Professional Documents
Culture Documents
Editors
Michael A. Hicks
Section of Geo-Engineering, Delft University of Technology, Delft, The Netherlands
Jelke Dijkstra
Section of Geo-Engineering, Delft University of Technology, Delft, The Netherlands
Marti Lloret-Cabot
Centre for Geotechnical and Materials Modelling, University of Newcastle, Newcastle,
Australia & Section of Geo-Engineering, Delft University of Technology, Delft,
The Netherlands
Minna Karstunen
Department of Civil and Environmental Engineering, Chalmers University of
Technology, Gothenburg, Sweden & University of Strathclyde, Glasgow, UK
CRC Press/Balkema is an imprint of the Taylor & Francis Group, an informa business
All rights reserved. No part of this publication or the information contained herein may be reproduced,
stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, by pho-
tocopying, recording or otherwise, without written prior permission from the publisher.
Although all care is taken to ensure integrity and the quality of this publication and the information
herein, no responsibility is assumed by the publishers nor the author for any damage to the property or
persons as a result of operation or use of this publication and/or the information contained herein.
Table of contents
Preface ix
Committees xi
Keynote speakers xiii
Computational methods
Large deformation analysis of cone penetration testing in undrained clay 1
L. Beuth & P.A. Vermeer
Adaptive Mesh Refinement for strain-softening materials in geomechanics 8
C.E. Heaney, R.B.J. Brinkgreve, P.G. Bonnier & M.A. Hicks
A dynamic material point method for geomechanics 15
I. Jassim, C. Coetzee & P.A. Vermeer
Pile penetration simulation with Material Point Method 24
L.J. Lim, A. Andreykiv & R.B.J. Brinkgreve
Coupling triangular plate and volume elements in analysis of geotechnical problems 31
S. Tan, M.A. Hicks & A. Rohe
Fracture growth in heterogeneous geomaterials 38
P.J. Vardon & J.D. Nuttall
Constitutive modelling
Modelling rate-dependent behaviour of structured clays 43
M. Karstunen, N. Sivasithamparam, R.B.J. Brinkgreve & P.G. Bonnier
FEM simulation of large vertical deformations caused by land subsidence and verification
of the results by using radar interferometry techniques 51
C. Loupasakis, D. Rozos, F. Raspini & S. Moretti
Numerical modeling of fracturing in soil mix material 58
G. Van Lysebetten, A. Vervoort, N. Denies, N. Huybrechts, J. Maertens,
F. De Cock & B. Lameire
A 3D practical constitutive model for predicting seismic liquefaction in sands 65
A. Petalas, V. Galavi & R.B.J. Brinkgreve
Comparison of two anisotropic creep models at element level 72
N. Sivasithamparam, M. Karstunen, R.B.J. Brinkgreve & P.G. Bonnier
Installation effects
Displacement pile installation effects in sand 79
A. Beijer Lundberg, J. Dijkstra & A.F. van Tol
Cyclic jacking of piles in silt and sand 86
F. Burali d’Arezzo, S.K. Haigh & Y. Ishihara
Influence of installation procedures on the response of capacitance water content sensors 92
M. Caruso, F. Avanzi & C. Jommi
v
The load capacity of driven cast in-situ piles derived from installation parameters 99
D. Egan
On the numerical modelling and incorporation of installation effects of jacked piles:
A practical approach 104
H.K. Engin, R.B.J. Brinkgreve & A.F. van Tol
Analytical and laboratory study of soil disturbance caused by mandrel driven prefabricated
vertical drains 111
A. Ghandeharioon
CEL: Simulations for soil plugging, screwed pile installation and deep vibration compaction 118
J. Grabe, S. Henke, T. Pucker & T. Hamann
Towards a framework for the prediction of installation rate effects 128
S. Robinson & M.J. Brown
Rate dependent shear strength of silt at low stresses 135
S. te Slaa & J. Dijkstra
Control of excess pore pressure development during pile installations in soft sensitive clay 140
T. Tefera, G. Tvedt & F. Oset
Seabed pipelines: The influence of installation effects 146
D.J. White
Soil improvement
Volume averaging technique in numerical modelling of floating deep mixed columns in soft soils 198
P. Becker & M. Karstunen
Comparison between theoretical procedures and field test results for the evaluation
of installation effects of vibro-stone columns 205
E. Carvajal, G. Vukotić, J. Castro & W. Wehr
Numerical analyses of stone column installation in Bothkennar clay 212
J. Castro, M. Karstunen, N. Sivasithamparam & C. Sagaseta
Execution of Springsol® deep mixed columns: Field trials 219
S. Melentijevic, F. Martin & L. Prieto
A method of modelling stone column installation for use in conjunction with unit cell analyses 226
B.G. Sexton & B.A. McCabe
Cement grout filtration in non-cohesive soils 233
X.A.L. Stodieck & T. Benz
The undrained cohesion of the soil as a criterion for column installation with a depth vibrator 241
J. Wehr
vi
Soil-structure interaction
Modeling of rock fall impact using Discrete Element Method (DEM) 245
G. Grimstad, O. Melhus, S. Degago & R. Ebeltoft
Investigation into the factors affecting the shaft resistance of driven piles in sands 252
D. Igoe, K. Gavin & L. Kirwan
Monitoring and risk assessment in EPB TBM’s in urban environments: High speed railway
tunnel Sants-Sagrera running next to Sagrada Familia Basilica (World Heritage) 258
J.E. Paris Fernández & J. Gómez Cabrera
Effect of roughness on keying of plate anchors 264
D. Wang, C. Han & C. Gaudin
vii
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
Preface
The partners of the European project GEO-INSTALL extend a warm welcome to all participants of the
International Conference on Installation Effects in Geotechnical Engineering (ICIEGE). This is the closing
conference of GEO-INSTALL (FP7/2007-2013, PIAG-GA-2009-230638), an Industry-Academia Path-
ways and Partnerships project funded by the European Community from the 7th Framework Programme.
Infrastructure construction involves the installation of structural elements, such as piles and various
ground improvement techniques for soils and rocks. The installation process itself can be quasi-static (for
example jacked piles) or dynamic (vibratory methods, such as stone columns and driven piles), and generally
involves very large deformations and changes in pore pressure. The fact that natural soils are complex geo-
materials, exhibiting structure and rate-dependent behaviour, makes analysis of such problems yet more chal-
lenging. In particular, the influence of installation on key design parameters, such as mobilised strength at the
soil-structure interface and soil stiffness, is difficult to quantify and, as yet, impossible to model. Numerical
analyses using the standard Finite Element Method (FEM) are unable to produce accurate descriptions of
large deformation problems due to excessive mesh distortions and novel techniques need to be developed.
The aim of the conference is to provide an international forum for presenting the latest developments
in monitoring, analysing and managing installation effects in geotechnical engineering. Active discussion
on important topics will be facilitated through invited keynote lectures, which set the scene for the main
theme of the conference. In addition, the partners of GEO-INSTALL will present selected highlights of
their joint research programme, which has been achieved through intense collaboration between industry
and academia.
The peer review papers contained in these proceedings were accepted for presentation at ICIEGE, held
in Rotterdam, The Netherlands 24–27 March 2013. They have been authored by academics, research-
ers and practitioners from many countries worldwide and cover numerous important aspects related to
installation effects in geotechnical engineering, ranging from large deformation modelling to real field
applications. The main topics are:
• Computational methods
• Constitutive modelling
• Installation effects
• Offshore construction and foundations
• Soil improvement
• Soil-structure interaction
The submitted abstracts were reviewed and the authors of those abstracts that fell within the scope
of the conference were invited to submit full papers for peer review. The editors would like to thank the
Scientific Committee who provided assistance in the review process. They would also like to thank the
keynote speakers, authors, participants and members of the Organising Committee. The editors are grate-
ful for the support of the European Community and the partner organisations of the GEO-INSTALL
project: University of Strathclyde, United Kingdom (Project Coordinator); Delft University of Technol-
ogy, The Netherlands; Deltares, The Netherlands; Keller Limited, United Kingdom; Norwegian Geotech-
nical Institute, Norway; Plaxis BV, The Netherlands; Stellenbosch University, South Africa; University
of Stuttgart, Germany.
On behalf of the partners of GEO-INSTALL, we welcome you to The Netherlands and hope that you
find the conference both enjoyable and inspiring.
Michael Hicks
Jelke Dijkstra
Marti Lloret-Cabot
Minna Karstunen
January 2013
ix
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
Committees
ORGANISING COMMITTEE
SCIENTIFIC COMMITTEE
xi
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
Keynote speakers
xiii
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
L. Beuth
Deltares, Delft, The Netherlands
P.A. Vermeer
Deltares, Delft, The Netherlands
University of Stuttgart, Germany
ABSTRACT: Cone penetration testing is a widely-used in-situ test for soil profiling as well as estimating soil
properties of strength and stiffness. In this paper, the relationship between the undrained shear strength of clay
and the measured cone tip resistance is investigated through numerical analysis. Such analyses serve to refine
and establish correlations between cone penetration testing measurements and soil properties; thus enabling
more reliable predictions of soil properties. The presented analyses are performed by means of a Material
Point Method that has been developed specifically for the analysis of quasi-static geotechnical problems
involving large deformations of soil. Both, the load-type dependency of the shear strength of undrained clay
as well as the influence of the anisotropic fabric of natural clay on the undrained shear strength are taken
into account through a new material model, the Anisotropic Undrained Clay model. Results indicate that
the deformation mechanism relevant for cone penetration in undrained normally-consolidated clay differs
significantly from predictions based on the Tresca model, but resulting cone factors appear to be useful.
1
compression and extension more accurately, a The Material Point Method discretises a solid
model based on critical state soil mechanics such body by means of a cloud of material points that
as the well-known Modified Cam Clay model move through a fixed finite element grid. Thereby,
might be used. This model takes into account the material points capture the arbitrary large
the shear-induced volume change of drained clay deformations of the solid body without the occur-
(Schofield & Wroth 1968). Indeed, for normally- rence of severe deformations of the finite elements.
consolidated clay as considered in this paper, this Material and state parameters of the solid body as
model correctly predicts lower undrained shear well as applied loads are stored in material points
strengths for triaxial compression and extension whereas the mesh does not store any permanent
than the Mohr-Coulomb model. information.
Due to the deposition process of clay, natural With the MPM, the underlying finite element
clay generally shows a different strength in hori- grid is used as with the UL-FEM to solve the sys-
zontal directions than in the vertical direction. tem of equilibrium equations for an applied load
In order to further increase the accuracy of CPT increment on the basis of those finite elements that
analyses, this anisotropic strength of clay must also contain material points at the considered loading
be taken into consideration. Therefore, the Aniso- step. Once displacement increments are mapped
tropic Undrained Clay (AUC) model is used in the from nodes to material points, once strain incre-
presented analyses. ments are computed at the locations of material
It implements the theory of critical state soil points, the mesh is usually reset into its original
mechanics and also considers the strength anisot- state. It might be changed arbitrarily.
ropy of natural clay (Vermeer et al. 2010; Beuth Obviously, the finite element grid used with the
2012). It is largely based on the S-CLAY1 model MPM must cover not only the solid in its initial con-
developed by (Wheeler et al. 2003) but consid- figuration as with the UL-FEM but the entire region
ers neither density and rotational hardening nor of space into which the solid is expected to move.
softening. It should be emphasised, that in contrast to
To the authors knowledge, usage of a quasi- mesh-less Lagrangian methods such as the Dis-
static MPM in combination with the AUC model crete Element Method, the material points repre-
for CPT analysis exceeds the accuracy of numerical sent subregions of a solid body and not individual
studies reported so far in literature, such as (Van particles such as sand grains. The mass and volume
den Berg 1994) and (Lu et al. 2004). of subregions is memorised with material points,
In the following section, the used numeri- but changes in the shape are not traced.
cal method will be introduced briefly. A detailed With one exception (Guilkey & Weiss 2003),
description of it would exceed the scope of this existing implementations of the MPM are dynamic
paper. For further information, the reader is there- codes that employ an explicit time integration
fore referred to (Beuth et al. 2007; Vermeer et al. scheme (Sulsky et al. 1994; Wi eckowski et al. 1999;
2009; Beuth et al. 2011) and (Beuth 2012). The con- Coetzee et al. 2005). Using these codes for the
stitutive modelling of undrained clay is treated in analysis of quasi-static problems is computation-
Section 3. Subsequently, the geometry and discreti- ally inefficient as explicit integration requires very
sation of the performed numerical analyses are pre- small time steps. The quasi-static MPM makes use
sented in Section 4. In Section 5, results obtained of an implicit integration scheme and thus circum-
for the analysis with the AUC model are compared vents the limitation on step size of dynamic codes.
to results for the Tresca model. The paper ends In recent years, the quasi-static MPM has been
with an outlook on possible future work. validated with numerous geotechnical benchmark
problems. Furthermore, it has been extended by a
contact formulation for modelling reduced frictional
2 QUASI-STATIC MATERIAL POINT or adhesive contact between structure and soil based
METHOD on interface elements (Vermeer et al. 2009).
2
K water
pw = εvol and
d p′ K′ εvol (1)
n
K water 3( vu v′ )
= K′ (2)
n ( − vu ) ( + v′ )
3
The height of the discretised space has been
chosen so as to accomodate a penetration depth
of up to 18 D measured from the base of the cone.
The lower mesh boundary is located 18 D below
the tip of the cone. Considering a rigid penetrom-
eter, there is no need to discretise the penetrometer
itself which simplifies the application of boundary
conditions along the penetrometer surface.
Rather than incrementally moving the pen-
etrometer downwards into the ground, the soil is
pushed upwards against the fixed penetrometer
surface. The upward displacement of the material
point discretisation is realised by prescribed dis-
placements applied on the bottom boundary of the
mesh. Material points are continuously fed along
the lower boundary into the mesh from a cloud
of material points that is initially located outside
the discretised region (feeder). Figure 2 shows the
material point discretisation including the feeder in
its initial and final configuration.
Along the sides of the mesh, displacements are
fixed normal to the vertical planes (roller bound-
aries). Only smooth contact between soil and
penetrometer is considered by prescribing roller
boundaries fixed normal to the penetrometer sur-
face. Reduced adhesive or frictional contact along
the penetrometer surface might be considered by
placing interface elements along the fixed bound-
ary of the penetrometer but this is not considered
here.
Soil weight is not considered with the presented
Figure 2. (a) Material point discretisation for the initial; computations since the gradient of vertical stress
(b) Final configuration at a penetration depth of 18 D. is not relevant at the greater depths of cone pen-
etration that are reached in the performed analy-
ses. Because the computation with the AUC model
requires initial stresses for initialisation of the extent
and orientation of the yield surface in principal
stress space, a surface traction of 10 kPa is applied.
K0 is set to 0.5. Thus, in an initial loading stage, a
vertical stress of 10 kPa and horizontal stresses of
5 kPa are realised. For the AUC model, the iso-
tropic preconsolidation pressure is then −7.1 kPa.
The computations were performed with
4-noded tetrahedral elements that were extended
by a strain-smoothing algorithm to prevent lock-
ing. A mesh with 15,678 degrees of freedom has
been used. Mesh re-finement has been applied in
the region around the penetrometer cone as shown
in Figure 3 (a) where large stress gradients occur.
Furthermore, a mesh re-finement is required on
the narrow edge along the longitudinal axis of the
discretised slice.
Initially, the soil surface coincides with element
boundaries adjacent to the cone base, see Figure 3
(b). An equal number of 10 material points is
initially placed inside each finite element. In total,
Figure 3. (a) FE discretisation with local mesh refine- 286,410 material points are used for the performed
ment along the cone tip; (b) Geometry of the cone tip. analyses.
4
5 RESULTS factor for the computations with the Tresca and
AUC model. For the Tresca model, a cone factor
FEM simulations of triaxial compression and of 12.5 is obtained, for the AUC model a cone fac-
extension tests were performed with the AUC model tor of 11.6.
for anisotropically consolidated soil under K0 = 0.5 The Tresca cone factor of 12.5 agrees well with
up to σp0 = 10 kPa. Similarly, a direct shear test was results of previous numerical studies based on
computed for K0-consolidation up to σp0 = 10 kPa in this model. Indeed, (Van den Berg 1994) reports
order to obtain the respective values of undrained for a smooth cone and a rigidity index Ir = G/
shear strength relevant for the performed CPT su = 600 a cone factor of approximately 13. (Lu
analyses. The resulting undrained shear strengths et al. 2004), who additionally take into considera-
are for triaxial compresssion su,comp = 3 kPa, for tion the initial stress state, report a matching cone
triaxial extension su,ext = 1.2 kPa and for simple factor of 12.5. Total stress analyses were consid-
shear su,ss = 2.15 kPa (see also Figure 1). For the ered in these studies whereas in the computations
computation with the Tresca model, an undrained of this paper distinction is made between effective
shear strength of 4 kPa is specified. An overview stresses and excess pore pressures as described in
of the undrained shear strengths relevant for the Subsection 3. Supplementary numerical analyses
two computations is provided in Table 1. show that the choice of analyses has little impact
In Figure 4 the obtained relative tip stresses σc/su,ss on the obtained cone factors, at least for und-
for the two computations are plotted over the rela- rained Poisson ratios above a value of 0.49 (Beuth
tive cone penetration. Once the influence of the soil 2012).
surface on the movement of soil particles around the Figure 5 (a) shows the loading types found within
penetrating structure subsides, a steady-state defor- the soil body at a penetration depth of 10 D; on the
mation process evolves, as can be seen in Figure 4. left half for the AUC model, on the mirrored right
The tip stress σc for the fully developed steady-state half for the Tresca model. Points colored dark grey
penetration process corresponds to the tip resistance represent regions of triaxial compression states-
qc measured during cone penetration testing. of-stress. Light grey points represent regions that
The relative tip resistance qc/su corresponds cannot be associated with a specific loading type.
to the cone factor Nc which is commonly used to For the Tresca model, triaxial compression states-
correlate the undrained shear strength of soil to of-stress dominate in the vicinity of the cone. The
measurements of cone tip resistance. Equating AUC model predicts triaxial compression states-
su to su,ss results in approximately the same cone of-stress to occur only in a small wedge below the
advancing cone.
Figure 5 (b) shows the effective vertical stresses
for the AUC model and the Tresca model at a pen-
Table 1. Undrained shear strengths [kPa]. etration depth of 10 D. Compressive stresses are
negative. Low vertical effective stresses occur next
su,comp su,ext su,ss to the cone for both material models forming wing-
shaped stress bulbs. This can be explained by the
Tresca 4.0 4.0 4.0 reduction of vertical compaction once soil leaves
the ‘shadow’ of the penetrating cone. Within the
AUC 3.0 1.2 2.15
shadow of the cone, soil is pushed to the sides and
downward, outside it, soil moves to the sides in an
upward direction as noted by Lu et al. Some of
the vertical compression previously applied on soil
particles is consequently released.
For the AUC model, as a consequence, mean
effective stresses decrease within the region of the
wing-shaped stress bulb. The stress paths obtained
for the AUC model correspond to soil yielding in
a critical state: stress states in this region follow
the critical state line. They do not correspond to
triaxial compression nor triaxial extension which
corresponds to the results shown in Figure 5 (a).
Soil yields in the vicinity of the cone with effec-
tive mean stresses decreasing down to −2.5 kPa
and deviatoric stresses q as low as 3 kPa which
Figure 4. Relative tip stress plotted over the relative explains the lower tip resistance found with the
cone displacement for the Tresca and AUC model. AUC model.
5
Figure 5. (a) Regions of triaxial compression states-of-stress colored dark gray; (b) Comparison of effective vertical
stresses σ ′yy [kPa].
6
Beuth, L., Benz, T., Vermeer, P., Coetzee, C., Bonnier, P., & Schofield, A., & Wroth, P. 1968. Critical state soil
Van Den Berg, P. 2007. Formulation and validation mechanics. McGraw–Hill New York.
of a quasi-static Material Point Method. In Proceed- Sulsky, D., Chen, Z., & Schreyer, H. 1994. A particle
ings of the 10th International Symposium on Numerical method for history–dependent materials. Compu-
Methods in Geomechanics, Volume 10, pp. 189–195. ter Methods in Applied Mechanics and Engineering
Taylor & Francis Group. 118(1–2), 179–196.
Beuth, L., Więckowski, Z., & Vermeer, P. 2011. Solution Van den Berg, P. 1994. Analysis of soil penetration.
of quasi-static large-strain problems by the mate- Ph. D. thesis, Delft University of Technology, The
rial point method. International Journal for Numeri- Netherlands.
cal and Analytical Methods in Geomechanics 35(13), Vermeer, P., Jassim, I., & Hamad, F. 2010. Need and per-
1451–1465. formance of a new undrained clay model.
Coetzee, C., Vermeer, P., & Basson, A. 2005. The mod- Vermeer, P., Yuan, Y., Beuth, L., & Bonnier, P. 2009.
elling of anchors using the material point method. Application of interface elements with the Material
International Journal for Numerical and Analytical Point Method. In Proceedings of the 18th Interna-
Methods in Geomechanics 29(9), 879–895. tional Conference on Computer Methods in Mechanics,
Guilkey, J., & Weiss, J. 2003. Implicit time integration Volume 18, pp. 477–478. Polish Academy of Sciences.
for the material point method: Quantitative and algo- Wheeler, S., Näätänen, A., Karstunen, M., & Lojander,
rithmic comparisons with the finite element method. M. 2003. An anisotropic elastoplastic model for soft
International Journal for Numerical Methods in Engi- clays. Canadian Geotechnical Journal 40(2), 403–418.
neering 57(9), 1323–1338. Więckowski, Z., Youn, S., & Yeon, J. 1999. A particle–in–
Lu, Q., Randolph, M., Hu, Y., & Bugarski, I. 2004. cell solution to the silo discharging problem. Interna-
A numerical study of cone penetration in clay. Geo- tional Journal for Numerical Methods in Engineering
technique 54(4), 257–267. 45(9), 1203–1225.
7
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
P.G. Bonnier
Plaxis BV, Delft, The Netherlands
M.A. Hicks
Geo-Engineering Section, Faculty of CITG, Delft University of Technology, Delft, The Netherlands
ABSTRACT: This paper describes the implementation of Adaptive Mesh Refinement (AMR) within
the geotechnical software package Plaxis 2D. The algorithm is recovery-based and aims to reduce the
discretisation error estimated as a measure of the incremental deviatoric strain. Once the global discre-
tisation error exceeds a user-defined tolerance, certain elements in the mesh are marked for refinement.
The refinement procedure is based on a combination of regular subdivision and longest-edge bisection.
Mapping from the old mesh to the new mesh is accomplished by using the recovered solutions at the
nodes and the shape functions. The AMR algorithm is demonstrated for a vertical cut problem for a
softening Drucker-Prager material. Regularisation prevents the mesh-dependency which would otherwise
be seen for such softening constitutive models.
8
now briefly discussed. A more detailed description especially suitable for geomechanical problems
can be found in Heaney et al. (2013). which exhibit large changes in strain.
The error estimator for element iEl is given by
2.1 Recovery
Variables defined at integration points are recov-
|| e ||iE1
Δγ ∫iiE1( γ* γ )2 , (3)
ered at the nodes in order to estimate the error
and to facilitate the mapping process. For this pur- and from this, the global error norm can then be
pose, Superconvergent Patch Recovery (SPR) is calculated:
used (Zienkiewicz & Zhu 1992a). It is an efficient,
local method involving the inversion of a relatively nE 1
small matrix (small when compared to the size of (|| ||Δγ )2 (|| ||Δi γ1 )2 . (4)
the global stiffness matrix). iE1 1
|| e || Δ γ
≥ η. (6)
2.2 Error estimation || Δγ * ||
Zienkiewicz & Zhu (1992b) prove that their
error estimator (described in Zienkiewicz & Zhu Introducing the assumption that the local errors
(1987)) is asymptotically exact for linear elastic are equally distributed over the mesh gives
problems. For elastoplastic problems Boroomand &
Zienkiewicz (1999) devised an error estimator based η || Δγ * ||
Δγ ≥
|| e ||local , (7)
on incremental energy. More recently Hicks (2000) nE1
employed an error estimator based upon a meas-
ure of incremental deviatoric strain. He defines the where nE1 is the total number of elements. So, once
incremental shear strain invariant, Δγ, as the global error exceeds the user-defined (global)
tolerance, then all elements whose error exceeds
η || Δγ *||
Δγ Δε ijdev Δε ijdev , (1) nE1
will be marked for refinement.
9
is that the solution over the entire domain must be
dk = 2 / 3 || d p
|| . (10)
transferred from the old to the new mesh. In order
to limit the numerical diffusion which may occur
as a result of transferring the solution between The regularisation employed here follows the
meshes, and to be more efficient, subdivision is non-local approach whereby strains are averaged
used. in the region around the integration point of
Regular bisection refers to the splitting of an interest. A weighted combination of the non-local
element into four elements by joining the mid- and local strains is used in the calculation of dκ.
points of the triangle’s edges (Bank et al., 1983). During the averaging process a parameter is intro-
Between a regularly-refined element and a non- duced which effectively controls the minimum
refined element will be a non-conforming edge width of the shear band. The interested reader is
with hanging nodes. A straightforward solution referred to Brinkgreve (1994) for further details.
to this is adopted here based on Rivara’s longest- This approach has already shown good results for
edge bisection method (Rivara 1984). One advan- standard FE calculations (Rolshoven & Jirásek
tage of using longest-edge bisection is that the (2003), Rezania et al. (2012)).
minimum internal angle of subsequent meshes
is never less than half of the smallest interior
angle of the original mesh (Rosenberg & Stenger 4 RESULTS
1975).
