A∗
Sliding damping γt 0.2γn
Rolling damping γr 0.05γn
Rolling damping γ0 0.05γn
Coulomb friction - sliding µs T.B.A∗
Coulomb friction - rolling µr T.B.A∗
Coulomb friction - torsion µ0 µr
∗
T.B.A.: To be assigned in Section 2.4.1

2.3.5 Description of control variables

To link the micromechanical normal and tangential forces from the particle interactions to
the macromechanical response of the granular material, several quantities including stress,
coordination number and anisotropy, are defined based on a discrete description. The average
stress tensor σ of a selected volume V of the 3D element is calculated following the definition
proposed by Rothenburg and Bathurst (1989)
1 X cc
σlk = F l , (2.22)
V c∈N l k
c

where Nc is the total number of contacts, Fl is l-component of the force vector, lk is the k-
component of the branch vector joining the centres of two particles in contact, and the sum
extends over all the contacts c between particles inside the selected representative volume
element with the total volume V . The triaxial mean effective and deviatoric stress, p and q,
are calculated from the stresses σxx , σyy and σzz obtained from Equation 2.22, according to
p = (σxx + σyy + σzz )/3 and q = (σzz − σxx ).
The strain tensor ideally should be derived from the available micromechanical
information, just like the stress tensor. However, in this study, the strain is used as an
indicator for the level of deformation and therefore it does not require an accurate

21
 Z
Y

Figure 2.5: Illustration of the DEM sample at the end of an isotropic triaxial compression
using six rigid walls. In this case Lx , Ly , Lz are the initial walls lengths prior to the cyclic
shearing

definition. The principal strains εxx , εyy and εzz are calculated from the boundary walls as
εxx = ∆lx /Lx , εyy = ∆ly /Ly , εzz = ∆lz /Lz , (2.23)

where ∆ly , ∆lx and ∆lz are the difference between the current and the initial wall lengths
prior to the cyclic shearing (see Figure 2.5). The triaxial deviatoric strain is defined as
εq = 2/3(εzz − εxx ).
In quantifying the microstructure network several definitions can be found in the
literature to determine the mean coordination number z; i.e., the average number of
contact per particle. Four definitions of the mean coordination number are used in this
study, as shown in Table 2.3 where N is total number of particles in the sample and N eff is
the effective number of particles. Another important quantity is the floating ratio, fr ,
defined by Bi et al. (2011) as
frβ = N − N eff /N,


(2.24)

it is used to quantify the number of particles that are excluded from the computation of z,
β stands for the same subindex I, II, III, and IV used for z in Table 2.3.
The floating ratio can be understood as the complement of the coordination number.
For example, if all particles N are considered to determine z, the effective number of
particles involved in the contacts is also N . In this case the floating ratio is frI = 0 and the
coordination number is zI = 2c/N . If only particles with more than three contacts are
considered, the effective number of particles is N eff = N − N h0i − N h1i − N h2i where N h0i

22
are the number of particles with zero contacts, also known as rattlers, and N h1i , and N h2i
are the number of particles with one and two contacts, respectively. The corresponding
frIV and zIV are calculated according to Equation 2.24 and Table 2.3.
It is also important to quantify the microstructure anisotropy in a granular assembly. To
this end, two anisotropies are distinguished: contact and mechanical anisotropy. The contact
anisotropy ac is used to give insight of the contact network, and it is defined according to
Azéma et al. (2013) as

5 F̄1 − F̄3
ac =
(2.25)
2 F̄1 + F̄2 + F̄3
where F̄1 , F̄2 and F̄3 are the principal values of the fabric tensor F . The contact anisotropy
defines the magnitude of anisotropy in the contact normal orientation. For instance, during
conventional triaxial compression a vertical major principal fabric yields a positive ac ; during
triaxial extension, a horizontal principal fabric outputs a negative ac . The fabric tensor used
here to quantify the normal contact orientation is defined by
1 X c c
Flk = nn , (2.26)
Nc c∈N l k
c

where ni (i = k, l) are the k and l-component of the unit branch vector, defined in Equation
2.7, joining the centre of the particle to the contact point c.
The mechanical anisotropy is mainly caused by external forces and depends on the
induced contact forces. It can be split into normal force anisotropy (due to the normal
force contact component) and tangential force anisotropy (due to the tangential force

Table 2.3: Alternative definitions for coordination number z according to the different
measurements of effective number of particles, N eff

Symbol N eff z Reference

2c
zI N −
N eff
2c
zII N − N h0i Thornton (2015)
N eff
2c − N h1i
zIII N − N h0i − N h1i Thornton and Antony (1998)
N eff
2c − N h1i − 2N h2i
zIV N − N h0i − N h1i − N h2i Kuhn (2017)
N eff

23
contact component). Using the same definition given by Azéma et al. (2013), one can
calculate the normal anisotropy (an ) and tangential anisotropy (at ) magnitudes by
5 (χ̄n1 − χ̄n3 )
an = − ac (2.27)
2 (χ̄n1 + χ̄n2 + χ̄n3 )

5 (χ̄1 − χ̄3 )
at = − ac − an (2.28)
2 (χ̄1 + χ̄2 + χ̄3 )
where χ̄ni and χ̄i (i = 1, 2, 3) are the principal values of the normal force tensor χn and
total force tensor χ. The normal and tangential force anisotropies define the magnitude of
anisotropy in the normal and tangential force respectively. For instance, during conventional
triaxial compression a vertical major principal of the normal and total force tensors yields
a positive an and at ; during triaxial extension, a horizontal principal of the normal and
total force tensors outputs a negative an and at . For a deeper insight into the calculation of
the fabric and force tensors, refer to (Azéma et al., 2013). The various definitions outlined
above have shown that each anisotropy magnitude can be used to characterize the anisotropic
behavior originating from each distinctive source.

2.4 Description of triaxial DEM test

Discrete element simulations of 3-dimensional spherical particles were conducted to
investigate the constant volume cyclic shearing behavior of granular materials. To this end,
the particles were confined within a triaxial configuration of six rigid walls as shown in
Figure 2.5. In this study, the frictional forces - and consequently shear forces - on the walls
were discarded. The boundaries were considered frictionless, i.e., the particle-to-wall
coefficients of sliding µs,wall , rolling µr,wall and torsion µ0,wall friction, were zero. This
guarantees principal stress conditions were always perpendicular to the wall. Therefore, the
particle-wall interactions were calculated by means of Equations 2.8 and 2.9. The normal
stiffness and normal damping parameters of the particle-to-wall contact were assigned the
same as the normal stiffness and normal damping parameters of the particle-to-particle
contact. Each simulation went through a rigorous sample preparation, isotropic
compression and constant volume cyclic shearing, which are explained in the following
sections. All simulations presented in this chapter were performed in the absence of gravity.

24
2.4.1 Sample generation and compression stage

The granular system consists of 4,913 polydisperse spherical particles. The radius of each
particle was assigned as suggested by Pöschel and Schwager (2005),

Rmin
R= v " (2.29)
u  2 #
u
t1 − nz 1 − Rmin
Rmax

where nz ∈ [0, 1] is a random number. Using Equation 2.29, the radii of the particles are
chosen from the interval (Rmax , Rmin ) such that the total mass of all the particles from
a certain size interval is the same for all size (Oquendo et al., 2011). Therefore, more
smaller particles were generated for each large particle so no specific size dominates the
mass system. The ratio Rmax : Rmin was set to 3:1. The contact model parameters and other
relevant simulation parameters are summarized in Table 2.4. Since the discrete element
method simulates “soft-particle” (employing particle overlap) with a finite hardness, the
dimensionless stiffness parameter κ proposed by Roux and Combe (2002) and defined as

kn
κ= ¯ (2.30)
dp
¯ and mean pressure p, was used to determine if the particles subjected
for kn , mean diameter d,
to the given p were considered rigid. Using kn = 1×104 kN/m, d¯ ≈ 0.0012 m and p = 100
kPa, in all of the simulations κ yields to 8 × 104 , a value well beyond the suggested rigid
grain limit κlimit ≈ 104 (da Cruz et al., 2005). The high value of kn implies a small value of
∆t, 1×10−9 s.
The compactness of the sample was represented by the solid fraction defined as Φ = Vs /V ,
where Vs is the volume occupied by the solid grains. The solid fraction can be related to the
void ratio e through
1
e= − 1. (2.31)
Φ
All of the samples were isotropically compressed following the same sample generation
procedure. First, the particles were initially placed at the nodes of a cubic grid of 16×16×16.
The length of each grid element was 2.12Rmax , guaranteeing no particle-to-particle contact or
particle-to-wall contact. Second, each grain was assigned a randomly-oriented initial velocity

25
Table 2.4: DEM model simulations and contact parameters used for the triaxial simulation

Parameter name Symbol Value

Number of particles N 4913
Particle radius (m) R [0.4-1.2]×10−3
Density (kg/m3 ) ρ 2650
Normal stiffness (kN/m) kn 1×104
Normal damping γn (en ) 0.2
Coulomb friction - sliding µs [0.15, 0.2, 0.25, 0.3]∗
Coulomb friction - rolling µr 0.05
∗
A different value is used during the constant volume cyclic shearing

with speed varying between 0 and 0.1 m/s. Third, all six walls were prescribed with the
target value of σw = 100 kPa of confinement stress using a servo-controlled algorithm to keep
the wall stress increments small to avoid any excessive particle interpenetration. Fourth, the
confinement stress σw = 100 kPa was kept constant until the static equilibrium criterion was
satisfied. It is common to define the static equilibrium when the external forces matched the
internal particles forces. However, in this study, the static equilibrium criterion was defined
when the average translational and angular speed for all particles were lower than 10−8 m/s
and 10−8 rad/s respectively, and the number of sliding contacts was less than 1%.
To obtain samples with different initial e0 , different sliding Coulomb friction coefficients
µs were used, as shown in Table 2.4. Once the sample reached the static equilibrium condition
during the isotropic compression, high µs values imply high e0 ; low µs values lead to low e0 ,
as shown in Figure 2.6(a). The values of the contact anisotropy at the end of the isotropic
compression are shown in Figure 2.6(b).

2.4.2 Constant volume cyclic shearing stage

Once the static equilibrium was reached, the samples were sheared under a constant volume
condition using µs =0.5. To preserve the constant volume condition, the lateral walls were

26
 (a) (b)

Figure 2.6: Variation of void ratio and contact anisotropy at the end of the compression
stage with sliding friction coefficient

moved with the prescribed wall velocity according to

vw,x = −0.5(lx /lz )vw,z (2.32)

vw,y = −0.5(ly /lz )vw,z (2.33)

where l is the length of the wall in the corresponding direction (x, y, z). It is imperative
to appropriately assign vw,z ; an unsuitable value will result in an expensive computational
simulation time (several days or months), and for a high value, the particle interactions
will result in the collisional regime associated with gas response according to da Cruz et al.
(2005). Therefore, to keep the simulation time within a reasonable duration, and to keep
the particle interactions inside the quasi-static regime (associated with soil response), the
inertial parameter I is used to back-calculate the wall velocity vw,z according to
q
I = ε̇z m/(dσ¯ w) (2.34)

where ε̇z = vw,z /lz . In this study, the top wall velocity vw,z was 0.001m/s. This guarantees
particle interactions were inside the quasi-static limit characterized by the condition I < 10−4
(GDR MiDi, 2004).
The cyclic shearing load was applied reversing the top wall velocity (positive in
compression, and negative in extension) whenever the cyclic deviatoric stress qamp = 30
kPa was reached. Table 2.5 presents all initial conditions (densities, confinement pressure,
coordination number and contact anisotropy) of the samples prior to the cyclic shearing.

27
Table 2.5: Initial conditions (initial void ratio e0 , initial confining pressure p0 , initial
coordination number z0 , and initial contact anisotropy ac0 ) for the triaxial tests

Test ID e0 p0 (kPa) z0 ac0

Triax-01 0.684 100 4.63 0.0306
Triax-02 0.702 100 4.55 0.0191
Triax-03 0.721 100 4.45 0.0249
Triax-04 0.732 98 4.23 -0.0286

2.5 Analysis of results

The numerical results are shown in three subsections. In the first subsection, the
macromechanical response is shown through excess pore water pressure ratio, stress-path
and stress-strain plots. The macromechanical results have similar trends to those observed
in laboratory testing. In the second subsection, the micromechanical response obtained
from particle-to-particle interaction is shown via coordination number, floating ratio and
the different anisotropic magnitudes. Interestingly, micromechanical results suggest a
bi-modal regime response: a solid like behavior (the expected soil response) and a
semifluidized like behavior. In the third subsection, the newly transitional semifluidized
state is described in detail: macroscopically via stress-path and stress-strain plots, and
microscopically by shear modulus, coordination number, floating ratio, and anisotropic
magnitude plots. Results suggest that the semifluidized state response is attributed to a
large number of suspended particles.

2.5.1 Macromechanical response: ru , stress-path and stress-strain

To focus on the liquefied states, a practical definition for liquefaction is needed for DEM
results to define when the mean stress p vanishes during liquefaction. Due to the fact that
the particles interact via collisions, and given that the simulations are performed numerically,
it is possible to measure small contact forces and other quantities even during liquefied states.
Here, liquefaction is defined when p is reduced to 0.99% of its initial mean stress.
During the shearing phase under constant volume conditions, the total confining stress
imposed by σxx and σyy remains constant, such that the pseudo-excess pore water pressure

28
ratio ru for a dry granular packing can be calculated as

(q/3 + p0 ) − p
ru = (2.35)
p0

where p0 is the initial mean stress before shearing. Figure 2.7 shows the typical mechanical
response in four samples with low to higher densities subjected to the same initial confinement
pressure (p0 = 100kPa). It can be observed in Figures 2.7(a),(d),(g) and (j), liquefaction
(ru ≈ 0.99) is reached regardless of their initial level of density. In the previous figures it is
also shown, for higher densities there is a slow accumulation of the excess of pore pressure
ratio to reach ru ≈ 0.99 with the number of cycles as oppose to low densities that require
few cycles to reach ru ≈ 0.99. A behavior similar to the trend observed in Section 2.2.
It is interesting to highlight that the Triax-03 test requires a higher number of cycles to
reach liquefaction than the Triax-02 test, which has a lower void ratio. This is presumably
due to its initial ac , as shown previously in Figure 2.6. In Figures 2.7(b),(e),(h), and (k), the
mean stress is reduced with cyclic loading until it reaches the failure line, the point in which
the stress path has the butterfly shape. The stress-strain loops are continuously increasing
in each half-cycle. Entering the liquefied states at low values of mean and deviatoric stress,
the level of axial strain generated in each half-cycle depends on its initial density as shown
in Figures 2.7(c),(f),(i), and (l). The post-liquefaction response is characterized by entering
and exiting the liquefied states (ru ≥ 0.99) twice in each loading cycle, which is a distinctive
feature of the liquefaction type cyclic mobility. The qualitative trends observed from the
cyclic triaxial results by Wichtmann (2016) compare favourably and validate the numerical
simulations shown herein. These simulation results compare favourably with other numerical
results found in the literature (Huang et al., 2018; Barrero et al., 2018).
Inspired by previous works that have shown large generations of shear strains at a very
small mean stress (Barrero et al., 2018; Wang et al., 2016; Wang and Wei, 2016), and
associating liquefaction to unstable states (Huang et al., 2018), it is encouraged a deeper
understanding of the origin of the large strain generation at low-stress values observed during
liquefied states from a particle-scale perspective.

29
 (a) (b) (c)

(d) (e) (f)

(g) (h) (i)

(j) (k) (l)

Figure 2.7: Results of DEM cyclic triaxial tests under constant volume for different initial
configurations: (a),(d),(g),(j) pore-pressure ratio history, (b),(e),(h),(k) stress path and
(c),(f),(i),(l) stress-strain curve

30
2.5.2 Micromechanical response: anisotropy and coordination
evolution

Researchers have previously shown the link between structural stability and the
coordination number (Cundall and Strack, 1983; Radjaı̈ and Dubois, 2011; Thornton, 2015;
Huang et al., 2018), for example, in a 3-dimensional system with µs = ∞, i.e. with no
sliding condition. The structural network can be analyzed by comparing its degrees of
freedom and its constraints. The number of degrees of freedom concerns any independent
motion in the system, such as translational or rotational displacements. The degrees of
freedom for a single spherical particle is six (three translation and three rotational), and
the degrees of freedom considering all the particles in the system is 6N . The constraints in
this context represent the number of unknown reactions at a single contact, which are three
(e.g., one normal force, one tangential force and a resulting moment) and considering all
contacts in the system result in 3c. If the total number of constraints is equal to the
degrees of freedom, i.e. 3c = 6N , the system is statically determinate (isostatic). Hence,
the critical coordination number using the isostatic condition is defined by
zc = 2c/N = 2(2N )/N = 4. The equation used to define zc corresponds to zI in Table 2.3.
According to Cundall and Strack (1983), the system is over-constrained (statically
indeterminate) when the coordination number goes above its critical value, z > zc . At this
point, the assembly has more contacts than its critical value to guarantee stability. On the
other hand, an under-constrained system is when the coordination number falls below its
critical value, z < zc . In this case, the unstable assembly has a lower number of contacts
than its critical value to ensure a stable structural network.
From a microscopic perspective, Figure 2.8 illustrates various contact configurations for
the blue spherical particle. Using the definition of zI given in Table 2.3, the lowest
coordination number represented by Figure 2.8(a) corresponds to the suspended state and
is associated to a liquefied state. In this state the soil response is characterized as a fluid
like behavior. The unstable states shown in Figures 2.8(b)-(d) represent transitional states
in which soil response can be described as a viscous fluid (Hwang et al., 2006), or also
referred to as semifluidized like behavior. The stable state shown in Figure 2.8(e) is the

31
 (a) z=0 (b) z=1 (c) z=2 (d) z=3 (e) z=4

Figure 2.8: Illustration of a blue spherical particle with different average number of contacts
(z): (a)-(d) scenarios associated with unstable states and (e) associated with stable state

non-liquefied state in which soil response is referred to as a solid-like behavior. Then, three
states are distinguished: (i) liquified state, (ii) a transitional state and (iii) a non-liquefied
state.
In Figures 2.9 and 2.10, the evolution of the average coordination number z and its
complementing floating ratio fr according to its different definitions given in Table 2.3 are
shown. In Figures 2.9(a),(d),(g), and (j), it is observed that the coordination number zIV is
reduced with cyclic loading. On the other hand, it has been observed in Figures
2.10(a),(d),(g),(j), that frIII is increased with cyclic loading, as expected. In more detail, it
is consistently observed the pace of reduction of zIV and frIII with the number of cycles is
slower for the Triax-01 test, as shown in Figures 2.9(a) and 2.10(a), and is much higher for
the Triax-04 test, shown in Figures 2.9(j) and 2.10(j). Figures 2.9(b),(e),(h), and (k) show
the coordination number evolution with the level of mean effective stress during cyclic
loading and the complementary Figures 2.10(b),(e),(h), and (k) show the evolution of the
floating ratio with the associated level of mean effective stress. For unstable systems
(z < zc ), it is observed that a drop in z and sudden increase in fr occurs at low mean stress
values, as shown in Figures 2.9 and 2.10. After the lowest z or highest (fr ), the material
enters the transitional state of the semifluidized state (Sf). The domain of the Sf state is
defined to extend until the level of stress shown with the red dotted line (pth ) in Figures
2.9(c)-(l) and 2.10(c)-(l). Then, if p < pth the material enters the Sf state (in grey in
Figures 2.9(c)-(l) and 2.10(c)-(l)), and if p > pth , the material is no longer inside the Sf
state. Notice that liquefied states (the lowest z) are included inside Sf state.
Using discrete element simulations, it is possible to explain the mechanical response of the
granular assembly as grains interact with each other using micromechanical quantities, e.g.
contact anisotropy and the mechanical anisotropy. The first link between the macroscopic

32
 (a) (b) (c)

(d) (e) (f)

(g) (h) (i)

(j) (k) (l)

Figure 2.9: Domain of the semifluidized state (0< p ≤ pth ) via coordination
number: (a),(d),(g),(j) coordination number evolution excluding particles with 0,1 and 2
contacts, (b),(e),(h),(k) coordination number evolution with mean stress, and (c),(f),(i),(l)
coordination number evolution with emphasis in low mean stress values

and the microscopic response was made by Rothenburg and Bathurst (1989). Using 2-
dimensional discrete simulations, Rothenburg and Bathurst established the first analytical
relation for disks between the shear strength, the contact anisotropy magnitude and the

33
 (a) (b) (c)

(d) (e) (f)

(g) (h) (i)

(j) (k) (l)

Figure 2.10: Domain of the semifluidized state (0< p ≤ pth ) via floaters ratio: (a),(d),(g),(j)
floaters ratio excluding particles with 0 and 1 contacts, (b),(e),(h),(k) floaters ratio evolution
with mean stress, and (c),(f),(i),(l) floaters ratio evolution with emphasis in low mean stress
values

mechanical anisotropy magnitudes, e.g. q/p = 1/2 (ac + an + at ). This equation opened
the possibility for researchers to understand how different anisotropy sources contributed

34
to the shear strength in granular materials. Later, using 3-dimensional simulations, Azéma
et al. (2013) proposed a generalized equation considering various angular polyhedral particle
shapes via
q 2
= (ac + an + at ) . (2.36)
p 5
Azéma et al. (2013) have included an additional term in Equation 2.36 to take into
account the contribution of the branch anisotropy, al . However, due to the low polydispersity
in this study the contribution of al is negligible (Azéma et al., 2013; Guo and Zhao, 2013).
Results of the Triax-04 test have been employed to validate the correlation in Equation 2.36.
Figure 2.11 shows the mobilized strength (q/p) defined using q and p described earlier in
Section 2.3.5 and using the sum of each anisotropy as described in Equation 2.36. The results
correlate well with Equation 2.36, even for the response inside the Sf state. To understand
contributions from anisotropy sources during the shearing process, each quantity (ac , an , at )
is explored separately.
Figure 2.12 shows the evolution of anisotropy sources (contact and mechanical). During
the cyclic loading, ac evolves cyclically with increasing amplitudes. The evolution of normal
anisotropy an follows the same trend as that observed for ac . The amplitude of ac is lower
than the amplitude of an , regardless of the number of cycles or initial level of density. The
tangential anisotropy, at , also follows a similar trend as that observed for an . Considerable
fluctuations are observed for at and an during liquefied and transitional states; a possible
signature for Sf state response. The observed fluctuations are presumably the consequence
of the microscopic ‘slip and stick’ mode, allowed by the slider used in the tangential contact
model. The development of anisotropy sources against mean stress is also considered but
not shown here due to lack of a clear trend. In the next section, the importance of Sf state
is explored to gain a deeper understanding of the origin of the large strain accumulation at
low stress values.