A vertical cut is modelled, the geometry of which can
be seen in Figure 1. The (effective) material param-
2.4 Mapping eters are shear modulus G = 3355 kNm−2, Poisson’s
New elements require values of displacement and ratio υ = 0.49, initial cohesion c0 = 5 kNm−2, mini-
stress, and also any state variables which are asso- mum cohesion cmin = 1 kNm−2 and softening modu-
ciated with the constitutive model. Displacements lus H = 100 kNm−2. The problem is undrained so the
can be mapped using the shape functions and the friction angle is zero and the yield surface becomes
nodal values of displacement. In order to map the a cylinder. Displacement is applied in increments
stresses, they are first recovered at the nodes using along half of the topsurface up to a total displace-
SPR, then they can be mapped to the new mesh ment of 0.03 m. Plane strain is assumed and self-
just as the displacements were. Any integration- weight is not considered.
point values can be mapped using this method. A number of meshes were used comprising
After mapping, the stress field will no longer be 6-noded triangular elements: the mesh densities
in equilibrium with the applied force. A zero-load are given in Table 1. First the problem was run for
increment can be applied in order to resolve this, each mesh with the standard FE method. Then
if necessary. the problem was run again for each mesh with
regularisation. Finally mesh 2 was used as a start-
3 REGULARISATION
f = q + α (p − c cot φ) = 0, (8)
10
Table 1. Numbers of elements and degrees of freedom
in each mesh used in the standard and the regularised FE
simulations.
11
Figure 3. The total reaction force (kNm-1)along the top surface against the applied displacement (m) for the vertical cut.
12
Figure 5. The deformed mesh at the end of the
AMR simulation. Displacements have been exaggerated
by a factor of 10.
5 CONCLUSIONS
13
ACKNOWLEDGEMENTS Mar, A. & Hicks, M.A. 1996. A benchmark computa-
tional study of finite element error estimation. Int.
The authors gratefully acknowledge the support of J. Numer. Meth. Engng 39(23), 3969–3983.
GEO-INSTALL: a project funded by the European Perić, D., Hochard, C., Dutko, M. & Owen, D.R.J. 1996.
Transfer operators for evolving meshes in small strain
Community through the Marie Curie IAPP pro- elasto-plasticity. Comput. Methods Appl. Mech. Engrg
gramme (Contract No. PIAP-GA-2009-230638). 137, 331–344.
Rezania, M., Bonnier, P.G., Brinkgreve, R.B.J. &
Karstunen, M. 2012. Non-local regularisation of
REFERENCES Drucker-Prager softening model. In Z. Yang (Ed.),
Proceedings of the 20th UK National Conference of
Bank, R., Sherman, A.H. & Weisser, A. 1983. Refine- ACME, Manchester (UK), pp. 275–278.
ment algorithms and data structures for regular local Rivara, M.C. 1984. Design and data structures of a fully
mesh refinement. In Scientific computing (IMACS adaptive multigrid finite element software. ACM
Transactions), North Holland, pp. 3–17. Transactions on Mathematical Software 10, 242–264.
Boroomand, B. & Zienkiewicz, O.C. 1999. Recovery Rolshoven, S. & Jirásek, M. 2003. Numerical aspects of
procedures in error estimation and adaptivity. Comput. nonlocal plasticity with strain softening. In Compu-
Methods Appl. Mech. Engrg 176, 127–146. tational Modelling of Concrete Structures, Austria,
Brinkgreve, R.B.J. 1994. Geomaterial models and numeri- pp. 305–314.
cal analysis of softening. Ph.D. thesis, Delft University Rosenberg, I.G. & Stenger, F. 1975. A lower bound on the
of Technology. angles of triangles constructed by bisecting the longest
Brinkgreve, R.B.J., Swolfs, W.M. & Engin, E. 2011. side. Mathematics of Computation 29, 390–395.
Plaxis 2D 2011. Plaxis BV. Zienkiewicz, O.C. & Zhu, J.Z. 1987. A simple error esti-
Heaney, C.E., Bonnier, P.G., Brinkgreve, R.B.J. & mator and adaptive procedure for practical engineering
Hicks, M.A. 2013. Adaptive mesh refinement with analysis. Int. J. Numer. Meth. Engng 24, 337–357.
application to geomaterials. In preparation for submis- Zienkiewicz, O.C. & Zhu, J.Z. 1992a. The superconver-
sion to Comput. Geotech. gent patch recovery and a posteriori error estimates.
Hicks, M.A. 2000. Coupled computations for an elastic- Int. J. Numer. Meth. Engng 33, 1331–1364.
perfectly plastic soil using adaptive mesh refinement. Zienkiewicz, O.C. & Zhu, J.Z. 1992b. The superconver-
Int. J. Numer. Anal. Meth. Geomech. 24, 453–476. gent patch recovery (SPR) and adaptive finite element
Hu, Y. & Randolph, M.F. 1998. H-adaptive FE analy- refinement. Comput. Methods Appl. Mech. Engrg 101,
sis of elasto-plastic non-homogeneous soil with large 207–224.
deformation. Comput. Geotech. 23, 61–83. Zienkiewicz, O.C. & Zhu, J.Z. 1995. Superconvergence
Kardani, M., Nazem, M., Abbo, A.J., Sheng, D. & and the superconvergent patch recovery. Finite Elem.
Sloan, S.W. 2012. Refined h-adaptive FE procedure Anal. Des. 19, 11–23.
for large deformation geotechnical problems. Comput.
Mech. 49, 21–33.
14
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
I. Jassim
Institut für Geotechnik, Universität Stuttgart, Germany
C. Coetzee
Department of Mechanical Engineering, University of Stellenbosch, South Africa
P.A. Vermeer
Deltares, Delft, The Netherlands
University of Stuttgart, Germany
ABSTRACT: A dynamic Material Point Method for use in Geomechanics is presented. Soil and structural
bodies are represented by (material) particles, which move inside an unstructured mesh of four-noded 3-D
tetrahedral elements. As such low-order elements tend to show locking for fully developed plastic flow,
a strain-enhancement remedy is described. As a first example, the penetration of a drop anchor into a
Mohr-Coulomb soil is considered. As both a soil body and a metal anchor are considered, an algorithm for
dynamic contact is used and described. An improved type of absorbing boundaries to avoid the reflection
of stress waves is also described. The second example consists of dynamic cone penetration. Finally, the
example of a collapsing tunnel is considered.
15
2 WEAK FORMULATION AND SPACE and storage advantages, is used instead of the con-
INTEGRATION OF EQUILIBRIUM sistent-mass matrix defined in Eq. (5). On denot-
ing the entries of the lumped-mass matrix, as mi
The Cauchy form of conservation of linear it yields
momentum is given by the equations
⎡ m1 0 ... 0 ⎤
ρ ∇ + ρ g and t σ n (1) ⎢0 m2 ... 0 ⎥
ML = ⎢ (6)
⎥
⎢0 0 ... mn ⎥⎦
where ρ is the material density, ü is the displace- ⎣
ment, a superposed dot denotes differentiation
with time, σ denotes the Cauchy stress tensor and where n denotes the number of degrees of freedom.
g is the gravitational acceleration vector. The sur- In order to conserve the mass of the continum,
face traction acting on the external boundary is particle-based integration is used i.e.
denoted by t and n is the outward unit normal of
the boundary. Applying the virtual work principle np
on a domain of volume V surrounded by boundary
S yields mi ∑ m p Nip and
p =1
np
∫δ ρ u dV = − ∫ δεε T σ dV + ∫ δ ρ g dV ∑ BTp σ p Vp
T T
F int = ∫ B σ dV ≈
T
(7)
V V V V p =1
+ ∫ δ uT t dS (2)
S
where np denotes the number of particles, mp is the
mass of particle p, Nip is the shape function evalu-
where δ implies a virtual quantity, ε is the strain ated at particle p and Vp is the volume associated
tensor and the script T denotes the transpose. with particle p. The drawback of using a lumped-
For space discretisation, the displacement field mass matrix is a slight numerical dissipation of the
u is approximated in terms of interpolation func- kinetic energy (Burgess et al. 1992).
tions N and nodal displacements a by u = Na. The
strain tensor is now written in vector notation as
3 TIME INTEGRATION
ε = Ba with B = LN and
ε=( )
T
(3) On applying Euler-forward time integration with
lumped-mass matrix, Eq. (4) yields
where B is the usual finite element strain-displace-
Δt −1
ment matrix, as computed from the linear dif- a t = a t + Δt at = ⎡⎣ MtL ⎤⎦
at , Ft (8)
ferential operator L and the shape functions N.
Substituting Eq. (3) into Eq. (2) gives
where Δt is the current time increment, a t and a t Δt
δ T
δa T
(F ext
F int
) or M
a= F are the nodal velocities at time t, t + Δt respectively.
The incremental nodal displacement is obtained by
with F F ext − F int (4) integrating the nodal velocity by the Euler-back-
ward rule and the position of the particles are sub-
in which sequently updated, i.e.
Δ Δt
Δat Δ at , xtp Δ xtp + N p Δat Δt
(9)
M ∫ N N F = ∫ ρ NT g dV + ∫ NT t dS
V V S
where xtp and xtp Δt are the particle positions at
and F int ∫ B σ dV
T
( )
V
time t and t + Δt respectively. Strains and stresses
at particles are updated using the same algorithms
as for Gaussian integration points within the
Equation (4) is identically used within FEM and standard FEM. In updated Lagrangian FEM, one
MPM. However, in the Material Point Method M would use Δa to update the finite element mesh,
can also change in size when particles move into but within the MPM only particles positions are
empty elements. In other words, the total number updated. Particles eventually cross element bound-
of degrees-of-freedom of the system can vary. A aries, which entails that the new element of a cross-
lumped-mass matrix, which offers computational ing particle has to be detected.
16
Figure 1. A torpedo-shaped drop anchor.
17
Figure 3. Penetration depth for different values of the
friction coefficient. Figure 4. Illustration of two bodies in contact for a
regular mesh.
Following Bardenhagen et al. (2000), Eq. (8) is where mA is the mass of a contact node computed
solved for the combined bodies A and B (see Fig. 4) only from body A as in Eq. (7). In frictional con-
as well as for each body separately, i.e. tact, the tangential force is limited by
18
6 MESH LOCKING AND STRAIN
SMOOTHENING
1 4
∑ (εv )k
εv ∑ εvi with εvn = k
(15)
4 i =1 ∑ ( )k
k
1
ε = ε d + ε v I with I = (1, 1, 1, 0, 0, 0 )T (16)
3
Figure 5. Flow chart of the contact algorithm applied This approach is applied in all examples of the
on body A. present paper.
mA ⎡ A
FtA max μ FnA = μ
Δt ⎣
a ( a A+B ⋅ n ⎤
⎦) (13)
7 ABSORBING BOUNDARIES
19
introduced by Lysmer and Kuhlemeyer (1969).
They proposed a solution in which the boundary is
supported on a dashpot. On denoting the normal
stress at a boundary node as tn and the shear stress
as ts, it yields
Ec =
( )E and G =
E shown in Figure 6. This displacement corresponds
(19) to δ = 0.5 m and αn = 2.5. Sensitivity study showed
( ))(( + ) 2( + ) that those are the best values for the problem con-
sidered. The final displacement of the bottom of the
The drawback of supporting the boundary by mesh can be calculated as an = (σ/Ec)δ. For σ = 1 kPa
dashpots is that the boundary will continuously and Ec = 100 kPa, this displacement is found to be
creep as long as the dashpot will receive stresses 0.05 m. Points 1,2 and 3 of Figure 6 indicate that
from the soil body. In order to limit the creep of the wave is just reflected after reaching the bottom
the boundary, a spring is added parallel to the of the mesh. Stress fronts correspond to point 1, 2
dashpot to obtain a Kelvin-Voigt type of bound- and 3 are (1.28, 1.86 and 1.14) kPa respectively.
ary response. Hence Eq. (17) is rewritten as
20
Figure 7. Applied load of dynamic penetration test.
21
Figure 9. Penetration as a function of number of blows.
10 CONCLUSIONS
22
bottom of the mesh. Hence, one usually employs material point method’’. International Journal of
so-called dashpots that will continually creep under Numerical and Analytical Methods in Geomechanics,
load. In order to limit such non-physical displace- Wiley Online Library (wileyonlinelibrary.com). DOI:
ments, the dashpot is combined with a spring to 10.1002/nag.965.
Borja, R.I. 1988, “Dynamics of pile driving by the finite
obtain a Kelvin-Voigt type of boundary response. element method’’. Computers and Geotechnics,
At present the main limitations of the code is the Vol. (5), 39–49.
lack of pore pressures and a soil model for cyclic Burgess, D. Sulsky, D. & Brackbill, J.U. 1992, “Mass
loading. Meanwhile, dynamic generation and dis- matrix formulation of the FLIP particle-in-cell
sipation of pore pressures is nearly finished and method’’. Journal of Computational Physics,
remains to be reported. The implementation of a Vol. (103), 1–15.
cyclic loading model for genuine simulations of Coetzee, C.J. Vermeer, P.A. & Basson, A.H. 2005,
pile driving has been planned. “The modelling of anchors using the material point
method’’. International Journal for Numerical and
Analytical Methods in Geomechanics, Vol. (29),
879–895.
ACKNOWLEDGEMENT Detournay, C. & Dzik, E. 2006, “Nodal mixed discretiza-
tion for tetrahedral elements’’, Proceedinh of ‘4 inter-
This research was carried out as a part of the ‘’GEO- national FLAC symposium on numerical modeling in
INSTALL‘’ project (Modelling Installation Effects geomechanics, Itasca Consulting Group.
in Geotechnical Engineering). It has received fund- Harlow, F.H. 1964, “The particle-in-cell computing
ing from the European Community through the method for fluid dynamics’’. Methods for Computa-
program (Marie Curie Industry-Academia Part- tional Physics, Vol. (3), 319–343.
nerships and Pathways) under grant agreement Lysmer, J. & Kuhlmeyer, R.L. 1969, “Finite dynamic
model for infinite media’’. Journal of the Engineering
no PIAP-GA-2009-230638. The PhD study of the Mechanics Division, Vol. (95), 859–877.
first author at Stuttgart University is funded by the Sulsky, D. Zhou, S.J. & Schreyer, H.L. 1995, “Application
DAAD (German Academic Exchange Service). of a particle-in-cell method to solid mechanics’’. Com-
puter Physics Communications, Vol. (87), 236–252.
Sulsky, D. & Schreyer, H.L. 1996, “Axisymmetric form
REFERENCES of the material point method with applications to
upsetting and Taylor impact problems’’. Computer
Bardenhagen, S.G., Brackbill, J.U. & Sulsky, D. 2000. Methods in Applied Mechanics and Engineering,
“The material-point method for granular materials’’. Vol. (139), 409–429.
Computer Methods in Applied Mechanics and Engi- Więckowski, Z. Youn, S.K. & Yeon, J.H. 1999, “A parti-
neering, Vol. (187), 529–541. cle-in-cell solution to the silo discharging problem’’.
Bathe, K.J. 1982, Finite Elements Procedures in Engi- International Journal for Numerical Methods in Engi-
neering Analysis, Prentice-Hall, Inc., New Jersey. neering, Vol. (45), 1203–1225.
Belytschko, T. Lu, Y.Y. & Gu, L. 1994, “Element-free Więckowski, Z. 2004, “The material point method
Galerkin methods’’. International Journal of Numeri- in large strain engineering problems’’. Computer
cal Methods in Engineering, Vol. (37), 229–256. Methods in Applied Mechanics and Engineering,
Beuth, L., Więckowski, Z. & Vermeer, P. 2010, “Solu- Vol. (193), 4417–4438.
tion of quasi-static large-strain problems with the
23
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
R.B.J. Brinkgreve
Delft University of Technology, Delft, The Netherlands
Plaxis BV, Delft, The Netherlands
ABSTRACT: Conventional finite element method (FEM) faces mesh distortion and mesh tangling
when it is used in the simulation of extreme deformation in pile penetration. To avoid the shortcoming
of FEM, material point method (MPM) is used owing its abilityto analyse engineering problems involv-
ing extreme deformation. However, MPM generates numerical noise in the calculation of stresses when
material points cross element boundary due to discontinuity of gradient of shape functions. Dual domain
material point method (DDMP), introduced earlier within an explicit framework, provides a continuity
of the gradient of shape functions which helps to eliminate the numerical noise in stress and strain fields.
In this paper we further extended the application of DDMP within an implicit scheme by formulating a
consistent tangent system. Additionally, we have presented a method to couple MPM and FEM analyses
in order to limit the application of MPM to the areas with extreme deformation, which allows to increase
computational efficiency. Numerical analysis results for a pile penetration problem have been presented
and compared with analytical solution for validation.
24
2 IMPLICIT FORMULATION OF MPM As mentioned in Zhang et al. 2011, the numeri-
AND DDMP cal noise is caused by the discontinuity of gradient
function of MPM in Eq. 5 when the internal force
2.1 Governing continuum equations is calculated.
Consider a continuous body Ω ⊂ Rn with a bound-
ary Γ = ∂Ω, the conservation equations for mass 2.3 Implicit time integration scheme
and linear momentum can be defined as,
During a load step, the computation grid will be
dρ solved using conventional FEM. The linearized
+ ρ∇⋅v = 0 (1) equation of motion during the Newton iteration
dt κ for current time step can be defined as (Wieck-
owski 2004),
ρa = ∇.σ + ρb (2)
k 1
where ρ is the mass density, v is the velocity, a is the ij du j ( fiext ) − (f
( fiint )k 1
m ij ⋅ a kj 1
Qkj (6)
acceleration, σ is the Cauchy stress tensor, and b is
the specific body force. where K is the stiffness matrix, mij =
∑ p =1 ρυυ pSi ( p ) S j ( p ) is the mass matrix, duj is
Np
N u kj = u kj −1 + duu j (7)
∑ ∫Ωc ρ Si S j d c j ∫Ωc ∇Si σ d Ωc
j =1
Meanwhile, the acceleration term can be calcu-
+ ∫ ρ Si d c ∫Γt tSi d Γc (3) lated by discretizing the time derivative with trap-
Ωc
ezoidal rule. The discretized acceleration term is
where N is the total number of nodes in a compu- given as,
tation domain Ωc, i, j are the nodal indices, aj is the
acceleration at node j, Si is the shape function of 4 k −1 4 0
a kj uj − vj a 0j (8)
node i, t is the surface traction, and Γt is the sur- Δt 2 Δt
face boundary of the computation domain. The
first term of the right hand side of Eq. 3 is defined where v 0j and a 0j terms are the nodal velocity and
as the internal force of the system, fiint. The sum of acceleration at the start of time step.
the second and the third terms of the right hand
side of Eq. 3 can also be defined as the external
force of the system, fiext . Comparable to conven- 2.4 Numerical implementation of implicit
tional FEM, the numerical integration of MPM MPM
over the computation domain Ωc is approximated At the start of a time step, the material points carry
by following, current state variables of acceleration a, velocity v,
Np mass m, volume V, and deformation gradient F.
∫Ωc F d Ωc ≈ ∑ υ p F ( p) (4) Meanwhile, the position of a material point can
p =1
be directly takenfrom the spatial position of mate-
rial points in the computation domain. These state
where F is an arbitrary function to be integrated variables are then interpolated to the computation
over the computation domain, xp is the location of grid using the standard shape function interpola-
material point p and υp is the volume of the mate- tion. The nodal velocity (and nodal acceleration)
rial point p. The internal force vector fiint can be can be interpolated by using conservation of
approximated by, momentum,
Np Np
fiint ∫Ωc ∇Si σ d ∑ p∇Si p (x p )
(5) mi i ∑ ρν pSi ( p )v p (9)
p =1 p =1
25
In the Updated Langrangian formulation, stiff- gradient term, ∇Si (x) with ∇S i ( ), the internal
ness matrix is defined by, force vector Eq. 5 can be defined as,
Np
K ijk −1 = ∫ ∇Si C T
∇S j d Ω
Ωc f i int = − ∑ v pσ p : ∇ S i (x p ) (13)
+ ∫ ∇Si σ ∇S j d Ω (10) p
Ωc
where Ωc is the current configuration of computa- To ensure conservation of energy, all the gra-
i ( ) consistently.
dient quantities have to use ∇S
tion domain, Cσ T is the Truesdell rate of elastic tan-
gent modulus and σ is the Cauchy stress tensor. The The gradient of displacement at location x can be
second identity of Eq. 10 also shows that stiffness calculated by,
matrix includes the term of material nonlinearity N
( ) = u ∇
(first term) and geometrical nonlinearity (second
term). The tangent modulus tensor depends on the
∇ ∑ i S i ( x ) (14)
i =1
type of constitutive model of the material and will
not be elaborated in here. Eq. 6 is solved to obtain The term ∇S i ( ) has enlarged influence to
incremental displacement du. Computation grid is neighbouring elements, where ∇ i ( ) ≠ 0 if com-
then deformed with du, and kinematics of the sys- putation grid element of node i and computation
tem is then updated before next iteration begins. grid element of location x are neighbouringele-
The update of velocity term is given by, ments. As a result, conventional FEM assembly
method cannot be used. In order to include the
2 k influence of neighbouring elements, the calcula-
v ik = v ik −1 + duu i v i0 (11)
Δt tion of gradient of displacement is done by using
two separate assembly procedures. First, the nodal
while the nodal acceleration a ik is updated by using gradient of displacement is calculated by mapping
Eq. 8. If the Newton procedure is converged with gradient of displacement to the nodes with,
defined convergence criteria, a convective stage of N
) = 1
current time step will be performed. In this stage,
the solution from the computation grids is inter-
∇ j
Vj
∑ ui ∫Ωc S j ∇Si d Ω (15)
i =1
polated back to material points by using shape
functions. By the end of the convective stage, the Subsequently, the gradient of displacement at x
computation grid will be disposed because all the is readily interpolated from the nodal gradient of
required state variables are carried by material displacements by using the shape function,
points to the subsequent time steps.
N
) S (x)
∇ ( ) = ∑ (∇ (16)
j j
j =1
3 FORMULATION OF THE STIFFNESS
IN DDMP
Eq. 16 can be computed by using the conven-
3.1 Continuous gradient of DDMP tional FEM local assembly, while Eq. 15 is com-
puted using global tensor multiplication. Because
DDMP calculates the value involving gradient of the integral in Eq. 15 is calculated independently,
the shape function by doing two-step mapping we have redefined it as a tensor,
procedure (Zhang et al. 2011). The method intro-
duces a continuous gradient of the shape function, 1
which is defined as,
Aij =
Vj ∫Ωc S j ∇Si d Ω (17)
N S ( )
i ( ) =
∑ V ∫Ωc S j ∇Si d Ω Meanwhile, the assembly of stiffness matrix also
j
∇ (12)
j =1 j requires the calculation of gradient quantities. By
using the same analogy of mapping material point
stress to the node of computation grid in Eq. 15,
where Vj, is the nodal volume of computation grid
( j ∫ ( x ) is continuous over the
)).. ∇
tangent modulus tensor of material stiffness at
Ωc j material point is mapped to the nodes of computa-
boundary of computation grid because it is calcu- tion grid by,
lated by interpolation of the shape function. This
continuous gradient is able to solve the problem of p N
1
numerical noise caused by discontinuity of gradi- ( στ
)mn ∑
VmVn p
p m(
στ
p )C ( p )Sn ( p) (18)
ent quantities. By substituting the conventional
26
With the nodal tangent modulus, (Cστ)mn, quan- from travelling out of the computation domain,
tity at location x can be approximated by using a buffer zone is provided on the free surface of the
shape function, MPM region.
1 N N
C T (x) = ∑ ∑ Sm ( x )(Cστ )mn Sn ( x )
VmVn m =1 n =1
(19)
5 MODELLING OF PILE PENETRATION
SIMULATION
Substituting Eq. 19 into the first term of
Eq. 10, 5.1 Two dimensional model
N
Fig. 1(Left) shows a symmetrical half of a sheet
1 N N p pile penetrated into a soil medium. The mesh was
Kij = ∑ ∑ ∑ ...
VmVn m =1 n =1 p generated using Gmsh (Geuzaine & Remacle 2009).
The model was fixed at the sides horizontally, and
× pSm ( p ∇Si ( p)(Cστ )mn n ( xp)∇Sj ( p) (20) fully fixed at the bottom. The analysis was carried
N N out by coupling MPM analysis and FEM analysis.
= ∑ ∑ A im ( )mn ATnj (21) MPM region was located exactly below the pile,
m =1 n =1 where extreme deformation during pile penetration
process. The mesh was progressively finer towards
The assembly of stiffness matrix is now simply the penetration region to simulate more accurate
a product of nodal tangent modulus tensor (Cστ)mn result. The mesh region with less deformation
and A tensors. Eq. 21 is calculated by first assem- was defined as FEM region. Conventional FEM
bling the global tensor (Cστ)ij using Eq. 18. This ten- method was used to calculate this region during
sor is a container of fourth order tensor described the analysis, and relatively coarse mesh was used to
on nodes i and j. A tensor is also assembled locally save calculation cost. In the MPM region and the
to form a global tensor. Because the size of both pile region, the deformed mesh was relaxed back
terms is large but the elements are mostly zero val- to its previous undeformed stage at the end of each
ued, sparse matrix is used. load step to prevent excessive deformation in the
The direct use of continuous gradient may cause mesh. A buffer zone of meshes was defined on
numerical instability because of zero-energy mode, top of the soil to contain the material points from
thus Zhang et al. 2011 introduced a modified gra- translating beyond the computation grid. Inthis
dient of shape function that combines both con- simulation, the contact between the pile and the
tinuous and conventional gradient quantities. This soil was simulated by the inherent no-slip contact
hybrid approach was implemented in this work as of MPM. A very fine layer of mesh was predefined
well. on the predicted soil and pile contact region to
increase the contact accuracy. The Mohr-Coulomb
failure criterion was used as the constitutive model
4 COUPLING OF MPM WITH FEM of the soil. We chose three soil models with differ-
ent cohesion values of 0.25 kN/m2, 0.50 kN/m2 and
The analysis of MPM requires more computation
effort than FEM. To couple the calculation process
between MPM and FEM, a mesh relaxation tech-
nique is introduced. In this technique, both MPM
and FEM are sharing the same computation mesh.
A special mesh relaxation procedure is carried out
only on MPM region at the end of the load step
calculation to recover the deformed mesh back to
the initial undeformed configuration. To perform
relaxation procedure, the deformed mesh is applied
with same boundary conditions but the external
force contribution is taken out during the relaxa-
tion process so that the mesh will be deforming
back to initial configuration. Addionally, the stiff-
ness of the FEM region is temporarily increased to
prevent the relaxation procedure from “relaxing”
the mesh on the FEM region. Thus, the relaxation Figure 1. (Left)2D plane-strain model of the pile
procedure will only recover the deformed mesh in penetration simulation. (Right) 3D model of the pile
the MPM region. To prevent the material points penetration simulation.