2.5.3 Transitional state: semifluidized state

To illustrate the transition response between Sf state (p ≤ pth ) and non-liquefied states
(p > pth ), the stress-path and stress-strain response for cycle 89 for test Triax-01 are shown in

35
Figure 2.11: Triax-04 test mean stress evolution and shear strength correlation with
anisotropy sources

Figures 2.13(a) and 2.13(b). The black arrows illustrate the trajectory followed. Figure 2.13
starts with the non-liquefied state shown with the red circle, marked as 1. The material
response from markers 1-2 and 4-5 represents the dilation stages. During the dilation stage,
the material hardnes until the deviatoric stress condition is reached qamp = 30 kPa, the
point at which the applied shear strain rate is reversed and the material is unloaded. The
material response from markers 2-3 and 5-6 represents the small unloading portion before
entering the liquefied states. The lowest mean stress values are observed at markers 3 and
6; at that time, the material enters the liquefied state. The material response from markers
3-4 illustrates the mechanical response inside the Sf state. Inside this special state, p and
q drop dramatically and remain low until the dilation stage is reached again in marker
4. The shear strains generated inside Sf state are much higher than those shear strains
generated outside the Sf state during the dilation phase. The low strains generated during
unloading stages are negligible. An interesting observation during cyclic mobility is the fact
that the stress path passes the point of “zero” effective stress twice per cycle, as seen in
Figure 2.13(a). It is commonly known inside the geotechnical community that the majority
of deformations in cyclic mobility are associated with the “zero” stress region. This explains
the “flat” region of the stress-strain curves (markers 3-4 in Figure 2.13(b)), in which the
soil presumably has low stiffness. To quantify stiffness values, the tangential shear modulus
defined as G = ∆q/(3∆εq ) is shown in Figure 2.14 for the 89th cycle for the Triax-01 test.
To focus only on the Sf state, peak deviatoric stresses represented by markers 2 and 5 are
not included in Figure 2.14. Results in Figure 2.14 suggests a very low shear modulus at

36
 (a) (b) (c)

(d) (e) (f)

(g) (h) (i)

(j) (k) (l)

Figure 2.12: Results of DEM constant cyclic triaxial test for different initial densities:
(a),(d),(g),(j)contact anisotropy, (b),(e),(h),(k) normal anisotropy and (c),(f),(i),(l)
tangential anisotropy

liquified state in marker 3, which slowly recovers during the Sf state as it approaches exit
marker 4.
Figure 2.15(a) shows the evolution of the average coordination number (zIV ) for the
89th cycle for Triax-01. The non-liquefied state in marker 1, zIV starts with an unstable

37
 (a) (b)

Figure 2.13: Post-liquefaction response for Triax-01 test: (a) stress-path and (b) stress-
strain. Red markers 1 and 4 are non-liquefied states. Red markers 2 and 5 correspond to
the peak deviatoric stresses. Red markers 3 and 6 are liquefied states

structure as zIV = 3.896. From marker 1 to 2, new contacts are established while the
deviatoric stress increases. As new contacts are stablished zIV increases towards its critical
coordination number value zc . The maximum value of z = 4.038 is reached when deviatoric
stress equals the stress limit, qamp = 30 kPa. After the load is reversed, the average number
of contacts dropped abruptly to well below 4 until it reaches its minimum value zIV = 3.132
in the liquefied state at marker 3. During the Sf state (shown from markers 3 to 4), zIV is
continuously increasing until the material exits the Sf state with zIV = 3.886, a similar value
obtained from marker 1. It is interesting to highlight that the exiting zIV value in markers 1
and 4 are lower than its critical value 4. This implies the material is in a non-liquified state
with collapsible structure due to the under-constrained condition.
The observed response of zIV from markers 4 to 5 is similar to the observed response

Figure 2.14: Tangent shear modulus during the post-liquefaction response for Triax-01 test

38
from markers 1-2. The trend described during and outside the SRF for zIV is a characteristic
response of cyclic mobility. Figure 2.15(b) shows the evolution of fr for the 89th cycle for
test Triax-01. Prior to the maximum deviatoric stress in marker 2, fr decreases until loading
reversal occurs, followed by a sudden increase until reaches the Sf state characterized by its
maximum value at marker 3 (fr = 0.978). The high value of fr is evidence of the dominant
presence of unstable configurations shown in Figures 2.8(a)-(b). Continuing the cyclic load,
new contacts are established, decreasing fr until eventually the sample exits the Sf state
with fr < 0.55 at marker 4. The observed response from markers 4 to 5 is similar to the
observed response from markers 1 to 2.
The contact anisotropy ac , shown in Figure 2.16(a), starts with a value of 0.2 and remains
constant from markers 1 to 2. After the loading reversal the response between markers 2-
3, the principal load is now horizontal and ac drops to accommodate to the new loading
condition. As the cyclic loading continues, ac continues varying until it exits the Sf state
at marker 4 with the same absolute value of 0.2 as previously observed at marker 1. The
response of ac during markers 4-5 remains constant at -0.2 until loading reversal occurs again.
The response of ac during non-liquified states oscillates within the value of 0.2 during triaxial
compression and -0.2 during triaxial extension as shown in Figure 2.16(a). The positive and
negative values observed for ac confirms the majority of the contacts are aligned in the
vertical and horizontal directions, respectively.
The evolution of normal anisotropy an is shown in Figure 2.16(b). Preceding the Sf state

(a) (b)

Figure 2.15: Post-liquefaction response for Triax-01 test: (a) coordination number excluding
particles with 0, 1 and 2 contacts and (b) floating ratio considering particles with 0 and 1

39
 (a) (b) (c)

Figure 2.16: Post-liquefaction anisotropy evolution for test Triax-01: (a) contact anisotropy,
(b) normal anisotropy and (c) tangential anisotropy

states in marker 3 and 6, the observed values of |an | from marker 1 to 2 and 4 to 5 are higher
than |ac |. From markers 1 to 2, the magnitude of the normal anisotropy increases with cyclic
loading until the load is reversed at marker 2. Subsequently, an drops dramatically until point
3 is reached. The observed response of an during the Sf state (shown with markers 3 to 4)
reflects the scattered behavior between positive and negative values that slowly converge
towards marker 4 as the cyclic load is continued. Marker 4 represents the beginning of the
non-liquefied state, again it can be seen the normal mechanical anisotropy dominates the
material response as |an | > |ac |.
The evolution of tangential anisotropy at is shown in Figure 2.16(c). Overall, the
evolution of tangential anisotropy at in Figure 2.16(c), follows the same trend as that
observed for an in Figure 2.16(b).
The new state bound by stress threshold pth called the semifluidized state, is a noticeable
link between the macroscopic “zero” stress region and the microscopic particle coordination
number to understand why the soil response is as a semifluidized behavior. During this state
from a macroscopic review, the pseudo-pore pressure ratio remains constant, while the mean
and deviatoric stress, p and q, remain small, resulting in a very low shear stiffness and a large
accumulation of shear strains. At the microscopic level, the liquefied state occurs around a
certain level of stress pth by a sudden drop in coordination number. The transitional state,
Sf, is characterized by a high presence of floaters together with the scattered response of the
mechanical anisotropy. These may be a signature of cyclic mobility.

40
2.6 Conclusion
Four triaxial numerical experiments were conducted using the open-source DEM code
developed by Oquendo (2013) for particle assemblies made up a polydisperse spheres with
different densities. The code has been modified to model the cyclic shearing under constant
volume condition. The objective was to explain the mechanism and origin of shear strains
during liquefied states when the granular material is subjected to cyclic loading. To this
end, simulation results were presented in terms of macromechanical and micromechanical
quantities. After interpreting the results, the novel concept of the semifluidized (Sf) state
is defined at macroscopic and microscopic levels.
At a macroscopic level, there is a considerable drop in the mean effective stress and shear
stiffness that occurs around a low level of stress, called the Sf stress pth . When the mean
effective stress is lower than pth , the material response is inside the Sf state. In this state,
the response of the material is characterized by a low shear stiffness, mean effective stress,
deviatoric stress and a large accumulation of shear strains. Once the material shear stiffness
increases, the material state exits the Sf state at the same level of stress given by pth .
From a microscopic point of view, the cyclic loading under constant volume conditions is
a detrimental type of load that reduces the number of contacts between particles, weakening
the particle network. If the granular material is subjected to to a high number of cycles
it might result in a considerable drop in the coordination number below its critical value
and a scattered response of mechanical anisotropy upon entering the semifluidized state.
During the semifluidized state, the unstable contacts govern the system response with a high
number of floating particles, resulting in a large accumulation of shear strain until the particle
network slowly recovers increasing its coordination number towards its critical coordination
number. The qualitative trend between liquefied and non-liquefied states described in terms
of macromechanical and micromechanical quantities sheds light onto the complex nature of
liquefaction in granular materials. The stress threshold pth used to defined the extend of
the semifluidized state might be a good stress indicator at which the shear stiffness and soil
dilation should be reduced to enable the large of shear strain accumulation.

41
3. Stress-strain constitutive modeling
of cyclic shearing response in the
semifluidized state

3.1 Introduction
Flow liquefaction often occurs in very loose granular media subjected to monotonic loading
and is characterized by a sudden loss of strength. In this phenomenon, soil enters a state
of instability, accompanied by a sudden increase in pore water pressure and shear strain.
On the other hand, cyclic mobility can occur in medium to dense granular media, and
is characterized by consecutive cycles of loss and gain of effective stress in cyclic loading,
resulting in temporary loss of stiffness, hence generation and accumulation of shear strains
in each load cycle that essentially happens at the state of very small effective stress. This
state will be addressed same as in Chapter 2, semifluidized, henceforth in this chapter.
The event where the effective stress vanishes for the first time is termed “initial
liquefaction” (Seed and Lee, 1966), which separates the whole liquefaction process into
“pre-liquefaction” stage and “post-liquefaction” stage. Large post-liquefaction shear
deformation is one of the major hazards encountered as a result of flow liquefaction and
cyclic mobility, and it has been extensively studied after observation in the Niigata
earthquake in 1964. Focusing only on cyclic mobility, laboratory studies on undrained
cyclic shearing of sand show excessive accumulation of shear strains in the post-liquefaction
response (Zhang and Wang, 2012; Rouholamin et al., 2017). The main goal of this chapter
is modeling the shear strains that occur in the semifluidized state during the
post-liquefaction cyclic loading in sands.

42
 Several constitutive models have been developed over the years to simulate the
stress-strain response of saturated sand under cyclic loading. These include, but are not
limited to, the generalized plasticity models (Zienkiewicz and Mroz, 1984; Pastor et al.,
1990), hypoplasticity models (von Wolffersdorff, 1996; Niemunis and Herle, 1997), nested
surface models (Elgamal et al., 2002), and bounding surface models (Dafalias and Manzari,
2004; Andrianopoulos et al., 2010; Boulanger and Ziotopoulou, 2013). In the generalized
plasticity model of Pastor et al. (1990), the cyclic mobility is modeled by superseding the
memory of the preceding cycles through the ratio of the maximum mobilized stress. This
has resulted in the model response to be softer than observed from the experiments. In the
hypoplastic model of Niemunis and Herle (1997), the concept of intergranular strain is used
to reduce the soil stiffness upon unloading-reloading scenarios for capturing the cyclic
degradation. The bounding surface plasticity model of Dafalias and Manzari (2004)
simulates the cyclic mobility by incorporating a fabric-dilatancy tensor, enhancing the
contractive behavior upon unloading that follows a dilation state, thereby dragging the
stress path to low effective stresses in undrained cyclic loading. Despite their elegant
mathematical formulations and their success in a number of different aspects of response in
the pre-liquefaction stage, the cyclic stress-strain loops in all of them tend to freeze very
rapidly after the first liquefaction cycle, contrary to what is seen from the experiments.
To model the level of increasing shear strains in cyclic loading, various constitutive
schemes have been proposed. In the nested surface model of Elgamal et al. (2002), when
the stress-ratio crosses the phase transformation line at low mean effective stresses (e.g.,
less than 10 kPa) – referred to as the “neutral phase” – shear strain develops in the form of
a perfectly plastic response, until a boundary defined in deviatoric strain space is reached.
Other models, e.g. Boulanger and Ziotopoulou (2013), have used a stiffness degradation as
a function of cumulative shear strains to achieve the continual cyclic degradation observed
in the experiments.
Extending the original model by Zhang and Wang (2012) in order to incorporate
critical state compliance following various examples (Manzari and Dafalias, 1997; Li and
Dafalias, 2000; Dafalias and Manzari, 2004), Wang et al. (2014) used various novel
concepts such as splitting the dilatancy into reversible (also split in two parts) and

43
irreversible components, and introducing the concept of a volumetric strain threshold
below which the soil is considered liquefied. The model is quite capable of simulating
several data including that of increasing shear strain amplitude successfully, but the
formulation stays in the side of one that is rather complicated.
These approaches have made practical contributions to the modeling of accumulation of
shear strain in cyclic mobility. However, they share a shortcoming common to several similar
models. They use either cumulative shear strain or cumulative irreversible volumetric strain
as part of their formulation that stays as permanent fixtures in the model because there is
no constitutive mechanism to eliminate their influence when it must. For example, if one
changes from undrained cyclic to drained monotonic, the existing cumulative shear strain
will affect the drained response unduly; and furthermore, when cyclic loading resumes after
drainage of a previous undrained cyclic event, these quantities that developed during the
previous cyclic loading will affect the new one as if it were a continuation of the former, not
a realistic consideration.
One may try to delete (forget) the existing cumulative shear or irreversible volumetric
strain upon initiating drained monotonic loading but then again what about if the drained
monotonic loading is very short in duration and cyclic shearing resumes right after; will
these pre-existing cumulative strains begin to accumulate from zero or not? If they do, an
unjustified response discontinuity will occur since a bit before and a bit after the very small
drained monotonic loading, the cumulative strains will be non-zero and zero, respectively.
On the other hand, if the shear strain, for example, is not forgotten when resuming drained
monotonic loading, it will adversely affect the response because its role is related to
undrained cyclic degradation that does not necessarily persist when drained monotonic
loading is resumed. In essence, the problem is that the cumulative shear or irreversible
(plastic) volumetric strain during undrained cyclic loading continues to accumulate and
affect the response unduly in subsequent drained monotonic or even cyclic loading unless a
mechanism for properly readjusting its effect is devised.
In this chapter, the modeling of sand cyclic stress-strain response under constant volume
conditions is explored by developing a novel constitutive tool aimed at degrading the stiffness
and dilatancy when the material enters a semifluidized state. Such state is reached when the

44
mean effective stress p acquires values within the so-called semifluidized range defined by a
relatively very small threshold mean confinement stress pth , i.e. when 0 < p < pth . This
stiffness and dilatancy degradation is removed without discontinuity when p abandons the
semifluidized range. It should be noted that the constant volume condition can (but does
not need to) correspond to the undrained behavior.
To this end, the chapter first evaluates the undrained cyclic simple shear response from
the laboratory test results of Bastidas (2016) to obtain a detailed description of the
semifluidized range. On that basis, a new state internal variable named strain liquefaction
factor is introduced whose constitutive role is to induce the aforementioned degradation,
and is formulated such that it evolves only at low mean effective stresses in the
semifluidized state under constant volume conditions, while it resumes progressively its
initial value upon subsequent loading associated with volume change and starts anew its
role when a new constant volume cyclic loading is imposed.
The proposed formulation is then incorporated into an existing stress-ratio controlled,
critical state compatible, bounding surface plasticity model by Dafalias and Manzari (2004),
that is well established for constitutive modeling of cyclic response of sands in the pre-
liquefaction stage, to simulate large but bounded shear strains in the post-liquefaction stage
of undrained cyclic simple shearing. The role of each one of the new model parameters is
explained through a series of sensitivity analyses. Finally, the improved performance of the
new model with the calibrated additional parameters is illustrated by simulating the pre-
and the post-liquefaction response of cyclic simple shear tests on Ottawa sand and a cyclic
torsional shear test on Toyoura sand.
For definiteness the new model will be referred to as SANISAND-Sf because it belongs
to the SANISAND family of models and incorporates the new constitutive notion of
Semifluidized (Sf) state.

45
3.2 Synthesis of observations from cyclic undrained
simple shear tests
Much research has been carried out to investigate the cyclic shearing response of sands using
laboratory experiments and numerical methods. In this section special attention is given to
evidence from a recent comprehensive database of experimental cyclic simple shear testing
results on Ottawa F-65 sand, and findings from a number of recent Discrete Element Method
(DEM) studies on assemblies of idealized grains.
Bastidas (2016) reported a comprehensive database of undrained cyclic simple shear test
on Ottawa F-65 sand involving cyclic mobility, where details of the pre- and post-liquefaction
response can be explored. Figures 3.1(a) and 3.1(b) show the stress path and stress-strain
loops in one of these tests at Dr =77%. The loading process can be divided into stages of
pre-liquefaction, shown in grey, and post-liquefaction, shown in black. In the pre-liquefaction
stage, the mean effective stress in each cycle progressively diminishes as the cyclic loading
continues. The 3rd cycle of loading shown with a dashed red line is the first cycle in which
the mean effective stress reaches a near-zero value that was introduced in Chapter 2 as the
semifluidized state. Beyond this cycle, the response is termed to be in the post-liquefaction
stage. In this stage, as shown in Figure 3.1(a), the stress path follows almost the same
“butterfly” shape, with the mean effective stress p entering and exiting the semifluidized
range at very small levels of mean effective stress. This is associated with a considerable and
progressive increase of shear strain amplitude between the 3rd and 23rd cycles, both shown
with red lines in Figure 3.1(b).
More details about the last loading cycle of this test are illustrated in Figure 3.2(a) and
3.2(b), with focus on what happens in the semifluidized state of response, where the mean
effective stress p falls below a corresponding threshold pth that in this particular case is
estimated to be around 10 kPa. The state inside the semifluidized range for this load cycle
corresponds to the marked symbols in the stress path from points A to B and C to D in the
increasing and decreasing directions of shear stress, respectively, as shown in Figure 3.2(a).
The corresponding shear strains generated in the semifluidized range of p are denoted by
semifluidized strain γsf , as shown in Figure 3.2(b). The evolution of γsf in each half-cycle

46
during the entire cyclic shearing process is presented in Figure 3.2(c). The level of γsf shows
an increase with the number of cycles, presumably toward a limiting value. This implies
that the rate of γsf should reduce with the number of cycles.
Such increasing large shear deformations in the post-liquefaction stage have been
observed in different undrained cyclic laboratory configurations such as
triaxial (Wichtmann and Triantafyllidis, 2016a; Rouholamin et al., 2017), torsional (Zhang
and Wang, 2012; Wahyudi et al., 2016) and direct simple shear (Bastidas, 2016). The
observations presented in Figure 3.2(c) follow a similar synthesis done by Wang et al.
(2016) on undrained cyclic torsional laboratory tests on Toyoura sand at a Dr =70%. While
a considerable level of information can be obtained from such experimental tests, they only
include the macroscopic measures of response such as stress, strain, density, and pore
pressure. Understanding the mechanical interaction at the particle level is key for
explaining the increase of γsf in undrained cyclic tests. Some meaningful micro-structural
measures of the granular assembly such as particle fabric during the course of loading are
left unexplored due to the equipment limitations in these physical experiments.
In recent years, numerical experiments on idealized particles using DEM have proven very
effective to understand micromechanics behind the complex response of granular material.

pre-liquefaction 3rd cycle

post-liquefaction 23rd cycle
20 20

10 10
t (kPa)

t (kPa)

0 0

-10 -10

-20 -20
0 40 80 120 -5 -2.5 0 2.5 5
sv (kPa) g (%)
(a) (b)

Figure 3.1: Experimental results of an undrained cyclic simple shear test on Ottawa F-65
sand at Dr = 77%: (a) stress-path and (b) stress-strain loops. The red dashed and red
continuous lines correspond to the 3rd and 23rd cycle, respectively, encompassing all cycles
in the post-liquefaction stage. (Data from Bastidas (2016))

47
 10 10
B 10
gsf
C B
5 5 8
C
t (kPa)

t (kPa)
6

|gsf| (%)
0 0
A 4
-5 -5 A
D gsf 2
D
-10 -10 0
0 4 8 12 -5 -2.5 0 2.5 5 0 10 20 30
sv (kPa) g (%) Cycle number
(a) (b) (c)

Figure 3.2: Details of shear strain accumulation in the semifluidized state: (a,b) stress-path
and the corresponding stress-strain loops in the 23rd loading cycle Fig. 3.1, illustrating the
shear strains generated within the semifluidized state, γsf , while increasing and decreasing
the shear stress between points A and B, and points C and D, respectively; and (c) absolute
values of the γsf in the half cycles of all loading cycles

These methods are used to carry out qualitative studies on granular assemblies to study the
evolution of particle fabric, anisotropy, and other important microscopic quantities, during
the course of loading. From the previous Chapter 2, DEM results have consistently shown
that shear strain amplitude increase occurs at low stresses and low coordination number
and can be correlated to appropriately defined quantities at grain scale, calculated via DEM
analysis.
Lastly, the value of pth at present cannot be exactly defined, because the semifluidized
state reflects a very complex situation. One could, for example, use such criteria for pth as the
value of p when the coordination number falls below the value of 4, observed in 3-dimensional
DEM analysis such as in Barrero et al. (2018). However, large shear strain could occur for
larger values of p. Hence, in this chapter the suggestion of pth = 10 kPa is a convenient
estimate that could eventually change if new evidence is provided, with no consequence on
the basic structure of the model.

48
3.3 Constitutive model formulation within the
semifluidized range

3.3.1 Reference constitutive model

The reference constitutive model in this study follows the basic premises of the original two-
surface plasticity model developed by Manzari and Dafalias (1997) and its sequel by Dafalias
and Manzari (2004), that formed the basis of what was later named the SANISAND class
of models (Taiebat and Dafalias, 2008).
The modeling approach follows the framework of bounding surface plasticity with
kinematic hardening of the yield surface (YS) and critical state soil mechanics concepts,
allowing for a unified description at any pressure and density by the same set of model
constants. The SANISAND class includes various extensions (Li and Dafalias, 2000;
Dafalias et al., 2004; Taiebat and Dafalias, 2008; Li and Dafalias, 2012; Dafalias and
Taiebat, 2016). The earlier works of Manzari and Dafalias (1997) and Dafalias and
Manzari (2004) represent the core of the constitutive model and the above-referenced
subsequent works build into the model different constitutive features that can be added to
the original model framework. To involve fewer model parameters and for simplicity, the
version with fabric change effects (Dafalias and Manzari, 2004) and an overshooting
correction scheme as described in Dafalias and Taiebat (2016) has been considered as the
reference soil constitutive model, and from here on it will be referred to as the reference
model. Aside from the two overshooting parameters for overshooting correction with
default values of ēpeq = 0.01% and n = 1, this reference model has 15 parameters listed in
Table 3.1. This version is recently used in a comprehensive study by Ramirez et al.
(2018a). An extensive description of the reference model can be found in the foregoing
references, and here a brief descriptive outline of only some relevant constitutive
ingredients of the reference model will be presented.
The major constitutive ingredients of the model are the use of bounding and dilatancy
surfaces, abbreviated as BS and DS, respectively, in multi-axial deviatoric stress space,
generalizing the peak stress ratio and dilatancy stress ratio (phase transformation line slope)

49
in p-q space, respectively. Both surfaces are made functions of the state parameter ψ (Been
and Jefferies, 1985) such that at the critical state where ψ = 0, the surfaces collapse onto
the fixed critical state surface that generalizes the critical state stress ratio q/p = M in
p-q space. Such dependence of BS on ψ allows the description of the softening response for
denser than critical samples, an idea first promoted by Wood et al. (1994). But even more
important is the dependence of DS on ψ, first introduced by Manzari and Dafalias (1997),
because it is the underlying reason why the model can simulate the response of both dense
and loose samples with the same set of constants while maintaining compatibility with the
premises of Critical State Theory.
Moreover, the model is able to capture the mean effective stress reduction due to the
gradual increase of pore water pressure during cyclic loading under undrained conditions,
such cyclic loading response described by means of kinematic hardening. The plastic
modulus and dilatancy of the model are controlled by two parameters symbolized by h0
and A0 , respectively. To better capture the sand response in cyclic loading, the contraction
tendency during reverse loading following a dilation phase is enhanced by a
fabric-dilatancy tensor, which accounts for the fabric changes during plastic dilation based
on evidence from microscopic studies. This enhances the pore pressure build-up toward the
liquefaction state and allows the model to capture the butterfly shape of the stress path as
observed in the experiments.
The model has proven successful in reproducing the monotonic and pre-liquefaction cyclic
response of sands for a wide range of pressures and densities (Dafalias and Manzari, 2004;
Taiebat et al., 2010; Ramirez et al., 2018a). However, under undrained cyclic simple shear
test, the stress-strain loops of the model in the post-liquefaction stage experience to lock-up
within a few cycles, preventing it to capture the increase of large cyclic shear strains with
the number of cycles as seen in the experiments. This aspect is intended to be improved
with the following proposed generic modification.

3.3.2 Proposed modification: the SANISAND-Sf model

The general idea underlying the proposed SANISAND-Sf model, is to introduce a small
threshold value of mean effective stress p, symbolized by pth , below which the granular

50
assembly is considered to be in a semifluidized state (hence the acronym Sf in the name of
the model), with “very” small plastic modulus and dilatancy ability. By defining the pressure
ratio pr = p/pth , this means that for pr < 1 one should have an increasing amplitude of shear
strain in each load cycle that can be attained by decreasing progressively the already small
value of the plastic modulus due to small p. This can be achieved by introducing an evolving
state internal variable called the “Strain Liquefaction Factor (SLF)” symbolized by l which
evolves continuously from an initial value (usually 0) to a final saturation value (e.g., 1), but
“only” when pr < 1, i.e. during the semifluidized state. This state internal variable will be
used for reducing the aforementioned plastic modulus and dilatancy parameters h0 and A0
inside this state. The important thing is that the scheme to be used in the range pr < 1 will
not significantly affect the response in terms of plastic modulus and dilatancy when we go
outside the semifluidized range, i.e. the SANISAND-Sf model will resume the response of
the reference model by a continuous transition to it.
In order to easily understand the role of the SLF, it is instructive to recall below the
analytical expressions of the plastic modulus Kp and dilatancy D of the reference model:

h i αb − α
: n
1/2 θ
Kp = h0 (2/3) G0 (1 − ch e) (p/pat ) (3.1)
(α − αin ) : n
D = A0 [1 + hz : ni] αdθ − α : n


(3.2)

where, h0 , A0 , G0 and ch are model parameters, e and pat are the void ratio and atmospheric
pressure, respectively, α is the back stress-ratio tensor, αin is the initial value of α at
the initiation of a new loading process taking place when the denominator of Equation 3.1
becomes negative according to the rules discussed by Dafalias (1986), αbθ and αdθ are the
image back stress-ratio tensors on the bounding and dilatancy surfaces, respectively, n is
the loading/unloading direction, and z is the fabric-dilatancy tensor variable. The reader
can refer to Appendix A for a more comprehensive explanation of the foregoing quantities,
while in regards to the present development the important thing in Equations 3.1 and 3.2 is
that Kp and D are proportional to the model parameters h0 and A0 , respectively.
With h00 and A00 denoting the constant parameters that multiply the plastic modulus and
dilatancy, respectively, in the reference model, and h0 and A0 representing their modified

51
and variable values in the SANISAND-Sf model, the evolving SLF l is then used to modify
the ratios h0 /h00 and A0 /A00 based on the expressions

n o
xl
h0 = h00
[1 − h1 − pr i] + fl (3.3)
n o
0 xl
A0 = A0 [1 − h1 − pr i] + fl (3.4)

where the Macauley brackets h i operate according to hAi = A if A > 0 and hAi = 0 if
A ≤ 0, while x is a positive model constant, and fl is a very small number. x and fl will
be discussed later. After entering the semifluidized range of pr < 1, the above equations
allow h0 and A0 to take smaller values than their typical constant values of h00 and A00 in
the reference model, leading to a softer shear modulus and equally smaller dilation, hence
maintaining the same level of volumetric strain rate.
The latter is true because in the reference model the plastic volumetric strain rate is
given by ε̇pv = D|ε̇pq |, with |ε̇pq | the absolute value of the plastic deviatoric strain rate, being
inversely proportional to plastic modulus, thus to h0 , while D is proportional to A0 ; hence,
when h0 and A0 are obtained by h00 and A00 multiplied by the same factor (the term in {}
in Equations 3.3 and 3.4), the ε̇pv remains the same. The foregoing can straightforwardly be
generalized to the multiaxial space where the equivalent plastic shear strain rate substitutes
for the triaxial quantity |ε̇pq |. The constant fl is a very small positive number (e.g., 0.01 in
the present work) and guarantees that h0 /h00 and A0 /A00 stay positive at p = 0.
When pr > 1, given the effect of the Macauley brackets h1 − pr i = 0. Equations 3.3 and
3.4 yield values for h0 and A0 equal to h00 and A00 multiplied by 1 + fl , which is very close to
1. This is the reason we opted to maintain the h0 and A0 notation as the general one for all
p.
On the contrary, when pr < 1 the values of h0 and A0 reduce significantly as l evolves
from 0 to 1 and as pr tends to 0; when pr = 0 one has that h0 = h00 fl and A0 = A00 fl , or
when l = 1 and pr < 1, h0 = h00 {pxr + fl } and A0 = A00 {pxr + fl }. Both scenarios lead to
very small positive values of h0 and A0 . Notice that based on Equation 3.1, Kp = 0 when
p = 0 irrespective of the value of h0 . This simple modification of the h0 and A0 parameters
through the newly proposed state internal variable l will guarantee the increasing amplitude

52
of shear strains in the semifluidized zone in subsequent cycles as l evolves without altering
the response outside the zone. The upper bound of 1 for the variable l will ensure its ultimate
limiting value, hence that of the strain amplitude. Figure 3.3 illustrates the variations of
h0 /h00 and A0 /A00 at different levels of state internal variable l and for different values of pr .
It will be shown subsequently that the x controls the max level of the shear strain
amplitude inside the semifluidized region and as such it can be made a function of other state
variables, for example relative density Dr . A simple linear expression such as x = x0 − cf Dr
can be tested with x0 and cf two constants. In this work, this important issue that may be
more complex than at first thought, will not be considered.
We will now address the very important evolution equation for the SLF l. Recall that
within the semifluidized range, the requirement compatible with Equations 3.3 and 3.4 is that
variable l increases from 0 to 1 during plastic loading. Its value will not affect the response
for pr > 1, as already mentioned, but there is an important point to observe. In subsequent
loading associated with volume change, the sample consolidates and if a constant volume
cyclic loading begins again after the consolidation, one cannot expect the prior acquired
value of l, possibly even l = 1, to have a full impact on this subsequent constant volume
cyclic loading. It is, therefore, necessary to have a mechanism of restoring the value of l back
to its initial value of 0 in a progressive way. Thus, a term of back-to-zero recovery must be
added to the evolution rate of l that will operate during drained loading taking place in a
re-consolidation after the end of a constant volume cyclic one. In order to address these two
requirements for the evolution of l, i.e. reaching a maximum value of 1, and returning to

1
l=0.0
0.8 0.2
h0/h'0, A0/A'0

0.4
0.6
0.6 0.8
1.0
0.4

0.2

0
0 0.2 0.4 0.6 0.8 1
pr

Figure 3.3: Variations of the h0 /h00 and A0 /A00 ratios with pr at different levels of l, illustrating
the upper and lower bounds of these ratios in the semifluidized state

53
0 upon drained (or semi-drained) deformation. The following rate equation of evolution is
proposed:
l˙ = hLi [pcyc cl h1 − pr i(1 − l)nl ] − cr l|ε̇v | (3.5)

where L is the loading index (or plastic multiplier) inside Macauley brackets indicating that
the first term of the right-hand side (RHS) operates only when plastic loading occurs. In
addition, the h1 − pr i indicates that this first term can only operate for pr < 1, i.e. in the
semifluidized state. The cl and nl are two positive model parameters controlling respectively
the pace and the nonlinearity of evolution of l towards 1. The pin is the mean effective stress
at the initiation of a new loading process, with the same criterion of updating as the αin in
the reference model, and the pinr and a are two positive model constants.
To better understand the role of these two parameters, consider the following. In a
constant volume cyclic simple shear loading, when the stress path reaches the “locked”
butterfly shape of subsequent contraction and dilation phases, depending on the level of
cyclic shear stress amplitude, the pin attains a certain value at the reversal points. The
corresponding value of pin in a test with a selected “reference” cyclic shear stress can be
considered as the pinr . Based on this pinr , the parameter a will influence the rate l˙ in
other levels of cyclic shear stress. The above argument is directly applicable to undrained
cyclic simple shear loading in which the locked stress path is almost symmetric. Other
cyclic shearing scenarios that are associated with asymmetric locked stress paths, e.g. cyclic
triaxial shearing, have two different levels of pin in each cycle. Furthermore, for random
cyclic loading the pin acquires different values along with the αin at the initiation of new
loading processes. In those cases, the above two parameters still play the same role, and
it is hoped and expected that this role will be similarly effective under these more general
loading processes as it was under almost symmetric cyclic shearing. This is of course typical
of the calibration of any constitutive model parameter from specific loading data, expected
to be appropriate for more general loading conditions.
The second term of the RHS in Equation 3.5 with the model constant cr provides the
return-to-zero of l in any loading scenario where |ε̇v | =
6 0, i.e. when drainage and re-
consolidation occur, with ε̇v being the total volumetric strain rate. The absolute value of
ε̇v was introduced in order to exclude the possibility of increasing the value of l above 1 in

54
case the drained deformation that may involve shearing, induces dilation (negative value of
ε̇v ). The existence of l itself in the second term of the RHS guarantees that once l = 0, it
stays at zero until the first term of the equation is activated again when 1 − pr > 0 inside
the semifluidized range. Then the evolution of l towards 1 takes place and is not affected by
the second term of the RHS of Equation 3.5 during undrained loading because ε̇v = 0.
Notice that if drained loading takes place at very small values of p such that 1 − pr > 0,
both terms of Equation 3.5 are active with a competing simultaneous evolution of l towards
1 (first term) or 0 (second term), which however cannot move outside its range from 0 to 1
under any circumstances. Such a drained loading at very small p is rather unusual and has
no serious effect in any case. In this work we will be dealing exclusively with simulations of
undrained cyclic loading and so the second return-to-zero term of the RHS of Equation 3.5
will be inactive. Nevertheless, a preliminary qualitative investigation without comparison
with experimental data will be conducted on the effect of return-to-zero term in Equation 3.5,
and accordingly on the ensuing calibration process of the related parameter cr .

3.4 Performance of the proposed modified model

Considering the foregoing modifications, the new model has two sets of additional
parameters: a set of five parameters listed in Table 3.2 that require detailed calibration and
for which sensitivity analysis is carried out, and a set of three parameters listed in Table
3.3 that are set to default values in the absence of detailed data. Details of the calibration
process including the effects of the calibrated additional parameters of the new model are
presented in this section, followed by the performance of the SANISAND-Sf model
compared to a number of different tests.
The performance of the proposed model is evaluated by the success of the simulation of
five undrained cyclic simple shear tests on Ottawa F-65 sand by Bastidas (2016) and one
undrained cyclic torsional test on Toyoura sand by Zhang (1997).
The selected undrained cyclic simple shear tests on Ottawa F-65 sand were carried out
at an initial vertical effective stress σv ' 100 kPa and a relative density of Dr ' 80%. The
selected undrained cyclic torsional test on Toyoura sand was carried out for an initial mean

55
 Table 3.1: Calibrated parameters for reference model

Description Symbol Ottawa F-65 sand∗ Toyoura sand†

Elasticity G0 125 125
ν 0.33 0.05
Critical state M 1.26 1.25
c 0.8 0.712
λc 0.0287 0..019
eref 0.78 0.934
ξ 0.7 0.7
Yield surface m 0.02 0.02
Plastic modulus h00 4.6 9
ch 0.968 0.968
nb 2.3 1.25
Dilatancy A00 0.5 0.4
nd 2.5 0.7
Fabric dilatancy zmax 25 20
cz 500 800
∗
Modified from Ramirez et al. (2018a)
†
Modified from Taiebat et al. (2010)

Table 3.2: Calibrated additional parameters of the new model

Parameter Sensitivity analysis∗ Ottawa F-65 sand Toyoura sand

x 4, 5, 6 5.4 4
cl 100, 200, 300 220 50
pinr 25 kPa 25 kPa 45 kPa
a 0, 2, 4, 8 8 0†
cr 0, 250, 1000 0‡ 0‡
∗
Common values for different sensitivity analyses are shown in bold.
†
To be calibrated upon availability of data for different levels of CSR.
‡
To be calibrated upon availability of detailed data for multiple-liquefaction stages.

56
 Table 3.3: Default additional parameters of the new model

Parameter Default value

pth 10 kPa
fl 0.01
nl 8

stress p ' 100 kPa and a relative density of Dr ' 70%.

For the Ottawa F-65 sand, the reference model parameters are adopted from the
calibration of Ramirez et al. (2018a) based on an extensive database of monotonic and
cyclic shear tests conducted on Ottawa F-65 sand at the University of Colorado Boulder
(Ramirez et al., 2017). Some of the parameters such as ν, h00 , A00 are adjusted from the
previous calibration to better represent the response for the experiments on the specific
batch of Ottawa F-65 sand by Bastidas (2016). The fabric dilatancy tensor zmax is also
modified to bring the stress path near the origin to better address the intended concept of
the semifluidized state in the present study. As for Toyoura sand, the reference model
parameters are adopted from Taiebat et al. (2010). Parameters such as h00 , A00 , nd and zmax
are again modified to better reproduce the experiments on the specific batch of Toyoura
sand by Zhang (1997). The results of the above calibration for the reference model are
shown in Table 3.1.
Using the calibrated parameters of the reference model for Ottawa F-65 sand, and
setting the additional parameters pth , fl and nl at their default values shown in Table 3.3,
the effects of the remaining five additional parameters that require calibration were
evaluated in simulating one of the target undrained cyclic simple shear tests. The results
were illustrated through a number of sensitivity analyses in Section 3.4.1. Then based on
the calibrated/optimized values of these five parameters, the simulation of the modified
SANISAND-Sf model is compared with that of the reference model and the experimental
data for this test, and presented in Section 3.4.2. This is followed by the simulation of the
remaining tests on Ottawa F-65 sand and Toyoura sand with the modified model and
comparing the results with the corresponding experimental data.

57
3.4.1 Effects of the new model parameters

Assuming an initial vertical effective stress of σv,0 = 102 kPa, an initial lateral effective
stress of σh,0 = 51 kPa, and an initial void ratio of e = 0.5836 (equivalent to Dr = 77%
for Ottawa F-65 sand), a number of sensitivity analyses were carried out on simulation of
an undrained cyclic simple shear loading with cyclic stress ratio CSR=τcyc /σv,0 =0.174. The
reference model parameters for these simulations were those of Ottawa F-65 sand listed in
Table 3.1. The additional parameters of the new model that require calibration, i.e. x, cl ,
cr , and a, were examined in a number of sensitivity analyses with the values listed in Table
3.2. The default model parameters were listed in Table 3.3.
Figures 3.4 - 3.5 and 3.6 show results of the sensitivity analyses on the effects of x, cl ,
and cr , respectively, and Figure 3.7 summarizes these sensitivity analyses. Three different
values were used for each of these target parameters in each figure while keeping all other
parameters constant. The simulations continued until reaching l = 0.6. Results are presented
in terms of shear strain γ, shear stress τ , strain liquefaction factor l, absolute value of the
semifluidized shear strain |γsf |, and the number of cycles.
Before going to details of the sensitivity analysis results, let us recall that the increase of
the shear strain amplitude in the post-liquefaction stage, which is governed by the reduced
shear stiffness in each cycle in the semifluidized state, is linked to the evolved value of l. As l
evolves from 0 and tends to values closer to 1, the amplitude of post-liquefaction shear strain
appears to reach a maximum value. The actual maximum value of γsf is only achieved when
l = 1. It would take a lot more cycles than what is shown in these figures to actually reach
l = 1; however, the presented simulation results show that even at l = 0.6 the γsf starts to
approach its limiting value.
Figures 3.4(a)(b)(c) show that x has a direct relation to this maximum value, i.e. as
x increases, the maximum attained level of post-liquefaction shear strain amplitude also
increases. It can also be clearly observed in Figures 3.4(d)(e)(f) that for x = 4, 5, and 6, at
l = 0.6 we reach γsf ' 3.5%, 6%, and 9%, respectively. The effect of x on |γsf | is summarized
in Figure 3.7(a). In fact in each case of x this figure shows the developed |γsf | for different
attained levels of l from 0 to 0.6. Figures 3.4(g)(h)(i) show that with lower levels of x it

58
takes a greater number of cycles to reach l = 0.6 than for higher levels of x.
The pace of evolution for the strain liquefaction factor l is controlled by the model
parameter cl . Figures 3.5(a)(b)(c) show an inverse relation between the value of cl and the
pace of evolution of the post-liquefaction shear strain toward its saturation level, i.e. as
cl increases the pace of evolution of γsf decreases. For these three simulations, the same
level of x = 5 is used, hence as expected at l = 0.6 all three cases reach the same post-
liquefaction shear strain amplitude, or the same γsf ' 6% as shown in Figures 3.5(d)(e)(f).
The effect of parameter cl in the pace of evolution of the semifluidized shear strain can also
be clearly observed in Figures 3.5(g)(h)(i) such that the cases of cl = 100, 200, 300 reach
the γsf ' 6%, in 23, 14, and 10 cycles, respectively; these are summarized in Figure 3.7(b)
to clearly highlight the effect of parameter cl . Again, in each case of cl this figure shows the
developed |γsf | for different attained levels of l from 0 to 0.6.
More specifically, Figures 3.7(a) and 3.7(b) summarize the key observations from the
sensitivity analyses on x and cl . Although both of these parameters affect in a synergistic
way the features of the post-liquefaction cyclic shear strain response, each one plays a
predominant role in regards to one of these features, guiding its calibration process: x
controls the maximum amplitude of γsf , and cl controls the number of cycles in the
post-liquefaction stage before reaching the maximum amplitude of γsf . To address the way
to go about calibrating the return-to-zero parameter cr , Figure 3.6 shows three sets of
simulations with different values for cr .
In these simulations, undrained cyclic simple shear loading leads to a first liquefaction
round and brings SLF to l = 0.6 as shown with grey symbols in Figures 3.6(d)(e)(f). It
takes around 16 cycles of shearing to reach to this level of l, and at this state, the double
amplitude shear strain reaches around 7% (see grey symbols in Figures 3.6(g)(h)(i). This is
followed by a drained re-consolidation stage to bring the vertical effective stress back to its
initial value. Using different levels of cr in this stage of loading leads to different levels of
recovery of l. The corresponding path of return of l to smaller values is not shown in the
figure. This is again followed by another undrained cyclic shearing to a second liquefaction
round and bringing SLF from its returned value at the end of the drained re-consolidation
stage again to l = 0.6 as shown with red symbols in Figures 3.6(d)(e)(f). At this state, the

59
 x=4 x=5 x=6
20 20 20

10 10 10

t (kPa)
t (kPa)

t (kPa)
0 0 0

-10 -10 -10

-20 -20 -20

-2 -1 0 1 2 -4 -2 0 2 4 -6 -3 0 3 6
g (%) g (%) g (%)

(a) (b) (c)

10 10 10

8 8 8

6 6 6
|gsf| (%)

|gsf| (%)

|gsf| (%)
4 4 4

2 2 2

0 0 0
0 0.2 0.4 0.6 0 0.2 0.4 0.6 0 0.2 0.4 0.6
SLF, l SLF, l SLF, l

(d) (e) (f)

10 10 10

8 8 8

6 6 6
|gsf| (%)

|gsf| (%)

|gsf| (%)

4 4 4

2 2 2

0 0 0
0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20
Cycle number Cycle number Cycle number

(g) (h) (i)

Figure 3.4: Illustration of the effect of x on the maximum amplitude of post-liquefaction

shear strain, in simulation of undrained cyclic simple shear loading: (a-c) stress-strain loops,
(d-f) evolution of γsf with the strain liquefaction factor up to l = 0.6, and (g-i) evolution of
the γsf with the number of loading cycles

double amplitude shear strain almost reaches again to the same level of 7% as in the first
liquefaction round. The stress-strain loops and the corresponding evolutions of l, γsf , the
number of cycles for the first and second undrained shearing rounds and the corresponding
post-liquefaction response are presented in different rows of Figure 3.6.
The cr = 0 represents the scenario where the back-to-zero recovery mechanism of SLF
is inactive and therefore l remains at 0.6. In this case, it takes only around 7 cycles in the

60
 cl=100 cl=200 cl=300
20 20 20

10 10 10
t (kPa)

t (kPa)

t (kPa)
0 0 0

-10 -10 -10

-20 -20 -20

-4 -2 0 2 4 -4 -2 0 2 4 -4 -2 0 2 4
g (%) g (%) g (%)

(a) (b) (c)

8 8 8

6 6 6
|gsf| (%)

|gsf| (%)

|gsf| (%)
4 4 4

2 2 2

0 0 0
0 0.2 0.4 0.6 0 0.2 0.4 0.6 0 0.2 0.4 0.6
SLF, l SLF, l SLF, l

(d) (e) (f)

8 8 8

6 6 6
|gsf| (%)

|gsf| (%)

|gsf| (%)

4 4 4

2 2 2

0 0 0
0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25
Cycle number Cycle number Cycle number
(g) (h) (i)

Figure 3.5: Illustration of the effect of cl on the pace of evolution of post-liquefaction shear
strain amplitude toward its saturation level, in simulation of undrained cyclic simple shear
loading: (a-c) stress-strain loops, (d-f) evolution of γsf with the strain liquefaction factor up
to l = 0.6, and (g-i) evolution of the γsf with the number of loading cycles

second round to reach the 7% double amplitude shear strain. For the cases of cr = 250 and
500, the SFL recovers during the drained re-consolidation stage to the values 0.46 and 0.34,
respectively. As a result of this, the corresponding number of cycles in the second round to
reach the 7% double amplitude of shear strain are 19 and 20. The higher the cr , the faster
the pace of recovery of l, and the greater number of cycles to reach the same level of shear
strain in the post-liquefaction of the second round compared to the first round.