27
1.00 kN/m2 to simulate this pile penetration proc- contact. The maximum shear stresslimit is exactly
ess. The soil has elastic stiffness, Es, of 100 kN/m2 the cohesion value of the soil. The shear stress
and Poisson ratio of v = 0.33. Meanwhile, the pile concentration starts at the corner of pile tip, and
stiffness, Ep, was set to 20,000 kN/m2 and the Pois- spreads out beneath the soil at about 45°. The soil
son ratio, v was chosen to be zero. The pile was is also yielded along the contact region between
defined as a hyperelastic material. Both the soil the side of the pile and the soil. The average ver-
and the pile were set to be weightless (γsoil = 0, tical stress on top of the pile has been measured
γpile = 0).The pile penetration was carried out by to determine the pressing force during penetration
applying prescribed displacement of 0.05 m per process. Fig. 3 shows the stress-penetration depth
step at the top of the pile until a maximum of plot of the simulation for soil samples with dif-
2.5 m penetration depth. Linear triangular ele- ferent cohesion values. All the soil samples show
ments were used for the meshes, and 12 material distinct elastic zone and hardening zone. In the
points per element were prescribed on the Gauss hardening zone, the vertical pile stress increases lin-
integration points initially. Separately a compari- early with the penetration depth. The contact gap
son of mesh fineness at contact region was carried between the side of the pile and the soil is directly
out with quadrilateral elements to investigate the influenced by the element sizes in the contact zone.
influence of rigid contact to the penetration force. Three different meshes with varying fineness are
Both MPM and DDMP analyses were carried out used to examine the effect of mesh refinement on
in these simulations. the analysis. Fig. 4 shows the plots of the vertical
stresses of the pile against the penetration depth
for the three samples of mesh refinements for the
5.2 Three dimensional model
first 1.0 m of penetration. The simulation with
Fig. 1(Right) shows the 3D model of the simula-
tion. The pile was penetrated into the soil by 0.02 m
per step until a maximum of 2.0 m, or about two
third of the actual pile length. Purely cohesive soil
with cohesion value of c = 1.0 kN/m2 was used
in this simulation. The analysis used linear tetra-
hedron elements, where the finest element size in
the contact area was set to 0.10 m, while the size
was progressively increasing in size to 0.50 m at the
boundary between the MPM region and the FEM
region. A total of 40 material points per element
initially was used.
6.1 2D Simulation Result Figure 3. Vertical stress values of the pile with increas-
ing penetration depth for all three soil samples with
Pile penetration for selected load steps is presented
different cohesion values.
in Fig. 2 for the simulation of soil with cohesion
of c = 0.5 kN/m2. High shear stress concentration
has been located at the corner of the pile and soil
28
coarser mesh has given stiffer response, where
higher transitional stress has been reached before
the soil starts to have hardening response. To vali-
date the stress-displacement response, an analytical
solution is established by calculating the pile load
capacity from the contribution of soil end bearing
capacity and the rigid skin friction. The Terzaghi’s
equation of calculating the bearing capacity for
shallow foundation is used (Das 2007). Meanwhile,
the skin friction of the pile is calculated by taking
the rigid friction resistance of the soil, which is the
cohesion of the soil. The skin friction resistance
can be readily approximated by taking the product
of soil cohesion and total contact area. Thus the
analytical solution plotted as the dashed line in Figure 5. Vertical stress of the soil, σyy, along the sec-
tion beneath the pile, for MPM and DDMP at selected
Fig. 4 is given as, pile penetration depths.
2csoil d
qt = qb + , (22)
b
6.1.1 Comparison of DDMP with MPM the figure, element discretization effect is clearly
In this case of comparison between DDMP and seen in the stress distribution in the MPM result,
MPM analyses, we have used the results obtained where clear discontinuity of stress values can be
from the model with quadrilateral mesh with finest seen. This discontinuity is reduced by the DDMP
mesh size of 0.0625 m at the contact region. The analysis.
initial number of material points per element is 16.
The soil is defined with cohesion, c = 0.5 kN/m2.
MPM analysis results show that vertical stress, σyy, 7 CONCLUSIONS
along the soil below the pile for selected penetration
depths has much higher numerical noise, which is The extreme deformation during pile penetration
reduced by DDMP analysis. This is depicted in has been successfully simulated using MPM and
Fig. 5. These numerical noisesin MPM has caused DDMP, which allow to overcome mesh tangling
the line search method in Newton iterative proce- problems of conventional FEM. In the simulation
dure to have slower convergence rate than DDMP. with purely cohesive soil samples, the pile resist-
ance of the simulation is compared with the ana-
lytical solution. The comparison has shown that
6.2 3D Simulation result
the simulation accurately predicted the pile stress,
Fig. 6 shows the shear stress colour maps of σxy at which is contributed by the end bearing of the pile
penetration depth of 1.00 m. The maximum shear and rigid contact resistance between the soil and
stress distribution, which is also the cohesion value the pile. The mesh refinement analysis meanwhile
of the soil, are spreading from the tip corner of the shows that coarser mesh has produced stiffer pile
pile. Maximum shear stress of the soil is activated resistance. Meanwhile, the DDMP analysis has
along the rigid contact next to side of the pile. In shown less numerical noise in the stress results. We
29
have also successfully coupled the MPM and the to extend applicability of the material point method
FEM analyses using the proposed mesh relaxa- for problems involving massive deformations.
tion method. The coupled system has shown an International Journal for Numerical Methods in
improvement in computation efficiency. Engineering 86, 1435–1456.
Sulsky, D., Zhou, S. & Schreyer, H. 1995. Application of
a particle-in-cell method to solid mechanics. Computer
Physics Communications 87, 236.
REFERENCES Wells, G.N. 2009. The Finite Element Method: An
Introduction. University of Cambridge and Delft
Bardenhagen, S.G. & Kober, E.M. 2004. The generalized University of Technology.
interpolation material point method. Computer Wieckowski, Z. 2004. The material point method in large
Modeling in Engineering and Sciences 5(3), 477–495. strain engineering problems. Computer Methods in
Das, B.M. 2007. Principles of Foundation Engineering. Applied Mechanics and Engineering 193, 4417–4438.
Cengage Publisher. Zhang, D.Z., Ma, X. & Giguere, P.T. 2011. Material
Geuzaine, C. & Remacle, J.F. 2009. Gmsh: a three- point method enhanced by modified gradient of
dimensional finite element mesh generator with built-in shape function. Journal of Computational Physics 230,
pre- and post-processing facilities. International Jour- 6379–6398.
nal for Numerical Methods in Engineering 79(79),
1309–1331.
Harlow, F.H. 1964. The particle-in-cell computing
method for fluid dynamics. Methods in Computational
Physics 3, 319.
Sadeghirad, A., Brannon, R.M. & Burghardt, J. 2011.
A convected particle domain interpolation technique
30
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
S. Tan
Geo-Engineering Section, Faculty of CITG, Delft University of Technology, Delft, The Netherlands
M.A. Hicks
Geo-Engineering Section, Faculty of CITG, Delft University of Technology, Delft, The Netherlands
Deltares, Delft, The Netherlands
A. Rohe
Deltares, Delft, The Netherlands
ABSTRACT: In geotechnics, it is common to have a thin layered material with high stiffness on top of
soil, to prevent damage from external loading or erosion. To model this numerically a very fine mesh is
often needed, which decreases the critical time step in explicit time integration algorithms, severely affect-
ing simulation performance. The use of 2D plate elements connected to 3D elements is investigated to
overcome this problem. The presented three-noded plate element is based on Kirchhoff thin plate theory,
and uses non-conforming polynomial shape functions. The lumped mass matrix, with both the transla-
tional and rotational degrees of freedom considered, is implemented with an explicit time integration
scheme. This plate element is coupled with volume elements and the implementation is tested for several
cases in which analytical or numerical solutions are available. All simulations show that the plate element
with the lumped mass matrix is working properly in geotechnical problems.
31
cular concrete plate resting on a homogeneous half
space. Finally, the conclusions are provided.
2 METHODOLOGY
32
where Ai indicates the area related to node i in an
element, as shown in Figure 2.
where the moments of inertia can be defined as, where t is the thickness of the plate and ν is Pois-
son’s ratio.
In order to reach an the equilibrium state for the
J x1 ∫∫ y12dx
d 1dy
d1 J y1 ∫∫ x12dx1dy1 dynamic problem, the local damping is introduced
A1 A1 as defined in FLAC (1998) by adding an artificial
damping term into equation (11), i.e.
∫∫ ∫∫ ρx2 dx2dy2
2 2
Jx 2 y2 dx2dy2 J y2 (10)
A2 A2
M L w = F ext Fint − F damping (16)
∫∫ ∫∫
2 2
Jx3 y3 dx
d 3dy
d 3 , J y3 x3 dx
d 3dy3
A3 A3 in which Fdamping is the damping force vector. Its
components are defined as,
33
3.2 Circular plate with simply supported edge
Fi damping α Fi ext − Fiint sig (wi ), i=1,2, Ndof
Figure 4 shows a simply supported circular plate
(17) with a point load applied at its centre. The plate is
modelled as linear elastic, with a Young’s modu-
where α is the damping factor and Ndof is the total
lus E = 200 MPa, and Poisson’s ratio ν = 0. The
number of degrees of freedom.
radius of the plate is a = 0.45 m and the thickness is
t = 0.1 m. The applied point load is q = 10 kN.
The analysis has been done using a local damp-
3 NUMERICAL SIMULATIONS ing factor of 0.5 to reach the equilibrium state.
OF 2D PROBLEMS Figure 5 shows the deflection of the loaded point
as a function of time, with the final deflection of
The plate element presented above has been imple- 7.24 mm being in close agreement to the analytical
mented in a 3D code and coupled with tetrahedral solution of 7.25 mm. Figure 6 shows the contour
elements for representing the underlying soil. The plot of the deflection of the plate.
implementation has been tested for both 2D and
3D cases in which analytical or numerical solutions
are available for validation.
34
4 NUMERICAL SIMULATIONS
OF 3D PROBLEMS
35
Figure 11. Vertical stresses in a two layer system
(Burmister, 1958).
36
analysis. The static problems considered in this Bazeley, G.P., Cheung, Y.K., Irons, B.M. & Zienkie-
paper were solved using a transient dynamic pro- wicz, O.C. 1965. Triangular elements in plate bending
cedure with the dynamic relaxation method. Dif- conforming and non-conforming solutions. Proceed-
ferent cases were studied in 2D and 3D, and the ings Conference on Matrix Methods in Structural
Mechanics, Wright Patterson A.F.B., Ohio. 547–576.
results were validated using analytical or numerical Burmister, D.M. 1945. The general theory of stresses and
solutions. All simulations showed that the lumped displacements in layered systems. International Jour-
mass matrix works properly with the plate ele- nal of Applied Physics, 16: 89–94.
ment. The plate element can be applied to layered Burmister, D.M. 1958. Evaluation of pavement systems
geotechnical problems, to simulate a thin layered of the WASHO road test by layered systems methods.
material instead of using volumetric elements. Highway Research Board Bulletin 177.
FLAC 1998. Fast lagrangian analysis of continua: theory
and background. Itasca Consultin Group, Inc., Min-
ACKNOWLEDGEMENT nesota, USA.
Reddy, J.N. 2007. Theory and analysis of elastic plates and
shells. CRC, Taylor and Francis.
This research is funded by the CSC (China Scholar- Specht, B. 1988. Modified shape functions for the three-
ship Council), TU Delft and Deltares. The authors node plate bending element passing the patch test.
appreciate the contributions of Professor Vermeer International Journal for Numerical Methods in Engi-
from Deltares, and Professor Wieckowski from neering, 26: 705–715.
Lodz University of Technology and Deltares, who Surana, K.S. 1978. Lumped mass matrices with non-zero
gave valuable advice while doing this research. inertia for general shell and axisymmetric shell ele-
ments. International Journal for Numerical Methods in
Engineering, 12: 1635–1650.
Yoder, E.J. & Witczak, M.W. 1975. Principles of pave-
REFERENCES ment design (2nd Edition). John Wiley & Sons.
Batoz, J.L., Bathe, K.J. & Ho, L.W. 1980. A study of
three-node triangular plate bending elements. Interna-
tional Journal for Numerical Methods in Engineering,
15: 1771–1812.
Bauchau, O.A. & Craig, J.I. 2009. Structural analysis
with applications to aerospace structures. Springer,
The Netherlands.
37
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
38
3.3 Random finite element method the method of local average sub-division (Fen-
ton & Vanmarcke 1990) to maintain global mean
To incorporate the effect of heterogeneous materials
properties.
into the finite element method, the Random Finite
The material property that is varied can be
Element Method (RFEM) has been developed,
expressed as:
where random spatial perturbations in material
properties are included in analyses via the incorpo-
ration of random fields e.g. Hicks and Samy 2002; u( ) = μ( z ) + σ( z )Z ( ) (3)
Hicks and Spencer 2010 and as described in sec-
tion 3.2. In this approach, the material properties where u is the material properties to be varied, μ(z)
are represented by a mean and standard deviation, is the mean, σ(z) is the standard deviation and Z(x)
and by a definition of the spatial variation and cor- is the local average. The local average is based upon
relation. This leads to a stochastic analysis, where an Gaussian distribution and a spatial correlation
multiple realisations (100 s or 1000s) are analysed function, giving a covariance function for an iso-
and compiled. The results of the analyses may be tropic random field, after Hicks & Samy (2002)
quantified in terms of risk and reliability, in con- and Hicks & Spencer (2010), of:
trast to traditional deterministic analyses which
β τ ) = σ 2 exp ⎛ − τ ⎞
generally result in a single characterisation e.g. a 2
(4)
single factor of safety. ⎝ θ ⎠
⎡ ⎛ 2 ⎞⎤
⎛ K ⎞
θc = 2 tan−1 ⎢⎢ ⎜ I − sign ⎥
1 K
i ( K II ) ⎜ I ⎟ + 8 ⎟ ⎥
4 ⎜ K II ⎝ K II ⎠ ⎟
⎢⎣ ⎝ ⎠ ⎥⎦
(2)
where KI and KII are stress intensity factors.
39
4.5 Monte Carlo analysis
Monte Carlo analyses can then be undertaken uti-
lising a series of possible realisations, each realisa-
tion valid based upon the material statistics, and
the results compiled. Each realisation consists of
the random field properties being mapped to the
finite elements and then a deterministic XFEM
analysis. The resulting analysis provides a possi-
ble outcome, therefore the compilation of results
provides the range of possible outcomes. Suffi-
cient realisations must be undertaken so that the
required response of the system, in terms of prob-
ability, has converged. In terms of crack propaga-
tion the required response is likely to be in terms of
geometry, i.e. length, direction, angle, etc.
4.6 Justification
Bordas et al. (2007) showed that the inclusion of soft
and hard inclusions, with contrasting Young’s mod-
uli, created deviations within the crack propagation
as shown in Figure 2, with the analyses repeated by
the Authors’ here for verification purposes.
The presence of the inclusions in Figure 2, illus- Figure 3. Plots of predicted crack propagation pathway
trate that the hard inclusion (Figure 2(a)) repels in the presence of a soft inclusion of varying radii (solid
crack propagation while that of the softer inclu- line propagation path and dashed line inclusion).
sion (Figure 2(b)) attracts the path. This numerical
experiment shows and justifies the need to investi-
gate further the affects of spatial variability within
material properties that vary continuously within a
material when analysing crack propagation.
40
a analogous effect, with the propagation paths vary- It can be seen that the two paths are signifi-
ing with the spatially varying Young’s modulus. cantly different and show that the spatially vary-
Figure 4 illustrates the predicted crack propaga- ing Young’s modulus has a significant effect on
tion in the presence of regularly placed regions of the direction and path of the crack propagation.
contrasting Young’s moduli. Figure 7 shows the propagated crack paths for 100
realisations of the methodology.
The predicted paths are shown to be relatively
5 RESULTS AND DISCUSSION smooth, as would be expected due to the local aver-
aging within the random field. This is a character-
5.1 Initial model istic that may change as the random fields become
more abruptly changing, with larger gradients
A model of uni-axial tensile stress was modelled
within the material properties, changes in the mate-
with a central edge crack within a 6 m × 4 m
rial behaviour models, or discrete changes in mate-
domain, as shown in Figure 5.
rial properties. In fact, the model illustrated here,
The boundary conditions on the base of the
shows a material with slowly changing properties,
domain restrict vertical displacement and horizon-
or as the results of material averaging, considering
tal displacement is restricted on the left corner. At
a rather large representative elementary volume
the top of the domain a uni-axial vertical load (of
(or at least significantly larger than grain size).
unity) is enforced. No explicit boundary condi-
Where grain level behaviour, or other local behav-
tions are enforced on the fracture surface allowing
iours, causes much stronger discontinuities these
displacement to occur, but not to support load.
effects can also be incorporated into this model,
The material variation is then model via the
either by altering properties on a element basis or
generation of a random field for Young’s modu-
via XFEM enrichment itself.
lus, E, with a log-normal spatial variation where
θ = 1.0 m, mean, Eμ = 1 × 109 kN/m2, and standard
deviation Eσ = 2.5 × 109 kN/m2. The variation is
then mapped to the domain, during each realisa-
tion of the analysis.
The crack is then allowed to propagate through
the domain using 40 iterative steps of 0.1 m with
the direction of the fracture determined via the cir-
cumferential stress as shown in equation (2).
41
Figure 7. Predicted crack propagation paths of 100
realisations. Figure 8. Percentage probability of a fracture occurring
at any location using 100 realisations.
This simple initial numerical experiment shows ing the positions of likely crack initiation, and the
that by spatially varying the Young’s modulus, by probability of its occurrence.
means of random fields, results in deviations in the
propagation paths of the cracks. There is likely to
be an infinite number of statistically relevant path- 6 CONCLUSIONS
way solutions due to the infinite number of possi-
ble random field variations. These results indicate A method to simulate the fracturing process in het-
that utilising a single deterministic analysis could erogeneous geomaterials is presented. Initial results
cause significant design oversights. Therefore a demonstrate that coupling random fields into the
probabilistic means evaluating problems of this eXtended Finite Element Method provides a range
nature would be advantageous. of equally valid solutions. Further analysis can
Probabilistically the methodology can used to yield a probabilistic depiction of the fracture and
provide predictions of pathway or network likeli- fracture growth process.
hood as illustrated in Figure 8. This may provide
significant design advantage, for example where
fracture paths tend towards formation bounda- REFERENCES
ries or are distributed (non-) uniformly around
structures. However this methodology can require Belytschko, T. & Black, T. 1999. Elastic crack growth in
several hundreds if not thousands of realisations finite elements with minimal remeshing. Int. J., Numer.
and therefore significant computational resources. Meth. Engng. 45: 601–620.
However, with more recent computational devel- Bordas, S., Nguyen, P.V., Dunant, C., Guidoum, A. &
Nguyen-Dang, H. 2007. An extended finite element
opments, where many cores can be situated on a library. Int. J. Num. Meth. Engng. 71 (6): 703–732.
single chip, these sort of analyses can become more Fenton, G.A. & Vanmarcke, E.H. 1990. Simulation of
commonplace. Furthermore many other probabil- random fields via Local Average Subdivision. ASCE J.
istic performance measures can be easily incorpo- Eng. Mech. 116(8): 1733–49.
rated into this methodology and therefore it can Hicks, M.A. editor. 2007. Risk and variability in geotech-
be introduced into reliability and risk based design nical engineering. London: Thomas Telford.
and project management tools. Hicks, M.A. & Samy, K. 2002. Influence of heterogeneity
on undrained clay slope stability. Quart. J. Eng. Geol.
5.2.1 Future work Hydrogeol. 35(1):41–9.
Hicks, M.A. & Spencer, W.A. 2010. Influence of het-
This work is an initial investigation into the feasi- erogeneity on the reliability and failure of a long 3D
bility, reliability and usefulness of the methodology slope. Computers and Geotechnics 37 (7–8): 948–955.
presented. It is anticipated to be extended into a Moes, N., Dolbow, J. & Belytschko, T. 1999. A finite ele-
coupled hydro-mechanical formulation to increase ment method for crack growth without remeshing.
validity to further geomechanical processes. Int. J. Numer. Meth. Engng. 46: 131–150.
The robustness of the methodology will be fur- Pais, M.J. 2010. MATLAB eXtended Finite Element
ther tested and compared with experimental results Method (MXFEM): User guide. Gainesville, Florida:
from the laboratory testing. University of Florida.
The Authors are also keen to incorporate crack Shih, C. & Asaro, R. 1988. Elastic-plastic analysis of
cracks on biomaterial interfaces: part I—small scale
initiation into the model, not only as this aspect yielding. J. App. Mech. 55: 299–316.
is of practical importance, but also as the meth-
odology should be of particular use when predict-
42
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
M. Karstunen
Chalmers University of Technology, Gothenburg, Sweden
University of Strathclyde, Glasgow, Scotland, UK
N. Sivasithamparam
Plaxis BV, Delft, The Netherlands
University of Strathclyde, Glasgow, Scotland, UK
R.B.J. Brinkgreve
Plaxis BV, Delft, The Netherlands
Delft University of Technology, Delft, The Netherlands
P.G. Bonnier
Plaxis BV, Delft, The Netherlands
ABSTRACT: Due to the desire of reducing the embedded CO2 in construction and the pressure in public
finances to get more value for money in big infrastructure projects, the demands for the accuracy of deforma-
tion predictions increase. Instead of piling, alternative environmentally friendly and cost effective solutions,
such as preloading via surcharge, vertical drains and column methods, such as deep-mixing, are becoming
increasingly attractive. Installation of piles and ground improvement into the ground will modify the state of
the soil. This is sometimes beneficial, and sometimes detrimental, and so far this effect has been rarely taken
into account. One reason for this is that the numerical techniques and the constitutive soil models have not
been able to represent the changes in soil structure and state in a satisfactory manner. The aim of the Europe-
an GEO-INSTALL project has been to develop numerical techniques that can be used to model installation
effects in geotechnical engineering. A key part of this has been constitutive model development, and their ro-
bust implementation. The aim of this paper is to discuss some recently developed rate-dependent constitutive
models for structured clays, which formed the basis for new developments, resulting in a new rate-dependent
model able to represent the complex rate-dependent stress-strain behaviour of soft structured clays. The im-
portance of modelling key features of soil behaviour in the context of rate-dependency are discussed in the
light of experimental evidence, and demonstrated through a series of numerical benchmark simulations.
43
any apparent bonding between the particles. The reconstituted clays samples given the difficulties
latter results from natural bio-chemical processes in micromechanical testing of intact natural clays.
such as the precipitation of calcites by bacteria in During irrecoverable straining for kaolin clay there
the soil and/or some complex geo-chemical pos- is evidence on re-orientation of particles, and
sesses associated with sedimentation environment changes in particle contacts, at microstructural
and subsequent geological and flow history. level (see e.g. Bai & Smart 1996, Hicher et al. 2000).
Particularly fascinating is this aspect are the so- This supports the macro-level evidence that the
called sensitive clays, which in extreme cases are anisotropy of clays, when undergoing large strains,
referred to as quick clays. Highly sensitive clays can can evolve.
be found in large areas of Northern Europe and Based on experimental studies on natural
North America (Rankka 2003), and they tend to Otaniemi clay, Wheeler et al. (2003) proposed an
change from rather stiff consistency to a flowing elastoplastic model, called S-CLAY1, with a rota-
liquid gel when disturbed, forming a potential geo- tional hardening law describing the changes in the
hazard. Sensitive clays were formed when clay parti- inclination of the yield surface due to irrecover-
cles and other fine fractions were sedimented in cold able straining. The calibration of the parameters
and salty marine environment after the Pleistocene for the S-CLAY1 model is rather straightforward
epoch. Due to the bi-polar electro-chemical charges and the model has been thoroughly validated
associated with the clay minerals, in salty sedimen- experimentally by Karstunen and her co-workers
tation environments the clay particles flocculate and (Karstunen & Koskinen, 2004, 2008).
form very open card-house structures (see e.g. Pusch As shown by i.e., Burland (1990) & Leroueil &
1970). Since their formation the clays have been con- Vaughan (1990), when natural clays are undergoing
solidating and creeping under their self-weight. Due large deformations, the initial apparent bonding is
to leaching, the salt ions at the surface of the clay progressively lost, and at large strains, soil starts to
particles have dissolved. The possible leaching proc- behave as a fully destructured material. The degra-
esses involve percolating water through the deposit, dation of bonding due to irrecoverable straining is
water seeping upwards through the deposit due to referred to as destructuration (Leroueil et al. 1979).
artesian pressure, and diffusion of salts towards Most constitutive models that attempt to account
zones with lower ion concentrations (Rankka et al. for the effect of bonding and destructuration, uti-
2004). This way the initially stable structure of the lize the conceptual ideas by Gens and Nova (1993).
clays has become meta-stable and sensitive. The same applies to the S-CLAY1S model (Kars-
When piles or any column-type of ground improve- tunen et al. 2005), in which the sudden collapse of
ment are installed in the ground, there are large defor- the sensitive clay structure upon yielding is rep-
mations in the soils, involving both shearing and the resented using the concept of intrinsic yield sur-
lateral expansion of the soil. Consequently, there are face (Gens and Nova 1993) in combination with
changes in the vertical and horizontal effective stresses, the rotational hardening of the S-CLAY1 model
and the parameters associated with the state of the soil, (Wheeler et al. 2003). This was the first constitutive
such as void ratio, preconsolidation pressure, density, clay model that was able to represent the changes
K0 (coefficient of earth pressure at rest) etc. are modi- in the state of the structured clays in a simple com-
fied. As demonstrated by Dijkstra et al. (2010, 2011) prehensive manner. The next stage was to incorpo-
using photoelastic techniques, installation causes rate rate-effects into the model.
significant density changes and rotations of princi- Given it was thought that anisotropy and its evo-
pal stresses. The former are of course beneficial, but lution has a major effect on the soft clay response,
effects of the latter on soil state are usually ignored. the S-CLAY1 model was extended to account for
In granular media, pile driving may also cause particle creep and rate effects by Leoni et al. (2008) using the
crushing Lobo-Guerrero & Vallejo (2005) and break- creep formulation by Vermeer and his co-workers
age of the apparent bonding in the soil, which in turn (Vermeer et al. 1998, Vermeer & Neher 1999). The
may have a detrimental effect on the pile capacity. resulting model, called ACM (Anisotropic Creep
In this paper, the focus is on structured natural model), has the advantage that the model param-
clays, and modelling the changes in their stress- eters are easy to derive. Most importantly, the
strain response due to loading and possible instal- concept of reference time or time shift (see Leoni
lation effects. et al. 2008) enables, for the first time a systematic
way to take into account the strain-rate in the tests
that are used for defining the model input param-
2 CONSTITUTIVE MODELLING eters. This is a very powerful feature of the model,
OF STRUCTURED CLAYS which gives it a major advantage over the so-called
overstress models, based on Perzyna’s (1963) over-
The response of clays is dominated by their micro- stress theory, proposed by e.g. Hinchberger & Gu
structure, which has been studied with the help of (2009) and Karstunen & Yin (2010).