61
 cr=0 cr=250 cr=500
20 20 20

10 10 10
t (kPa)

t (kPa)
t (kPa)
0 0 0

-10 -10 -10

-20 -20 -20

-5 -2.5 0 2.5 5 -5 -2.5 0 2.5 5 -5 -2.5 0 2.5 5
g (%) g (%) g (%)
(a) (b) (c)

8 8 8

6 6 6

|gsf| (%)
|gsf| (%)

|gsf| (%)

4 4 4

2 2 2

0 0 0
0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8
SLF, l SLF, l SLF, l
(d) (e) (f)

8 8 8

6 6 |gsf| (%) 6
|gsf| (%)

|gsf| (%)

4 4 4

2 2 2

0 0 0
0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20
Cycle number Cycle number Cycle number
(g) (h) (i)

Figure 3.6: Illustration of the effect of cr on the post-liquefaction response after a drained
re-consolidation stage, where the first undrained cyclic shearing is shown in grey and the
second undrained cyclic shearing is shown in red: (a-c) stress-strain loops, (d-f) evolution of
γsf with the strain liquefaction factor up to l = 0.6, and (g-i) evolution of the γsf with the
number of loading cycles

The outcomes of the sensitivity analysis on cr are summarized in Figure 3.7(c) for reaching
different levels of |γsf |. This figure, in fact, elaborates on the constitutive role of the back-
to-zero recovery parameter cr , and shows a systematic way for calibration of this parameter
provided that such type of experimental data becomes available.
The slope of plots of CSR vs. number of cycles to reach a certain value of shear strain in

62
 10 25 20
l=0.6 |gsf|=6%
l=0.6
8 20 16

Cycle number

Cycle number
5%
6 0.5 15 12 4%
|gsf| (%)

0.5 3%
4 10 8 2%
0.4 0.4 1%
0.3 5 0.3 4
2 0.2
0.2
0.0 0.0 0
0 0
3 4 5 6 7 0 100 200 300 400 0 250 500 750
x cl cr
(a) (b) (c)

Figure 3.7: Summary of the effects of x, cl , and cr on different aspects of the post-liquefaction
response from Figure3.4, 3.5 and 3.6: (a) effect of x on the attained values of |γsf | at different
levels of l, (b) effect of cl on the number of cycles to reach different levels of l, and (c) effect
of cr on the number of cycles to reach different levels of |γsf | in the second liquefaction round

the post-liquefaction stage (CSR-N) can be controlled using parameters pinr and a. These
parameters directly influence the term (pin /pinr )a in Equation 3.5, and therefore the rate of
evolution of l in the semifluidized state.
The effect of this term is elaborated in Figure 3.8 for the plot of CSR vs. number of
cycles to reach 3% of single amplitude shear strain (NγSA =3% ). Here for pinr = 25 kPa the
parameter a has been varied from 0, that is equivalent to (pin /pinr )a = 1 hence no effect of
this term in l,˙ to 8 that is equivalent to different levels of (pin /pinr )a depending on the CSR.
In general, for a non-zero value of a, the lower the CSR, the higher the NγSA =3% . The slope
of CSR-N becomes flatter at higher values of a.
Parameter pth was set at a default value of 10 kPa. Sensitivity analysis (not shown here)
reveals that the value of pth controls the minimum thickness of the cyclic shear stress-strain
loops. It is believed that the value of pth should be very small. However, a default value of
10 kPa appears to provide reasonable results as shown in the next section.

3.4.2 Simulation of undrained cyclic response on Ottawa F-65

and Toyoura sand

Based on the trends obtained in the sensitivity analyses of the previous section, the additional
parameters of the new model that require calibration were optimized for the data of Ottawa
F-65 sand (Bastidas, 2016) using a trial-and-error process. For this purpose, given the

63
 0.24
a=0
0.22 a=2
a=4
0.2 a=8

CSR
Fitted line
0.18

0.16

0.14
10 20 30 40
Number of cycles to gS.A=3%

Figure 3.8: Illustration of the effect of a on the number of cycles to reach 3% of single
amplitude shear strain (NγSA =3% ) in simulation of undrained cyclic simple shear loading
with different level of cyclic stress ratio CSR

corresponding experimental data of undrained cyclic simple shear tests with CSR= 0.174
as shown in Figures3.9(a) and 3.9(b), and more specifically with reference to the stress-
strain loops shown in Figure 3.9(f), the parameters x and cl were calibrated to 5.4 and
220, respectively. Recalling the combined role that parameters pinr and a play in the term
(pin /pinr )a that affects the pace of evolution of SLF in Equation 3.5, the calibrated value of
parameter a depends on the choice of parameter pinr which is a selected reference value of
the mean effective stress at the reversal point of the stress path in the corresponding locked
butterfly loop. Based on pinr = 25 kPa, the post-liquefaction strains at different level of CSR
for this sand are best captured using a = 8. Finally, the cr parameter is set to zero as we do
not have data of multiple liquefaction tests on this soil.
Based on the model parameters presented in Tables 3.1,3.2 and 3.3, results of the
undrained cyclic simple shear test on Ottawa F-65 sand at an initial vertical effective stress
of 102 kPa and relative density of 77% as shown in Figures 3.9(a) and 3.9(b), are compared
with the corresponding simulation results using the reference model as shown in Figures
3.9(c) and 3.9(d), and those from the new model accounting for the semifluidized state as
shown in Figures 3.9(e) and 3.9(f). While both simulations show a similar level of
performance in the pre-liquefaction state and in the overall stress path, the new model
shows a very clear improvement in capturing the accumulation of shear strain amplitudes
in the post-liquefaction cycles as it tends towards a limiting level. Moreover, the
performance of the new model is compared with the experiments at various shear stress
amplitudes until reaching 3% of single-amplitude shear strain in Figures 3.10 and 3.11. The

64
 20 20

10 10

t (kPa)
t (kPa) 0 0

-10 -10
Experiment
-20 -20
0 40 80 120 -5 -2.5 0 2.5 5
sv (kPa) g (%)
(a) (b)

20 20

10 10

t (kPa)
t (kPa)

0 0

-10 -10
Base model
Reference model
-20 -20
0 40 80 120 -5 -2.5 0 2.5 5
sv (kPa) g (%)
(c) (d)

20 20

10 10
t (kPa)
t (kPa)

0 0

-10 -10
Extended model
-20 -20
0 40 80 120 -5 -2.5 0 2.5 5
sv (kPa) g (%)
(e) (f)

Figure 3.9: Stress-path and stress-strain loops of an undrained cyclic simple shear test on
Ottawa F-65 sand at Dr = 77%: (a-b) Experimental results from Bastidas (2016), (c-d)
reference model simulation results, (e-f) extended model accounting for semifluidized strains

corresponding results of CSR versus NγSA =3% are summarized in the CSR-N plot in Figure
3.12; the level of match between the experiments and the simulations in this plot is an
indication of the success of the proposed modification. It should be noted that the
reference model cannot even reach the γSA = 3% in any of these CSR levels, as the
stress-strain loops get locked at smaller levels of shear strain.

65
 Experiment Simulation
20 20

10 10

t (kPa)

t (kPa)
0 0

-10 -10
CSR = 0.16
-20 -20
0 40 80 120 0 40 80 120
sv (kPa) sv (kPa)
20 20

10 10
t (kPa)

t (kPa)
0 0

-10 -10
CSR = 0.174
-20 -20
0 40 80 120 0 40 80 120
sv (kPa) sv (kPa)
20 20

10 10
t (kPa)

t (kPa)

0 0

-10 -10
CSR = 0.192
-20 -20
0 40 80 120 0 40 80 120
sv (kPa) sv (kPa)
30 30
20 20
10 10
t (kPa)

t (kPa)

0 0
-10 -10
-20 -20
CSR = 0.212
-30 -30
0 40 80 120 0 40 80 120
sv (kPa) sv (kPa)
30 30
20 20
10 10
t (kPa)

t (kPa)

0 0
-10 -10
-20 -20
CSR = 0.224
-30 -30
0 40 80 120 0 40 80 120
sv (kPa) sv (kPa)

Figure 3.10: Stress-path for different CSR levels on undrained cyclic simple shear test
on Ottawa F-65 sand at Dr = 77%: Experimental and extended model results, data
from Bastidas (2016)

66
 Experiment Simulation
20 20

10 10

t (kPa)

t (kPa)
0 0

-10 -10
CSR = 0.16
-20 -20
-4 -2 0 2 4 -4 -2 0 2 4
g (%) g (%)
20 20

10 10
t (kPa)

t (kPa)
0 0

-10 -10
CSR = 0.174
-20 -20
-4 -2 0 2 4 -4 -2 0 2 4
g (%) g (%)
20 20

10 10
t (kPa)

t (kPa)

0 0

-10 -10
CSR = 0.192
-20 -20
-4 -2 0 2 4 -4 -2 0 2 4
g (%) g (%)
30 30
20 20
10 10
t (kPa)

t (kPa)

0 0
-10 -10
-20 -20
CSR = 0.212
-30 -30
-4 -2 0 2 4 -4 -2 0 2 4
g (%) g (%)
30 30
20 20
10 10
t (kPa)

t (kPa)

0 0
-10 -10
-20 -20
CSR = 0.224
-30 -30
-4 -2 0 2 4 -4 -2 0 2 4
g (%) g (%)

Figure 3.11: Stress-strain loops for different CSR levels on undrained cyclic simple shear
test on Ottawa F-65 sand at Dr = 77%: Experimental and extended model results, data
from Bastidas (2016)

67
 0.24
Experiment
0.22 Simulation
Fitted line
0.2

CSR
0.18

0.16

0.14
0 10 20 30
Number of cycles to gS.A=3%

Figure 3.12: Comparing the experimental results and model simulations in terms of cyclic
stress ratio versus number of loading cycles to reach 3% single-amplitude of shear strain in
undrained simple shear tests on Ottawa sand F-65 sand at Dr =77%

The performance of the SANISAND-Sf model is also examined by simulating an

undrained cyclic torsional test on Toyoura sand at initial mean effective stress of 100 kPa
and relative density of 70% by Zhang (1997), as shown in Figure 3.13.
The related model parameters are again listed in Tables 3.1, 3.2 and 3.3. The model
shows an excellent performance in capturing both the pre-liquefaction and post-liquefaction

Experiment Simulation
40 40

20 20
t (kPa)

t (kPa)

0 0

-20 -20

-40 -40
0 40 80 120 0 40 80 120
p (kPa) p (kPa)
40 40

20 20
t (kPa)

t (kPa)

0 0

-20 -20

-40 -40
-10 -5 0 5 10 -10 -5 0 5 10
g (%) g (%)
(a) (b)

Figure 3.13: Stress-path and stress-strain loops of an undrained cyclic torsional shear test on
Toyoura sand at Dr = 70%: (a) Experimental results from Zhang (1997), and (b) extended
model accounting for semifluidized strains

68
response of this test. Note that the parameter a and cr were set to 0 in the simulation of
this test. However, upon the availability of data for different levels of CSR and multiple-
liquefaction tests, these parameters can be calibrated to their representative values.

3.5 Conclusion
Post-liquefaction constant volume cyclic shearing response of sands is associated with a
considerable and progressive increase of shear strain amplitudes. This response is linked to
reduced plastic shear modulus and dilation tendency in the semifluidized state at low levels
of mean effective stress. A new state internal variable named strain liquefaction factor is
introduced that evolves from 0 to 1 within the semifluidized range of low mean effective
stresses. Its constitutive role is to reduce the values of parameters controlling the plastic
modulus and dilatancy, while maintaining the same plastic volumetric strain rate, in the
semifluidized range.
The evolution rate equation of the strain liquefaction factor includes a back-to-zero
recovery term under drained loading, that allows the process of its progressive influence on
shear strain amplitude to be repeated upon resuming cyclic loading after drainage.
Although not examined in detail, this is a much sought-after constitutive feature that is
absent from other constitutive models that address the issue of increasing cyclic shear
strain amplitude under constant volume loading by permanently decreasing the stiffness by
means of an ever increasing function, usually of cumulative shear strain. This new
constitutive ingredient of the strain liquefaction factor is added to a reference bounding
surface sand plasticity model (Dafalias and Manzari, 2004) which is the precursor of the
SANISAND family of models, that has proven success in the simulation of monotonic and
cyclic response of sand in the pre-liquefaction state. Consequently, the name
SANISAND-Sf was introduced for the new model indicated by the initials Sf, the
consideration of the response in the semifluidized state.
The new model has two sets of additional parameters: the ones that require calibration
and the ones that can be set at default values. Each parameter has been identified to
be responsible for certain particular characteristic of response that helps its calibration.

69
Performance of both the reference model and the SANISAND-Sf model are then assessed
in the simulation of the stress-path and stress-strain response of an undrained cyclic simple
shear test on Ottawa F-65 sand. While the reference model fails to capture the progressive
increase of shear strain amplitudes in the post-liquefaction stage, simulation of this aspect
of response has been significantly improved in the new extended model.
Simulations in this chapter, have shown the successful performance of the new model in
simulation of stress-strain response of more undrained cyclic simple shear tests on Ottawa
F-65 sand and an undrained cyclic simple torsional shear test on Toyoura sand. Nevertheless
one must observe that while the overall fitting of the CSR-N curve is satisfactory, this is often
achieved by balancing the under and over prediction of N in the pre and post semifluidized
range, respectively. The ideal would be to have separate satisfactory simulations of the
CSR-N curves for these two ranges, and this will be a subject for future investigation.
Sand is a granular medium with no viscous effects like clays, but when it reaches the
semifluidized state it is expected that rate sensitivity of this intermediate state between solid
and fluid material phase will appear. Nevertheless some preliminary investigations by means
of DEM has shown that the rates required for the mechanical response to exhibit noticeable
rate sensitivity are much larger than those typically encountered under cyclic loading. Thus,
it is believed that the rate independence assumed also during the semifluidized state is a
realistic assumption.
Further detailed and targeted experimental or DEM studies might suggest several
improvements in the proposed formulation for the role and evolution of the newly proposed
state internal variable l. In particular, having defined the specific effect of each one of the
introduced new parameters on the response within the semifluidized range, new data may
guide their dependence on relative density, initial pressure and amplitude of the shear
stress. In addition, DEM investigations can offer a better understanding of the
return-to-zero formulation of the strain liquefaction factor, as proposed in Equation 3.5,
during the consolidation stage of loading between different cyclic liquefaction events. A
final note is that the ideas presented by Equations 3.3, 3.4 and 3.5 are generic in concept
and do not apply only to the reference model but to any model that has expressions for
plastic modulus and dilatancy multiplied by coefficients like h0 and A0 .

70
4. Application of the constitutive
model in dynamic continuum
modeling of slope

4.1 Introduction
In geotechnical engineering, numerical modeling has become significantly more popular
with the development of modeling tools and techniques. The key component of various
continuum mechanics-based numerical platforms for solving boundary value problems
(BVP) is properly capturing the material stress-strain response, or the material
constitutive model. In recent decades, many advanced constitutive models have been
developed, evaluated against laboratory element tests, and used in analysis of dynamic
problems related to soil liquefaction. Soil constitutive models are in a wide spectrum of
sophistication. The developers often seek the balance between including sufficient features
to capture the key aspects of the mechanical response of soil and also keeping the models
and their calibration sufficiently simple and straightforward. This is not an easy task,
particularly for a complex natural material like soil.
In the big picture, the goal of modeling in this field is successful simulation of BVP to
provide analysis-based insight for practical cases in geotechnical engineering. As such, the
experimental tests on representative BVP are essential in the evaluation of the numerical
modeling methods, including the role of the material constitutive models and their range of
applicability and limitations. Therefore, this chapter is dedicated to evaluating the
performance of the extended model, previously shown in Chapter 3, on a representative
BVP.

71
 For geotechnical problems, centrifuge tests are of particular interest in providing insights
into response of soil system BVP. The recent, Liquefaction Experiments and Analysis Project
(LEAP) aims specifically at the combined use of advanced centrifuge testing and numerical
modeling of soil liquefaction. The experimental data obtained in this project provided a
unique opportunity to assess the capabilities of a number of numerical modeling techniques
currently available for the analysis of soil liquefaction. In total twenty-four centrifuge tests
were carried out by nine centrifuge facilities around the world, for a prediction exercise in
which the University of British Columbia was invited to participate with other universities as
part of the numerical simulations teams. Then naturally, LEAP is a candidate to evaluate
the extended model capabilities for predicting the soil liquefaction response during cyclic
loading in a BVP.
This chapter consists of simulating the response of a saturated slope subjected to seismic
loading using the reference and extended constitutive models as described in Chapter 3.
Selected numerical and the centrifuge experimental results from LEAP-UCD-2017 (Kutter
et al., 2019b), referred from here on as LEAP, are presented and compared to evaluate
the predictive capabilities of the extended constitutive model for simulating more realistic
boundary value problems and highlighting possible shortcomings in the model formulation.
In particular, the comparison of the results between the reference and extended model is
expected to highlight the contribution of the newly proposed strain liquefaction factor for
modeling the post-liquefaction shear deformations.
To carry out the numerical simulations, the models are implemented in the finite
difference platform FLAC 3D (Itasca, 2012a) for 3-dimensional simulations. This platform
uses an explicit time-integration scheme to solve the equilibrium equation and includes a
fully coupled solid-fluid interaction formulation for simulating the pore pressure build up
and dissipation during and after dynamic excitation. Details of the model implementation
and verification in the finite difference platform are presented in Appendix A.
Section 4.2 presents a brief summary of the LEAP experiments. It also includes a
detailed description of the centrifuge model geometry, material characterization and model
instrumentation presented in the prototype scale. Section 4.3 explains the key aspects of
the finite difference model used for simulation of the LEAP centrifuge experiments. The

72
centrifuge and numerical results are presented and compared in Section 4.4 and the
contribution of the strain liquefaction factor is explored in Section 4.5. Section 4.6 includes
conclusions and important outlooks.

4.2 Liquefaction Experiments and Analysis project:

LEAP
A primary need for all numerical modelers is to have a reliable experimental database to
validate their constitutive models. The need for validation specifically in modeling soil
liquefaction was first acknowledged in the pioneering project VELACS (verification of
liquefaction analysis by centrifuge studies) (Arulanandan and Scott, 1993a,b). In this
project, the experimental results obtained from the tests performed by seven centrifuge
facilities showed the difficulty in obtaining reliable data. Also, the project offered a good
opportunity to verify the accuracy of various constitutive models in simulating soil
liquefaction before the actual experiments were performed.
The next important project on soil liquefaction is the Canadian Liquefaction Experiment
(CANLEX) (Robertson et al., 2000). The main objective of this project was to study the
phenomenon of soil liquefaction by means of in situ and laboratory testing on six sites in
Western Canada. Extensive in situ testing was performed at each site, including standard
penetration tests, cone penetration tests (CPT), seismic CPT, self-boring pressuremeter
testing, and geophysical logging. Laboratory testing was carried out on both reconstituted
and undisturbed samples. The laboratory testing program included triaxial and simple shear
tests under drained and undrained conditions, subjected to monotonic and cyclic loadings.
The outcome of CANLEX project suggested an appropriate site characterization according
to the level of risk.
Following the need for validation, LEAP starts a new international effort to demonstrate
repeatability, variability and sensitivity from the experimental and numerical results. The
LEAP project was carried out by invited groups of leading researchers with expertise in either
experimental modeling or numerical simulation of geotechnical earthquake engineering. The

73
experimental modeling groups were tasked to establish a reliable laboratory and centrifuge
testing database for the evaluation, calibration and validation of constitutive models for
analysis of soil liquefaction.
The Theoretical & Applied Geomechanics (TAG) research group of UBC, along with
another eleven international teams was invited to participate in numerical modeling of the
LEAP element test and centrifuge experiments in 2017-2019. The numerical modeling groups
were tasked to perform Type-B and Type-C predictions of soil liquefaction response during
the centrifuge testing. In a Type-B prediction exercise, the modelers are given the centrifuge
experiment geometry, the dynamic input motion, and the achieved level of density prior to
the shaking (Lambe, 1973). Once the centrifuge experiments were completed, in a Type-C
prediction exercises the modelers are allowed to perform a second round of simulations to
refine their model predictions knowing the actual centrifuge results.
In order to assist the numerical effort, the modelers were provided with a large number of
laboratory tests performed to characterize the physical and mechanical response of Ottawa F-
65 sand and to define the stress-strain response of this soil in cyclic loading. These laboratory
tests were used by the numerical modeling groups to calibrate and validate their constitutive
models. The outcomes of the LEAP project were presented in Yang et al. (2019) and Reyes
et al. (2019).
In order to explore the contribution of the strain liquefaction factor in the BVP results,
the experimental results and Type-C numerical predictions are presented and discussed in
this chapter. The results are limited to centrifuge test, carried out at the University of
California at Davis (UCD) (Carey et al., 2019), as the type of sand used for the centrifuge
test correspond to the same batch of sand used for the element tests shown in Chapter 3.
This type of prediction allows comparison and evaluation of the extended model capabilities
in predicting the undrained soil response when subjected to cyclic loading with knowledge
of the experimental centrifuge test results.

4.2.1 Model geometry and material characterization

The soil used in LEAP was Ottawa F-65 sand, the same type of sand used in Chapter 3.
Ottawa F-65 sand is a clean (less than 5% fines) sub-rounded to sub-angular silica sand

74
classified as SP in the Unified Soil Classification System. The average specific gravity (Gs ≈
2.665 ± 0.012) reported by Kutter et al. (2018) and the grain size parameters specifically
obtained at the centrifuge facility of University of California at Davis (UCD) (Carey et al.,
2019) are summarized in Table 4.1. A large database of laboratory experiments including
triaxial, simple shear and permeability tests are presented in the archive of Carey et al.
(2016). Some of the cyclic simple shear test results found in the archive are the experiments
conducted by Bastidas (2016) previously mentioned in Chapter 3.
The UCD facility reported a dry density of 1658 kg/m3 (Dr = 69%), and the average
minimum and maximum densities, 1,475 and 1,756 kg/m3 , respectively. The baseline
schematic geometry of the centrifuge bucket for shaking in the circumferential direction
shown in Figure 4.1(a), represents a side view of the centrifuge bucket used for LEAP. The
sloping ground is 4 m high at the center of the slope, 20 m long and 9 m wide sand deposit
with a gentle 5◦ slope. The observed shape of the curve of the soil surface in Figure 4.1(a)
progressively straightens with increasing g-field after it starts spinning. At the target
g-field, the unintended curve is eliminated. The saturation of the model is carried out
before spinning using a de-aired viscous pore fluid.