44
As discussed the in following, and pointed out
by Grimstad et al. (2010) and Karstunen & Yin
(2010), some of the key assumptions in the ACM
model are, however, both inconsistent with experi-
mental evidence and fundamentally wrong (see
also the companion paper by Sivasithamparam
et al. 2013). As a consequence, in practical context,
the ACM model often ends up grossly overpre-
dicting deformations in structured soft soils (see
Karstunen et al. in press), when the values of soil
constants are objectively selected. This affects the
predictive ability of the model and prevents its use
in practical context on structured soil deposits.
The work by Hinchberger & Gu (2009), Kars-
tunen & Yin (2010) and Yin et al. (2011), demon-
strates that for predicting certain features of natural
clay behavior, such as tertiary creep and creep rup-
ture, or indeed the phenomenon of progressive
failure, it is also necessary to account for the effects
of bonding and destructuration. In the following
some features of the ACM model are highlighted,
which necessitated the development of a new rate- Figure 1. Current state surface (CSS) and normal con-
dependent model Creep-SCLAY1 (for details see solidation surfaces (NCC) of the ACM model and the
the companion paper Sivasithamparam et al. direction of viscoplastic strains (triaxial stress space).
2013). The importance of modelling key features
of soil behaviour in the context of rate-dependency
are discussed and demonstrated through a series of In the ACM model, there is a separate failure sur-
numerical benchmark simulations. face on which zero dilatancy condition is imposed
in order to comply with the condition of zero
volumetric strain rate at critical state. Hence, the
3 ACM AND CREEP-SCLAY1 MODELS there is no “dry side” of the critical state. Because
an associated flow is assumed on the CSS, there
The creep formulation in the ACM model (Leoni is actually a “jump” in the volumetric creep rate
et al. 2008) is based on the idea of a Normal Con- when approaching critical state (see Figure 1): the
solidation Surface (NCS), see Figure 1, which is constant volumetric creep rate at the CSS surface
treated as the contour of constant volumetric creep suddenly reduces to zero (results of simulations
strain rate. For simplicity, the model is plotted in showing this can be found in Sivasithamparam
Figure 1 in triaxial stress space, in terms of mean et al. 2013).
effective stress p’ and deviator stress q. The scalar Most importantly, because the volumetric creep
α represents the current degree of anisotropy, M strain rate is assumed to be constant, large volu-
is the stress ratio at critical state and ψ is the dila- metric creep strains and very large deviatoric creep
tancy angle. strains are predicted in the stress space between
The volumetric creep strain rate is given by a the K0 line and critical state line. As shown in the
simple power law as follows: simulations in this paper, this results in severe
overprediction of deformations at boundary value
problems. Because of these assumptions, quite
β
μ ⎛ peq
′ ⎞ λ κ* significant apparent strain-softening is predicted
εvc = ⎜ ⎟ with β = (1) in undrained simulations of shearing of normally
τ ⎝ p′p ⎠ μ* consolidated or lightly overconsolidated samples.
Furthermore, as shown by Sivasithamparam et al.
where μ* is the modified creep index, λ* in the (2013), the ACM model cannot model the isotach
modified compression index, κ * is the modified behaviour that is observed in soft clays caused by a
swelling index and τ is the reference time (see Leoni step-change in strain rate. This is not satisfactory,
et al. 2008 for details). The ratio in the brackets is and these insufficiencies inspired further model
an inverse of a generalised yield stress ratio. Hence, developments.
creep is occurring even within the overconsolidated In the Creep-SCLAY1 model (Figure 2), instead
region, when the current stress surface (CSS) in of assuming constant volumetric creep strain rate,
Figure 1, is smaller than the NCS. similarly to Grimstad et al. (2010) and Yin et al.
45
Figure 2. Current state surface (CSS) and normal con- Figure 3. The Creep-SCLAY1 model (general stress
solidation surfaces (NCC) of the Creep-SCLAY1 model space).
and the direction of viscoplastic strains (triaxial stress
space).
the general stress space as the CSS and NCS rotate
(see Figure 3). In order to incorporate the effect
(2011), it is assumed that the viscoplastic multi- of bonding and destructuration in some of the
plier is constant along the CSS. The viscoplastic simulations, an intrinsic surface has been added
multiplier is defined in such a way that the model together with a destructuration law that is analo-
gives identical volumetric creep strain rate to Eq. gous with the S-CLAY1S model (Karstunen et al.
(1) under oedometric loading (see Grimstad et al. 2005). In the following, this version of the model
(2010) for details) as: is referred to as the Creep-SCLAY1S model. It is
basically a rate-dependent model that accounts for
β
μ ⎛ peq
′ ⎞ ⎛ M 2 − α 2 NC ⎞ the changes in fabric arrangement and bonding,
K0
εvc = ⎜ ⎟ ⎜ 2 ⎟ (2) enabling a rather complete representation of the
τ ⎝ p′p ⎠ ⎜⎝ M − ηK2 NC ⎟⎠ stress-strain behaviour of structured clays.
0
The usability of a constitutive model in a practi-
cal context relies on its robust implementation into
Where η = q / p′ is the stress ratio, and subscript
the finite elements. The Creep-SCLAY1 model
K0NC refers to the normally consolidated K0 stress
is implemented into the PLAXIS (Brinkgreve
path. The corresponding strain rate vectors are
et al. 2010) finite element code as a user-defined
plotted in Figure 2. Due to the evolution of ani-
soil model (UDSM) using a Modified Newton-
sotropy, the Creep-SCLAY1 model predicts some
Raphson scheme (Sivasithamparam 2012), whilst
minor apparent strain softening in undrained
the ACM model is using a fully implicit Newton
shearing in normally consolidated region. Because
iteration (Leoni et al. 2008). In the following the
there is now a “dry side”, with increasing strain
model is applied to a typical benchmark problem,
rate it is also possible for the stress path to pass
a simple embankment.
over the critical state line, which is in agreement
with experimental evidence (see simulations in
Sivasithamparam et al. 2013).
4 BENCHMARK SIMULATIONS
For finite element analyses, the model needs to
be generalised. Instead of a scalar α, the anisot-
4.1 FE model and model input
ropy is described with a fabric tensor, and invari-
ants can no longer be used (see Wheeler et al. 2003 The performance the ACM model and the Creep-
for details). In order to account for Lode angle SCLAY1 model are demonstrated in a benchmark
dependency, instead of the Drucker-Prager model problem using the 2D PLAXIS finite element
that assumes constant M, the formulation of Sheng code. An embankment constructed on soft soil is
et al. (2000) has been adapted, in order to have a assumed to be 2 m high, with a width at the top
smooth variation of the Lode angle of the fabric in of 10 m and the side slopes with a gradient of 1:2.
46
The soft soil is assumed to have the properties of Such a model would of course not be able to repro-
soft Bothkennar clay (Symposium 1992), which duce the stress-strain response in cases where there
extends to a depth of 30 m. The groundwater table are major changes in the stress path directions, and
assumed to be located at 1 m below the ground sur- in principle would not be able to reproduce the ele-
face. At the surface there is a 1 m depth over-con- ment level tests done by Karstunen & Koskinen
solidated dry crust. The problem has been analysed (2008).
as a small strain simulation. The values for the
model parameters for the soft clay layer are shown
4.2 Results
in Table 1. The embankment, assumed to be made
of granular material, was modelled with a simple The predicted vertical displacements by ACM
Mohr Coulomb model and the same applies to and Creep-SCLAY 1 as function of time are plot-
the dry crust layer (see Table 2 for material param- ted in Figure 4 and the corresponding settlement
eters). This problem is expected to be dominated through at the end of construction and at the end
by the soft soil response and is not sensitive to the of consolidation have been plotted in Figure 5.
embankment and crust parameters. The settlement predicted by both models are huge
The most difficult model parameter to deter- (unrealistically large considering that the embank-
mine is constant ω representing the evolution of ment is only 2 m high. The results clearly demon-
anisotropy. Hence, for comparison the simulation strate that the problem is not a one-dimensional
is also run with a version of Creep-SCLAY1 model problem, as for K0 consolidation the predictions by
where the anisotropy has been fixed, i.e. only initial the two models would have been the same.
anisotropy is considered, with no evolution of ani- The introduction of the constant viscoplastic
sotropy. This assumption was made e.g. by Bodas multiplier, is clearly improving the results, and this
Freitas et al. (2011) in their rate-dependent model. is even more apparent in Figure 6, where the lateral
Parameter e0 OCR K0 α0 χ0
γ
Parameter [kN/m3] κ* ν′ λ* M
Parameter λ*i ωd ω ζ ζd
Figure 4. Predicted vertical displacements as a function
Soft clay 0.06 1.0 50 9 0.2 of time by ACM and Creep-SCLAY1.
τ
Layer μ*i μ* [days]
γ Ε ϕ′ c′ ψ
Layer [kN/m3] [kPa] [°] [kPa] ν′ [°]
47
Figure 7. Predicted vertical displacements as a func-
tion of time by Creep-SCLAY1, Creep-SCLAY1 with no
Figure 6. Predicted lateral displacements by ACM and rotation and Creep-SCLAY1S.
Creep-SCLAY1 at the end of consolidation.
48
these are constants for structured natural clays. tural effects by advanced macroscopic constitutive
Due to the gradual degradation of bonds, both models. These models have also the advantage that
the apparent compression index and the appar- they enable us to improve our understanding of
ent creep index have maximum values just after what might be happening, when piles and ground
the onset of yield (see e.g. data by Graham et al. improvement elements are installed into soft struc-
1983). Indeed, even their ratio (and consequently tured clays.
exponent β in Eq. 1) is not constant. As shown by
Karstunen & Yin (2010) this type of behaviour
can be represented with a rate-dependent model ACKNOWLEDGEMENTS
that accounts for the effect of bonding and its
gradual destruction. The work presented was carried out as part of a
In order to explore the effect of bonding and Marie Curie Industry-Academia Partnerships and
destructuration at boundary value level, the bench- Pathways project GEO-INSTALL (PIAP-GA-
mark was simulated with Creep-SCLAY1S model, 2009-230638). The first author is also benefitting
i.e. an extension of the Creep-SCLAY1 model that from the financial support by Marie Curie Indus-
accounts additionally for initial bonding and the try-Academia Partnerships and Pathways project
subsequent degradation of bonding. The results CREEP (PIAP-GA-2011-286397). The experimen-
have been included in Figures 7, 8 and 9, labelled tal work facilitating the model developments was
Creep-SCLAY1S. The predicted vertical and hori- done at Aalto University sponsored by Academy
zontal deformations have reduced by nearly 45% of Finland (Grant 128459).
and 55% respectively, and most importantly the
unrealistic large deformations due to the in situ
stresses only have virtually vanished. REFERENCES
49
Hicher, P.Y., Wahyudi, H. & Tessier, D. 2000. Microstruc- Pusch, R. 1970. Clay microstructure. A study of the
tural analysis of inherent and induced anisotropy in microstructure of soft clays with special reference to
clay. Mechanics of Cohesive-Frictional Materials 5(5): their physical properties. Swedish Geotechnical Insti-
341–371. tute, Proceedings No 24. Stockholm.
Hinchberger, S.D. & Qu, G. 2009. Viscoplastic constitu- Pusch, R. Personal communication, 2012.
tive approach for rate-sensitive structured clays. Cana- Rankka, K. 2003. Kviklera—billdning och egenskaper. SGI
dian Geotech. J. 46(6): 609–626. Varia 526. Swedish Geotechnical Institute, Linköping.
Karstunen, M. & Koskinen, M. 2004. Anisotropy and Rankka, K., Andersson-Sköld, Y. Hulten C., Larsson, R.
destructuration of Murro clay.” Proc., Advances Lerocex, V. & Dahlin, T. 2004. Quick clays in Sweden SGI
in Geotechn. Eng. Skempton Conf., London, UK, Report 65. Swedish Geotechnical Institute, Linköping.
1: 476–487. SGI. 1995. Geotekniska skadekostnader och behov av
Karstunen, M. & Koskinen M. 2008. Plastic anisotropy ökad geoteknisk kunskap. Internal Report K94/1825/3.
of soft reconstituted clays. Canadian Geotechnical In Swedish, not publicly available.
Journal 45: 314–328. Sheng, D., Sloan, S. & Yu, H. 2000. Aspects of finite ele-
Karstunen, M., Krenn, H., Wheeler, S.J. Koskinen, M. & ment implementation of critical state models. Compu-
Zentar, R. 2005. Effect of anisotropy and destructu- tational Mechanics 26: 185–196.
ration on the behaviour of Murro test embankment. Sivasithamparam, N. 2012. Modelling creep behaviour
ASCE International Journal of Geomechanics 5(2): of soft soils. Internal report. Plaxis B.V. & University
87–97. of Strathclyde, not publicly available.
Karstunen, M., Rezania, M, Sivasithamparam, S. & Sivasithamparam, N., Karstunen, M., Brinkgreve, R.B.J., &
Yin, Z.-Y. (in press). Comparison of anisotropic Bonnier P.G. 2013. Comparison of two anisotropic
rate-dependent model for modelling consolidation rate dependent models at element level. Proc. Interna-
of soft clays. International Journal of Geomechanics tional Conference on Installation Effects in Geotechni-
doi:10.1061/(ASCE)GM.1943–5622.0000267. cal Engineering, Rotterdam, NL, 23–27 March 2013.
Karstunen, M. & Yin, Z.-Y. 2010. Modelling time-de- Symposium. 1992. Bothkennar soft clay test site: Charac-
pendent behaviour of Murro test embankment. Géo- terization and lessons learned (Géotechnique sympo-
technique 60(10): 735–749. sium in print). Géotechnique 42(2): 161–380.
Leoni, M., Karstunen, M. & Vermeer, P.A. 2008. Aniso- Vermeer P.A. & Neher H.P. 1999. A soft soil model that
tropic creep model for soft soils. Géotechnique, 58(3): accounts for creep. Proc. Int. Symp. Beyond 2000 in
215–226. Computational Geotechnics. Amsterdam. Balkema,
Leroueil S., Tavenas F., Brucy F., La Rochelle P. & Rotterdam, 249–261.
Roy M. 1979. Behaviour of destructured natural clays. Vermeer P.A., Stolle D.F.E. & Bonnier P.G. 1998. From
ASCE Journal of Geotechnical Engineering, 105(6): the classical theory of secondary compression to
759–778. modern creep analysis. Proc. Computer Methods and
Leroueil. S. & Vaughan, P.R. 1990. The general and con- Advances in Geomechanics. Balkema, Rotterdam.
gruent effects of structure in natural soils and weak Wheeler, S.J., Näätänen A., Karstunen M. & Lojander M.
rocks. Géotechnique, 40(3): 467–488. 2003. An anisotropic elasto-plastic model for natural
Lobo-Guerrero, S. & Vallejo, L.E. 2005. DEM analysis soft clays. Canadian Geotechnical Journal 40(2):
of crushing around driven piles in granular materials. 403–418.
Géotechnique 55(8): 617–623. Yin, Z.-Y., Karstunen, M., Chang, C.S., Koskinen, M. &
Perzyna, P. 1963. The constitutive equations for work- Lojander, M. 2011. Modelling time-dependent behav-
hardening and rate sensitive plastic materials. Proc. iour of soft sensitive clay. J. Geotech. Geoenviron.
Vibration Problems Warsaw 3, 281–290. Eng., 137(11): 1103–1113.
50
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
ABSTRACT: Land subsidence induced by the over-exploitation of aquifers is a very common phenom-
enon affecting extensive areas worldwide. Knowledge about the range and the rate of the deformations is
necessary for the installation and the protection of constructions. A detailed study was conducted aim-
ing to validate the efficiency of the Mohr–Coulomb and the Hardening soil models, introduced in the
PLAXIS 2D finite-element code, for the simulation of land subsidence. The data used for this case study
came from the Kalochori region on the west side of Thessaloniki, Northern Greece. The rapid develop-
ment of the area and the extensive need of water led to the development of surface subsidence, reaching
maximum values of 3–4 m, in several parts of the study area. The validation of the simulation results was
conducted by using all available subsidence indications as well as a land motion mapping produced by PSI
(Persistent Scatterer Interferometry) analysis.
1 INTRODUCTION
51
2 THE STUDY AREA
52
affected by subsidence phenomena, no surface The main limitations encountered in single-pair
ruptures were located throughout the entire study interferograms can be overcome through PSI (Per-
area. This is clearly related to the absence of tec- sistent Scatterer Interferometry) techniques, a family
tonic lines as well as the big thickness and the rela- of different approaches based on processing several
tive uniformity of the Quaternary deposits. multi-temporal space-borne SAR imagery (at least
The total amount of subsidence deformations 15–20, or more) of the same target area. The PSI
was estimated to exceed 3 m in several locations. approach uses long stacks of radar dataset (as those
This aspect was supported by Stiros (2001), who in the ESA ERS1/2 archive) and analyzes the signals
compared the 1:5000 topographic diagrams for the backscattered from the observed scene to estimate
period 1955–1980. This comparison proved that, and remove atmospheric artifacts. The electromag-
until the 1980’s, the maximum subsidence regions netic returns of electromagnetically stable, highly
(subsidence > 3 m) were located within the areas reflective point-wise targets (the so-called Perma-
containing the productive wells of the Water Com- nent Scatterers, (PS) is statistically processed and
pany of Thessaloniki (Fig. 2) as well as the south- analyzed to retrieve estimates of the displacements
western section of the industrial area containing occurred between different acquisitions (Ferretti et
the most water-consuming industries. Currently, al., 2000; 2001). PS usually correspond to man-made
along the embankments the inland is at least 4 m structures (i.e. buildings, roads, bridge, monuments,
below the sea level. pylons), as well as natural reflectors, such as out-
cropping rocks. Thanks to this multi-interferometric
approach, at each Permanent Scatterer, the relevant
3.2 Interferometric techniques
LOS velocity can be estimated with unprecedented
3.2.1 Basic principles accuracy, sometimes even better than 0.1 mm/year.
Assessment of ground motion is historically based Displacement time series can be retrieved, acquisi-
on conventional geodetic methods (GPS, level- tion per acquisition, with accuracy on single meas-
ling network above everything else), or on timely urements usually ranging from 1 to 3 mm (Colesanti
updated surveying. These methods represent some et al., 2003). Each measurement is temporally and
of the most reliable geodetic measurements tools, spatially referred to a unique reference image and to
but are time consuming and resource intensive. In a stable reference point, respectively.
addition to conventional geodetic monitoring sys- Since the pioneering studies of Massonnet et al.
tems, Earth Observation (EO) techniques have suc- (1993; 1994; 1995; 1996) and Singhroy et al. (1998),
cessfully demonstrated, in recent years, to be highly satellite SAR data have been fully exploited to
valuable in measuring land motion in a wide range retrieve ground deformations related to geohazards.
of application fields (Tralli et al., 2005).
The study of land subsidence in the wider
3.3 SAR data used
Thessaloniki Plain exploited the PSI technique,
a remote sensing technology based on multi- Within the framework of the ESA GMES Ter-
temporal satellite SAR (Synthetic Aperture rafirma project (GMES—Global Monitoring for
Radar) imagery. Environment and Security), 9 satellite image frames
A single SAR image is of no practical use, since it were processed using a special semi-automated proc-
is not possible to distinguish the different phase con- essor to produce a PSI ground motion map covering
tributions related to atmosphere, topography and a 65,000 km2 area of Greece. This WAP (Wide Area
noise. Differential Interferometry SAR (DInSAR), Product) over Greece is based on stripmap ERS1/2
a technique that exploits the phase variations (or images, obtained from the European Space Agency
interference) of two SAR images gathered at differ- (ESA). Among these 9 satellite image frames, track
ent times on the same target area by the same satel- 7_3, made of a dataset of 63 SAR imagery in C
lite using the same acquisition mode and properties band (5.6 cm wavelength), acquired along descend-
(beam, orbit, off-nadir angle, etc), is a far more suit- ing orbits in 02/05/1992–30/12/2003, was employed.
able approach (Zebker et al., 1986; Massonnet & The reference points of the stack—to which the
Feigl, 1998, Rosen et al., 2000). Since the 90s sig- displacement estimates are referred—was selected
nificant results have been obtained in detecting, in the southernmost part of Thessaloniki’s urban
mapping and quantifying surface deformation with fabric, in a sector unaffected by ground motions.
millimeter precision over large, contiguous areas The master image for Track 7_3 dataset has been
along the sensor’s Line-Of-Sight (LOS) direction. chosen on December 31st, 1996.
Conventional DInSAR is limited by temporal and
geometrical decorrelation (Zebker and Villasenor,
3.4 Evaluation of the PSI data
1992). Moreover, phase distortions introduced by
atmospheric effects, can compromise the accuracy The analysis of the mean annual displacement
of the results (Massonnet & Feigl, 1995). velocities, retrieved by PSI processing, provide
53
essential information. As expected, the density of a large part of the urban fabric is built over Neogene
measurement points is higher for the urban area formations the above observation is reasonable. On
of Thessaloniki (maximum density up to about the contrary the Sindos-Kalochori area presents sig-
300 PS/km2), and over the urban fabric of the nificant deformations. Particularly, the PSI results
Sindos and Kalochori villages (up to hundred of show that subsidence was widespread throughout
PS/km2), but lower for agricultural and vegetated the western suburbs of the Thessaloniki metropoli-
terrains (few point/km2). Nevertheless, more than tan area in the time period from 1995 to 2001. Land
1.000 point-wise measurements targets have been subsidence can be clearly identified in the coastal
yielded for the 15 km2 wide-Kalochori village and area of Kalochori, where the highest LOS velocities
its industrial area. An overview of the WAP results range between 15 and 25 mm/yr in 1995–2001, with
for ERS1/2 datasets is presented in Figure 4, peaks of 40 mm/yr recorded in the SW of the vil-
as measured by the SAR sensor along the satellite’s lage. The observed subsidence rate in Kalochori is
line of sight and expressed in mm/y. The black dots extended to another subsidence bowl affecting the
refer to stable areas. The triangles pointing south village of Sindos, on the right side of the Gallikos
represent increasing deformation rates, moving River. In this area the LOS deformation rate ranges
away from the sensor (subsidence). between 5 and 15 mm/yr, with peaks of 46 mm/yr.
The PSI results reveal that the city of Thessalo- These two well defined areas of land deformation
niki shows very low LOS deformation rates, ranging are abruptly truncated along a NW-SE alignment:
between −1.5 and 1.5 mm/yr (close to the PSI tech- northeast of Kalochori and Sindos.
nique detection sensitivity), indicating relatively sta- The potential of repeat-pass space-borne SAR
ble ground conditions since the 90s. Considering that interferometry can be exploited not only to map
Figure 4. The LOS deformation rates in Kalochori—Sindos between 1995 and 2001, obtained through the PSI
analysis. Time series for selected PS A, B, C, D and E are shown in Figure 5.
54
Figure 6. The simulation profile. The soil layers are
indicated in respect to their reference at Table 1.
55
Table 1. Soil data sets parameters coming from the reevaluation of the available laboratory and in situ tests.
Drained
[1] YB [2] BG Silty [3] B Sand [4] BG Silty [5] B Sand [6] B Sand
Draining conditions Sand sand Silty Sand Sand II Silty Sand II Silty Sand III
Table 2. FEM vertical displacements. Furthermore the current research presents the
applicability of space-borne interferometry on the
150 m deep 100 m deep study of land subsidence phenomena.
profile profile
5 CONCLUSIONS REFERENCES
Concluding the application of the Mohr– Andronopoulos, V. 1979. Geological and geotechnical
Coulomb soil model for the direct simulation of study in the Kalochori (Thessaloniki) area. Institute
the land subsidence mechanism can be consid- of Geology and Mineral Exploration Report, Athens,
p. 90.
ered accurate. The required values of the stiffness Andronopoulos, V., Rozos, D. & Hatzinakos I. 1990.
parameters can be obtained from oedometer tests Geotechnical study of ground settlement in the Kalo-
and particularly from the loading stages refer- chori area, Thessaloniki District. Institute of Geology
ring to the active stresses applied after the ground and Mineral Exploration Report, Athens, p. 45.
water drawdown. Andronopoulos, V., Rozos, D., Hatzinakos, I. 1991.
Considering that most of the times land sub- Subsidence phenomena in the industrial area of
sidence phenomena cannot be back calculated, Thessaloniki, Greece. In: Johnson, A. (Ed.). Land
because of the lack of accurate displacement Subsidence, pp. 59–69.
records, the implementation of a verified simula- Brinkgreve, R.B.J., Al-Khoury, R., Bakker, K.J. 2002.
Plaxis, Fine Element Code for Soil and Rock Analysis,
tion procedure cannot be established. In those 2D–Version 8. Balkema, Rotterdam.
cases, the knowledge that the Mohr–Coulomb Colesanti, C., Ferretti, A., Prati, C. & Rocca, F., 2003.
soil model, a simple model with low requirements Monitoring landslides and tectonic motions with the
on mechanical parameters, can provide accurate Permanent Scatterers Technique. Engineering Geol-
results can be proved valuable. ogy (68) pp. 3–14.
56
Ferretti, A., Prati, C. & Rocca, F., 2000. Nonlinear sub- Nicolau, S. & Nicolaidis, M. 1987. Geoelectric study in
sidence rate estimation using Permanent Scatterers in Kalochori village of Thessaloniki. Report, Institute of
differential SAR interferometry. IEEE Trans. Geosci. Geology and Mineral Exploration, Athens, 10.
Remote Sens. (38) pp. 2202–2212. Psimoulis P., Ghilardi M., Fouache E. & Stiros S., 2007.