4.2.2 Model instrumentation and recorded data

The locations of the accelerometers and the pore water transducers (denoted with AH and
P) are shown in Figure 4.1(a). In total, LEAP recommended instrumenting the centrifuge
model with twelve horizontal accelerometers, eight pore pressure transducers and two
vertical accelerometers. Horizontal accelerometers AH11 and AH12 located at the base of

Table 4.1: Specific gravity (Kutter et al., 2018) and grain size parameters for Ottawa F-65
sand obtained from (Carey et al., 2019)

Parameter Units Mean

Gs - 2.665
D10 mm 0.13
D30 mm 0.16
D50 mm 0.19
D60 mm 0.20

75
 (a) (b)

Figure 4.1: Schematic geometry for LEAP-UCD-2017 experiment for shaking in the
circumferential direction of the centrifuge. Taken from Kutter et al. (2019a)

the container, measure the acceleration input motion. Horizontal accelerometers through
AH1-AH4 and pore pressure transducers P1-P4 are located at the centre midline to reduce
possible boundary effects from the rigid walls. The distance of 1.5m in the transverse
direction between accelerometers AH1-AH4 and pore pressure transducers P1-P4 shown in
Figure 4.1(b), is intended to reduce the sensor-to-sensor interaction. Horizontal
accelerometers AH5-AH10 and pore pressure transducers P5-P10 are located near the
boundaries to evaluate the effect of the rigid boundaries. The vertical accelerometers AV1
and AV2 located at the top of the container, measure the container rocking and Coriolis
acceleration. The horizontal surface displacements are measured using surface markers

(a) (b)

Figure 4.2: Acceleration time history and response spectrum (5% damped) of the input
harmonic motion at the base of the numerical model

76
placed at the intersecting nodes of a 2-m-by-2-m grid. However, at the UCD centrifuge
facility the limited capacity of the data acquisition system only allowed the recordings at
pore pressure transducers P1-P4, P9-P10, and accelerometers AH1-AH4, AH11-AH12,
AV1-AV2. Moreover, issues measuring the surface marker displacements during the shaking
limited the experimental data only to one surface marker D1 at the middle of the slope.
The target motion used in LEAP is a 1-Hz, 16 cycles, ramped sine wave, with its peak
being the target PGA of the motion. According to the data recordings from the horizontal
accelerometers AH11 and AH12 reported by the UCD facility for the centrifuge test three, the
achieved input acceleration time history with a PGA of 0.193g and the acceleration response
spectra (with 5% damping) are illustrated in Figures 4.2(a) and 4.2(b) (Carey et al., 2019).

4.3 Description of the numerical model

This section begins with a brief introduction to the numerical modeling platform used in
the present chapter. It is followed by a description of the model setup, including the spatial
discretization, mechanical and fluid boundary conditions, simulation procedures related to
initial condition setup, dynamic analysis parameters such as temporal discretization and
numerical damping, and other provisions followed in the numerical simulation.

4.3.1 General numerical platform

The analysis platform used in simulating the centrifuge experiment is the FLAC 3D
software developed by Itasca (2012a). FLAC 3D is a 3-dimensional explicit finite-difference
program for engineering mechanics computation, an extension of FLAC (Itasca, 2012b) to
simulate the behavior of 3-dimensional structures built of soil, rock and other materials
that undergo plastic flow when their yield limits are reached. The spatial discretization of
the continuum domain is represented by polyhedral elements within a 3-dimensional grid
that can be adjusted by the user to fit the shape of the object to be modeled. Each
element behaves according to a prescribed linear or nonlinear stress/strain law in response
to the applied boundary restraints.
The explicit Lagrangian calculation scheme and the mixed-discretization zoning

77
technique are used in FLAC 3D to ensure proper modeling of the plastic flow of the
material. Each element (hexahedron), also referred to as a zone in this program consists of
two overlays of five sub-zones (tetrahedra), and in each sub-zone a constant strain rate is
assumed. Thus, quantities such as displacements and pore pressures are accessible at the
grid points while stresses and strains are only accessible at the sub-zones. Two analysis
configurations exist in FLAC 3D : Fluid-Mechanical Interaction analysis and Dynamic
analysis. The formulation of coupled deformation-fluid diffusion processes in FLAC 3D is
done within the framework of the quasi-static Biot 3-dimensional theory and can be
applied to problems involving single-phase Darcy flow in a porous medium. The
3-dimensional formulation of the fluid-mechanical interaction is the main advantage of
FLAC 3D over F LAC, and it represents the main reason to simulate the centrifuge test
with FLAC 3D . The dynamic analysis option extends the FLAC 3D analysis capability to a
wide range of dynamic problems in disciplines such as earthquake engineering, seismology
and mine rock bursts. The calculation is based on the explicit finite difference scheme to
solve the equation of motion using lumped grid point masses derived from the real density
of surrounding zones. The dynamic feature can be coupled with the fluid-mechanical
interaction feature, which makes FLAC 3D capable of modeling dynamic pore-pressure
generation leading to liquefaction in cyclic loading processes.

4.3.2 Model geometry, initial and boundary conditions, and

input motion

The numerical model prepared to represent the prototype scale of the centrifuge tests
conducted in LEAP, includes a 3-dimensional finite difference mesh with only one zone
along the slope strike direction to model the centrifuge experiment. The mesh is not
allowed to deform in the slope strike direction, and therefore works similar to a plane strain
condition. The mesh density, boundary conditions, and locations of the recording
sensors/control points are all presented in the 2-dimensional view of the model geometry as
shown in Figure 4.3. In this Figure, AH1-AH4, P1-P4, and D1 refer to the horizontal
acceleration, pore pressure, and horizontal displacement control points, respectively.

78
 Water pressure applied on the top nodes
Fixed pore
pressure
D1
Impermeable boundary

Impermeable boundary
Horizontally fixed &
Horizontally fixed &

AH4
P4
AH3
P3
AH2
z P2
AH1
x P1

Dynamic input & Impermeable boundary

Figure 4.3: 2-dimensional view of the grid with the boundary conditions, and the recordings
locations for the pore pressure (P), horizontal acceleration (AH), and horizontal displacement
(D)

The mesh size for the numerical model should be selected to provide accurate wave
transmission. The maximum element length is obtained using the shear wave velocity
measurement prior to the dynamic shaking, 120 m/s (Okamura and Sjafruddin, 2019), and
the maximum frequency content of the input motion conservatively taken as fmax =20 Hz.
Using a factor of 10 to consider soil softening, the maximum element length in the
numerical model should be less than 0.6 m.
The mesh consists of forty zones in the slope dip direction, eight zones in the height
direction, and one zone in the slope strike direction. The dimensions of the zones are 0.5 m
in the slope dip direction and 0.39-0.61 m in the height direction. In total, there are 320
zones and 738 grid points in this model. The grid points at the base of the model are fully
constrained in all three directions, while the grid points on the side walls are constrained
laterally, and the grid points on the surface allow full drainage with fixed values of pore
pressure to model the submerged surface of the slope and replicate the actual centrifuge test.
After establishing the self-weight and the related initial conditions, the recorded acceleration
time history shown in Figure 4.2(a) is applied as the input motion at the base grid points
of the model for simulating the dynamic response of the slope. The details are presented in
the following section.

79
 Horizontal
stress (kPa)
0
2
4
6
8
10
12
14
16
18
20
22

Figure 4.4: Contour of effective horizontal stress before the dynamic shaking

4.3.3 Sequence of analysis

In order to establish a reasonable initial stress state of the model in the prototype scale, the
numerical simulation starts with staged construction of the slope, where the dry soil deposit
is constructed using a layer-by-layer technique and is allowed to establish the geostatic
equilibrium stress state under a gravity of 1 g. In this process, the mechanical boundary
conditions are as shown in Figure 4.3, except for the stress boundary condition on the top
surface of the slope. A simple Mohr-Coulomb material model with the bulk modulus 622
MPa, shear modulus 239 MPa, friction angle 31◦ and dilation angle 0◦ are assigned to the
sand layers to represent their stress-strain response. After the mechanical equilibrium is
achieved for all layers of the slope deposit, the fluid-mechanical interaction is switched on,
and a normal stress gradient representing the target submerged pressures of water is applied
on the top surface of the slope, as shown in Figure 4.3. The pore pressures at the top surface
are fixed to the submerged pressures of water during the analysis. When both mechanical and
fluid equilibrium are obtained, the sand constitutive model is switched to the elastoplastic
model SANISAND with the target density of the soil deposit given by the centrifuge facility.
The effective stress distribution for the horizontal and vertical effective stress prior to the
dynamic shaking are shown in Figures 4.4 and 4.5, respectively.
The dynamic analysis feature embedded in the numerical platform is utilized, and the
“processed” acceleration input motion is applied at the base grid points of the slope. This
processing of the input motion includes baseline correction and filtering of the high frequency

80
 Vertical
stress (kPa)
0
5
10
15
20
25
30
35
40
45
50

Figure 4.5: Contour of effective vertical stress before the dynamic shaking

content above 25 Hz. The simulation continues well beyond the end of the base excitation
to dissipate excess pore pressure and allow for resulting settlement of the soil deposit.
FLAC 3D adopts central finite difference approximation method for the time integration,
and the convergence criterion is conservatively chosen when the maximum unbalanced force
compared to the sum of applied forces is less than 10−6 . In the solid-fluid interaction, the
water bulk modulus is chosen as 2.2×106 kPa. Soil hydraulic conductivity used during
the static stage of the analysis for establishing self-weight was first estimated based on
the average initial void ratio of the centrifuge tests and the constant head permeability
tests conducted by Ghoraiby et al. (2019) as k = 1.15 × 10−4 m/s. Note that variation
of hydraulic conductivity with the initial void ratio was negligible. In dynamic loading, a
Rayleigh damping of 1% with the central frequency of 1 Hz is adopted. During the dynamic
stage of the analyses, the hydraulic conductivity was increased to account for the increase
of soil hydraulic conductivity during liquefaction, as suggested by Taiebat et al. (2007) and
Ueng et al. (2015). The increased value of hydraulic conductivity, k = 2.29 × 10−4 m/s, was
kept constant throughout the dynamic stage.

81
4.4 Comparison between measured and simulated
results
For the Type-B simulations, the information available prior to the dynamic shaking,
including the data provided by the centrifuge facility, such as the soil characterization and
the input motion, were provided to the numerical model to better represent the soil
conditions at the actual centrifuge experiment. As for the constitutive models, the
calibrated set of representative model parameters to reproduce Ottawa F-65 sand response
as shown in Table 3.1 were used for the reference and extended model. The calibrated set
of model parameters for the semifluidized state are shown in Tables 3.2 and 3.3. Results of
the numerical simulations are presented and compared with the experimental data in terms
of acceleration response spectra, pore water pressures, and horizontal displacements at
selected control points.

Acceleration response

Figure 4.6 depicts horizontal acceleration response at certain locations indicated in Figure
4.3. The peaks observed in the acceleration time history represent dilation pulses. The
dilation peaks in Figures 4.6(a)(c)(e) and (g) from the simulation results using the reference
model, appeared to be stronger than those measured in the experiment at all locations.
The simulation results using the reference model shown in Figures 4.7(a)(c)(e) and (g),
underpredicted the experimental spectral acceleration around the predominant frequency
of the input motion (1 Hz). The high spectral acceleration at lower periods suggests the
soil stiffness in the simulation using the reference model might be higher than in the actual
centrifuge experiment.
The peaks observed in Figures 4.6(b)(d)(f) and (h), from the simulations results using the
extended model are also stronger than those measured from the experiment. The simulation
results using the extended model shown in Figures 4.7(b)(d)(f) and (h), matched with high
accuracy the experimental spectral acceleration around the predominant frequency of the
input motion (1 Hz), except for the experimental measurement located at AH4. The spectral

82
acceleration at lower periods for the simulation results using the extended model are closer
to the experimental measurements. This suggests the soil stiffness in the simulation using
the extended model is closer to the actual conditions in the centrifuge experiment.

Pore water pressure response

Figure 4.8 depicts excess pore water pressure at the locations indicated in Figure 4.3.
The pore pressure ratio ru defined, with the excess of pore water pressure and the initial
vertical effective stress measured prior to the dynamic shaking, as ru = ∆u/σz,0 is used to
determine if liquefaction occurred during the dynamic shaking. Figure 4.8 indicates both,
reference and extended model, could capture the trend of pore pressure response observed in
the experiment. Figure 4.8 also suggests liquefaction occurred in the actual experiment and
in both simulations using the reference and extended model. The drops in the excess pore
pressure plot indicate both experiments and simulations show presence of dilation pulses.
The rate of generation of the excess of pore water pressure for both simulations, using the
reference and the extended model, are overestimated. This suggests the model performance
for both, reference and extended model, are poor in capturing the buildup and dissipation
of pore water pressure.
A simple solution is to explore new model parameters to better represent the soil response
in the centrifuge test. However, the new set of model parameters might not accurately
represent the soil response observed in the laboratory test as shown by Ramirez et al. (2018a).
Therefore, the approach to calibrate the constitutive model parameters to represent the soil
response during the centrifuge experiment is not considered because it goes beyond the scope
of work of the present study.
The rate of generation and dissipation of pore water pressure is strongly influenced by
hydraulic conductivity. For these simulations, a constant increased hydraulic conductivity is
used; however, in reality, this value changes throughout the dynamic load and it can increase
up to four times its initial value during soil liquefaction as shown in the work of Ueng et al.
(2015). One general approach suggested by previous researchers to improve the prediction
of rate of generation and dissipation of pore water pressure is to vary hydraulic conductivity
of the soil as a function of time or ru (Shahir et al., 2012). In this study, however, a constant

83
 (a) (b)

(c) (d)

(e) (f)

(g) (h)

Figure 4.6: Experimentally measured and numerically computed acceleration time histories
from accelerometers located at different depths of the centre midline

84
 (a) (b)

(c) (d)

(e) (f)

(g) (h)

Figure 4.7: Experimentally measured and numerically computed response spectra (5%
damped) time histories from accelerometers located at different depths of the centre midline

85
hydraulic conductivity is used to fairly compare and evaluate the shortcomings of models.

Horizontal displacement

Figures 4.9 and 4.10 show the horizontal nodal displacement contours and the displacement
vectors at the end of shaking for the simulations using the reference and the extended model,
respectively.
The maximum displacement vector of 0.092 m recorded from the simulation using the
reference model occurred at the center of the slope, as shown in Figure 4.10. The
displacement vectors in Figure 4.10, suggests the up-slope elements mainly settled and
moved slightly horizontal towards down-slope. The center-slope elements moved
horizontally towards down-slope. The majority of the down-slope elements heaved and
moved horizontally towards the centrifuge box wall. The horizontal displacement at marker
D1 shown in Figure 4.11(a), suggests the simulation using the reference model
underpredicts by about half the measurement obtained in the experiment.
Different from the reference model, the maximum displacement vector of 0.293 m recorded
from the simulation using the extended model occurred at the down-slope elements. This
value is at least three times more than the result obtained using the reference model. Using
the extended model, the pattern of the displacement vectors (see Figure 4.10) is similar to
those shown previously by its former model in Figure 4.9. However, the elements in between
the up-slope and center-slope seemed to heave, meanwhile, the elements in between the
center-slope and down-slope seemed to settle slightly. The horizontal displacement at marker
D1 shown in Figure 4.11(b), suggests the prediction of the simulation using the extended
model comes close to the measurement obtained in the experiment. Figure 4.11(b) is clear
evidence that the contribution of the novel mechanism embedded in the strain liquefaction
factor, aimed to reduce soil stiffness and enable a higher generation shear of strains during
liquefaction states. It is suitable for simulation of dynamic centrifuge tests. To illustrate
the contribution of the strain liquefaction factor to the improved prediction, an additional
detailed analysis is carried out in the following section.

86
 (a) (b)

(c) (d)

(e) (f)

(g) (h)

Figure 4.8: Experimentally measured and numerically computed excess pore water pressure
(EPWP) time histories from pressuremeters located at different depths of the centre midline

87
4.5 Exploring the strain liquefaction factor
To shed light on the strain liquefaction factor (l), and the impact of the semifluidized state
on the horizontal displacements, time histories of variations of mean effective stress p, strain
liquefaction factor l, xz-component of the stress tensor τ , and xz-component of the strain
tensor, γ = 2εxz are presented and discussed.

Horizontal Disp. (m)

Displacement vector (m)

Figure 4.9: Reference model horizontal displacements and displacement vectors at the end
of the dynamic shaking

Horizontal Disp. (m)

Displacement vector (m)

Figure 4.10: Extended model horizontal displacements and displacement vectors at the end
of the dynamic shaking

88
 (a) (b)

Figure 4.11: Experimentally measured and numerically computed horizontal displacements

at centre midline displacement marker

Time history of mean effective stress p

The mean effective stress p is used to illustrate if the material state experiences the
semifluidized state. For this purpose, if p falls below the stress threshold, pth (shown with
the black dotted line in Figure 4.12), the state is in the semifluidized state, and is
highlighted with the grey shadow in Figure 4.12. This figure, shows the evolution of the
mean stress time history until the end of the shaking. Figures 4.12(a) and 4.12(d) illustrate
the evolution for the center-slope elements at the bottom of the submerged deposit.
Figures 4.12(b) and 4.12(e) illustrate the evolution at mid-depth of the submerged deposit.
Figures 4.12(c) and 4.12(f) illustrate the evolution near the top of the submerged deposit.
The red dotted line corresponds to the cyclic stress threshold pinr , included but used later
to highlight important aspects in the evolution of the strain liquefaction.
Results shown in Figures 4.12(a), 4.12(b) and 4.12(c) correspond to the simulation results
using the reference model that excludes the formulation of the semifluidized state. The
results using the reference model revealed the maximum stress peak observed at all depths
occurs near twelve seconds after the input motion is applied. Figures 4.12(d), 4.12(e) and
4.12(f) show the simulation results using the extended model. For all depths, the material
experienced the semifluidized state multiple times. In particular, elements near the top
experienced the semifluidized state for the majority of the duration in the dynamic load.
The results using the extended model suggested that the maximum stress peak happens
near sixteen seconds after the input motion is applied for the bottom elements, and for the

89
 (a) (b) (c)

(d) (e) (f)

Figure 4.12: Numerically computed mean effective stress p at different depths from the centre
midline: (a)(d) bottom, (b)(e)mid-height and (c)(f) surface. The grey shaded region denotes
areas in which the semifluidized state is experienced. The black and red dashed lines are the
values for the pth and pinr , respectively

mid and top elements occurs around twelve seconds. The timing in the maximum stress
peaks coincides with the maximum drops previously observed in Figure 4.8. Then, the
maximum peak stress correspond to dilation pulses in which the soil recovers its stiffness
after a liquefied state.

Time history for strain liquefaction factor l

The time history of the strain liquefaction factor l, at different depths are shown in Figure
4.13. Recall, l is only allowed to evolve when the material experiences the semifluidized
state (e.g. p < pth ). Figures 4.13(a),(b) and 4.13(c) confirms l evolved only inside the
semifluidized state, as intended. However, the evolution of l at different depths is rather
far from homogeneous. For example, take the elements near the soil surface for which the
material state experiences the semifluidized state for the majority of the shaking. In this
particular case, the low value of l is directly influenced by the term pcyc . Remember, pcyc is
defined as (pin /pinr )a , where pin is the mean effective stress at the initiation of a new loading

90
 (a) (b) (c)

Figure 4.13: Numerically computed strain liquefaction factor, l, at different depths from the
centre midline. The grey shaded region denotes areas in which the semifluidized state is
active

process (e.g. when the material is unloaded) and, pinr and α are positive material model
parameters. Therefore, it is fair to assume pin is equal to the peak stress shown previously in
Figure 4.12(f). Clearly from Figure 4.12(f), pin remains below the red dotted-line, pinr . As a
consequence, the top elements with very low pcyc have a very low, or nearly absent, value of
l as seen in Figure 4.13(c).
Take now the bottom elements, the maximum stress peak shown in Figure 4.12(d) is
higher than the cyclic stress threshold, pinr . The resulting pcyc is sufficiently large such that
l evolves significantly as shown on Figure 4.13(a). The reference model does not account
for the semifluidized deformations, therefore the values of l are not shown. It is essential to
understand the impact of l in the stress-strain response to have a general idea of the effects
for this particular BVP. To this end, the shear stress τ and γ are explored separately.

Time history of shear stress τ

Figure 4.14 depicts the time history of the shear stress τ at different depths. Positive shear
stress values occur when the material is sheared in the slope dip direction. Meanwhile,
negative shear stress values occur when the material is sheared in the opposite direction to
the slope dip direction (i.e. towards the up-slope elements). The trends in the shear stress
shown for simulations using both, reference and extended model are similar.
The amplitude of the positive and negative peaks in the time history of the shear stress
for the simulation results using the reference model, is continuously increased until the

91
 (a) (b) (c)

(d) (e) (f)

Figure 4.14: Numerically computed shear stress τ at different depths from the centre midline.
The grey shaded region denotes areas in which the semifluidized state is active

maximum amplitude point at which the amplitude decreases. Using the extended model,
the level of shear stress is affected by the semifluidized state. For example take the bottom
elements, when the material is sheared opposite to the slope dip direction and l has already
evolved similarly as shown in Figure 4.13(a). The model response near the thirteen second
mark indicates a minimum shear stress level near zero, as shown in Figure 4.14(d). This is
expected, because as l is increased the model parameter h0 is reduced, leading to a softer
shear modulus that ultimately reduces the level of the shear stress peaks. However, the level
of positive shear stress peaks is increased presumably due to the slope dip direction.

Time history of shear strain γ

Figure 4.15 shows the time history of shear strain γ at different locations. Figures 4.15(a)
and 4.15(d) illustrate the evolution for the center-slope elements at the bottom of the deposit,
subsequent Figures from 4.15(b) to 4.15(e) illustrates the evolution at a middle depth and
Figures 4.15(c) and 4.15(f) illustrate the evolution near the top of the deposit. The positive
value of γ at the end of the shaking, suggests the material is moved in the slope dip direction.

92
 (a) (b) (c)

(d) (e) (f)

Figure 4.15: Numerically computed shear strain γ at different depths from the centre midline.
The grey shaded region denotes areas in which the semifluidized state is active

The simulation results with the reference model revealed the maximum level of γ = 3%
occur at the middle depth elements, as shown in Figure 4.15(b). As for the simulation
results using the extended model, suggests the strain increments occurred when the material
experiences the semifluidized state. The level of γ obtained at the end of the dynamic input
for the mid and top depth elements are higher than the level obtained using the reference
model. However, the level of γ at the bottom depth is lower than the level obtained with
the reference model. Presumably this is due to the significant drops around the thirteen and
fourteen second point of the shear strain time history as shown in Figure 4.15(d). The results
presented in this chapter suggests the proposed mechanism for l to reduce the soil shear
modulus and dilation is a successful mechanism to improve the reference model capabilities
beyond the first instance of liquefaction during cyclic loading.

93
4.6 Conclusion
The novel extended version of the SANISAND model is used in this chapter to simulate
a boundary value problem, representing the complex behavior of a submerged and sloping
sand soil deposit when subjected to dynamic base excitation. The extended constitutive
model includes an elegant mathematical formulation to reduce the shear stiffness and soil
dilation when the material experiences the semifluidized state to allow for the development
and accumulation of large post-liquefaction shear deformations. In order to evaluate the
performance of the extended model and with reference to the reliable data from one of
the centrifuge experiments performed for the liquefaction experiments and analysis projects
(LEAP), two numerical simulations were performed using the former (referred from here on
as the reference model) and the extended version of SANISAND. To this end, the constitutive
models were implemented in the finite difference platform F LAC 3D to simulate the centrifuge
test performed at the University of California at Davis. Both constitutive models, reference
and extended, take the set of model parameters calibrated based on undrained cyclic simple
shear tests to represent the Ottawa F-65 sand response.
The simulation results revealed that both the reference and extended model are adequate
to capture the soil response in the centrifuge experiment. In particular, the extended model
outperforms the reference model in capturing with high accuracy the acceleration response
of the propagated shear wave. The pore water pressure is captured to a satisfactory level.
However, the poor pore water pressure buildup and dissipation revealed the need for future
improvements. The displacement measurement from the actual centrifuge test corresponds to
one point located at the mid-slope, the ideal would be to have more horizontal displacement
measurements on the slope surface to compare with the simulation results. However, the
horizontal displacement observed for the extended model at the end of dynamic loading is
remarkably close to the recorded measurement from the centrifuge experiment.
In general, the results and comparisons presented in this chapter indicate that the
extended model is suitable for simulating boundary value problems (e.g. centrifuge testing)
and capable of capturing several aspects of the undrained behavior of sand soil response
when subjected to dynamic base excitation. The centrifuge results were valuable in

94
identifying the shortcomings of the extended model to properly capture the pore water
pressure buildup and dissipation. This deficiency can be improved by revisiting the
extended model response to stiffen the material response prior to liquefaction. This is an
area in need of investigation, not explored on this dissertation, for future constitutive
modelers.

95
5. Summary, Conclusions and Future
work

5.1 Summary
This dissertation contributes to the modeling of the cyclic shearing and liquefaction
response of granular soils. Inspired by micromechanics level results of a series of discrete
element simulations for cyclic triaxial shearing of granular assemblies, a useful concept of
semifluidized state is introduced and used later to extend the numerical capabilities of a
reference bounding surface plasticity model for modeling the cyclic shear strains in the
post-liquefaction. This extended model is implemented in a finite difference framework,
using the cutting-plane algorithm with sub-stepping as the stress integration scheme and
other numerical treatments to handle low-stress states and to reduce the integration error.
The outcome of numerical implementation allows the extended model to be used in the
simulation of boundary value problems and future applications in other research contexts.
More detailed summary of different components of the thesis is presented below.
In this dissertation, the discrete element method is used to simulate a series of cyclic
triaxial tests carried out under constant-volume conditions. At the sample generation stage,
the numerical samples composed of different spherical particles are isotropically compressed
with the same initial confinement pressure meanwhile varying their initial density. At the
shearing stage, the numerical samples are subjected to cyclic loading while maintaining
the volume constant. The element simulation results well capture the qualitative trend
observed in laboratory tests on granular material response subjected to cyclic loading. The
element simulation results are shown using quantities to describe the bulk (macroscale) and
the particle (microscale) response. The macroscale results refer to the common triaxial

96
quantities used in laboratory triaxial tests, such as deviatoric and mean stress, axial strain,
and pore pressure ratio. The microscale results are presented using the coordination number
and floating ratio, with three anisotropic magnitude measurements: contact, normal and
tangential. The micro and macro observations are used to define the semifluidized state,
further used to explain the origin of large shear strain developed during cyclic mobility.
The observed stress-strain response inside the semifluidized state is incorporated as an
extension of the SANISAND family of models to account for the semifluidized
deformations. The contribution of the extended model is to introduce a small threshold of
stress, pth , below which the sample is considered to be in a semifluidized state with small
plastic modulus and dilatancy. This is achieved by introducing the strain liquefaction
factor, l, that evolves continuously inside the semifluidized state, reducing the
aforementioned plastic modulus and dilatancy parameters h0 and A0 . Additionally, the
novel strain liquefaction evolution accounts for a healing mechanism to return l to its
initial zero value in any loading scenario under partially or fully-drained conditions. In
order to understand the contribution of each constitutive parameter to describe l, a series
of sensitivity analyses on the new model parameters x, cl , and a are carried out to assess
their effects on the stress-strain response. A comparison between the base and the
extended model shows the improved predictive capabilities in estimates of accumulating
strain during post-liquefaction response. Importantly, the model results have been
validated using the stress-path and stress-strain response in undrained cyclic simple shear
tests on Ottawa F-65 sand response for different shearing levels, and the liquefaction
resistance curve, which represents a more practical approach. The model results are also
validated using the stress-path and stress-strain response in undrained cyclic torsional test
on Toyoura sand response.
Continuing with the new model validation, simulations of a submerged slope subjected
to earthquake loading in centrifuge experimental testing on Ottawa F-65 sand are
performed in Chapter 4. In this chapter, numerical simulations are performed using the
base and extended model parameters calibrated from Chapter 3, which are intended to test
the model implementation performance and accuracy. Specifically, the propagation of the
input shear wave main frequency of 1Hz is compared by means of horizontal acceleration

97
time histories, excess of pore pressure and surface horizontal displacements, measure from
the actual centrifuge experiment. The contribution of the semifluidized state during
dynamic propagation is highlighted by comparing results of the mean stress, strain
liquefaction factor, and shear strain time histories using the extended and reference model.
None of these applications would have been possible without a robust numerical model
implementation.

5.2 Conclusion
Important findings and developments generated via this dissertation are herein highlighted.
The proposed model has a simple formulation for capturing several challenging aspects of
the sand cyclic shearing behavior under undrained conditions. The model formulation is
developed after comparison to the observed experimental results on cyclic shearing on sand
and discrete simulations results. The discrete element simulations are found to be an
extremely powerful numerical tool for exploring granular material response at the particle
level. The numerical results go beyond the classical macroscopic quantities used in triaxial
testing, and use advanced microscopic descriptors to quantify the response of the particle
network.
Broadly, the macroscale results presented herein confirmed the experimental observations
that granular material is susceptible to develop cyclic mobility regardless of its initial level
of density and contact anisotropy. On the microscale, the results allow to understand the
liquefaction phenomena at the particle scale. The coordination number (used to measure
the stability of the particle network) suggests that the average number of contacts in the
particle network are reduced in each cycle of the cyclic shearing under constant volume
conditions. A critical coordination number is defined to establish the boundary between
a stable and unstable network. Liquefaction occurred in unstable particle network with
a sudden drop in the coordination number. With a low average number of contacts, the
complementary floating ratio quantity reveals particles are suspended or floating, resulting
in large accumulation of strains in the granular assembly. The coordination number and
floating ratio lead to a new regime named as the semifluidized state, which is depicted by

98
a low coordination number, as well as by a high floating ratio, resulting in the reduction of
shear stiffness to a very low level. The complementary anisotropic magnitudes are suitable for
understanding the contribution of each anisotropic quantity and their effect on the mobilized
strength. The calculated results suggested the geometric anisotropy magnitude oscillated
between bounds well defined during triaxial compression and extension, whereas the normal
and tangential force anisotropy magnitude plots suggested a variable response during the
semifluidized state. The proposed state offers a new approach of the mechanism behind
granular material liquefaction and gives insight into the origin of the strain accumulation
from a particle point of view.
The continuum model incorporates the semifluidized deformations into an existing
constitutive model by successfully integrating the newly proposed variable: the strain
liquefaction factor. To this end, the formulation of the proposed extended model developed
in Chapter 3 is inspired from the semifluidized state described in Chapter 2. The proposed
formulation is incorporated inside the formulation of the existing reference model
by Dafalias and Manzari (2004). The reference model, known for its excellent performance
in capturing sand monotonic responses for a wide range of confining pressures and initial
densities meanwhile using the same set of parameters. However, the performance of the
stress-strain curve seems to lock-up within a few cycles during the post-liquefaction
response when simulating undrained cyclic simple shear tests. Hence, a major challenge is
to capture the large increase of cyclic shear strains with the number of cycles as seen in the
experiments.
The main contribution of the new proposed variable enables the calculation of the
progressive accumulation of strain, as is observed in post-liquefaction response from
experimental tests. To achieve this, the strain liquefaction factor evolves asymptotically
from zero to one only inside the semifluidized state. Inside the new state, the strain
liquefaction factor decreases the soil plastic modulus and dilatancy capability, allowing the
model to progressively generate strains with increasing number of cycles. However, the
detrimental effect on the plastic modulus and dilatancy is not permanent as is assumed in
other constitutive models. A healing mechanism allows the strain liquefaction factor to
reset to its initial zero value in any loading scenario under partially or fully drained

99
conditions. The five new primary model parameters are the main model parameters to
calibrate using cyclic simple shear or cyclic torsional undrained tests. Secondary model
parameters might be modified according to the modeler’s need. The strain liquefaction
factor seems to be a technique to allow estimates of the strain generated as the cyclic
loading proceeds according to the observed experimental results.
The model implementation in the finite difference framework F LAC 3D , tested with a
rigorous verification process, allows to simulate any laboratory element test, or geotechnical
boundary value problems (e.g., site response analysis, tailing dams, and centrifuge
experiments). In particular, the implemented model is used to predict the deformations
developed during the centrifuge experiment on a submerged slope subjected to dynamic
shaking. The extended model shows a great capability to predict the horizontal
acceleration and horizontal surface displacements measured during centrifuge testing, while
using the same set of model parameters calibrated from laboratory element tests. These
results show that the incorporated strain liquefaction factor improves in capturing several
aspects of the soil response during the centrifuge testing. The findings suggest that the
model implementation is able to handle the geotechnical boundary value problems
simulated through out this dissertation, and more importantly, they show the model
accuracy in predicting sand cyclic response during physical modeling in the centrifuge.

5.3 Recommendations for future research

After studying liquefaction for a few years, many research caveats remain that could be
explored for future research:

• The DEM simulations presented in this work were performed under constant volume
conditions. In order to expand the current findings, it is interesting to couple the
DEM method with lattice Boltzmann to incorporate the fluid material into the
simulations. This will properly capture the laboratory undrained condition, and
enable direct comparison between discrete simulations and laboratory tests.
Furthermore, incorporating the fluid material during undrained simulations will allow
exploration of other types of boundaries beyond rigid walls such as membranes or

100
 flexible walls. However, it is recommended to explore alternative numerical options,
such as parallel computing or mass-scaling, in order to reduce the simulation time.

• The work shown in Chapter 2 covers just few post-liquefaction cycles. Future
research directions could focus on simulating more post-liquefaction cycles with the
hope of exploring the evolution of the macroscopic and microscopic quantities. The
main outcome would consist in obtaining the evolution and level of strain developed
during post-liquefaction cycles. Specifically, exploring the limiting value of the
semifluidized strain for different initial densities could lead to a significant
contribution to the future development of the linear expression x = x0 − cf Dr of the
constitutive formulation shown in Chapter 3.

• The work shown in Chapter 3 covers the constitutive model response for cyclic simple
shear and cyclic torsional shear under constant volume conditions. Future research
could focus on simulating the stress-strain response of cyclic triaxial shearing under
constant volume conditions. It is known that the observed material post-liquefaction
response from laboratory cyclic triaxial tests under undrained conditions suggests
that the stress-strain loops are continuously increasing in half cycle. However, the
strain level developed during triaxial compression usually differs from the one which
is developed during triaxial extension. Taking for example the laboratory test results
shown in Figure 2.1, the cyclic shearing is carried out to the same deviatoric level
during compression and extension. The level of strain developed in each half cycle
during extension seems to be higher than the level of strain developed in each half
cycle during compression, as shown in Figure 2.1(e). The stress-strain loops
generated during post-liquefaction response using the SANISAND model ratchets
towards the triaxial extension in few cycles, since the distance between the stress
point and the bounding surface is smaller during triaxial extension than during
triaxial compression. This asymmetric strain response during post-liquefaction results
into a challenging material response to be reproduced with the current formulation of
SANISAND model. In order to have more representative stress-strain loops during
post-liquefaction response, the linear relation of the distance between the stress point

101
 and the bounding surface could be revised. Research contributions in this topic are
expected to extend the model capabilities in simulating laboratory element response.

• The model has been validated in Chapter 4 with the centrifuge experiment of a
submerged slope subjected to dynamic shaking. However, the numerical model was
prepared to represent the prototype scale of the centrifuge test. Future numerical
simulations could focus in representing the scale model. To this end, one could adjust
the length of the box container as well as follow both construction and spinning
stages occurring in the centrifuge experiment prior to the shaking stage. Finally, the
fluid parameters could be modified in order to better represent represent the viscous
fluid used in the actual experiment. The numerical simulation results obtained by
using the mentioned model scale would enable the direct comparison between
centrifuge and numerical model results. Contributions on this topic are expected to
extend the reliability of the numerical results thanks to the use of the model scale.

102
Bibliography

Adinata, J., Reyes, A., Barrero, A. R., and Taiebat, M. (2018). An investigation on the
effect of bidirectional seismic loading on volumetric strain and surface settlement of sand
deposits. In Fifth geotechnical earthquake engineering and soil dynamics conference, pages
237–246, Austin, TX, US. Earthquake Engineering and Soil Dynamics.

Alarcon-Guzman, A., Leonards, G., and Chameau, J. (1988). Undrained monotonic and
cyclic strength of sands. Journal of geotechnical engineering, 114(10):1089–1109.

Amorosi, A., Boldini, D., and Germano, V. (2008). Implicit integration of a mixed isotropic–
kinematic hardening plasticity model for structured clays. International Journal for
Numerical and Analytical Methods in Geomechanics, 32(10):1173–1203.

Andrade, J. E., Ramos, A. M., and Lizcano, A. (2013). Criterion for flow liquefaction
instability. Acta Geotechnica, 8(5):525–535.

Andrianopoulos, K. I., Papadimitriou, A. G., and Bouckovalas, G. D. (2010).

Explicit integration of bounding surface model for the analysis of earthquake
soil liquefaction. International Journal for Numerical and Analytical Methods in
Geomechanics, 34(15):1586–1614.

Arulanandan, K. and Scott, R. F., editors (1993a). Verification of numerical procedures

for the analysis of soil liquefaction problems, volume 1. International Conference on the
Verification of Numerical Procedures for the Analysis of Soil Liquefaction Problems,
Rotterdam.

Arulanandan, K. and Scott, R. F., editors (1993b). Verification of numerical procedures

for the analysis of soil liquefaction problems, volume 2. International Conference on the

103
 Verification of Numerical Procedures for the Analysis of Soil Liquefaction Problems,
Rotterdam.

Azéma, E., Radjai, F., and Dubois, F. (2013). Packings of irregular polyhedral particles:
Strength, structure, and effects of angularity. Physical Review E, 87(6):062203.

Bardet, J.-P., Mace, N., and Tobita, T. (1999). Liquefaction-induced ground deformation
and failure.

Barrero, A. R., Oquendo, W. F., Taiebat, M., and Lizcano, A. (2018). Cyclic shearing
response of granular material in the semi-fluidized regime. In Fifth geotechnical earthquake
engineering and soil dynamics conference, pages 100–107, Austin, TX, US. Earthquake
Engineering and Soil Dynamics.

Barrero, A. R., Taiebat, M., and Lizcano, A. (2015). Application of an advanced constitutive
model in nonlinear dynamic analysis of tailings dam. In Proceedings of the Sixty Eighth
Canadian Geotechnical Conference, pages Paper ID: 748 , 7 pages, Quebec, QC, Canada.

Bastidas, A. M. P. (2016). Ottawa F-65 Sand Characterization. PhD thesis, University of

California, Davis.

Beaty, M. H. and Byrne, P. M. (2011). Ubcsand constitutive model version 904ar. Itasca
UDM Web Site, 69.

Been, K. and Jefferies, M. G. (1985). A state parameter for sands. Géotechnique, 35(2):99–
112.

Bi, D., Zhang, J., Chakraborty, B., and Behringer, R. P. (2011). Jamming by shear. Nature,
480:355–358.

Boton, M., Azéma, E., Estrada, N., Radjaı̈, F., and Lizcano, A. (2013). Quasistatic rheology
and microstructural description of sheared granular materials composed of platy particles.
Phys. Rev. E, 87:032206.

104
Boulanger, R. and Ziotopoulou, K. (2013). Formulation of a sand plasticity plane-
strain model for earthquake engineering applications. Soil Dynamics and Earthquake
Engineering, 53:254 – 267.

Carey, T. J., Kutter, B. L., Manzari, M. T., and Zeghal, M. (2016). Leap soil properties and
element test data. Accessed: http://doi.org/10.17603/DS2WC7W.

Carey, T. J., Stone, N., Bonab, M. H., and Kutter, B. L. (2019). Leap-ucd-2017 centrifuge
test at university of california, davis. In Kutter, B. L., Manzari, M. T., and Zeghal,
M., editors, Model tests and numerical simulations of liquefaction and lateral spreading:
LEAP-UCD-2017, pages 253–274. Springer.

Casagrande, A. (1976). Liquefaction and cyclic deformation of sands - a critical review. In

Harvard Soil Mechanics Series, number 88.

Castro, G. (1975). Liquefaction and cyclic mobility of saturated sands. Journal of the
Geotechnical Engineering Division, 101(6):551–569.

Cundall, P. and Strack, O. (1983). Modeling of microscopic mechanisms in granular material.

In Jenkins, J. T. and Satake, M., editors, Mechanics of Granular Materials, volume 7 of
Studies in Applied Mechanics, pages 137 – 149. Elsevier.

Cundall, P. A. and Strack, O. D. L. (1979). A discrete numerical model for granular

assemblies. Géotechnique, 29:47–65(18).

da Cruz, F., Emam, S., Prochnow, M., Roux, J.-N., and Chevoir, F. (2005). Rheophysics
of dense granular materials: Discrete simulation of plane shear flows. Physical Review E,
72:021309.

Dafalias, Y. F. (1986). Bounding surface plasticity. i: Mathematical foundation and

hypoplasticity. Journal of Engineering Mechanics, 112(9):966–987.

Dafalias, Y. F. and Manzari, M. T. (2004). Simple plasticity sand model accounting for
fabric change effects. Journal of engineering mechanics, 130(6):622–634.

105
Dafalias, Y. F., Papadimitriou, A. G., and Li, X. S. (2004). Sand plasticity model accounting
for inherent fabric anisotropy. Journal of Engineering Mechanics, 130(11):1319–1333.

Dafalias, Y. F. and Taiebat, M. (2016). Sanisand-z: Zero elastic range sand plasticity model.
Géotechnique, 66(12):999–1013.

Elgamal, A., Yang, Z., and Parra, E. (2002). Computational modeling of cyclic mobility and
post-liquefaction site response. Soil Dynamics and Earthquake Engineering, 22(4):259 –
271.

GDR MiDi (2004). On dense granular flows. The European Physical Journal E, 14(4):341–
365.

Ghoraiby, M. E., Park, H., and Manzari, M. T. (2019). Physical and mechanical properties
of ottawa f-65 sand. In Kutter, B. L., Manzari, M. T., and Zeghal, M., editors, Model Tests
and Numerical Simulations of Liquefaction and Lateral Spreading, pages 41–62. Springer.

Guo, N. and Zhao, J. (2013). The signature of shear-induced anisotropy in granular media.
Computers and Geotechnics, 47:1 – 15.

Huang, X., Kwok, C.-y., Hanley, K. J., and Zhang, Z. (2018). Dem analysis of the onset
of flow deformation of sands: linking monotonic and cyclic undrained behaviours. Acta
Geotechnica, 13(5):1061–1074.

Hwang, J.-I., Kim, C.-Y., Chung, C.-K., and Kim, M.-M. (2006). Viscous fluid characteristics
of liquefied soils and behavior of piles subjected to flow of liquefied soils. Soil Dynamics
and Earthquake Engineering, 26(2):313–323.

Ishihara, K., Tatsuoka, F., and Yasuda, S. (1975). Undrained deformation and liquefaction
of sand under cyclic stresses. Soils and Foundations, 15(1):29–44.

Itasca (2012a). FLAC: Fast Lagrangian Analysis of Continua in 3 Dimensions, Version 5.0.
Itasca Consulting Group, Inc., Minneapolis, Minnesota.

Itasca (2012b). FLAC: Fast Lagrangian Analysis of Continua, Version 7.0. Itasca Consulting
Group, Inc., Minneapolis, Minnesota.

106
Jean, M. and Moreau, J. J. (1987). Dynamics in the presence of unilateral contacts
and dry friction: A numerical approach. In Piero, G. D. and Maceri, F., editors,
Unilateral Problems in Structural Analysis — 2. International Centre for Mechanical
Sciences (Courses and Lectures), volume 304. Springer.

Kramer, S. L. (1996). Geotechnical earthquake engineering. Pearson Education.

Kuhn, M. R. (2017). 4 - loading, movement, and strength. In Kuhn, M. R., editor, Granular
Geomechanics, pages 153 – 227. Elsevier.

Kutter, B. L., Carey, T. J., Hashimoto, T., Zeghal, M., Abdoun, T., Kokkali, P., Madabhushi,
G., Haigh, S. K., d’Arezzo, F. B., Madabhushi, S., Hung, W.-Y., Lee, C.-J., Cheng, H.-C.,
Iai, S., Tobita, T., Ashino, T., Ren, J., Zhou, Y.-G., Chen, Y.-M., Sun, Z.-B., and Manzari,
M. T. (2018). Leap-gwu-2015 experiment specifications, results, and comparisons. Soil
Dynamics and Earthquake Engineering, 113:616 – 628.

Kutter, B. L., Carey, T. J., Stone, N., Bonab, M. H., Manzari, M. T., Zeghal, M., Escoffier,
S., Haigh, S., Madabhushi, G., Hung, W.-Y., Kim, D. S., Kim, N. R., Okamura, M., Tobita,
T., Ueda, K., and Zhou, Y.-G. (2019a). Leap-ucd-2017 v. 1.01 model specifications. In
Kutter, B., Manzari, M., and Zeghal, M., editors, Model Tests and Numerical Simulations
of Liquefaction and Lateral Spreading, pages 3–26. Springer.