Ferretti, A., Prati, C. & Rocca, F., 2001. Permanent Scat- Subsidence and evolution of the Thessaloniki plain,
terers in SAR interferometry. IEEE Trans. Geosci. Greece, based on historical leveling and GPS data.
Remote Sens. (39) pp. 8–20. Engineering Geology (90) pp. 55–70.
Hatzinakos, I., Rozos, D., Apostolidis, E. (1990). Engi- Rosen, P.A., Hensley, S., Joughin, I.R., Li, F.K., Madsen, S.N.,
neering geological mapping and related geotechnical Rodriguez, E. & Goldstein, R.M., 2000. Synthetic aper-
problems in the wider industrial area of Thessalo- ture radar interferometry. Proc. I.E.E.E. (88).
niki, Greece. In: Price, D. (Ed.), Proceedings of Sixth Rozos, D. & Hadzinakos, I. 1993. Geological conditions
International IAEG Congress, Amsterdam, Balkema, and geomechanical behaviour of the neogene sediments
pp. 127–134. in the area west of Thessaloniki (Greece). Proc. Int.
Loupasakis, C. & Rozos, D., (2009) Land Subsidence Symp. on Geotechnical Engineering of Hard Soils—
Induced by Water Pumping in Kalochori Village Soft Rocks, Greece, Anagnostopoulos et al (Eds), A.A.
(North Greece)—Simulation of the Phenomenon by Balkema, Vol.1, pp. 269–274, 1993 Rotterdam.
Means of the Finite Element Method, Quarterly Jour- Rozos, D. Apostolidis, E. & Hadzinakos, I. 2004. Engi-
nal of Engineering Geology and Hydrogeology, Geo- neering geological map of the wider Thessaloniki area,
logical Society of London, (42) pp. 369–382. Greece. Bulletin of Intern. Assoc. of Eng. Geol. & the
Massonnet, D., Rossi, M., Carmona, C., Adragna, F., Environment, Springer—Verlag, Vol. 63, pp.103–108,
Peltzer, G., Feigl, K. & Rabaute, T., 1993. The dis- 2004. Rotterdam.
placement field of the Landers earthquake mapped by Singhroy, V., Mattar, K.E. & Gray, A.L. 1998. Landslide
radar interferometry. Nature (364). characterisation in Canada using interferometric SAR
Massonnet, D., Feigl, K.L., Rossi, M. & Adragna, F. and combined SAR and TM images. Advances in
1994. Radar interferometric mapping of deforma- Space Research (21) pp. 465–476.
tion in the year after the Landers earthquake. Nature Soulios, G. 1999. Research for the development of the
(369). aquifers in the low lands on the west of Thessaloniki
Massonnet, D. & Feigl, K.L., 1995. Discrimination of for the interests of the Water Company of Thessalo-
geophysical phenomena in satellite radar interfero- niki. unpublished technical report, Aristotle Univer-
grams. Geophys. Res. Lett., (22). sity of Thessaloniki, p. 99.
Massonnet, D., Briole, P. & Arnaud, A. 1995. Defla- Stiros, S.C. 2001. Subsidence of the Thessaloniki (north-
tion of Mount Etna monitored by Spaceborne Radar ern Greece) coastal plain, 1960–1999. Engineering
Interferometry. Nature, (375). Geology, (61) pp. 243–256.
Massonnet, D., Thatcher, W. & Vadon, H. 1996. Detec- Tralli, D.M, Blom, R.G., Zlotnicki, V., Donnellan, A.,
tion of postseismic fault zone collapse following the Evans, D.L. 2005. Satellite remote sensing of earth-
Landers earthquake. Nature (382). quake, volcano, flood, landslide and coastal inunda-
Massonnet, D. & Feigl, K.L., 1998. Radar interferometry tion hazards. Journal of Photogrammetry and Remote
and its application to changes in the Earth’s surface. Sensing (59) pp. 185–198.
Rev. Geophys., (36). Zebker, H.A. & Goldstein, R.M. 1986. Topographic
Mouratidis A., Briole P., Astaras A., Pavlidis S., Mapping From Interferometric Synthetic Aperture
Tsakiri M., Ilieva M., Rolandone F. & Katsambalos, K. Radar Observations. Journal of Geophysical Research
2010. Contribution of InSAR and kinematic Gps data (91), pp. 4993–4999.
to subsidence and geohazard monitoring in Central Zebker, H.A. & Villasenor, J. 1992. Decorrelation in
Macedonia (N. Greece), Scientific Annals, School of interferometric radar echoes. IEEE Trans. Geosci.
Geology, Aristotle University of Thessaloniki, Pro- Remote Sens. (30).
ceedings of the XIX CBGA Congress, Thessaloniki,
Greece (100) pp. 535–545.
57
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
J. Maertens
Jan Maertens BVBA, Belgium
KU Leuven, Leuven, Belgium
F. De Cock
Geotechnical Expert Office GEO.BE, Lennik, Belgium
B. Lameire
Belgian Association of Foundation Contractors ABEF, Brussels, Belgium
ABSTRACT: The deep soil mixing technique consists of an in situ mechanical mixing of the soil with
an injected binder (e.g. cement). However, the presence of soil inclusions (poorly or even unmixed soil)
in the artificial material is unavoidable. This heterogeneous character of soil mix material makes it differ-
ent from traditional building materials. The presented research investigates the influence of the volume
percentage of inclusions on the strength, stiffness, stress-strain behaviour and fracture pattern of soil mix
material. 2D numerical simulations are conducted using a Discrete Element Program (UDEC) and the
results are compared with experimental data. It is observed that the reduction of the strength and stiffness
of a sample is significantly larger than the weighted average of the UCS and Young’s modulus, taking into
account the volumes of the well mixed material and the softer inclusions. However, the strength is remark-
ably more affected by the volume percentage of inclusions than the stiffness. Moreover, other parameters
than the percentage of weak material are also important (e.g. shape, size and relative position of the inclu-
sions) and result in wide ranges of resulting strength and stiffness.
58
as large blocks, complementary numerical simula- adjacent blocks does not represent a physical crack
tions (2D) are conducted in UDEC, a numerical as long as it is not activated. An example of a mesh
program based on the discrete element method. of triangles for a rectangular medium is shown in
The followed approach allows simulating frac- Figure 1a. Since fracture growth is limited to the
ture initiation and growth in and around the soft available contacts, the distribution of the contacts
soil inclusions. This allows a correct distinction should be as uniformly as possible so that there is
between shear and tensile fractures, but helps also no bias by the mesh. Moreover, on a larger scale a
to understand and quantify the effect of heteroge- given fracture with a certain overall direction tends
neities in the studied material. A large number of to develop as a combination of activated individ-
simulations have been executed to study the influ- ual contacts with sometimes very different orienta-
ence of the volume percentage, shape, number and tions, as illustrated in Figures 1b and 1c for a 60°
relative position of the inclusions. inclined crack.
In this study, the blocks only deform elastically.
The strength parameters of the material are thus
1.3 Advantage of discrete simulations not explicitly implemented in the blocks. However,
During the past decades, several numerical the contacts are modelled with a Mohr-Coulomb
approaches have been elaborated to realize the sim- failure criterion with tension cut-off. After activa-
ulation of crack initiation and propagation. Some tion, the strength parameters are set to residual
of these codes are based on finite elements, on values. In addition to this, UDEC models the con-
finite differences or on displacement discontinuity tacts with stiffness, in shear and normal direction
boundary elements. Though, it is found that the (respectively ks and kn). This stiffness allows defor-
Distinct Element Method (DEM) is a more than mation of contacts prior to activation. The defor-
valuable alternative (Debecker et al. 2006). Several mation due to this contact stiffness has an effect on
DEM codes have been developed. In the Parti- the stress distribution within the sample (Debecker
cle Flow Code (PFC) materials are modeled as a et al. 2006).
dense packing of rigid spherical elements, bonded
together at their contact points (Potyondy & 2.2 Calibration of model parameters
Cundall 2004). For this study, the Universal Dis- On the one hand, material properties as obtained
tinct Element Code (UDEC) is chosen. This code from laboratory tests are assigned to the triangu-
is originally developed for the simulation of frac- lar blocks (Table 1). On the other hand, the stiff-
tured rock mass behavior, e.g. slope stability and ness and strength parameters of the contacts are
rock fall which depend on the activation of existing not physically measurable and their values are
fractures (Itasca 2004). However, in the past it has deduced from a calibration process. First, the con-
been successfully used for the numerical modeling tact properties for a homogeneous model of well
of fracture initiation and growth in rock (Debecker mixed material are calibrated based on the stress-
2009, Tempone & Lavrov 2008). strain behavior and typical fracture patterns dur-
ing uniaxial compression tests. These properties
are all well-known from laboratory tests (e.g. peak
2 NUMERICAL SIMULATIONS strength, Young’s modulus, post-peak behavior,
59
Table 1. Material properties of the blocks correspond- predefined shape and location of the inclusions
ing to the matrix material and the soft inclusions (Van are incorporated in a (random) mesh of triangles.
Lysebetten 2011). Since UDEC does not provide a mesh generator
that can meet these requirements, a mesh generator
Material properties Matrix Inclusions
was implemented in Matlab. The algorithm divides
Density ρ [kg/m3] 1820 1820 the sample in multiple small rectangular sectors in
Young’s modulus E [GPa] 11.6 0.165 which grid nodes are randomly generated around
Poisson’s ratio υ [−] 0.3 0.4 the predefined nodes of the inclusions. Finally,
Bulk modulus K [GPa] 9.67 0.275 Delaunay triangulation is performed on the entire
Shear modulus G [GPa] 4.46 0.059 collection of nodes (i.e. nodes that define the
boundary of the inclusions as well as the randomly
generated nodes). The triangulation algorithm is
based on the Quickhull algorithm for constructing
convex hulls (Barber et al. 1996). To keep calcula-
tion time in UDEC reasonable, the number of tri-
angles is limited to approximately 8000. Of course,
this limits the minimal representative length of
the individual contacts of the discrete model and
consequently the minimum fracture length that
can be simulated (Debecker & Vervoort 2006). The
quality of the generated mesh is largely dependent
on the choice of the input parameters, such as the
minimum distance between grid nodes in relation
to the dimensions of the sectors.
An appropriate mesh for fracture simulation
Figure 2. (a) 2D sample (120 × 240 mm) from a real soil must meet several geometrical requirements. First,
mix column cross-section. (b) Mesh of the basic model. the boundary of the inclusions must be respected
by the edges of the generated triangles (the edges
are called ‘elements’ from now on). Second, the
failure by 60 to 70° inclined shear fractures, distribution of the element orientation should be
etc.). This time-consuming process finally leads as uniformly as possible, as discussed in paragraph
to the matrix—matrix contact strength and stiff- 2.1. Third, adjacent triangles should be of approx-
ness parameters. Next, the strength properties of imately the same size and sharp triangles should
matrix—inclusion and inclusion—inclusion con- be avoided, both to guarantee reasonable solution
tacts are easily deduced from the matrix—matrix accuracy in UDEC. In order to check the gener-
contact strength by applying the strength ratios of ated triangles, quality parameters have been intro-
mixed and unmixed material. duced (Debecker & Vervoort 2006):
• ‘Minimum triangle angle’: the smallest angle of
2.3 Basic model the triangle and a degree to evaluate the skinni-
ness of the triangles. Its aimed interval is [20°,
The starting point for the soil mix samples is a sec-
60°].
tion through a real soil mix column, of which an
• ‘Triangle aspect ratio’: a measure for the maxi-
area of 120 × 240 mm is considered (Fig. 2a). This
mum edge size to the minimum altitude of the
2D sample contains 11 inclusions which account
triangle. Its ideal interval is [1, 2.5].
for 11 vol%. Based on this sample, the basic model
• ‘Triangle quality’ Q: the ratio of the smallest
is assembled. It consists of 9 of the 11 inclusions
angle to the largest angle of the triangle. The
or 10 vol% of inclusions (Fig. 2b). Note that the
ideal interval of Q is [0.4, 0.9].
term vol% is justified, since the 2D model in fact
has a unit thickness. Moreover, using this termi- The present algorithm provides an overview of
nology the link with the measurements on real soil all quality parameters after the mesh has been gen-
mix material is more easily made. erated. Note that all quality parameters are only
indicative and not stringent.
2.4 Mesh generation algorithm
2.5 Results of the simulations
As illustrated by the basic model in Figure 2b, the
mesh of a soil mix sample is built up by a com- In order to study the influence of several inclusion
bination of a predefined structure and a certain properties (e.g. vol%, shape, size, relative position),
degree of randomness. Or in other words, the about 200 uniaxial compression tests on soil mix
60
models (120 × 240 mm) with different inclusion reached (Fig. 3b), also shear fractures are induced.
geometries have been simulated. Finally, this results in a shear zone that extends
from the upper left corner to the lower right corner
2.5.1 Influence of volume percentage, shape of the sample (Fig. 3c).
and size of the inclusions Figure 3d presents three detailed fracture pat-
On the basis of the real 2D section and the basic terns of tested samples, cored from a real soil mix
model (Figs. 2a and 2b), 68 additional models are wall with several soil inclusions. The inclusions
generated with 1, 5, 10 and 20 vol% inclusions. For are clearly visible, as well as the induced fractures.
these models, the relative position of the inclu- Moreover, the observed fractures are comparable
sions is varied as well as the shape (e.g. explic- to the simulated fracture patterns around inclu-
itly rounded or sharp-ended) and the number of sions. For example in the left picture, two fractures
inclusions (e.g. 1 large rounded inclusion against are induced at both extremities of the bottom part
3 smaller rounded inclusions). of the inclusion, which is also observed around
The evolution of the fracture pattern of the basic several inclusions in the simulations. In the central
model (Fig. 2b) is presented in Figures 3a, b and c and right pictures, fractures are observed some-
for 3 different loading steps (just before, at and just where along the middle of the inclusion or at the
after peak strength). The basic model has a UCS most top or bottom point of the inclusion. Again,
value of 4.4 MPa against a UCS value of 11.7 MPa these fractures are also visible in the simulations.
for the pure soil mix sample without inclusions. The Figure 4 compares laboratory data of in situ
9 inclusions (a total of 10 vol%) reduce the strength cored samples from 12 different Belgian construc-
thus with more than 60%. At relatively low vertical tion sites with the results of the numerical simula-
loads (less than 50% of the peak strength) a part of tions (69 models). Both sets of results show a similar
the material inside the inclusions starts to fail. This downward trend in function of the volume percent-
is logical since the inclusions have lower strength age of inclusions. The results of the simulations are
properties. However, failure is clearly limited to clearly situated within the (wider) range of labora-
specific parts of the inclusions. At 2.6 MPa vertical tory results. The fact that the experimental data have
load several vertical splitting type of fractures are a wider range is logical, as the soil type at the 12 con-
induced in the matrix at the top and at the bottom structions sites is different. Clayey, as well as loamy
of the grouped inclusions (Fig. 3a). At higher ver- and sandy soils were encountered, and of course the
tical loads, especially after the UCS value has been mechanical characteristics of the different soil mix
materials are different (Denies et al. 2012b). For the
numerical simulations one single value is considered
for the model without any inclusions, i.e. a strength
value of 11.7 MPa for the homogeneous sample.
Other reasons for the wider range are that in reality a
real soil mix sample is never perfectly homogeneous
61
and that two samples even without any inclusions and stiffness of a sample (Van Lysebetten et al.
(or with exactly the same amount of inclusions) still 2012). For example, it is observed that strength
have different strength values because of variations and stiffness decrease if the inclusions are located
in e.g. cement content. along the diagonal of the sample. This configu-
Therefore, the numerical simulations are more ration probably facilitates a global shear failure.
suitable to study the effect of the volume percent- Figure 5a shows that there is a clear negative trend
age of unmixed material on the strength of a sam- between strength and stiffness and wi/t: the ratio
ple. The numerical results of Figure 4 are based of the total width of inclusions, i, and the sample
on the basic model and the 68 additional models width, t (Fig. 5b). However, a significant overlap is
(diamond-shaped markers). First, the decrease also observed, e.g. the largest UCS value for a width
of strength is clearly larger than the percentage ratio of 60% is still larger than the smallest value for
of inclusions in the sample. For a mere 1% of 20% ratio. For the stiffness, the overlap is probably
unmixed material strength is reduced by 13% on even larger. In these additional simulations, it was
average. For 10% of inclusions even half of the also observed that the strength and stiffness also
strength disappears on average (Van Lysebetten decrease if the vertical distance between the inclu-
2011). The same can be concluded for the stiffness sions decreases, since inclusions tend to act as one
(Vervoort et al. 2012), though it is less influenced larger inclusion. However, this effect is less impor-
(for 1 and 10 vol% of inclusions the stiffness is tant than the influence of the ratio wi/t.
reduced on average with respectively 3 and 32%).
Second, there is an overlapping zone between the 2.5.3 Link with simulated fracture patterns
strengths of 5 and 10% and of 10 and 20%. This Figs 6a, b, c and d show the evolution of the frac-
overlap is caused by the relatively large range for ture pattern of 4 models for successive loading
each volume percentage of inclusions. For the stiff- steps. At relative low stress levels (less than 50% of
ness, there is just no overlap between successive the strength) vertical extension type of fractures
volume percentages. initiate at the lower and upper parts of the inclu-
Strength and stiffness of a sample are clearly sions (see also Figs 3a, b and c for the evolution
not only determined by the volume percentage. of the fracture pattern of the basic model). Only at
This is further analyzed for additional models with stress levels near the UCS value of the sample, shear
10% of inclusions. It is observed that strength and fractures are observed at the sides of the inclusions
stiffness are reduced more by sharp-ended inclu- after which the sample finally fails due to a combi-
sions than by rounded inclusions. Apart from this, nation of shear and extension type of fractures.
the strength of a sample also drastically decreases Most of the described effects of inclusions can
when less inclusions (but with the same shape and be related to and explained by the observed fracture
the same total percentage of inclusions) are present patterns. For example, stress peaks near sharp-
(Vervoort et al. 2012). ended inclusions are higher than close to rounded
inclusions, causing fracture initiation at lower
2.5.2 Influence of the relative position
of the inclusions
Apart from the volume percentage, size and number
of inclusions, the relative position has also an influ-
ence on strength and stiffness. 123 additional mod-
els are generated to study this effect. All additional
models represent 10% of inclusions and are com-
posed of two sets of respectively 3 and 5 inclusions.
The main difference between all models is the relative
position of these inclusions. In some models, they
are concentrated, in others they are spread as wide
as possible, or they are nicely situated along a vertical
line or along a diagonal, etc. The ranges of strength
(3.8–8.1 MPa) and stiffness (5–8.7 GPa) increase in
comparison with the results of the 69 models for
the study of the influence of the volume percent-
age, shape and number of inclusions (respectively
3.1–9.2 MPa and 6.3–7.9 GPa). This is logical since
the more simulations, the greater the probability that
new minimum and maximum values are found. Figure 5. (a) Relation between the strength (left)
The simulations clearly show the influence of the and stiffness (right) and wi/t. (b) Definition of the ratio
relative position of the inclusions on the strength wi/t = i/t.
62
Figure 6. Illustration of the fracture pattern at successive loading steps of 4 samples of the original set each with
10% of inclusions. Shape and number of inclusions are different. The stiffness of the models is respectively 7.2 GPa,
8.0 GPa, 6.7 GPa and 7.4 GPa.
stress levels. Finally, this results in lower strength the sample is damaged. Again, this results in lower
and stiffness (compare Figs 6a and c). Fractur- strength and stiffness.
ing also initiates at lower stress levels when 1 large
inclusion is present instead of 3 smaller inclusions,
3 CONCLUSIONS
at least if these 3 inclusions are vertically aligned
(compare Figs 6a, b and d). Again, this is caused
Although calibration of the numerical discrete
by higher stress peaks near the largest inclusions
model is very time-consuming, the results show
(i.e. because a higher horizontal distance has to be
that numerical simulations are very interesting
bridged).
to perform relatively easily sensitivity analyses.
However, the fact that the 3 inclusions are verti-
The influence of soil inclusions on the strength
cally aligned is very important. Figure 6d shows the
and stiffness of soil mix material has been inves-
fracture pattern of a sample with the same 3 inclu-
tigated by a total of 192 simulations. It has been
sions positioned diagonally. This sample results in
observed that soft inclusions have a large impact
about the same strength and stiffness as the sample
on the resulting strength and stiffness of a sam-
with 1 large rounded inclusion. The large reduc-
ple. However, not only the volume percentage is
tion of strength in comparison with the model
important, but also the relative position of the
with 3 vertically aligned inclusions is caused by
inclusions, their shape and number, etc. Moreover,
the diagonal orientation of the inclusions result-
UDEC has shown to be suitable for the simula-
ing in a larger wi/t ratio. Moreover, the stress peaks
tion of fracture initiation and growth in soil mix
around the stress relieved zones interfere and prob-
material with soft inclusions. Similar trends of
ably amplify each other.
the influence of the volume percentage of inclu-
Finally, if the vertical spacing between the inclu-
sions on the strength are observed for the results
sions of Figure 6b is decreased, the inclusions
of laboratory data of in situ cored samples from
tend to form one larger inclusion. Overlapping
different Belgian construction sites and the results
zones with increased stresses expand again and the
of numerical simulations. The simulated fracture
amplifying effect grows. This causes earlier fracture
patterns are also comparable with these observed
initiation (at lower vertical loads), but the vertical
in real soil mix samples.
extension type fractures around the central inclu-
sion also reach faster the upper and lower inclu-
sions. The travelled distance through the stronger ACKNOWLEDGEMENTS
material before reaching other weak inclusions is
thus lower. Because propagation through these This project has been made possible by the finan-
inclusions goes much faster, a much larger part of cial support of the Flemish government agency for
63
Innovation by Science and Technology (IWT). It Denies N., Huybrechts N., De Cock F., Lameire B.,
is the result of cooperation between the Belgian Vervoort A., Maertens J. 2012b. Soil mix walls as retain-
Building Research Institute (BBRI), the Belgian ing structures, mechanical characterization. Proceedings
Association of Foundation Contractors (ABEF) of the TC 211 International Symposium on Ground
Improvement, Brussels (Belgium), Vol. 3, pp. 99–115.
and KU Leuven. Ganne P., Denies N., Huybrechts N., Vervoort A.,
Tavallali A., Maertens J., Lameire B., De Cock F. 2011.
Soil Mix: influence of soil inclusions on the structural
REFERENCES behavior. Proceedings of the 15th European conference
on soil mechanics and geotechnical engineering, Athens
Barber, C.B., Dobkin, D.P, Huhdanpaa, H. 1996. The (Greece), pp. 977–982.
Quickhull algorithm for convex hulls. ACM Trans- Itasca 2004. UDEC v4.0 manual. Itasca Consulting
actions on Mathematical Software, Vol. 22 (4), Group, Inc., Minnesota, USA.
pp. 469–483. Potyondy D.O. & Cundall P.A. 2004. A Bonded-
Cundall P.A. 1971. A computer model for simulating Particle Model for Rock. International Journal for
progressive large scale movements in block rock sys- Rock Mechanics & Mineral Sciences, Vol. 41(8), pp.
tems. Proceedings of the International Society for Rock 1329–1364.
Mechanics Symposium, paper II-8. Tempone P. & Lavrov A. 2008. DEM modeling of mud-
Cundall P.A. & Board M. 1988. A microcomputer pro- losses into single fractures and fracture networks.
gram for modeling large-strain plasticity problems. Proceedings of the 12th international conference
Proceedings of the 6th International Conference on of the international association for computer meth-
Numerical Methods in Geomechanics. Rotterdam ods and advances in geomechanics, Goa (India),
(Netherlands), pp. 2101–2108. pp. 2475–2482.
Debecker B. 2009. Influence of planar heterogeneities on Van Lysebetten G., Vervoort A., Maertens J.,
the fracture behavior of rock, Ph.D. Dissertation, KU Huybrechts N. 2012. Discrete modelling for the study
Leuven, Leuven (Belgium). of the effect of soft inclusions on the behaviour of soil
Debecker B., Vervoort A., Napier J.A.L. 2006. Fractur- mix material. In preparation.
ing in and around a natural discontinuity in rock: a Van Lysebetten G. 2011. Soil Mix for Construction pur-
comparison between boundary and discrete element poses: Quality control. M.Sc. Thesis (Geotechnical
models. Proceedings of the 5th International Confer- and Mining Engineering) KU Leuven, Leuven.
ence on Engineering Computational Technology, Las Vervoort A., Van Lysebetten G., Tavallali A. 2012.
Palmas de Gran Canaria (Spain), paper 168. Numerical modeling of fracturing around soft
Debecker B. & Vervoort A. 2006. A 2D triangular Delau- inclusions. Proceedings of the Southern Hemisphere
nay grid generator for the simulation of rock features. International Rock Mechanics Symposium, Sun City
Proceedings of the 5th International Conference on (South-Africa), pp. 33–46.
Engineering Computational Technology, Las Palmas
de Gran Canaria (Spain), paper 220.
Denies N., Huybrechts N., De Cock F., Lameire B.,
Maertens J., Vervoort A. 2012a. Soil mix walls as
retaining structures, Belgian practice. Proceed-
ings of TC211 International Symposium on Ground
Improvement, Brussels (Belgium), Vol. 3, pp 83–97.
64
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
R.B.J. Brinkgreve
Plaxis BV, Delft, The Netherlands
Delft University of Technology, Delft, The Netherlands
ABSTRACT: This paper presents a three dimensional formulation of a simple and practical constitutive
model developed for evaluating seismic liquefaction in sands. The model is an extension of the two dimen-
sional UBC-SAND model developed at University of British Colombia (Beaty & Byrne 1998) which
utilises isotropic and simplified kinematic hardening rules for primary and secondary yield surfaces, in
order to take into account the effect of soil densification and predict a smooth transition into the liquefied
state during undrained cyclic loading. By means of a simplified Rowe stress-dilatancy theory the model is
capable of modelling cyclic liquefaction for different stress paths. To show the capability of the model to
predict cyclic liquefaction in soils, the mechanical behaviour of some sands is numerically studied under
direct simple shear conditions and compared with experimental data. The effect of densification and
parameters selection on the results is discussed. Finally, the model is used for simulating an experimental
dynamic centrifuge test and the numerical results are compared with the real measurements.