Kutter, B. L., Chen, Y.-R., and Shen, C. K. (1994). Triaxial and torsional shear test results
for sand. Contract report, Department of civil and environmental engineering, University
of Califronia, Davis.

Kutter, B. L., Manzari, M. T., and Zeghal, M., editors (2019b). Model tests and numerical
simulations of liquefaction and lateral spreading: LEAP-UCD-2017. Springer.

Kutter, B. L., Manzari, M. T., Zeghal, M., Arduino, P., Barrero, A. R., Carey, T. J.,
Chen, L., Elgamal, A., Ghofrani, A., Montgomery, J., Ozutsumi, O., Qiu, Z., Taiebat,
M., Tobita, T., Travasarou, T., Tsiaousi, D., Ueda, K., Ugalde, J., Wada, T., Yang, M.,
Zheng, B. L., and Ziotopoulou, K. (2019c). Numerical sensitivity study compared to
trend of experiments for leap-ucd-2017. In Kutter, B. L., Manzari, M. T., and Zeghal, M.,

107
 editors, Model Tests and Numerical Simulations of Liquefaction and Lateral Spreading,
pages 209–225. Springer.

Lade, P. V. (1992). Static instability and liquefaction of loose fine sandy slopes. Journal of
Geotechnical Engineering, 118(1):51–71.

Lambe, T. W. (1973). Predictions in soil engineering. Géotechnique, 23(2):151–202.

LeBlanc, C., Hededal, O., and Ibsen, L. B. (2008). A modified critical state two-surface
plasticity model for sand: Theory and implementation. Technical memorandum No. 8.

Lee, J. and Herrmann, H. J. (1993). Angle of repose and angle of marginal stability:
molecular dynamics of granular particles. J. Phy. A, 26(2):373.

Li, X. S. and Dafalias, Y. F. (2000). Dilatancy for cohesionless soils. Géotechnique, 50(4):449–
460.

Li, X. S. and Dafalias, Y. F. (2012). Anisotropic critical state theory: role of fabric. Journal
of Engineering Mechanics, 138(3):263–275.

Li, X. S. and Wang, Y. (1998). Linear representation of steady-state line for sand. Journal
of Geotechnical and Geoenvironmental Engineering, 124(12):1215–1217.

Luding, S. (2008). Cohesive, frictional powders: contact models for tension. Granular Matter,
10(4):235–246.

Manzari, M. T. and Dafalias, Y. F. (1997). A critical state two-surface plasticity model for
sands. Géotechnique, 47(2):255–272.

Manzari, M. T., Ghoraiby, M. E., Zeghal, M., Kutter, B. L., Arduino, P., Barrero, A. R.,
Bilotta, E., Chen, L., Chen, R., Chiaradonna, A., Elgamal, A., Fasano, G., Fukutake,
K., Fuentes, W., Ghofrani, A., Haigh, S., Hung, W.-Y., Ichii, K., Kim, D. S., Kiriyama,
T., Lascarro, C., Madabhushi, M. S. P. G., Mercado, V., Montgomery, J., Okamura, M.,
Ozutsumi, O., Qiu, Z., Taiebat, M., Tobita, T., Travasarou, T., Tsiaousi, D., Ueda, K.,
Ugalde, J., Wada, T., Wang, R., Yang, M., Zhang, J.-M., Zhou, Y.-G., and Ziotopoulou,
K. (2019a). Leap-2017: Comparison of the type-b numerical simulations with centrifuge

108
 test results. In Kutter, B. L., Manzari, M. T., and Zeghal, M., editors, Model Tests and
Numerical Simulations of Liquefaction and Lateral Spreading, pages 175–207. Springer.

Manzari, M. T., Ghoraiby, M. E., Zeghal, M., Kutter, B. L., Arduino, P., Barrero, A. R.,
Bilotta, E., Chen, L., Chen, R., Chiaradonna, A., Elgamal, A., Fasano, G., Fukutake, K.,
Fuentes, W., Ghofrani, A., Ichii, K., Kim, D. S., Kiriyama, T., Lascarro, C., Mercado,
V., Montgomery, J., Ozutsumi, O., Qiu, Z., Taiebat, M., Travasarou, T., Tsiaousi, D.,
Ueda, K., Ugalde, J., Wada, T., Wang, R., Yang, M., Zhang, J.-M., and Ziotopoulou, K.
(2019b). Leap-2017 simulation exercise: Calibration of constitutive models and simulation
of the element tests. In Kutter, B. L., Manzari, M. T., and Zeghal, M., editors, Model
Tests and Numerical Simulations of Liquefaction and Lateral Spreading, pages 153–173.
Springer.

Manzari, M. T. and Prachathananukit, R. (2001). On integration of a cyclic soil plasticity

model. International Journal for Numerical and Analytical Methods in Geomechanics,
25(6):525–549.

Marti, J. and Cundall, P. (1982). Mixed discretization procedure for accurate modelling
of plastic collapse. International Journal for Numerical and Analytical Methods in
Geomechanics, 6(1):129–139.

Nezami, E. G., Hashash, Y. M., Zhao, D., and Ghaboussi, J. (2004). A fast contact detection
algorithm for 3-d discrete element method. Computers and Geotechnics, 31(7):575 – 587.

Niemunis, A. and Herle, I. (1997). Hypoplastic model for cohesionless soils with elastic strain
range. Mechanics of Cohesive-frictional Materials, 2(4):279–299.

Okamura, M. and Sjafruddin, A. N. (2019). Leap-ucd-2017 centrifuge test at ehime

university. In Kutter, B. L., Manzari, M. T., and Zeghal, M., editors, Model tests and
numerical simulations of liquefaction and lateral spreading: LEAP-UCD-2017, pages 275–
287. Springer.

Oquendo, W. F. (2013). Micromechanical Statistical Study of the Critical State in Soil

Mechanics. PhD thesis, Universidad Nacional de Colombia.

109
Oquendo, W. F., Muñoz, J. D., and Radjai, F. (2016). An equation of state for granular
media at the limit state of isotropic compression. EPL (Europhysics Letters), 114(1):14004.

Oquendo, W. F., Muñoz, J. D., and Lizcano, A. (2011). Influence of rotations on the critical
state of soil mechanics. Computer Physics Communications, 182(9):1860 – 1865.

Ortiz, M. and Popov, E. P. (1985). Accuracy and stability of integration algorithms

for elastoplastic constitutive relations. International Journal for Numerical Methods in
Engineering, 21(9):1561–1576.

Ortiz, M. and Simo, J. C. (1986). An analysis of a new class of integration algorithms

for elastoplastic constitutive relations. International Journal for Numerical Methods in
Engineering, 23(3):353–366.

Pastor, M., Zienkiewicz, O. C., and Chan, A. H. C. (1990). Generalized plasticity and the
modelling of soil behaviour. International Journal for Numerical and Analytical Methods
in Geomechanics, 14(3):151–190.

Pöschel, T. and Schwager, T. (2005). Computational Granular Dynamics: Models and

Algorithms. Springer-Verlag Berlin Heidelberg, 1 edition.

Radjaı̈, F. and Dubois, F. (2011). Discrete-element modeling of granular materials. Wiley-

Iste.

Ramirez, J., Badanagki, M., Abkenar, M. R., ElGhoraiby, M., Manzari, M., Dashti, S.,
Barrero, A., Taiebat, M., Ziotopoulou, K., and Liel, A. (2017). Seismic performance of a
layered liquefiable site: Validation of numerical simulations using centrifuge modeling. In
Geotechnical Frontiers 2017, pages 332–341.

Ramirez, J., Barrero, A. R., Chen, L., Dashti, S., Ghofrani, A., Taiebat, M., and Arduino,
P. (2018a). Site response in a layered liquefiable deposit: Evaluation of different numerical
tools and methodologies with centrifuge experimental results. Journal of Geotechnical and
Geoenvironmental Engineering, 144(10):04018073.

110
Ramirez, J., Barrero, A. R., Chen, L., Ghofrani, A., Dashti, S., Taiebat, M., and Arduino,
P. (2018b). Capabilities and limitations of different numerical tools in capturing seismic
site performance in a layered liquefiable site. In Fifth geotechnical earthquake engineering
and soil dynamics conference, pages 79–88, Austin, TX, US. Earthquake Engineering and
Soil Dynamics.

Reyes, A., Yang, M., Barrero, A. R., and Taiebat, M. (Submitted in December
2019). Numerical modeling of soil liquefaction using the sanisand-sf model in the leap
experiments. Soil Dynamics and Earthquake Engineering.

Richard, F., Woods, R., and Hall Jr, J. (1970). Vibration of soils and foundations. Prentice
Hall.

Robertson, P. K., Wride, C. E., List, B. R., Atukorala, U., Biggar, K. W., Byrne, P. M.,
Campanella, R. G., Cathro, D. C., Chan, D. H., Czajewski, K., Finn, W., Gu, W. H.,
Hammamji, Y., Hofmann, B. A., Howie, J. A., Hughes, J., Imrie, A. S., Konrad, J.-M.,
Küpper, A., Law, T., Lord, E. R., Monahan, P. A., Morgenstern, N. R., Phillips, R.,
Piché, R., Plewes, H. D., Scott, D., Sego, D. C., Sobkowicz, J., Stewart, R. A., Watts,
B. D., Woeller, D. J., Youd, T. L., and Zavodni, Z. (2000). The canlex project: summary
and conclusions. Canadian Geotechnical Journal, 37(3):563–591.

Rothenburg, L. and Bathurst, R. (1989). Analytical study of induced anisotropy in idealized

granular materials. Geotechnique, 39(4):601–614.

Rouholamin, M., Bhattacharya, S., and Orense, R. P. (2017). Effect of initial relative density
on the post-liquefaction behaviour of sand. Soil Dynamics and Earthquake Engineering,
97:25 – 36.

Roux, J.-N. and Combe, G. (2002). Quasistatic rheology and the origins of strain. Comptes
Rendus Physique, 3(2):131 – 140.

Seed, B. and Lee, K. L. (1966). Liquefaction of saturated sands during cyclic loading. Journal
of Soil Mechanics & Foundations Div, 92.

111
Shahir, H., Pak, A., Taiebat, M., and Jeremić, B. (2012). Evaluation of variation of
permeability in liquefiable soil under earthquake loading. Computers and Geotechnics,
40:74 – 88.

Simo, J. and Ortiz, M. (1985). A unified approach to finite deformation elastoplastic analysis
based on the use of hyperelastic constitutive equations. Computer methods in applied
mechanics and engineering, 49(2):221–245.

Sloan, S. W., Abbo, A. J., and Sheng, D. (2001). Refined explicit integration of elastoplastic
models with automatic error control. Engineering Computations, 18(1/2):121–194.

Taiebat, M. and Dafalias, Y. F. (2008). Sanisand: simple anisotropic sand plasticity model.
International Journal for Numerical and Analytical Methods in Geomechanics, 32(8):915–
948.

Taiebat, M., Jeremić, B., Dafalias, Y. F., Kaynia, A. M., and Cheng, Z. (2010). Propagation
of seismic waves through liquefied soils. Soil Dynamics and Earthquake Engineering,
30(4):236 – 257.

Taiebat, M., Shahir, H., and Pak, A. (2007). Study of pore pressure variation during
liquefaction using two constitutive models for sand. Soil Dynamics and Earthquake
Engineering, 27(1):60 – 72.

Thornton, C. (2015). Granular Dynamics, Contact Mechanics and Particle System

Simulations: A DEM study, volume 24 of Particle Technology Series. Springer
International Publishing, Cham.

Thornton, C. and Antony, S. (1998). Quasi–static deformation of particulate media.

Philosophical Transactions of the Royal Society of London. Series A: Mathematical,
Physical and Engineering Sciences, 356(1747):2763–2782.

Ueng, T. S., Wang, Z. F., Chu, M. C., and Ge, L. (2015). Tests of permeability in saturated
sand during liquefaction. In Sixth International Conference on Earthquake Geotechnical
Engineering.

112
Vaid, Y., Chung, E., and Kuerbis, R. (1990). Stress path and steady state. Canadian
Geotechnical Journal, 27(1):1–7.

von Wolffersdorff, P.-A. (1996). A hypoplastic relation for granular materials with a
predefined limit state surface. Mechanics of Cohesive-frictional Materials, 1(3):251–271.

Wahyudi, S., Koseki, J., Sato, T., and Chiaro, G. (2016). Multiple-liquefaction behavior
of sand in cyclic simple stacked-ring shear tests. International Journal of Geomechanics,
16(5).

Walton, O. R. and Braun, R. L. (1986). Viscosity, granular-temperature, and stress

calculations for shearing assemblies of inelastic, frictional disks. Journal of Rheology,
30(5):949–980.

Wang, G. and Wei, J. (2016). Microstructure evolution of granular soils in cyclic mobility
and post-liquefaction process. Granular Matter, 18(3):51.

Wang, R., Fu, P., Zhang, J.-M., and Dafalias, Y. F. (2016). Dem study of fabric features
governing undrained post-liquefaction shear deformation of sand. Acta Geotechnica,
11(6):1321–1337.

Wang, R., Zhang, J.-M., and Wang, G. (2014). A unified plasticity model for large post-
liquefaction shear deformation of sand. Computers and Geotechnics, 59:54 – 66.

Wichtmann, T. (2016). Soil behaviour under cyclic loading-experimental observations,

constitutive description and applications. Heft 181, Karlsruher Institut für Technologie.

Wichtmann, T. and Triantafyllidis, T. (2016a). An experimental database for the

development, calibration and verification of constitutive models for sand with focus to
cyclic loading: part i—tests with monotonic loading and stress cycles. Acta Geotechnica,
11(4):739–761.

Wichtmann, T. and Triantafyllidis, T. (2016b). An experimental database for the

development, calibration and verification of constitutive models for sand with focus to

113
 cyclic loading: part ii—tests with strain cycles and combined loading. Acta Geotechnica,
11(4):763–774.

Wood, D. M., Belkheir, K., and Liu, D. F. (1994). Strain softening and state parameter for
sand modeling. Géotechnique, 42(2):335–339.

Yamamuro, J. A. and Covert, K. M. (2001). Monotonic and cyclic liquefaction of very loose
sands with high silt content. Journal of Geotechnical and Geoenvironmental Engineering,
127(4):314–324.

Yang, M., Barrero, A. R., and Taiebat, M. (2019). Application of a sanisand model for
numerical simulations of the leap 2017 experiments. In Kutter, B. L., Manzari, M. T.,
and Zeghal, M., editors, Model tests and numerical simulations of liquefaction and lateral
spreading: LEAP-UCD-2017, pages 563–578. Springer.

Zhang, J.-M. (1997). Cyclic critical stress state theory of sand with its application to
geotechnical problems. PhD thesis, Tokyo Institute of Technology, Tokyo, Japan.

Zhang, J.-M. and Wang, G. (2012). Large post-liquefaction deformation of sand, part i:
physical mechanism, constitutive description and numerical algorithm. Acta Geotechnica,
7(2):69–113.

Zienkiewicz, O. and Mroz, Z. (1984). Generalized plasticity formulation and applications to

geomechanics. Mechanics of engineering materials, 44(3):655–680.

114
Appendices

115
A. Constitutive model
implementation

A.1 Introduction
This Appendix describes the implementation of the SANISAND-Sf model using a multiaxial
stress space. The model is implemented in the finite difference framework F LAC 3D (Itasca,
2012a), using the cutting-plane algorithm with a pressure dependent sub-stepping as the
stress integration scheme, and other numerical treatments to handle low effective stress
states and the mixed-discretization. F LAC 3D is a three-dimensional explicit finite-difference
platform specially designed for geotechnical applications. The model implementation in
F LAC 3D is also compatible with its former 2-dimensional platform, F LAC (Itasca, 2012b).
The verification of the model implementation is evaluated simulating of drained and constant
volume triaxial monotonic compression tests, constant volume cyclic triaxial tests along with
the summarized iso-error maps.

A.2 Model overview

An overlook of the formulation of the reference model (Dafalias and Manzari, 2004) with an
overshooting correction scheme as described in Dafalias and Taiebat (2016) and incorporating
the new contitutive notion of semifluidized state shown in Chapter 3, are presented in this
section. Both stress and strain quantities are assumed to be positive in compression. All the
stress components are considered effective.

116
A.2.1 Elastic and plastic strains

The nonlinear elastic response of the SANISAND model is assumed to be hypoelastic. The
total strain increment ε̇, consists of an elastic component ε̇e and a plastic component ε̇p .
The elastic component can be further decomposed into deviatoric strain ėe and volumetric
strain ε̇ev which are defined by

ṡ ṗ
ėe = , ε̇ev = (A.1)
2G K

where p = tr(σ)/3 is the mean effective stress, with σ being the effective stress tensor,
s = σ − pI is the deviatoric stress tensor, K and G are the hypoelastic bulk and shear
moduli that are calculated according to Richard et al. (1970) and Li and Dafalias (2000) as:

(1/2)
(2.97 − e)2

p 2(1 + ν)
G = G0 patm , K= G (A.2)
(1 + e) pat 3(1 − 2ν)
where G0 is a dimensionless material constant, ν is the Poisson’s ratio, e is the void ratio,
and patm is the atmospheric pressure used for normalization.

A.2.2 Critical state

SANISAND is formulated within the critical state soil mechanics framework. The location
of the critical state line in the e − p space that defines the critical void ratio ec is given by
the power relation after Li and Wang (1998):

 ξ
pc
ec = ec0 − λc (A.3)
pat

where ec0 is the void ratio at pc = 0, λc and ξ are dimensionless positive material constants.
The model uses the concept of the state parameter ψ proposed by Been and Jefferies (1985)
to define the distance between the current void ratio and the critical void ratio:

ψ = e − ec (A.4)

117
A.2.3 Yield surface

The yield surface can be visualized as a small cone in the multiaxial stress, which is expressed
by:
p
f = [(s − pα) : (s − pα)]1/2 − 2/3mp = 0 (A.5)

where the m is a dimensionless positive material constant representing the opening size of
the yield surface and α is the deviatoric back-stress ratio representing the bisector of the
cone. The gradient of the yield surface in the stress space at the current stress state is
obtained from Equation A.5 as

∂f 1 r−α
= n − (n : r)I, n= p (A.6)
∂σ 3 (r − α) : (r − α)

where n is the devitoric unit tensor (n : n = 1) that is normal to f = 0, and r = s/p is the
deviatoric stress ratio tensor. The tensor n in Equation A.6 is used to calculate the Lode
angle θ as
√
cos(3θ) = 6tr[n3 ) (A.7)

The critical stress ratio for a given θ is denoted as Mθ . The value is calculated by interpolating
between the values of Mc at θ = 0 (during triaxial compression) and Me at θ = π/3 (during
triaxial extension), according to

2c Me
Mθ = g(θ, c)Mc , g(θ, c) = , c= (A.8)
(1 + c) − (1 − c) cos 3θ Mc

A.2.4 Critical, dilatancy and bounding surfaces

The model uses three concentric and homologous surfaces: the critical, dilatancy and
bounding surface, that are considered in the deviatoric stress ratio π-plane. The image
points αcθ , αdθ , αbθ on these surfaces are defined by central mapping from the origin along

118
the direction n as

p
αcθ = 2/3αθc n, αθc = g(θ, c)M − m (A.9)
p
αdθ = 2/3αθd n, αθd = g(θ, c)M exp(nd ψ) − m (A.10)
p
αbθ = 2/3αθb n, αθb = g(θ, c)M exp(−nb ψ) − m (A.11)

where M, nd and nb are dimensionless positive material constants.

A.2.5 Plastic flow

The model includes a non-associative flow rule allowing realistic evaluations of plastic strain
increments according to
ε̇p = hLiR (A.12)

where L is the loading index calculated later in Section A.3.2, R is the direction of the ε̇p
and h i is the MacCauley brackets such that hLi = L if L > 0 and hLi = 0 if L ≤ 0. The
deviatoric part R0 of the R tensor is defined to be normal to the critical surface at the image
point αcθ . The volumetric part of the R tensor is defined by the scalar dilatancy coefficient
D. The R tensor is given by
R = R0 + 1/3DI (A.13)

where
R0 = Bn − C(n2 − 1/3I), (A.14)

B and C are defined as

r
31−c 31−c
B =1+ g(θ, c) cos 3θ, C=3 g(θ, c). (A.15)
2 c 2 c

The scalar dilatancy coefficient D is defined as

D = Ad (αdθ − α) : n (A.16)

119
where Ad is defined as
Ad = A0 (1 + hz : ni) (A.17)

and A0 is

n o
xl
A0 = A00 [1 − h1 − pr i] + fl (A.18)

where l is the strain liquefaction factor, pr = p/pth is the pressure ratio, A00 , x, fl and pth are
positive model parameters. The fabric dilatancy internal variable z is introduced to account
for the effect of fabric changes during the reverse loading followed by a plastic dilatant phase.

A.2.6 Evolution laws

The model has three internal variables; α, z and l. The evolution law of α is a function of
the distance between of image back-stress ratio tensor on the bounding surface and current
back-stress ratio in the form of

α̇ = hLi2/3h(αbθ − α) (A.19)

where h is the hardening coefficient. In order to handle efficiently the nonlinear response
and the reverse loading, the hardening coefficient is a function of e, p and α as

b0
h= (A.20)
(α − αin ) : n

b0 = G0 h0 (1 − ch e)(p/patm )−1/2 (A.21)

in which h0 is modified accordingly to

n o
xl
h0 = h00 [1 − h1 − pr i] + fl (A.22)

where h00 is a positive model parameter, and αin is the value of α at initiation of a new
loading process and it must be updated following the description given in Section A.2.7.

120
The variable ch is a positive material constant.
The evolution law for the fabric-dilatancy tensor z is introduced by

ż = −cz h−hLiDi(zmax n + z) (A.23)

where cz and zmax are positive model paremeters. Ultimately, the proposed evolution law for
the strain liquefaction factor is

l˙ = hLi [pcyc cl h1 − pr i(1 − l)nl ] − cr l|ε̇v | (A.24)

where pcyc = (pin /pinr )a is the cyclic ratio, and cl , cr , pinr and a are positive model parameters.

A.2.7 Treatment for overshooting response

The presence of the reference back-stress ratio tensor αin is necessary to obtain a smooth
transition at initiation of an elasto-plastic loading event, that is when the reference back-
stress ratio tensor takes the value of the current back-stress ratio tensor as αin = α, it also
creates the so-called overshooting response during loading and re-loading scenarios associated
with its updating. Loading and re-loading scenarios are very common in any form of cyclic
loading, e.g., during earthquake loading. Overshooting is a well known phenomenon for most
of the constitutive models using the bounding surface formulation. In order to prevent the
model from overshooting, the smooth transition from αin to α proposed by Dafalias and
Taiebat (2016) is incorporated to the model formulation.
To this end, when the aforementioned signal (α − αin ) : n ≤ 0 for updating is received (that
is whenever the denominator in Equation A.20 becomes negative), the value of αin is weighted
between the current α and the previous value of the reference back-stress ratio tensor, α∗in
according to

αin = fα α∗in + (1 − fα ) αn (A.25)

where fα is function of the equivalent deviatoric plastic strain epeq , obtained by integration
p
of (2/3)ėp : ėp , and the deviatoric strain threshold ēpeq . The function fα takes the values

121
between 0 and 1 according to

n 
fα = 1 − epeq /ēpeq

(A.26)

where n is a material constant (by default = 1).

A.3 Numerical implementation

The constitutive model is implemented into the finite difference framework, using the
cutting-plane algorithm proposed by Simo and Ortiz (1985) with a pressure dependent
sub-stepping as described in LeBlanc et al. (2008) for the stress integration scheme. A
numerical treatment for zero effective stress state is adopted as in LeBlanc et al. (2008)
and the yield-surface drift correction by Sloan et al. (2001) are added to the model
implementation to avoid numerical difficulties at low stress levels and to correct the error
associated to the mixed-discretization, respectively. In the following sections the stress
integration scheme, sub-stepping and several numerical treatments used in the numerical
implementation are described in detail. The equations are written such that they can be
implemented into any finite element or finite difference platform. There is a small section
dedicated to the model implementation treatments associated exclusively to the framework
of FLAC 3D and FLAC. In what follows, the subscripts n and n + 1 are the current step
and next strep, respectively.

A.3.1 Pressure dependent sub-stepping

In general, explicit integration methods are prone to face numerical stability challenges for
large strain increments. It is well known that implicit integration schemes are unconditionally
stable regardless of the size of strain increment. However, for implicit integrations there is a
high computational cost (i.e., long simulation time) and usually their complex formulation
make the explicit approach a suitable solution for practical purposes. Therefore, in order
to use an explicit integration we have to reduce the global strain increment given by the

122
numerical platform to assure the numerical convergence of the integration method. To
avoid possible non-converging scenarios due to large strain increments, LeBlanc et al. (2008)
proposed a pressure dependent sub-stepping. At the beginning of each step, the strain
increment ∆εn+1 is divided into several strain sub-increments k. The total number of sub-
increments is determined based on the strain tolerance criterion m . The idea consist of
diving the strain increment ∆εn+1 into k number of sub-increments if the magnitude of
trial
the elastic trial stress increment |∆σn+1 | exceeds the current mean effective stress pn by the
trial
correspoding tolerance as |∆σn+1 |/pn > s . This process is continued, and k keeps increasing
trial
until the new strain increment satisfy |∆σn+1 |/(kpn ) ≤ s .
Due to the conical wedge shape of yield surface, it is obvious to expect a high number of
sub-increments near the yield surface apex, or at low mean pressures. Therefore, the mean
effective stress level inversely affect the number of sub-increments, 1/p ∝ k. The main steps
of the algorithm are as follow:

1. Initialize the local number of sub-increments k = 1.

2. Calculate sub-increments of the void ratio ekn+1 , the increment of volumetric strain
(∆εv )kn+1 and deviatoric strain (∆e)kn+1

ekn+1 = (1 + en ) exp[tr(∆εn+1 )/k] (A.27)

(∆εv )kn+1 = tr(∆εn+1 )/k (A.28)

1
∆ekn+1 = ∆εn+1 /k − (∆εkv )n+1 I (A.29)
3

3. Calculate the trial mean effective stress ptrial trial

n+1 and the trial deviatoric stress sn+1 , as

ptrial k


n+1 = g (∆εv )n+1 (A.30)
k 2
2
1/2 (2.97 − en+1 )


1/2 1+ν
g x = pn + G0 pat x (A.31)
3(1 − 2ν) (1 + ekn+1 )
k
strial
n+1 = sn + 2Gn+1 (∆e)n+1 (A.32)

4. Calculate the allowable magnitude of stress increment with the current mean effective

123
 stress and the sub-stepping tolerance s and the magnitude of the trial elastic prediction
trial
of stress increment |∆σn+1 |, as

σ limit = pn s (A.33)
trial
|∆σn+1 | = |strial trial
n+1 + pn+1 I| − σn (A.34)

trial
if |∆σn+1 | > σ limit , the number of sub-increments is doubled k = 2k, go to step 2; else
go to step 5.

5. Update the volumetric strain increment (∆εv )n+1 and deviatoric strain increment
∆en+1

(∆εv )n+1 = (∆εv )kn+1 (A.35)

∆en+1 = ∆ekn+1 (A.36)

A.3.2 Stress integration scheme

The cutting-plane algorithm was chosen for its simplicity and accuracy, and being a semi-
explicit integration scheme it requires several iterations to obtain the final solution. The
main concept of the cutting-plane algorithm is to first make an elastic estimate of the stress
increment and then bring the stress back to the solution through several iterations by using
a first order Taylor expansion of the consistency condition (f˙ = 0) during each iteration.
The main steps of the algorithm are as follows:

1. Initialize the local iteration number i, plastic volumetric strain ∆εpv and deviatoric
strain ∆ep increments

i = 0; (∆εpv )in+1 = 0; (∆ep )in+1 = O (A.37)

where O is the second order zero tensor.

124
2. Calculate the void ratio e at the next step as

en+1 = (1 + en ) exp[tr(∆εn+1 )] (A.38)

3. Perform an elastic prediction for the mean stress p and deviatoric stress s components
and consider the value for the back-stress ratio tensor of the current step the same for
the next step as

pin+1 = g (∆εv )n+1

(A.39)
 2
2

1/2 1+ν 1/2 (2.97 − en+1 )
g x = pn + G0 pat x (A.40)
3(1 − 2ν) (1 + en+1 )
sin+1 = sn + 2Gin+1 ∆en+1 (A.41)

αin+1 = αn (A.42)

4. Evaluate the material state on the yield surface to determine whether plastic loading
occurs

p
f (sin+1 , pin+1 , αin+1 ) = [(sin+1 −pin+1 αin+1 ) : (sin+1 −pin+1 αin+1 )]1/2 − 2/3mpin+1 (A.43)

if the evaluated material state, f (sin+1 , pin+1 , αin+1 ), is higher than the mean stress
reduced by the yield surface tolerance f as f (sin+1 , pin+1 , αin+1 )/pin+1 > f , plastic
loading is induced, go to step 4; else, the material state remains inside the yield surface.
The material response inside the yield surface is assumed to be elastic, go to step 8.

5. Calculate the loading index increment and update the loading index as

i f (sin+1 , pin+1 , αin+1 )

∆L =  i (A.44)
∂f
: (Ce )in+1 : Rin+1 + 23 pin+1 hin+1 (αbθ − α)in+1 : nin+1
∂σ n+1
Li+1 = Li + ∆Li


(A.45)

125
6. Update stress, strain and internal variables

(∆εpv )i+1
n+1 = L
i+1 i
Dn+1 (A.46)

(∆ep )i+1
n+1 = L
i+1
(R0 )in+1 (A.47)

pi+1 p i+1


n+1 = g (∆εv )n+1 − (∆εv )n+1 (A.48)

si+1 i p i+1
 
n+1 = sn + 2G n+1 ∆e n+1 − (∆e )n+1 (A.49)

αi+1
n+1 = αn + L
i+1
(2/3)hin+1 (αbθ − α)in+1 (A.50)
i+1
= zn − cz −(∆εpv )i+1 i i


zn+1 n+1 (zmax nn+1 + zn+1 ) (A.51)
i+1 i
= ln + Li+1 pcyc cl h1 − pr i(1 − ln+1 i
)nl − cr ln+1
 
ln+1 |(∆εv )n+1 | (A.52)

7. Check the residual on the yield surface to determine if its lower than f , as

p
f (si+1 i+1 i+1 i+1 i+1 i+1 i+1 i+1 i+1 1/2
n+1 , pn+1 , αn+1 ) = [(sn+1 −pn+1 αn+1 ) : (sn+1 −pn+1 αn+1 )] − 2/3mpi+1
n+1 (A.53)

if f (si+1 i+1 i+1 i+1

n+1 , pn+1 , αn+1 )/pn+1 < f , convergence is reached go to step 8; else i = i + 1 and

go to step 5.

8. Update the stress, strain and internal variables

(∆εpv )n+1 = (∆εpv )i+1

n+1 (A.54)

(∆ep )n+1 = (∆ep )i+1

n+1 (A.55)

pn+1 = pi+1
n+1 (A.56)

sn+1 = si+1
n+1 (A.57)

αn+1 = αi+1
n+1 (A.58)
i+1
zn+1 = zn+1 (A.59)
i+1
ln+1 = ln+1 (A.60)

126
A.3.3 Numerical treatment for low effective stress state

In the implementation of the model, to avoid numerical difficulties at low effective stress
during soil liquefaction a minimum effective stress pmin and minimum deviatoric stress smin
are defined using the stress tolerance m and patm according to LeBlanc et al. (2008) as

pmin = patm m (A.61)

 
patm m
smin = sn+1 (A.62)
pn+1

where pn+1 and sn+1 are the integrated mean effective stress and deviatoric stress,
respectively. The idea of this approach is to bring the stress back to the levels of pmin and
smin . The main steps of this approach are as follows:

1. Calculate the minimum levels of mean effective stress pmin and deviatoric stress smin as

pmin = patm m (A.63)

 
patm m
smin = sn+1 (A.64)
pn+1

2. Update the material state if pn+1 < pmin , as

pn+1 = pmin (A.65)

sn+1 = smin (A.66)

The effectiveness of this approach has been verified in Section A.4.2.

A.3.4 Numerical treatment for mixed-discretazation

As far the appendix A goes, a detailed description of the constitutive model in multi-axial
formulation is given in section A.2. In order to verify the solution of the integrated model
equations a series of laboratory element tests are simulated. Finally, the robustness of the
model implementation and the optimization of the implementation parameters are shown
in section A.4. FLAC 3D and FLAC uses a mixed discretization technique in which each

127
hexahedra zone (or quadrilateral zone) is divided internally by five tetrahedron (or two
triangles) in two overlaid sets of constant strain rate elements. The main idea of this
technique is to share the volumetric stress and strain components over all sub-zones, while
the deviatoric components are treated separately for each hexahedra (Marti and Cundall,
1982). This implies the stress state for any sub-zone is unlikely to satisfy the consistency
condition of elasto-plastic models, as a consequence the new integrated material state will
not necessarily lie on the yield surface. To bring the stress state back to the yield surface,
an additional step is added after the integration scheme in the model implementation. An
iterative drift correction proposed by Sloan et al. (2001) applied at the end of all sub-zones
operations, is adopted to ensure the average stress and internal variables satisfy the
consistency condition. The main steps of this algorithm are explained as followed:

1. Initialize the local iteration counter j, stress and internal variables

j=1 (A.67)

pjn+1 = pn+1 (A.68)

sjn+1 = sn+1 (A.69)

αjn+1 = αn+1 (A.70)

2. Calculated the plastic multiplier L, trial stress components and trial internal variables
as

f (sjn+1 , pin+1 , αjn+1 )

Lj =  j (A.71)
∂f
: (Ce )jn+1 : Rjn+1 + 23 pjn+1 hjn+1 (αbθ − α)jn+1 : njn+1
∂σ n+1

σ trial = (sjn+1 + pjn+1 I) − Lj (Ce )jn+1 : Rjn+1 (A.72)

αtrial = (αjn+1 ) + Lj (2/3)hjn+1 (αbθ − α)jn+1 (A.73)

3. Evaluate if the trial stress and trial internal variables are closer to the yield surface
than the previous values, |f (strial , ptrial , αtrial )| < |f (sj , pj , αj )|, go to step 4; else, the

128
 trial stress and trial internal variables must corrected as

f (sjn+1 , pin+1 , αjn+1 )

Lj =  j (A.74)
∂f
: (Ce )jn+1 : Rjn+1 + 23 pjn+1 hjn+1 (αbθ − α)jn+1 : njn+1
∂σ n+1
 j
∂f
σ trial
= (sjn+1 + pjn+1 I) −L j
(A.75)
∂σ n+1

αtrial = (αjn+1 ) (A.76)

4. Determine if the residual of the trial stress and trial internal variable evaluated are
lower than the drift tolerance F T OL, |f (strial , ptrial , αtrial )| ≤ F T OL, go to step 5; else,
update the stress components, internal variables and local iteration counter as

j =j+1 (A.77)

pjn+1 = ptrial (A.78)

sjn+1 = strial (A.79)

αjn+1 = αtrial (A.80)

and go to step 2.

5. Update the stress components and internal variables

pn+1 = ptrial (A.81)

sn+1 = strial (A.82)

αn+1 = αtrial (A.83)

The corrected stress and internal variables are assigned to the zone, hence at the beginning
of each step all sub-zones share the same volumetric and deviatoric components for both
stress and internal variables.

129
A.4 Verification of the model implementation
The verification process of the model implementation, the numerical treatments for
overshooting response, sub-stepping and low-effective stress states are here evaluated by
conducting a series of triaxial element simulations. All the simulations results presented in
this section were carried out using one hexahedron element and one quadrilateral element
for the FLAC 3D and FLAC implementation, respectively.

A.4.1 Verification of reference model implementation

This section presents the verification of the numerical implementation of the reference
model. The verification process builds on the work of Taiebat et al. (2010) to implement
the reference model usiung a stand-alone constitutive driver. The model implementations
using the constitutive driver, FLAC 3D and FLAC framework, are used here to simulate a
series of monotonic and cyclic triaxial tests under drained and constant volume conditions.
The simulation results for the constitutive driver are carried out using one single Gauss
point. In what follows, the solid line, diamond and cross symbols represents the results
obtained using the constitutive driver, FLAC 3D and FLAC, respectively.
Based on the model parameters for Ottawa-F65 sand presented in Tables 2.4, the
monotonic triaxial tests simulations are performed using axial strain increments ∆εa of
1.0e − 5, and sheared until 30% of axial strain is reached. These simulations are performed
on samples with initial void ratio from 0.555 to 0.698, and an initial confining pressure
ranging from 100 to 300 kPa and 50 to 200 kPa, for drained and constant volume
conditions respectively. Figures A.1 and A.2 show the simulations results for both triaxial
compression tests under drained and constant volume conditions, respectively. For the
drained condition, The simulations results using FLAC 3D and FLAC are able to achieve
identical stress-strain curves shown in Figures A.1(a)(b)(c) and volumetric strain evolution
shown in Figures A.1(d)(e)(f) obtained using the constitutive driver. For the constant
volumen condition, the simulation results using FLAC 3D and FLAC show good agreement
with constitutive driver results.

130
 (a) (b) (c)

(d) (e) (f)

Figure A.1: Comparisons between simulations of monotonic drained triaxial compression

tests with initial confinements of 100 kPa (black color), 200 kPa (red color) and 300 kPa
(blue color) and initial void ratios of 0.555, 0.64 and 0.698, using different numerical platforms

(a) (b) (c)

(d) (e) (f)

Figure A.2: Comparisons between simulations monotonic constant volume triaxial

compression with initial confinements of 50 kPa (green color), 100 kPa (black color) and
200 kPa (red color) and initial void ratios of 0.555, 0.64 and 0.698, using different numerical
platforms

131
 (a) (b)

Figure A.3: Comparisons of results from different numerical platforms simulating a cyclic
constant volume triaxial test

The cyclic performance of the model is evaluated using the model parameters for Toyoura
sand shown in Table 2.4. For this particular simulation, the model parameters
m, h0 , nb , A0 , nd , zmax and cz were modified according to the values reported in the work
of Dafalias and Manzari (2004). The cyclic loading is applied by reversing the axial strain
increment ∆εa whenever the cyclic deviatoric stress qamp of 114.2 kPa is reached. In Figure
A.3 the results for an initial confinement pressure of 293 kPa and initial void ratio of 0.808,
are shown in terms of stress-strain loops in Figure A.3(a) and stress path in Figure A.3(b).
The results show a close match between constitutive driver and the implementations in
F LAC 3D and F LAC. Notice, the results shown in Figure A.3 are also in good agreement
with those published by the original authors (Dafalias and Manzari, 2004). These
simulations results, shown in Figures A.1, A.2 and A.3 are in good agreement with those
obtained using the constitutive driver, verifying the reference model implementation in
FLAC 3D and FLAC under monotonic and cyclic loading for a wide range of initial
confinements and void ratios.

A.4.2 Verification of the treatment for overshooting response

The overshooting treatment is verified with the work reported by Dafalias and Taiebat
(2016). To this end, three triaxial compression tests under drained conditions are simulated

132
 (a) (b) (c)

Figure A.4: Overshooting performance for a drained triaxial compression test on isotropically
consolidated samples subjected to (a) small unloading, (b) medium level of unloading, and
(c) moderate level of unloading

using a single hexadron zone in F LAC 3D . The model parameters for Toyoura Sand as
shown in Table 2.4, are assigned to the zone. Each simulation is performed with an initial
confinement of 100 kPa and an initial void ratio of 0.833, and sheared until 2% of axial
strain is reached. The applied loading follows the same unloading-reloading scenario as
explained in Dafalias and Taiebat (2016), every 0.5% of axial strain the sample is subjected
to a unloading and re-loading scenario. The stress-strain response for all three triaxial
compressions are shown in Figure A.4. The results shown in blue are obtained by
incorporating the overshooting treatment to the model implementation, while the results in
red correspond to a model implementation without the overshooting treatment. An
additional monotonic loading shown in green, is added in each figure for comparison
purposes. Figures A.4(a),A.4(b) and A.4(c) show the stress-strain response considering a
small, medium and large unloading scenario, respectively.
The model response with the overshooting treatment is in good agreement with the
worked reported in Dafalias and Taiebat (2016), verifying the incorporated overshooting
treatment to the model formulation.

133
A.4.3 Verification of the treatment for pressure dependent
sub-stepping

A drained monotonic triaxial compression test was simulated to evaluate the impact of s
on the implementation results. The test, with an initial confinement of 200 kPa and an
initial void ratio of 0.555 is carried out using an axial strain increment of ∆εa = 1.0e − 4
until εa = 10% is reached. The model parameters used correspond to the Ottawa F-65
sand provided on Table 2.4. A sensitivity analysis is carried out varying the sub-stepping
tolerance s with 1.0e-1, 1.0e-2 and 1.0e-3.
Figures A.5(a) and A.5(b) shows the stress-strain curve and volumetric strain evolution.
The results indicate s has a direct impact on the model implementation accuracy, i.e., as s
decreases the simulated results converge into a single curve. However, this accuracy comes
with the cost of an increase in the simulation duration (time). This is expected due to the
high number of sub-increments, k, when using low values of s . Iso-error maps will be used
to quantify the level of implementation accuracy.

A.4.4 Accuracy of the model implementation

To evaluate the accuracy of the model implementation a series of simulations are performed
by modeling a triaxial test. The sample with an initial confinement 100 kPa and an initial

(a) (b)

Figure A.5: Sensitivity analyses using different s on a drained triaxial compression test

134
void ratio e0 = 0.555, is subjected to a series of strain increment combinations up to 0.1%
in magnitude in both axial ∆εa and radial ∆εr strains. Simulations performed using s =
1.0e − 3 and ∞ are compared with a reference simulation considered as the “exact” solution.
Notice that setting s = ∞ suggests the sub-stepping treatment is disabled. The exact
solution is obtained using strain increments 1.0e-7 in both axial ∆εa and radial ∆εr strains.
All the foregoing simulations are performed with the model parameters appropriate for the
simulation Ottawa sand and presented in Table 2.4. The computed stress tensor σ are
compared with the results obtained from the exact solution σ ∗ to determine the relative
error δ defined as
s
(σ − σ ∗ ) : (σ − σ ∗ )
δ= . (A.84)
(σ : σ )

Iso-error maps are used here to assess the overall accuracy of the model implementation,
a procedure used in the past by numerous researchers (Ortiz and Popov, 1985; Ortiz and
Simo, 1986; Manzari and Prachathananukit, 2001; Amorosi et al., 2008; Andrianopoulos
et al., 2010). Figures A.6(a) and A.6(b) presents the iso-error maps using s = 1.0e − 3
and ∞, respectively. Observe that when sub-stepping is disabled and with moderate strain
increments in the range of 10−4 , the level of error δ is greater than 5%. The accuracy of
the implementation is enhanced when the sub-stepping is active, with the relative error δ
becoming less than 1% for a wider range of strains increments. Hence, it is suggested to set
s = 1.0e − 3 for future simulations.

A.4.5 Verification of the treatment for low effective stress state

The simplest efficient method to reduce the simulation time is to incorporate a stress cut-off,
such that the minimum allowable stress is controlled with m to avoid possible negative or
very low values of p during soil liquefaction. In order to evaluate the impact of m in the
numerical results three cyclic triaxial simulations are carried out using three different values
of m = 1.0e − 1, 1.0e − 2, and 1.0e − 3. The cyclic triaxial simulations are performed on
samples considering an initial isotropic confinement of 100kPa and an initial void ratio of
0.65. Figures A.7(a) and A.7(b) show the stress-strain loops and stress-path response. The

135
 8
6
2

0.3
0.3 5
0.1

0.2
0.2
0.1

5
0.0

5
5
2
4 6 8

(a) (b)

Figure A.6: ISO-error maps for the initial stress state at p0 = 100 kPa and e0 = 0.555
showing the reduction of the realative error with the sub-stepping tolerance s

results show m have a negligible influence on the model implementation results. However,
Table A.1 shows m has a direct influence in the simulation time, i.e. as m decreases the
simulation time is increased.

(a) (b)

Figure A.7: Sensitivity analyses on m simulating a undrained cyclic triaxial test

136
 Table A.1: Analysis of computational cost for different m

m Time (min)
1.0e-2 0.895
1.0e-3 0.908
1.0e-4 0.973

A.5 Summary of the constitutive model

implementation
This chapter presents the formulation and numerical implementation of the SANISAND-Sf
model accounting for semifluidized state. The overshooting correction proposed by Dafalias
and Taiebat (2016) is incorporated to the multiaxial formulation of the model to prevent
unrealistic stiffening in the unloading-reloading scenarios. Using a cutting-plane algorithm
with sub-stepping as the stress integration scheme, and numerical treatments for low
effective stress state and mixed-discretization, the model was implemented in the finite
difference frameworks: F LAC 3D , making it available to other members of the geotechnical
community. The constitutive model implementation is verified against drained and
constant volume triaxial tests simulations, under monotonic and cyclic loading, showing
the implementation accuracy in simulating soil response for a range of strain probes. The
model implementation has been used to simulate successfully geotechnical boundary value
problems, e.g., seismic shaking on a tailing dam (Barrero et al., 2015) and numerous site
response analyses (Ramirez et al., 2018a; Adinata et al., 2018), which provides the
confidence of generalizing its applications in seismic site response analysis.

137