65
2 KEY FEATURES OF THE UBC3D-PLM 2
⎛ p′ ⎞
np
⎧⎪ ⎛ sin φ ′ ⎞ ⎫⎪
G *
kGp ⎜ ⎨1 − ⎜
mob
m
⎟ RF ⎬ (5)
2.1 Yield surfaces ⎝ PA ⎟⎠ ⎪⎩ ⎝ sin φ p′ ⎠ ⎪⎭
The UBC3D-PLM model uses the well-known
Mohr-Coulomb yield function generalized in 3-D where kGp is the plastic shear modulus number; dλ
principal stress space (Tsegaye 2010). The formula- stands for the plastic multiplier; np is the plastic
tion of the surface is given bellow: shear modulus exponent; φmob is the mobilized fric-
tion angle; φpeak is the peak friction angle; and Rf is
σ ma
′ x −σm
′ in
fm = (1) the failure ratio qf /nult based on the Duncan-Chang
2 model, ranging from 0.5 to 1.0, where qf is the devi-
⎛σ ′ + σm ′ in ⎞ atoric stress ratio at failure and qult is the ultimate
− ⎜ max c ′ cot φ p′ ⎟ i φm
′
⎝ 2 ⎠ mob
stress ratio from the best fit hyperbola. The hard-
ening rule relates the amound of plastic strain with
where σ ma′ x is the maximum principal effective
the mobilized friction angle which is computed
stress and σ mi
′ n is the minimum principal effective
at each step. A schematization of the hyperbolic
stress, c′ is the cohesion of the soil, φ p′ is the
shear hardening rule is depicted in Figure 1.
peak effective friction angle of the soil and φm′ is
the mobilized friction angle during hardening.
The intermediate stress does not influence the yield 2.4 Plastic potential function and flow rule
surface in three dimensional stress space.
A non-associated flow rule based on the Drucker-
Prager plastic potential function is formulated in
2.2 Elastic behaviour 3-D space as follows:
The elastic behaviour which occurs within the yield
6 sinψ m
surface is governed by a stress dependent non-lin- g q− ( p + c cot φ p ) (6)
ear rule defined by Puebla et al. (1997). The bulk 3 − sinψ m
modulus K and the elastic shear modulus G are
stress dependent and are given by the following where, g is a conical surface passing through the tri-
equations: axial compression point and is independent of the
Lode angle θL. The q and p terms are the deviatoric
me
⎛ p⎞ and the isotropic parts of the stress tensor. The influ-
Ke kBe PA ⎜ ⎟ (2) ence of the Lode angle in different stress paths was
⎝ PA ⎠ investigated with a Lode angle dependent formula-
ne tion of the plastic potential function and proved
⎛ p⎞ not to play a key role during modelling the onset of
Ge kGe PA ⎜ ⎟ (3)
⎝ PA ⎠ liquefaction.
In the UBC3D-PLM model the mobilized dila-
where kBe and kGe are the bulk and the shear moduli tancy angle is computed based on the flow rule
numbers respectively, at a reference stress level. The
factors ne and me are parameters to define the rate
of stress dependency of stiffness. In the literature,
the reference stress level Pref is commonly taken
as the atmospheric pressure (PA = 100 kPa). Pure
elastic behaviour is predicted by the model during
unloading.
66
simplified kimematic hardening rule is used for the
secondary surface.
The plastic shear modulus number kGp during
primary loading is identical with the one entered
as input parameter. However, during secondary
loading it increases as a function of the number of
cycles in order to capture the effect of soil densifi-
cation as follows:
⎛ n ⎞
KGp KGp 4 + rev ⎟ × hard f achard (11)
⎝ 2 ⎠
67
Table 1. Input parameters for the validation of the
UBC3D-PLM in modelling monotonic element tests,
cyclic element tests and a dynamic centrifuge test.
Cyclic and
Monotonic tests dynamic tests
68
and the direct simple shear loading path. This issue
has to be taken into account during the calibration
of the model according to the most critical stress
path expected in a specific case.
69
Figure 9. Evolution of shear strains during undrained
simple shearing. Fraser sand (RD = 40%). CSR = 0.1.
συ = 100 kPa.
70
REFERENCES predicting liquefaction in sands. Proceedings of the
22nd european young geotechnical engineers conference,
Beaty, M. & Byrne, P. 1998. An effective stress model Gothenburg, Sweden., 167–172.
for predicting liquefaction behaviour of sand. Geo- Puebla, H., Byrne, M. & Phillips, P. 1997. Analysis of
technical Earthquake Engineering and Soil Dynamics can-lex liquefaction embankments prototype and
III ASCE Geotechnical Special Publication No.75. 1, centrifuge models. Canadian Geotechnical Journal 34,
766–777. 641–657.
Beaty, M. & Byrne, P. 2011. Ubcsand constitutive model Rowe, P.W. 1962. The stress-dilatancy relation for static
version 904ar. Itasca UDM Web Site, 69. equilibrium of an assembly of particles in contact.
Byrne, P.M., Park, S.S., Beaty, M., Sharp, M., Proceedings of the Royal Society of London. Series A,
Gonzales, L. & Abdoun, T. 2004. Numerical model- Mathematical and Physical Sciences 269A, 500–527.
ling of dynamic centrifuge tests. 13th World Confer- Sriskandakumar, S. 2004. Cyclic loading response of
ence in Earthquake Engineering. fraser sand for validation of numerical models simu-
Martin, G., Finn, W. & Seed, H. 1975. Fundamentals of lating centrifuge tests. Master’s thesis, The University
liquefaction under cyclic loading. Journal of the Geo- of British Columbia, Department of Civil Engineering.
technical Engineering Division, ASCE 101. Tsegaye, A. 2010. Plaxis liqueafaction model. report
Petalas, A., Galavi, V. & Bringkreve, R. 2012. Validation no. 1. PLAXIS knowledge base.
and verification of a practical constitutive model for
71
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
N. Sivasithamparam
Plaxis BV, Delft, The Netherlands
University of Strathclyde, Glasgow, Scotland, UK
M. Karstunen
Chalmers University of Technology, Gothenburg, Sweden
University of Strathclyde, Glasgow, Scotland, UK
R.B.J. Brinkgreve
Plaxis BV, Delft, The Netherlands
Delft University of Technology, Delft, The Netherlands
P.G. Bonnier
Plaxis BV, Delft, The Netherlands
ABSTRACT: This paper presents a comparison of two anisotropic creep models, ACM and Creep-
SCLAY1, which differ in their formulation of creep strain rate. Creep is formulated in ACM using
the concept of contours of constant volumetric creep strain rate, whereas the newly developed Creep-
SCLAY1 model uses the concept of a constant rate of visco-plastic multiplier. The two models are identical
in the way the initial anisotropy and the evolution of anisotropy are simulated. A key assumption of both
models is that there is no purely elastic domain. The models are compared at element level. The numerical
simulations show that the Creep-SCLAY1 model is able to give a better representation of natural clay
behaviour at element level.
72
which can be easily derived from standard labora-
tory tests. This paper shows a direct comparison
of both models and their prediction capability at
element level.
The first part of this paper gives a short descrip-
tion of the ACM and Creep-SCLAY1 models in
triaxial stress space. In further sections the single
element simulations results obtained using the
SoilTest facility in the finite element code PLAXIS
(Brinkgreve et al. 2012) are presented, followed by
brief conclusions.
c
∈= ∈e + ∈ (1)
2.1 ACM
Leoni et al. (2008) proposed the Anisotropic Creep
Model (ACM) extending from a previously devel-
oped isotropic creep model (Vermeer et al. 1998;
Vermeer & Neher 1999) which is based on ellipses
of Modified Cam Clay (Roscoe & Burland 1968).
An extract of the mathematical formulation from
Leoni et al. (2008) is presented below. The outer
rotated ellipse defines the normal consolidation
surface (NCS) and the size of this ellipse evolves
with volumetric creep strains according to the Figure 1. Current state surface and normal consolida-
hardening law tion surface in triaxial stress space.
⎛ ∈c ⎞
p′p = p′p 0 exp ⎜ * v * ⎟ (2) (CSS). The intersection of the CSS with the hori-
⎝ λ −κ ⎠
zontal axis is called the equivalent mean stress p′eq,
and it is defined as
where λ* and κ* are the modified compression
index and modified swelling index respectively. The
( q − p ′ )2
intersection of the vertical tangent to the ellipse ′ = p′ +
peq (3)
with p′ axis is the isotropic preconsolidation pres- ( M 2 − 2 ) p′
sure p′p (see Figure 2.1).
The inner ellipse passes through the current state where M is the stress ratio at critical state and a
of effective stress called the current stress surface scalar quantity α is used to describe the orientation
73
of the normal consolidation surface and current
stress surface.
The volumetric creep strain rate is given by a
power law as follows:
β
μ * ⎛ peq
′ ⎞
vc =
∈ ⎜ ⎟ (4)
τ ⎝ p′p ⎠
λ* κ *
β= (5)
μ*
2.2 Creep-SCLAY1
In Creep-SCLAY1, Eq. (4) is modified to an
expression that gives the rate of the visco-plastic
multiplier as follows:
β
′ ⎞ ⎛ M − α K 0NC ⎞
2 2
μ * ⎛ peq
Λ= ⎜ ⎟ ⎜ 2 ⎟ (6)
τ ⎝ p′p ⎠ ⎜⎝ M − ηK2 NC ⎟⎠
0
74
λ* (modified compression index) and κ* (modified PLAXIS as user-defined soil models. The Creep-
swelling index). Furthermore, the initial value for SCLAY1 model has been implemented by the
a state variable p′p (initial size of the ellipse) is first author and the ACM has been implemented
required. In the context of finite element analy- by Leoni et al. (2008). Single element simulations
ses, the initial value of p′p is calculated based on were done using the PLAXIS SoilTest facility to
the OCR (vertical overconsolidation ratio) or highlight the similarities and the differences in the
POP (pre-overburden pressure), normally con- model predictions. Parameters used for these simu-
solidated K 0NC value (estimated by Jaky’s formula) lations corresponding to Both-kennar clay parame-
and the initial vertical effective stress. ters (Symposium 1992) are summarized in Table 1.
Parameters describing initial anisotropy (α0) Firstly, Creep-SCLAY1 and ACM were com-
and its evolution, include soil constants ω (rate pared in undrained compression simulations with
of rotation of the surfaces) and ωd (relative rate two strain rates (10% per day and 100% per day).
of surface rotation). The scalar value α0 and ωd Initial effective stress σ 3′ = 100 kPa and K0 = 0.5
can be theoretically derived based on M values were assumed and 10% maximum strain was
(see Wheeler et al. (2003) for details) as follows: applied. Figures 4 and 4 show the stress paths and
deviatoric stress versus axial strain predicted by
η02 + 3η0 − M 2 the two models. Though both models are able to
α0 = (7) predict dependence on strain rate, the peak und-
3 rained strength predicted by ACM is lower than
3 4M 2 − 4η02 3η0 that predicted by the Creep-SCLAY1 model. In
ωd = (8) contrast to Creep-SCLAY1, ACM predicts stress
8 η02 M + 2η0
path approaches the CSL with reducing p′ and q,
converging towards the stress origin due to the
where η0 = 3(
3 1−
1 NC
0 ) /(1 2 K 0NC ) . assumption of constant volumetric creep strains.
The parameter ω can be estimated based on In the ACM simulations, jumbs were observed
initial anisotropy (α0), modified compression as highlighted in Figure 3 due to the transition
index (λ*), M and ωd (see Leoni et al. (2008) for between current state surface to Mohr-Coulomb
details) as follows: failure surface. Furthermore, the ACM cannot
reach to a critical state condition with shearing at
constant volume and effective stresses.
1 10M 2 − 2α 0ω d
ω= ln (9) Several publications (e.g. Graham et al. (1983),
λ* M 2 − 2α 0ω d Tatsuoka et al. (2002), Leroueil & Marques (1996),
Tavenas et al. (1978) and Vaid & Campanella
In derivation of Eq. (9), a number of assump- (1977)) showed the influence of step changes in
tions has been made (see Leoni et al. (2008)). strain rate on the stress-strain behaviour of soft soil
Consequently, with certain parameter combina- in undrained triaxial compression. Immediately
tions Eq. (9) might result with a negative value, after an increase in strain rate the stress-strain path
which makes no physical sense. As an alternative, is seen to jump upwards and show an initial stiff
an empirical formula suggested by Zentar et al. response. If the strain rate is reduced back to the
(2002) to estimate the ω value can be used: original strain rate then a downwards stress jump
is observed after which the path rejoins the original
10 20 curve defined by the lower strain rate. The paths
≤ω ≤ (10)
λ λ in stress-strain curves are indicated to be uniquely
Table 1. Bothkennar clay parameters.
μ* (modified creep index) can be obtained by
measuring the volumetric strain on the long Parameters value
term and plotting it against the logarithmic
time. τ (the reference time, which is linked to the λ* 0.1
definition of vertical preconsolidation stress) v′ 0.2
can usually be taken to equal one day (see κ* 0.00667
Brinkgreve et al. (2012) for details). M 1.5
OCR 1.5
α0 (initial anisotropy coefficient) 0.59
4 NUMERICAL SIMULATION ω (anisotropy coefficient) 50.0
ωd (anisotropy coefficient) 1.0
This section discusses the performance of both μ* (viscosity coefficient) 5.07 × 10−3
models in a single element simulation. Both models τ (viscosity coefficient) 1.0 day
are implemented into the finite element code
75
Figure 3. Simulation of undrained triaxial compression with varying strain rate.
76
defined by the strain rate and the effects of strain 5 CONCLUSIONS
rate changes are observed to be persistent, which
is a characteristic of isotach behaviour, i.e., there This paper studies the performance of two aniso-
is a unique stress-strain strain-rate relation for tropic creep constitutive models at element level.
a given soil. Most soft clays in both undisturbed In ACM, the creep strain rate is formulated using
and reconstituted states, undisturbed natural stiff contours of volumetric creep strain rates whereas in
clays and cases of soft rock all show isotach vis- the newly developed Creep-SCLAY1 model, creep
cous behaviour. Figure 4 shows a stepwise change strain rate is formulated using the concept of rate
in strain rate undrained compression simulations of visco-plastic multiplier. The model simulations
using Creep-SCLAY1 and ACM to verify the demonstrate that the new formulation results in
capability of both models to predict the isotach a better prediction of natural soft soil behaviour.
behaviour. Figure 4 clearly demonstrates that The following observations are made from the
ACM cannot properly simulate the isotach behav- comparison:
iour observed in natural soft clays under a stepwise Though both models are able to predict rate
change in strain rate. Furthermore, the stress path effect dependence in undrained compression simu-
simulated by ACM cannot overpass the critical lation, in contrast to the Creep-SCLAY1 model,
state as shown in Figure 4. This too is not in agree- the ACM predicts stress paths which approach the
ment with experimental observations for slightly CSL with reducing p′ and q, converging towards
structured or reconstituted clays (Yin et al. 2010). the stress origin. ACM cannot reach a critical state
There is a mathematical difference between the condition with shearing at constant volume and
two models to calculate the creep strain compo- effective stresses. Undrained compression using
nents in general stress space. In ACM and Creep- stepwise change in strain rate simulations demon-
SCLAY1, creep strains are calculated as follows: strate that the ACM cannot reproduce the isotach
ACM: behaviour observed in natural soft soils. Further-
more, ACM cannot overpass the CSL; this may
∈ vc ∂peq
′ not be in agreement with experimental observa-
cij =
∈ ∂peq′ tions for slightly structured or reconstituted clays.
∂σ iij
∂p ′
(11) There is a mathematical difficulty in ACM. When
calculating creep strain rates, the value of ∂p′eq/∂p′
Creep-SCLAY1: can reach infinity when η/M becomes to 1, i.e., the
stress condition reaches a critical state.
vc
∈ ∂peq
′ Further work will involve comparing the per-
cij =
∈
( )
(12) formance of the models against experimental data
∂peq
′ ∂σ iij and instrumented test structures.
∂p ′
K 0NC
REFERENCES
77
destructuration on the behaviour of Murro test Tavenas, F., Leroueil, S., La Rochelle, P. & Roy, M. 1978.
embankment. Int. J. of Geomechanics (ASCE) 5(2), Creep behaviour of an undisturbed lightly overcon-
87–97. soli-dated clay. Can. Geot. J. 15(3), 402–423.
Karstunen, M. & Yin, Z.Y. 2010. Modelling time-de- Vaid, Y. & Campanella, R. 1977. Time-dependent behav-
pendent behaviour of Murro test embankment. Geo- iour of undisturbed clay. ASCE J Geotech Eng Div
technique 29, 1–34. 103(7), 693–709.
Leoni, M., Karstunen, M. & Vermeer, P. 2008. Aniso- Vermeer, P.A. & Neher, H. 1999. A soft soil model that
tropic creep model for soft soils. Gotechnique 58 (3), accounts for creep. Beyond 2000 in Computational Geo-
215–226. technics, R.B.J. Brinkgreve(eds), Rotter-dam. 4, 249–261.
Leroueil, S. & Marques, M. 1996. State of art: Impor- Vermeer, P.A., Stolle, D.F.E. & Bonnier, P.G. 1998. From
tance of strain rate and temperature effects in geo- classical theory of secondary compression to modern
technical engineering. measuring and modelling time creep analysis. Proc. 9th Int. Conf. Comp. Meth. and
dependent behaviour of soils. ASCE, Geotechnical Adv. Geomech., Yuan(eds) 4, 2469–2478.
Special Publication 61, 1–60. Wheeler, S., Näätänen, A., Karstunen, M. & Lojander,
Roscoe, K. & Burland, J. 1968. On the generalised stress- M. ¨ 2003. An anisotropic elasto-plastic model for soft
strain behaviour of wet clay. Engineering Plasticirv, clays. Can. Geot. J. 40, 403–418.
535–609. Yin, Z.Y., Chang, C.S., Karstunen, M. & Hicher, P.Y.
Sekiguchi, H. & Ohta, H. 1977. Induced anisotropy and 2010. An anisotropic elastic-viscoplastic model for
time dependency in clays. 9th ICSMFE, Tokyo, Con- soft clays. Int. J. of Solids and Structures 47, 665–677.
stitutive equations of Soils 17, 229–238. Zentar, R., Karstunen, M., Wiltafafsky, C., Schweiger, H.F. &
Sivasithamparam, N. 2012. Modelling creep behaviour Koskinen, M. 2002. Comparison of two approaches for
of soft soils. Internal report Plaxis B.V. Symposium modelling anisotropy of soft clays. Proc. 8th Int. Symp.
1992. Bothkennar soft clay test site: Characteriza- on Numerical Models in Geomech. (NU-MOG VIII),
tion and lessons learned (Géotechnique symposium in 115–121.
print). Géotechnique 42(2), 161–380. Zhou, C., Yin, J.-H., Zhu, J.-G. & Cheng, C.-M. 2006.
Tatsuoka, F., Ishihara, M., Di Benedetto, H. & Kuwano, Elastic anisotropic viscoplastic modeling of the strain-
R. 2002. Time-dependent shear deformation charac- rate dependent stress–strain behaviour of K0-consoli-
teristics of geomaterials and their simulation. Soils & dated natural marine clays in triaxial shear test. Int. J.
Foundations 42(2), 103–138. Geomech 5(3), 218–232.
78
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
ABSTRACT: Installation of jacked or driven displacement piles imposes large deformations in the soil.
These installation effects will influence the subsequent load-deformation response of the installed piles
and should be taken into account in the description of the soil-structure interaction. A series of model
pile tests were carried out in the geotechnical centrifuge at TU Delft. The tests combined horizontal con-
tact stress measurements on an instrumented model pile and visual observation of the soil deformation
adjacent to the pile. The model scaling, experimental set-up and results of pile installation in samples with
various initial relative densities are discussed. The results indicated good consistency and reproduced the
friction fatigue effect, the lateral loads at rest acting on the pile show surprisingly small influences from
the pile installation.
79
A series of geotechnical centrifuge tests in sand
samples with different initial relative densities have
therefore been performed to bring some further
clarification of this phenomenon.
Stress measurements on the pile-soil contact
and soil deformation measurements adjacent to
the pile have been made. These tests have been per-
formed with a small instrumented model pile. The
influence of the scaling conditions on the effect
of friction fatigue and the difference between the
model and prototype scale is also discussed.
4G Δh 2 TESTING PROCEDURE
Δσ n′ = (1)
D
The pile installation was modelled in the TU Delft
where G is the shear modulus, Δh is the width of geotechnical centrifuge (Allersma, 1994). The tests
the dilating interface and D is the pile diameter. were carried out at 50 g. A 380 mm × 150 mm ×
The size of the dilating interface depends on the 144 mm strongbox was used for the test. The sam-
grain size and is assumed to be relatively constant. ple container was equipped with two Plexiglas
Hence, the latter will not scale in the centrifuge. An windows where visual observations were made.
interface width of 0.02 mm has been proposed in The Plexiglas edges were marked at regular inter-
design procedures and has been seen to correlate val for correction of lens distortion. The test pro-
well with field measurements (Lehane et al, 2005, cedure consisted of stress measurements and visual
Axelsson, 1998). measurements. These were carried out pairwise in
Smaller model piles will have a significantly the same sample. The stress measurements were
higher additional normal stress at full mobilization. performed in the center of the sample container
Figure 1 shows the additional normal stress for with a 10 mm diameter instrumented model pile
different width of the dilating interface for a 10 mm connected to a data acquisition system. The soil
diameter pile with 50 MPa shear modulus. deformation measurements were performed in
The much higher interface stress during the half-space at one side of the Plexiglas borders with
shearing load cycles may result in a different a 5 × 10 mm non-instrumented pile. The experi-
magnitude of the friction fatigue effect in the mental set-up is shown in Figure 2.
model tests. The length of the frictional interface A machine vision camera (VC 4012Nano) was
is smaller in the centrifuge tests. Ring shear tests mounted on the centrifuge for the visual meas-
show very large mobilization length is needed to urements. The sample container was illuminated
reach a constant interface friction angle (Klotz & with portable LEDs attached to the centrifuge
Coop, 2001, Jardine et al, 2012). The ultimate shaft swing.
capacity will therefore be much larger in the model Stress measurements were conducted on an
test compared to the prototype pile, which has instrumented model pile. The model pile had
been observed in centrifuge tests (White & Lehane, two levels of normal contact stress sensors situ-
2004). Also, the pile diameter will slightly increase ated at 2.5 D and 7.5 D distance from the pile tip.
during pile installation because of the Poisson’s Each consisted of two separate sets of full-bridge
ratio effect. However, for small stiff piles which are strain gauges embedded inside the pile in sensing
80
3 TEST SERIES
81
4 RESULTS
82
Figure 8. Normal stress measurements for the higher Figure 11. Average normal stress normalized by the pile
and lower stress sensor in medium dense sand. base stress qb.
83
Figure 13. Displacement paths for loose, medium dense and dense initial density.
5 DISCUSSION & CONCLUSIONS of normal stress distribution along the shaft and
friction fatigue are similar at all densities. If the
The results in Figures 4 and 5 show good similarity different magnitudes of the material behavior can
with other centrifuge measurements (Klotz & Coop be taken into account, the same conceptual model
2001, White & Lehane, 2004), hence supporting the would therefore be applicable for pile design at any
quality of the setup. The small amount of meas- densities. The results also show that the cone resist-
urements points for the horizontal measurements ance is a relatively good tool for normalization of
in Figure 6 create somewhat non-continuous stress the measurements, at least when pile installation in
distribution, but a clear trend with increasing depth silica sand is modeled.
and dependence on higher initial density is shown.
A friction fatigue phenomenon is also observed in
the current tests (Figures 7–9). Interestingly, the REFERENCES
magnitude of the stress reduction scales well with
the initial stress as registered during the first pas- Allersma, H.G.B., 1994, The university of Delft Geo-
sage of the pile, where the initial stress is density technical centrifuge, Proc. Int. Conf 1994.
Axelsson, Gary, 1998, Long-term increase in shaft capac-
dependent. Also, the normal stress measurements ity of driven piles in sand, Proc. 4th Int. Conf. on Case
in Figure 10 show a surprisingly good fit to the Histories in Geotech. Engng., St. Louis, Missouri.
preferred design values for K0 in Kraft, 1991, espe- Beijer Lundberg, A., J. Djikstra, and K. van Beek, 2012,
cially, when taking into account the complex load- Measurements of soil contact stress in a harsh envi-
ing path experienced by the soil near the interface ronment, Instrumentation and Measurement Technol-
during installation. Currently, it is unclear whether ogy Conference (I2MTC), 2012 IEEE International.
this is a test specific result or more generally appli- IEEE, 2012.
cable. A more detailed interpretation of the test Boulon, M. & Foray, P., 1986, Physical and numeri-
data is required to verify the general applicability. cal simulation of lateral shaft friction along offshore
piles in sand, In numerical methods in offshore pileing,
Also, the unloaded horizontal stress have been 3rd international conference, Nantes, 1986.
used for this analysis. DeJong, J.T., White, D.J. and Randolph, M.F. (2006).
The normal stress measurements are normalized Microscale observation and modeling of soil-structure
by the base stress in Figure 1. This normalization, interface behavior using particle image velocimetry,
which can be assumed to be proportional to the Soils and Foundations, 46(1): 15–28.
cone resistance, shows that there is a clear connec- De Nicola, A. & Randolph, M.F., 1993, Tensile and com-
tion between the normal stress and the base stress. pressive shaft capacity of piles in sand, Journal of geo-
The displacement path measurements in Figure 13 technical engineering 119(12), 1952–1973.
show the larger horizontal displacement for dense Eslami, A. and Fellenius, B.H., 1997, Pile capacity by direct
CPT and CPTu methods applied to 102 case histories.
sand, which could influence the final normal stress Canadian Geotechnical Journal. 34(6): 886–904.
around the pile tip. Houlsby, G.T., 1991, How the dilatancy of soils affects
The measurements show a relatively continuous their behavior, Proceedings of the 10th European
change in governing mechanism between loose to Conference on Soil Mechanics and Foundation
dense sand. This implies and that the mechanisms Engineering.
84
Huy, N.Q., Dijkstra, J, and Tol, A.F. van, 2005, Influence Lehane, B.M., J.A. Schneider, and X. Xu, 2005, The
of loading rate on the bearing capacity of piles in UWA-05 method for prediction of axial capacity of
sand. Proceedings of the 16th International Confer- driven piles in sand, Proc., 1st Int. Symp. on Frontiers
ence on Soil Mechanics and Geotechnical Engineering, in Offshore Geotechnics. Perth, Australia: Balkema.
pp. 2125–2128. Randolph, M.F., Dolwyn, J. & Beck, R., 1994, Design of
Jardine, R.J., Zhu, B.T., Foray, P. & Yang,. Z.X., 2012, driven piles in sand, Geotechnique, 44(3), 427–448.
Measurements of stresses around closed ended dis- Vennemann, P., Lindken R. & Westerweel J., 2007, In
placement piles in sand, Geotechnique, 56(9), 1–17. vivo whole-field blood velocity measurement tech-
Klotz, E.U. & Coop, M.R., 2001, An investigation of the niques, Experiments in fluids, 42(4), 495–511.
effect of soil state on the capacity of driven piles in White, D.J., 2005, A general framework for shaft resist-
sands, Geotechnique, 51(9), 733–751. ance on displacement piles in sand, Proceedings of the
Kraft, L.M., 1991, Performance of axially loaded pipe 1st International Symposium on Frontiers in Offshore
piles in sand, J. of Geotech. Engrg, 117(2), 272–296. Geotechnics, ISFOG 2005. Taylor and Francis.
Lee, J.H., and R. Salgado, 1999, Determination of pile White, D.J., and B.M. Lehane, 2004, Friction fatigue
base resistance in sands, Journal of Geotechnical and on displacement piles in sand, Géotechnique, 54(10),
Geoenvironmental Engineering, 125(8), 673–683. 645–658.
Lehane, B.M., Jardine, R., Bond, A. & Frank, R., 1993,
Mechanisms of Shaft friction in Sand from instru-
mented pile tests, J. of Geotech. Engrg. 119(1), 1–19.
85
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
Y. Ishihara
Giken Seisakusho Co. Ltd, Kochi, Japan
ABSTRACT: Jacked piles are becoming a valuable installation method due to the low noise and
vibration involved in the installation procedure. Cyclic jacking may be used in an attempt to decrease the
required installation force. Small scale models of jacked piles were tested in sand and silt in a 10 m beam
centrifuge. Two different piles were tested: smooth and rough. Piles were driven in two ways with monot-
onic and cyclically jacked installations. The cyclically jacked installation involves displacement reversal
at certain depth for a fixed number of cycles. The depth of reversal and amplitude of the cycle vary for
different tests. Data show that the base resistance increases during cyclic jacking due to soil compaction
at the pile toe. On the other hand, shaft load decreases with the number of cycles applied due to densifica-
tion of soil next to the pile shaft. Cyclic jacking may be used in unplugged tubular piles to decrease the
required installation load.
86
Cyclically jacked installation can be used to scaling effect can be reduced with a ratio Dp /d50 > 50.
promote friction fatigue and hence reduce the jack- Fioravante (2002) also observed that the shear resist-
ing load by applying displacement reversal at a cer- ance mobilised depends on the normalised roughness
tain depth for a fixed number of cycles. The paper RN = Rt/d50 being Rt the pile surface roughness meas-
presents centrifuge tests on jacked piles installed in ured over a length of 0.8 to 2.5 mm. For monotonic
flight with cyclic jacking. The depth of reversal and loading no dilation is expected for RN < 0.02 (fully
amplitude of the cycles vary for different tests. The smooth) and for RN > 0.1 (fully rough) the interface
paper examines the effect of stroke amplitude and dilates and high values of τsu are expected.
installation rate on friction fatigue. Pile stiffness and The experimental work presented in this paper
strength after the installation are also assessed. was carried out in a 10 m beam centrifuge at
Cambridge University. Two different (Dp /d50) ratios
were tested in centrifuge using sand (Dp /d50 = 60)
2 CENTRIFUGE MODELLING and silica flour (Dp /d50 = 100). The aim of the
modelling is to prove whether silica flour can be
Centrifuge modelling is widely used in soil mechan- used in centrifuge in replacement of sand in the
ics due to the possibility of reproducing at model attempt of reducing scaling effects. Also the influ-
scale the correct stress level. The geotechnical cen- ence of the normalised roughness was analysed
trifuge increases the gravitational acceleration in by using two different piles: one rough and one
the model by spinning it at high centrifugal accel- smooth.
eration. The increase in acceleration compensates Tests were performed in a circular strong box of
the reduction in the model size. Scaling laws are diameter 850 mm and high of 400 mm. The height
derived for correspondence between the model and of the soil in the tub was 320 mm for all tests. Tests
the prototype (Taylor 1995). A detailed description were performed at a centrifugal acceleration of
of the geotechnical centrifuge at Cambridge uni- 50 g and piles were installed in flight by means of
versity may be found in Schofield (1980). an automatic 2D actuator. Full details of the test
Problems may occur when modelling of piles in programme and pile installation procedure will be
centrifuge, especially of shaft piles in centrifuge, given in the next paragraphs.
is affected by scaling effects related to the ratio of
shear band thickness/grain size on the mobilised
shear stress which does not scale in centrifuge. The 2.1 Test programme
thickness of the shear band depends on the sur- Monotonic and cyclically jacked installations were
face roughness and varies between 2–5 times d50 for performed in saturated silica flour and in dry sand.
smooth piles 10–15 times d50 for rough piles being In total 14 pile installations were installed, 6 in silica
d50 the mean grain size. flour and 8 in sand. A summary of the pile instal-
Fioravante (2002) studied the effect of the ratio lations programme is presented in Table 1. For the
of the pile diameter/mean grain size (Dp /d50) on the cyclically jacked tests the displacement was reversed
shear resistance mobilised τsu and observed that the at three different depths: 25 mm, 75 mm and 125 mm
87
for the number of cycles. The stroke is the maxi- cal properties of silica flour are summarized in
mum displacement in one cycle. Figure 2 shows the Table 2. No standard preparation exist for prepara-
typical displacement history applied during cyclic tion of centrifuge models with silica flour. Due to
jacking, the stroke varies as indicated in Table 1. the impossibility of pouring dry silica flour, silica
The cyclic jacking speed was 1.0 mm/s for sand and flour was mixed with water and pumped into the
0.5 mm/s for silica flour. The rates of penetration are tub (Silva (2005)). A layer of sand was used as bot-
comparable to the rate used at prototype scale. tom drainage. In the attempt to obtain a very dense
model the tub was vibrated on a vibration table
during filling. The operation also was necessary
2.2 Materials
for the release of air bubbles eventually trapped in
2.2.1 Marine Quartz sand during mixing.
Fine sand and silica flour were used in the attempt
of reducing scale effects. The sand is called Marine
2.3 Testing apparatus
Quartz sand and was supplied by Specialist Aggre-
gates Ltd.1 The sand is mostly silica at 96.2% 2.3.1 Model pile
and its mean grain size is between 0.1–0.3 mm. A stainless steel model pile was used for the tests.
Figure 3 shows the particle size distribution (PSD) The pile has an outside diameter of 12 mm, inside
obtained both by sieving and by SPOS (Single diameter of 10 mm and length of 160 mm. The
particle Optical Sizing) (White 2003). The PSD bending stiffness of the model pile is not reliant
obtained by SPOS overestimates the grain size of on accurate scaling. The purpose of having a very
20–30% as described by White (2003). Sand was stiff pile in centrifuge was primarily due to avoid
poured dry with an automatic sand pourer at a pile buckling during driving. Strain gauges were
relative density of 90%. A dense sample was aimed positioned at the pile base and at the pile head
in order to enhance friction fatigue under cycling to measure loads experienced during the tests.
shearing (Mortara 2007). The strain gauges consisted of a full Wheatstone
bridge circuit, compensating for any tempera-
2.2.2 Silica flour ture changes in the pile material. Finally, in order
Silica flour is >99% silica and it has a mean grain to obtain the rough pile, on the outside surface
size of 12 μm that classifies it as silt. The mechani- of the pile sand grains were stuck with gel glue.
Figure 4 shows the pile before and after the proce-
dure was terminated.
Values of the maximum surface roughness of the
pile were obtained by using an Taylor & Hobson
Form Talysurf series inductive profilometer over a
pile length of 2.5 mm. The value of Rt obtained for
the smooth pile was 25 μm. Due to the full scale of
Property Value
88
the instrument, it was impossible to measure the
surface roughness of the rough pile in the same way.
Rt can roughly be estimated as Rt = d50. From this
values, values for RN can be calculated as described
in section 2, which are reported in Table 1.
2.5 Instrumentation
Miniature pore pressure transducers (PPT) were
used for measuring excess pore pressure induced
by pile installation in silt. The transducers have a
maximum pressure reading of 700 kPa. Sensors Figure 6. Mobilised shaft load vs base load.
were placed at 20 mm from the pile centre after
the consolidation phase was terminated. An hol-
low tube was used for drilling holes in the model
and for driving the transducers in the holes easier. flour and (b) sand. As expected the loads required
Data from all channels were acquired at a sampling for installation in silica flour were considerably
frequency of 500 Hz. lower than those recorded in sand. The maximum
load recorded for silt was 289 N whilst in sand was
4.9 kN. The inconstant trend of the installation
loads recorded for Si_M03 in Figure 5(a) is prob-
3 RESULTS
ably due to the presence of in homogeneities in the
soil body. For the tests in sand, Figure 5(b) clearly
3.1 Monotonic installations
shows that the load required for installing the rough
Five monotonic installations were performed in pile (Sd_M01) is 80% higher than for the smooth
total: 3 in saturated silt and 2 in dry sand. In silt pile (Sd_M02). This is believed to be primarily to the
the installations were carried out at different instal- increase in the effective base area after roughening.
lation rate as indicated in Table 1, in sand an instal- It is interesting to observe the amount of
lation rate of 1.4 mm/s was used in both tests but shaft load mobilised during the installation.
two different piles were used. Figure 6 shows the maximum base and shaft load
Figure 5 shows the installation loads recorded recorded at the end of each installation normalised
by the strain gauges on the pile head for (a) silica by the total head load. It can be observed that,
89
for all installations, the shaft load is always less
than 50% of the head load and it is maximum for
Sd_M01 (30%). The lowest value is 8% for Sd_M02.
As observed by Uesugi & Kishida (1987) in direct
shear tests, the interface friction angle can be cor-
related to the normalised surface roughness RN of
the interface sand/steel. In the tests performed RN
is 0.12 for smooth piles in sand, 1 for rough piles
in sand and 1.6 for smooth piles in silica flour. The
interface friction angle increases with normalised
roughness. Figure 8. Head, Base and Shaft load—Sd_S01 at
z = 75 mm.
3.2 Cyclically jacking
As described in Section 2, cyclically jacking cycle whilst the shaft load reduces. As the shaft
involves pile displacement reversal at a given depth load is only a small percentage of the total head
for a fixed number of cycles. Typical results for load, for closed-ended piles, cyclic jacking leads to
silica flour and sand are shown in Figure 7. For a overall increase in the head load.
simplicity only the installations with the largest Nevertheless it is interesting to analyse shaft load
stroke are shown. reduction which becomes important when install-
After displacement reversal it is clear that for ing unplugged tubular piles. The authors believe
all installation the head load required for installing a phenomenon similar to friction fatigue occurs
the pile increases more than the monotonic value. when the pile is sheared cyclically. In fact after
The extra load required diminishes when the pile the pile is being installed, as described by White &
is pushed deeper and goes toward its monotonic Bolton (2004), an interface of crushed grains
value. Typically the area influenced by cycli- develops between the pile shaft and the surround-
cally loading is between 2–3 pile diameter. This is ing soil. During cyclic shearing the grains in the
believed to be due to compaction occurring under interface rearrange and, according to stress level,
the pile toe during cyclic jacking. there may be additional particles breakage and vol-
Although Figure 7 clearly shows the effect of ume contraction. If volume contraction occurs, the
cyclic jacking on the required force for installation, horizontal stress acting on the pile shaft decreases
it does not give information on the behaviour of leading to a decrease in the shaft friction.
shaft and base load evolution during cycling. For A parameter called ’degradation ratio’ (R) can
this aim, Figure 8 shows a typical history of head, be defined in order to evaluate the reduction of
base and shaft load during cyclic jacking. For com- shaft friction during cyclically jacking. The param-
parison base and shaft loads are normalised by the eter R is the ratio between the peak values recorded
value at the first cycle. It can be observed that dur- in the shaft load for each cycle n to the peak value
ing cyclic jacking, the base load increases for each recorded in the first cycle.
Fshafft ( n )
R= (1)
Fshafft ( )
90
ACKNOWLEDGEMENT
91
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
C. Jommi
Politecnico di Milano, Milano, Italy
Delft University of Technology, Delft, The Netherlands
ABSTRACT: The effects of installation procedures on the performance of capacitance probes for moni-
toring water content changes in surficial soils are analysed numerically. The sensors measure the resonant
frequency of an inductive-capacitive circuit, which includes the surrounding soil as capacitive element.
Literature calibration curves are most often adopted to determine the soil water content from the measured
resonant frequency. Alternatively, specific calibration for a given soil is performed in the laboratory. In both
cases, reference is made to ideal conditions, in which perfect contact is assured between the probes access
tube and the soil sample, prepared at uniform void ratio. Installation procedures in the field affect the sys-
tem and they may hinder correct estimation of volumetric water content in the field. Numerical results are
presented to quantify the influence of soil density changes promoted by installation operations and of the
saturated kaolin/cement paste used to guarantee continuous contact between the access tube and the soil.
92
dipoles of water in the dielectric medium (i.e. the 2 CAPACITANCE SENSORS
soil) are aligned by electric field and become polar-
ized. The capacitor is arranged to be part of an The EnviroSCAN® capacitance sensors consist
oscillator circuit, so that the frequency of oscilla- of two brass rings (50.5 mm outer diameter and
tion is a direct measure of capacitance (Gardner 25 mm height) mounted on a plastic sensor sup-
et al. 1991). Capacitance probes are a fast, safe and port (Fig. 1). The probes are inserted inside a PVC
relatively inexpensive way of measuring the relative access tube, strictly in contact with the rings and the
permittivity of soils, which in turn can be used to surrounding soil. Up to 16 probes can be mounted
estimate soil water content (Robinson et al. 1998, on a plastic guide at different depths. Minimum
Kelleners et al. 2004). Response time of the probes, relative distance between probes is 10 cm.
less than 1 second, is one of the most attractive The resonant frequency of the inductive-capacitive
features of capacitance probes. circuit, in which the probe is inserted, Fr, is linked to
This contribution focuses on the performance the inductive L and capacitive C elements,
of capacitance sensors (Sentek® 2001), which can
−1
be mounted in series into an access tube plugged
into the soil, down to two meters from the soil sur-
Fr ⎡2 p
⎣ ( LC ) ⎤⎦ , (1)
face. Monitoring depth can reach 5 meters below
the ground level, by means of appropriate install- where L is the circuit inductance and C = Cs + Cp is
ing procedures (Caruso, 2007). the total capacitance, including the soil component,
The sensors measure the resonant frequency Cs, together with probe constants, Cp (Kelleners
of the inductive-capacitive circuit, which includes et al. 2004). The volume of soil which effectively
the surrounding soil as capacitive element. Cali- contributes to the resonant frequency of the probe
bration relationships are then adopted to infer is a cylinder of about 20 cm diameter and 10 cm
soil water content from the measured resonance height, and its capacitance is a function of its rela-
frequency. tive dielectric permittivity, ε r.
As soil capacitance strongly depends on volu-
metric water content more than on any other prop-
erty, literature calibration curves are most often
adopted (e.g. Robinson et al. 1998, Kelleners et al.
2004). Better practice reckons on specific calibra-
tion for a given soil performed in the laboratory,
on samples prepared at uniform void ratio and
water content (Schwank et al. 2006).
In both previous cases, calibration refers to ideal
conditions, in which perfect contact is assured
between the probes access tube and a soil sample,
which is assumed to have a uniform void ratio in
the whole representative volume contributing to
the system response.
When the access tube is installed in situ, two
different procedures may be adopted: tipically, it
can be driven into the soil (to a maximum depth of
one to two meters, depending on the soil), or it can
be inserted into a drilled hole of slightly smaller
diameter. In the latter case, a mortar made of
kaolin, cement and water is added in the borehole
before inserting the access tube, in order to assure
perfect contact between the tube and the surround-
ing soil.
Both procedures create a disturbance in the
soil, which affects the resonant frequency of the
equipment, which in turn may hinder correct esti-
mation of volumetric water content in the field
based on ideal calibration procedures. Relevant
results of numerical analyses performed to quan-
tify the influence of installation procedures on
the response of the system are presented in the Figure 1. Capacitance sensors, internal circuit and
following. plastic support.
93
Since the sensors output primarily varies with
the air/water ratio, the resonant frequency is scaled
to the maximum measurement range. The scaled
frequency, SF, is defined as
SF = ( Fa Fr ) ( Fa Fw )
−1
(2)
3 IMPLEMENTATION
Figure 3. General view of the discretisation domain.
To analyse the possible effects of installation
procedures on the system response, three differ-
ent experimental configurations were simulated the system imposing a fixed potential difference
numerically. of 10 V between the brass rings, and solving for
The analyses were performed by means of full medium capacitance. Afterwards, the resonant
3D finite element discretisation, implemented in frequency of the circuit was derived based on its
Comsol Multiphysics®. Preliminary analyses had definition (eq. 1).
shown that, due to the shape of the plastic sup- A reference system was first analysed
port (Fig. 2), axisymmetric scheme should not be (configuration 1), representing the ideal labora-
adopted in the simulation, because of the relevant tory calibration protocol, where perfect contact is
error induced in the simulated electrical response. assured between the access tube and a sample hav-
Charge conservation throughout the domain ing uniform void ratio in the whole representative
and null electrical fluxes across the external volume. The results of this analysis were compared
boundaries were imposed. The radius of the dis- to the two numerical schemes representing, respec-
cretisation domain was fixed in all the analyses to tively, the driven access tube (configuration 2),
220 mm (Fig. 3). Different discretisation meshes and the access tube plugged into a pre-drilled hole
were adopted, up to a maximum of roughly filled with mortar (configuration 3).
150000 tetrahedral elements. Maxwell laws were The resonant circuit of a probe was simulated
solved with reference to steady state response of by reconstructing the real geometry of the relevant
portion of the system, composed by the sensor, the
access tube and the PVC internal support, with a
total height of 212 mm.
As for configuration 1, the ideal installation
procedure has been modeled by defining a simple
porous media of average properties around the
tube, with a porosity of 0.45.
As for configuration 2, the driven procedure
has been reproduced by decreasing the void ratio
in an inner portion of the domain (Fig. 4), fol-
lowing field and model experience on driven
piles (Randolph et al. 1979, Dijkstra et al. 2012).
Configuration 3 entails the definition of an inner
zone of 1.5 mm filled with kaolin, cement and water
around the access tube, while the surrounding soil
volume was assigned an undisturbed porosity of
0.45, as a first approximation (Fig. 5).
Figure 2. Top view of the inner part of the mesh, As for the relative dielectric constants, values
showing the plastic support section. of 1 for air (εr,a), of 88 for water (εr,w), of 3 for
94
To analyse the behaviour of sensors in contact
with moist soil, a mixture law had to be adopted to
assign dielectric properties to the soil layers. Two
different mixture models were tested, the one pro-
posed by Dirksen (Hilhorst et al. 2000),
ε S (θ ) θ ε + ( − n) ε ( θ) ε , (3)
r r,w r,s r,a
1
S (θ ) = (4)
Figure 4. Schematic top-view of driven installation
3( n − θ )
scheme.
and the simple model by Topp et al. (1980):
4 REFERENCE MODEL
95
between the soil volume prior to and after instal-
lation was imposed to determine the final aver-
age porosity in the disturbed volume. A porosity
profile for the disturbed area was determine
imposing it had this average value, and the initial
porosity n0 = 0.45 at the external boundary of the
disturbed soil volume. The final porosity profile
reads:
⎛ R2 ⎞
n(r ) = 0.4558 − 0.144 ⎜ 20 ⎟ (6)
⎝r ⎠
5.3 Results
Simulated resonant frequencies for the three con-
figurations as a function of volumetric water
content of the undisturbed soil, θ0, are shown in
Figure 9.
Figure 8. Electric field in an ideal soil with θ equal to
0.45 (mixture model by Topp et al. 1980).
5 INSTALLATION EFFECTS
96
As evident, the driven access tube procedure 6 CONCLUSIONS
entails a change in the resonant frequency of the
system, which increases with the undisturbed volu- A finite element model was designed to assess
metric water content, θ0. The actual resonant fre- the influence of installation procedures on the
quency is always greater than that of the ideal case, response of capacitance water content sensors.
at the same θ0. The results allow for quantifying the sensitivity
The difference between the estimation of θ0, of the system to porosity changes, relevant for the
with respect to the ideal reference scheme, is in driven installation procedure, and to the presence
the range between 10% and 20%. The maximum of the kaolin mortar adopted in the drilled instal-
difference corresponds to θ0 equal to 0.2, which lation procedure.
means around a degree of saturation of about If disturbance induced by installation is disre-
0.5. garded, erros up to about 20% of the actual value
As for the drilled procedure, 3, the effect of the may affect the estimation of water content. In
kaolin mortar changes with θ0, since it represents both cases, the estimated water content is gener-
a wetter layer in dry soils, and a drier layer in wet ally lower than the actual one, although for differ-
soils. Therefore, the results for the ideal scheme ent reasons and with different relevance at varying
and for the drileld scheme intersect, at a value of water content.
θ0 which is roughly equal to the volumetric water The conclusions of this numerical study were
content of the mortar. confirmed by recent preliminary data coming
The difference between the estimations of θ0 in from a dedicated experimental investigation in the
this case, with respect to the ideal reference scheme, field.
may be higher than for the driven procedure, and
ranges from 0 to 40%. The maximum difference
occurs for dry soil. REFERENCES
These conclusions are confirmed also by the
elaboration in Figure 10, in which the comparison Caruso, M. 2007. A theoretical, experimental and numer-
between the actual θ0 and the values of θ one would ical approach to predict the hydrologic balance in the
estimate relying on a ideal scheme are compared. upper unsaturated soil layers (in Italian). Doctorate
The latter were calculated by scaling the simulated Thesis, Università di Parma.
resonant frequencies with the values of resonant Charlesworth, P. 2005. Soil water monitoring. An infor-
frequency in air and water, as suggested in a stand- mation package. 2nd ed. Irrigation Insight No 1.
ard interpretation procedure. Dijkstra, J., Broere, W. & Van Tol, A.F. 2012. Electrical
resistivity method for the measurement of density
changes near a probe. Géotechnique 62(8): 721–725.
Gardner, C.M.K., Bell J.P., Cooper J.D., Dean T.J.,
Hodnett M.G. & Gardner N. 1991. Soil Water Content.
In Smith R.A., Mullings C.E., (eds) Soil Analysis—
Physical Methods. Marcel Dekker, New York.
Hilhorst, M.A., Dirksen, C., Kampers, F.W.H. &
Feddes, R.A. 2000. New dielectric mixture equation for
porous materials based on depolarization factors. Soil
Science Society of America Journal. 64: 1581–1587.
Kelleners, T.J., Soppe, R.W.O., Robinson, D.A., Schaap, M.G.,
Ayars, J.E. & Skaggs, T.H. 2004. Calibration of capaci-
tance probe sensors using electric circuit theory. Soil
Science Society of America Journal. 68: 430–439.
Mojid, M.A., Wyseure, G.C.L. & Rose, D.A. 2003.
Electrical conductivity problems associated with
time-domain reflectometry (TDR) measurement in
geotechnical engineering. Geotechnical and Geological
Engineering 21: 243–258.
Noborio, K. 2001. Measurement of soil water content
and electrical conductivity by time domain reflectom-
etry: a review. Comp and Elec in Agric 31: 213–237.
Paltineanu, I.C. & Starr, J.L. 1997. Real time soil water
dynamics using multisensors capacitance probes. Soil
Science Society of America Journal 61: 1576–1585.
Figure 10. Simulation results for the three models in Randoplh, M.F., Carter, J.P. & Wroth, C.P. 1979. Driven
terms of inferred volumetric water content as function piles in clay—the effects of installation and subse-
of θ0. quent consolidation. Géotechnique 29(4): 361–393.
97
Robinson, D.A., Gardner, C.M.K., Evans, J., Schwank, M., Green, T.R., Mätzler, C., Benedickter, H. &
Cooper, J.D., Hodnett, M.G. & Bell, J.P. 1998. The Flürer, H. 2006. Laboratory characterization of a com-
dielectric calibration of capacitance probes for soil mercial capacitance sensor for estimating permittivity
hydrology using an oscillation frequency response and inferring soil water content. Vadose Zone Journal.
model. Hydrology and Earth System Sciences 2 (1): 5: 1048–1064.
111–120. Sentek 2001. Calibration of Sentek Pty Ltd Soil Moisture
Scanlon, B.R, Andraski, B.J. & Bilskie, J. 2002. Miscel- Sensors. Sentek Pty Ltd, Stepney, South Australia.
laneous methods for measuring matric or water poten- Topp, G.C., Davis, J.L. & Annan, A.P. 1980. Electromag-
tial. In Dane, J.H. & Topp, G.C. (eds), Methods of Soil netic determination in soil-water content: measure-
Analysis, part 4, Physical Methods. Soil Sci Soc. Am: ment in coaxial transmission lines. Water Resources
643–670. Research. 16: 574–582.
Schwank, M. & Green, T.R. 2007. Simulated effects of
Soil Temperature and Salinity on Capacitance Sensor
Measurements. Sensors, 7(4): 548–577.
98
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
The load capacity of driven cast in-situ piles derived from installation
parameters
D. Egan
Keller Foundations, UK
ABSTRACT: The theoretical estimation of the ultimate capacity and serviceability performance of
common displacement foundation systems such Driven Cast In-Situ (DCIS) piles is difficult due to the
huge disruption in soil structure and insitu stress regime caused by the installation process. Even though
much research effort is expended on complex numerical modeling and reduced scale laboratory or centri-
fuge modeling there remains the difficulty of translating the knowledge gained into practical prediction
tools appropriate for routine design and installation of the full size product in the field. So to advance
and validate the conclusions drawn from numerical and small scale research the third strand of measuring
and analysing full size field behaviour must be added. This paper will summarise recent advances made in
the field measurement and analysis of installation parameters to predict the load capacity of driven cast
in-situ piles. The results from installation and testing of a DCIS pile is used to illustrate the methodology
now being routinely by Keller Foundations in the UK. The conclusions drawn from this paper are already
raising the standard of reliability, efficiency and sustainability of DCIS piles on routine projects.
99
pile load capacity. While this approach has been
used satisfactorily for the installation of hundreds
of thousands of piles over many decades, mod-
ern instrumentation now allows both robust and
detailed control and feedback of the pile driving
process, in real time, for every pile installed.
On modern instrumented DCIS piling rigs the
energy delivered to each pile can now be meas-
ured and used to validate that pile’s ultimate load
capacity. As the pile is driven the driving param-
eters (drive energy, drive tube depth and rate of
advancement) are displayed to the rig driver giving
a high degree of control to the installation proc-
ess. Additional benefits of an automated electronic
data capture system include a reduction of manual
recording and on-site paperwork (which requires
a finite manpower resource), robust archiving of
records and real time access to installation data
remote from the work site.
From the experience of developing a large
database of systematically acquired pile instal-
lation data it has been possible to map the proc-
ess required to allow real time assessment of pile
capacity. Initially focused on piles founded in
granular soils an empirical approach has been
developed which has been used to validate, in near
real time, the adequacy of DCIS piles as they are
driven. Of course these advances should be seen
in the context of dynamic pile driving formulae
which have been used on preformed piles for many
decades. Not withstanding widespread the use of Figure 1. Typical borehole from the test site.
set calculations for pre-formed piles, their appli-
cability (and reliability) for driven cast in-situ pil-
ing, where the drive tube is firstly driven into the and Alluvium were not considered as contribut-
ground and then removed and replaced by cast in ing to the pile geotechnical capacity, but rather a
place concrete, is highly questionable. source of potential negative skin friction, they are
The development of real time assessment of ignored in the assessment of ultimate pile capacity
DCIS pile capacity for piles driven into coarse for design purposes.
grained soils is presented in detail below. This
method has been developed over a period of time
and is based on drive data and load test data 4 DRIVEN CAST INSITU PILE
carried out on a series of sites. However to bet- INSTALLATION
ter illustrate the process reference to a single test
pile (TP3) installed at a site in Erith, London is Figure 2 illustrates the pertinent features of the
made. DCIS system. A steel drive tube, closed at the base by
a sacrificial base plate, is top driven to the required
depth. The steel tube is then charged with a free
3 DESCRIPTION OF TEST SITE flowing concrete and withdrawn from the ground.
Steel reinforcement can be inserted in the pile bore
The test site was located within the Thames basin before or after the concrete is placed. The require-
to the south of the river at Erith, London. The ments for the execution of DCIS piles are set out
ground conditions comprised a mantle of Made in EN 12699 ‘Execution of special geotechnical
Ground, overlying soft and very soft clay and works—Displacement piles’. In addition Keller
peat Alluvium. The piles were founded in coarse monitor all aspects of the installation process by
grained River Terrace Deposits (RTD) comprising onboard computer, with the key parameters for
medium dense to very dense gravelly sand. Thanet capacity assessment being the hammer energy
Sand was found beneath the RTD. Figure 1 shows imparted for each blow, the rate of advancement
a typical bore hole for the site. The Made Ground of the tube per blow and the tube and base plate
100
Figure 2. The DCIS installation process.
5 RECORDED INSTALLATION
PARAMETERS
101
ground into which the pile is being driven can be
obtained. In developing the analogy between qpc
and CPT qc factors such as energy losses within the
drive/tube system, frictional losses along the drive
tube shaft, rate effects and size effects have all to
be considered.
An extensive series of field tests using dynamic
pile driving analysis has been undertaken to assess
the loss in energy between the point of impact of
the hammer at the top of the drive tube and the
pile base plate.
The ratio of the area of the base plates of differ-
ent sizes of DCIS pile and a standard CPT range
between 26 and 6, and taken in the round this is
considered beneficial. One of the disadvantages
of the CPT, especially in coarse soils, is the pro-
pensity for the tip resistance to be influenced by
larger soil particles (or localised weaker or softer
layers) in the soil. In contrast the larger diameter
of the DCIS base plates stresses a greater depth
of soil leading to somewhat of lower bound (and
safe) smoothing effect when compared to the CPT
qc. Indeed the zone of soil stressed beneath the
pile toe at the end of driving is exactly that which
will be stressed by the finished pile in the working
condition. (This is in the context of a single pile,
102
The above equation is readily implemented in the procedure to assess the ultimate capacity of DCIS
commercially available software or spreadsheets. piles located in coarse grained soils based in pile
The program Oasys Pile, using the Berezantzev installation data is available.
formulation for the bearing capacity factor, Nq,
(Oasys Pile user manual, 2012) was used to give:-
8 CONCLUSIONS
Qult , socket kN kN kN (5)
Driven cast in-situ piling has proved a popular and
efficient piling method over many decades, which,
Test Pile 3 was subjected to a maintained in suitable ground conditions, produces piles of
extended proof load test to failure following the high capacity and efficiency. Recent developments
procedure outlined in the ICE Specification for in rig instrumentation have enabled the routine and
Piling and Embedded Retaining Walls (2007). systematic acquisition of high quality installation
Figure 6 shows the load settlement response. records. The benefits of an automated electronic
A plunging failure of the pile under an applied data capture system include a reduction of manual
load, Qf, of 1947 kN and at a displacement of recording and on-site paperwork (which requires
45 mm was observed. a finite manpower resource), robust archiving of
In the analysis of the pile test results the tem- records and real time access to installation data
porary hold up resistance generated over the pile remote from the work site. In addition Keller has
shaft passing through the made ground and allu- demonstrated the development of a real time vali-
vium was estimated at 80 kN, which must be added dation method for DCIS piles is viable. A rational
to Qult,socket as calculated in pile to give the true link between the installation parameters, funda-
anticipated theoretical pile capacity of 2060 kN. mental soil parameters relevant after pile installa-
Thus the pile under test achieved 94% of its the- tion, the physical mechanisms of load generation
oretical capacity which is considered an excellent and observed load capacity has been established.
correlation between the calculated capacity and These developments in the driven cast in-situ
that shown by the static load test. piling process have lead to improved efficiency
The results of the illustration of the installation (and hence sustainability) of the DCIS method.
and load testing of pile TP3 show that a rational
REFERENCES
103
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
H.K. Engin
Geo-Engineering Section, Delft University of Technology, Delft, The Netherlands
R.B.J. Brinkgreve
Geo-Engineering Section, Delft University of Technology, Delft, The Netherlands
Plaxis BV, Delft, The Netherlands
ABSTRACT: The installation process of a displacement pile causes a considerable amount of soil dis-
placement and high levels of stresses, and therefore alters the soil state and properties around the pile.
These installation effects may have important consequences on the performance of the pile in its service
life (e.g. load—displacement behaviour) and on the neighbourhood (e.g. vibrations, nuisance). A more
realistic behaviour and therefore an improved design would be achieved by considering the installation
effects in the analyses. In current practice, the installation effects are taken into account by some empiri-
cal design methods in order to estimate the bearing capacity of foundation piles. The objective of this
numerical study is to investigate and model the installation effects of pile jacking in sand in a numerical
framework. In the first part of the study a simplified numerical technique was employed to investigate
the installation effects. The results are approximated by nonlinear regression. Despite the limitations and
simplifications, it was shown that the installation effects can be represented in terms of functional forms
reasonably well. Furthermore, these functions can be easily applied in a standard FE analysis.
104
the final equalisation states have been analysed, i.e. behaviour. A simple sketch of the geotechnical
the situation after the jack force has been removed. problem considered and the corresponding axisym-
For the approximation of the installation effects metric FE model employed in the numerical analy-
a nonlinear regression algorithm is employed. By ses is given in Figure 2. The soil profile consist of
the approximated form the state around a pile that approximately 15m thick very soft clay layer over-
is modelled as wished-in-place, can be altered to lying a sand layer, which is a typical profile in The
incorporate the installation effects without simu- Netherlands and places having similar geological
lating the installation process. Finally a compari- sequences (e.g. Pleistocene glacial and interglacial
son of the results obtained by the approximated periods). The clay layer is modelled as an elas-
effects and the simulation results using the Press- tic layer of 1 m thickness having an effective unit
Replace technique are given. weight of γ ′ = 100κN/m3. The purpose of modelling
the clay layer as such is to facilitate the generation
of geostatic (K0) stresses in the sand layer. There-
2 NUMERICAL MODELLING fore the clay layer is considered to generate a proper
OF PILE JACKING stress level on top of the sand layer. The interaction
between the pile and the clay layer is also ignored
The installation effects around a jacked pile have due to very low cohesion of the clay layer.
been investigated using the the ‘Press-Replace’ (PR) To model sand behaviour, hypoplasticity (von
technique (Engin & van Tol 2011). The technique Wolffersdorff 1996) was used in the analyses due
was introduced to promote the modelling of large to its state dependent stiffness and strength for-
deformation effects using small deformation for- mulation. Having a state dependent stiffness and
mulations. It was shown that the installation effects strength results in different levels of stiffness and
could be modelled reasonably well. In Figure 1, a strength mobilisations at different levels of pen-
general view of the model and four subsequent cal- etration depths. Similarly, different levels of mobi-
culation phases are shown. The calculation phases lisation of stiffness and strength is obtained for
continue until the desired level of penetration of the different pile diameters and different sand densi-
pile is achieved. In order to facilitate the PR tech- ties. As a result, the effect of change in stiffness is
nique, an axisymmetric FE mesh with small slices taken into account in the analyses. Table 1 presents
is introduced in the region where the pile will be the hypoplastic model parameters of Baskarp sand
jacked. Interface elements are also defined between used in the analyses. The soil state (stress and den-
the pile tip and the underlying soil to model proper sity) around the pile is altered as a result of the dis-
interaction with the continuum. In order to avoid placement of the soil dictated by the pile jacking
stress oscillations at the corner of the pile tip, inter- operations. At the same time, the stiffness of the
faces should be extended into the soil volume (van soil changes. Hence, these three major installation
Langen 1991). Based on the preliminary study, the effects should be assessed in order to quantify the
vertical as well as horizontal interface extension new state around the jacked pile. A state dependent
int
lengths equal to the slice thickness (lext tslice ) were
suggested. Furthermore, an optimum slice thick-
ness in terms of numerical accuracy and CPU time,
tslice = 0.10 ⋅ Dp was found.
In this study, the PR technique is used to model
continuous pile jacking in sand to investigate the
installation effects for different pile geometry and
soil density variations. Due to lack of space, only
one variation (D = 0.40m; L = 10D; Id = 0.80) is
presented in this paper. The focus is on the change
in state around the pile due to the installation
process, which also dictates the load displacement
Figure 2. a) Sketch of the geotechnical problem
modelled b) General view of the FE model.
ϕc hs
(°) (MPa) n ed0 ec0 ei0 a β
Figure 1. Details on the Press-Replace modelling tech-
nique and progress of penetration of the pile (Engin 30 4000 0.42 0.548 0.929 1.080 0.12 0.96
et al., 2011).
105
constitutive model such as hypoplasticity, reduces All length scales in the model were also
the number of unknowns since the stiffness is normalised with the size of pile diameter, D. This
defined in terms of stress and void ratio. Therefore, normalisation enables a more objective evaluation
to assess the installation effects for each case ana- of the effects of pile diameter and length on the
lysed, the Cartesian stresses at the end of equalisa- Rrr, Rzz, Rθθ, and Rrz and Re values.
tion, σ rr′ f , σ zz
′ f , σ θθ
′ f , and σ rz
′ f and the void ratio, ef
distributions around the jacked pile are normalised
with the corresponding K0 states as: 3 APPROXIMATION OF INSTALLATION
EFFECTS
σ rr′ f σ′f σ′f
Rrr = , Rzz = zz , Rθθ = θθ , In this part of the study the possibility of incor-
p0′ p0′ p0′
porating the installation effects around a wished-
σ rz
′f ef in-place pile to account for the installation effects
Rrz = , Re = (1)
p0′ eK 0 without simulating the whole penetration process
is investigated. A practical way to represent the
For the K0 state, installation effects is using model functions for
each variable and applying a regression analysis to
σ rr′ ,0 σ rr′ ,0 determine the coefficients of the model functions.
Rrr,0 = = Once having the fit, less data, i.e. only the fitting
p0′ 1
( rr,,
rr zz,,0
zz θθ ,0 ) coefficients, are required to represent a continuous
3 field of installation effects.
3K 0
= (2) Based on the preliminary study the following
1 2K0 model functions, in which the length scales are also
normalised by the pile diameter D, are proposed
Similarly, for the rest of the normalised Carte- for the representation of each normalised Carte-
sian stresses the following can be obtained: sian stress and void ratio field:
3 3K 0 ∧
n
Rzz ,0 = , Rθθ ,0 = (3) ψ m = Rm, + ∑ aim ⋅ e − ρ im (5)
1 2K0 1 2K0
i =1
Since σ rz ,0 = 0,
0 and e eK 0 , and,
Rrz,0 = 0, Re,0 = 1 (4)
ρim = biim (r / D riim )2 + cim ( z / D + L / D zim )2 (6)
Since the initial (K0) state is known, one can
determine the altered state by multiplying the where m (m = 1, 2,…, 5) represents the variable
normalised values by mean effective stress at considered and n represents the number of expo-
K0 state ( p0′ ). Based on the new stress and void nential terms used for that variable; such that
ratio, the hypoplastic model spontaneously alters m = 1 represents the radial, m = 2, the vertical,
the stiffness to a consistent level as a result of its m = 3, the tangential, m = 4, the shear stress com-
efficient stiffness formulation. ponents and m = 5 represents the void ratio. The
approximation functions have n = 3, 4, 3, 6, and 4
exponential terms, respectively. Rm,0 is the constant
term calculated for the K0 state of corresponding
variable (see Equations(2)–(4)). aim, bim and cim are
the fitting coefficients representing the magnitude
and spread of the exponential forms and rim and
zim are the fitting coefficients representing the loca-
tion of the Gaussian forms of corresponding vari-
able m. For the sake of compatibility for different
L/D analyses, the vertical position of the Gaussian
form is given relative to the location of pile tip (e.g.
the term in the exponent cim(z/D + L/D + zim)2).
The PR results of each case have a different dis-
cretisation level. Most of the effects are pronounced
near the pile, as expected. To have a more reliable
correction factor and a consistent distribution of
Figure 3. Description of the surface plots. the data points, the PR field data were interpolated
106
onto a uniform grid. The grid spacing is smaller
near the pile, in the influence region, where the
installation field shows clear deviation from the K0
state, and larger otherwise. As a result of using the
new grid the regression analysis is preconditioned
to fit better in the influence region. Furthermore, in
order to eliminate the spurious concentrations and
to obtain the underlying functions that describe
the installation fields, median filtering was applied
on the interpolated PR analysis results.
The model functions have in common two spatial
variables r and z, one constant K0 and five fitting
coefficients aim, bim, cim, rim and zim. Since the model
functions are not linear in r and z, the resulting
regression analysis applied on the surface data is
a bivariate non-linear regression. Figures 4–8 give
a comparison of the approximate surface and the
∧
PR simulation results (data points). Figure 4. Fitting results of normalised radial stresses ψ rr.
After obtaining the fitting coefficients and
substituting the normalised forms back into
Equations (1)–(5), the following forms can be
obtained for the Cartesian stresses, σ rr′ , σ zz ′ , σ θθ
′
and σ rz′ and the void ratio, e:
⎡ 3 ⎤
σ rr′ = p0′ ⋅ ⎢Rrrr,0 + ∑ ai1 ⋅ e − ρi 1 ⎥ (7)
⎣ i =1 ⎦
⎡ 4 ⎤
′ = p0′ ⋅ ⎢Rzzz ,0 + ∑ ai 2 ⋅ e − ρi 2 ⎥
σ zz (8)
⎣ i =1 ⎦
⎡ 3 ⎤
′ = p0′ ⋅ ⎢Rθθθ ,0 + ∑ ai 3 ⋅ e − ρi 3 ⎥
σ θθ (9)
⎣ i =1 ⎦
⎡ 6 ⎤
′ = p0′ ⋅ ⎢Rrrz ,0 + ∑ ai 4 ⋅ e − ρi 4 ⎥
σ rz (10)
⎣ i =1 ⎦
∧
Figure 5. Fitting results of normalised vertical stresses ψzz.
⎡ 4 ⎤
e eK 0 ⋅ ⎢1 + ∑ ai 5 ⋅ e − ρi 5 ⎥ (11)
⎣ i =1 ⎦
107
Figure 10. Comparison of the surface fit of σ ′zz with
∧
Figure 7. Fitting results of normalised shear stresses ψ rz. PR data.
Figure 9. Comparison of the surface fit of σ ′rr with Figure 12. Comparison of the surface fit of σ ′rz with
PR data. PR data.
108
obtained using the formulas was on the conserva-
tive side of the curve obtained from the PR results.
An improvement factor, η at any displacement
level can be defined using the total reaction force,
Ft of each case as:
FTΨ
4 DISCUSSION OF RESULTS
AND CONCLUSIONS
109
REFERENCES Mahutka, K.P., Kőnig, F., & Grabe, J. 2006. Numeri-
cal modelling of pile jacking, driving and vibratory
Baligh, M.M. 1975. Theory of deep static cone penetra- driving. In Proceedings of International Conference
tion resistance. Technical report, MIT Dept. of Civil on Numerical Simulations of Construction Processes
Eng. in Geotech-nical Engineering for Urban Environment
Bauer, E. 1996. Calibration of a comprehensive hypo- (NSC06), pp. 235–246.
plastic model for granular materials. Japanese Geo- Pham, H.D., Engin, H.K., Brinkgreve, R.B.J. & van Tol, A.F.
technical Society 36(1), 13-26. 2010. Modelling of installation effects of driven piles
Broere, W. & van Tol, A. 2006. Modelling the bearing using hypoplasticity. In T. Benz and S. Nordal (Eds.),
capacity of displacement piles in sand. In Proceedings Numerical Methods in Geotechnical Engineering,
of the ICE—Geotechnical Engineering, Volume 159, Numerical Methods in Geotechnical Engineering 2010:
pp. 195–206. Proceedings of the Seventh European Conference on
Chow, F.C. 1996. Investigations into the behaviour of dis- Numerical Methods in Geotechnical Engineering,
placement piles for offshore foundations. Ph.D. thesis. pp. 261–266. Taylor & Francis.
Dijkstra, J. 2009. On the Modelling of Pile Installation. Said, I., De Gennaro, V., & Frank, R. 2008. Axisymmet-
Ph.D. thesis. ric finite element analysis of pile loading tests. Com-
Einav, I. & Randolph, M.F. 2005. Combining upper puters and Geotechnics 36(1–2), 6–19.
bound and strain path methods for evaluating pen- Teh, C.I. & Houlsby, G.T. 1991. An analytical study of
etration resistance. Int.J. Num. Meth. Engng. 63(14), the cone penetration test in clay. Géotechnique 41(1),
1991–2016. 17–34.
Engin, H.K., B.R.B.J. & van Tol, A.F. 2011. Numerical van Langen, H. 1991. Numerical Analysis of Soil-Struc-
analysis of installation effects of pile jacking in sand. ture Interaction. Ph.D. thesis.
In G. Pietruszczak, S. & Pande (Ed.), In proceedings von Wolffersdorff, P.A. 1996. A hypoplastic relation for
of International Symposium on Computational Geome- granular materials with a predefined limit state sur-
chanics, ComGeo-II, pp. 744–755. face. Mechanics of Cohesive-frictional Materials 1(3),
Lehane, B. 1992. Experimental investigations of pile 251–271.
behaviour using instrumented field piles. Ph.D. thesis. White, D.J. 2002. An investigation into the behaviour of
Liu, W. 2010. Axisymmetric centrifuge modelling of deep pressed-in piles. Ph.D. thesis.
penetration in sand. Ph.D. thesis.
110
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
A. Ghandeharioon
Department of Civil Engineering, Faculty of Science and Engineering, Laval University,
Quebec, Canada
ABSTRACT: Analytical investigations and large-scale laboratory experiments were conducted to study the
soil disturbance due to the installation of mandrel-driven Prefabricated Vertical Drains (PVDs) in soft saturated
clays. Considering the mandrels commonly used for installing PVDs in the field, an elliptical Cavity Expan-
sion Theory (CET) was formulated to analyze the shear strain and pore pressure developed in soil during the
mandrel installation. The elliptical CET was developed using modified Cam clay parameters for the undrained
analysis of PVDs installed in soft soil deposits. This formulation identifies a critical zone, a plastic zone, and an
elastic zone around mandrel-driven prefabricated vertical drains. The large-scale laboratory tests consider the
effects of in-situ stresses using a specially designed consolidometer, and a rate controlled installation machine.
The pore water pressure was measured at various locations during the installation of a PVD and withdrawal
of the mandrel. The analytically predicted pore pressures agreed with the measurements in the laboratory. The
results of moisture content tests were also analyzed to verify the concept of an elliptical smear zone around
drains, and to derive a relationship between in-situ effective stresses and the extent of the smear zone.
111
2 ELLIPTICAL CAVITY EXPANSION In terms of effective stresses, Equations (1)
THEORY become:
⎛ P − σ 0 ⎞ ⎛ r12 ⎞
U =⎜ i ⎜ ⎟, V τ rθ 0, u=0 (5)
⎝ 2 G ⎟⎠ ⎝ r ⎠
2
⎛r ⎞
Figure 1. Expansion of an elliptical cavity in an infinite Δσ r Δσ r′ = − Δσ θ′ Δσ θ = − ( Pi − σ ) ⎜ 1 ⎟ (6)
⎝ r⎠
soft saturated clay, shown in polar coordinates.
112
where, r is the position of a soil element measured where, σrp is the total radial stress at the elastic-
from the center of the cavity and r1 is the radius of plastic boundary.
an elliptical cavity. G is the shear modulus of soil. As shown by Cao et al. (2001), a closed form
The pressure-expansion relationship in the elas- solution of the total stresses may be derived by
tic phase is developed as: taking the deviator stress in the plastic zone equal
to the deviator stress at failure (qf). By incorpo-
2G rating the logarithmic strains to justify the large-
r1 = r (7)
2G (Pi − 0 )0 strain effects, the internal pressure in cavity is
expressed as:
in which, r0 (θ ) a0 b0 b02 (a02 b02 ) 2
θ is the ini-
tial radius of the elliptical cavity. np 1 qf 3G (r − r )
Pi p0 − Mp0′ − ln (12)
3 3 r12 q f
2.3 Analysis of the plastic state
As the mandrel penetrates the soil, internal pres- M ′ ( n p /2 ) λ
in which, q f Mp
sure in the cavity increases and as a result, the devi- According to Ghandeharioon (2010), the rela-
ator stress (q) of every soil particle increases. The tionship between radial distance and its associated
total stress parameters are used in this research to deviator stress can be determined by:
analyze the cavity expansion in the context of criti-
cal state. In the plane strain condition: ⎛ r2 − r2 ⎞
ln ⎜1 − 1 2 0 ⎟
⎝ r ⎠
p ( + )/ q = 3( − )/ 2 (8) κΛ ⎡ η⎞ ⎤
1⎛
r r q
= −2 3 ⎢ζ − tan ⎝ M ⎠ np − 1⎥
1
t
tan
3G υM ⎣ ⎦
The constant deformation in soil establishes a
link between q p′ and the initial state of effective (13)
stress, p0′. Based on Cao et al. (2001), the effective
stress path in the p′ : q plane in modified Cam clay The principle of effective stress denotes that
context under an undrained condition is given by: the excess pore water pressure is the difference
between changes in the states of total and effec-
1 tive stresses:
M ′ n p ( p′ p′ )
−
q Mp Λ −1 (9)
Δu = Δp − Δp' (14)
where, n p pp′y0 /pp0′ and Λ = 1 −κ λ are the isotropic
overconsolidation ratio and plastic volumetric Based on Equations (13) and (12), q and p can
strain ratio, respectively. py′0 is the maximum iso- be calculated at any soil element inside the plas-
tropic preconsolidation stress. κ is the slope of tic region. Equation (11) is then used to derive
elastic swelling line and λ is the slope of the normal the state of total stress at that particular posi-
compression line in the ln p′:υ plane where υ is the tion, while noting that p σ r q/ 3 . Finally,
specific volume. by using Equation (14), the value of excess pore
Using Equation (9), the minimum internal pres- pressure can be determined at the location being
sure required to yield the soil elements adjacent to considered.
the wall of the cavity is determined as:
113
The deviator stress of a particular soil element porous brass tips were used to monitor changes
can then be related to its radial distance from the in pore pressure during initial consolidation and
center of the cavity by: during installation of the mandrel and its with-
r1 ⎛ q κΛ ⎡ −11 ⎛
η⎞ ⎤⎞
= 1 − exp ⎜ −2 3 ⎢ζ − tan ⎜⎝ M ⎟⎠ + ta np − 1⎥ ⎟
1
(16)
r ⎝ 3G υM ⎣ ⎦⎠
114
apparatus, a sample of undisturbed saturated clay After the PVD was installed, a consolidation
was deemed to be impractical. Therefore, a com- pressure was applied to the soil deposit. Once con-
mercially available reconstituted clay was used for solidation reached 90%, 54 samples were cored
the tests. The reconstituted soft clay was assessed to from various locations along three different direc-
have a moisture content of 60% (1.1 times its liquid tions, to determine the extent of the smear zone.
limit) and a plasticity index of 28%. The PVD and More details of the large-scale laboratory tests are
mandrel used in the tests were 100 mm × 4 mm and presented in Ghandeharioon (2012).
125 mm × 28 mm in cross section, respectively.
115
for tests conducted under surcharge pressure of
20 kPa, 32.5 kPa and 50 kPa, respectively.
The changes in excess pore water pressure show
that the maximum pore pressure at any given level
develops owing to the compression ahead of the
mandrel during the installation process just before
the base of the mandrel passes the observation
level. When the base of the mandrel passes the
observation level, the vertical compression changes
because of a shearing force developed at the inter-
face of the mandrel and soil, and consequently
the pore pressure drops off until the PVD is fully
installed. Once the mandrel is started to be with-
drawn the pore pressure drops due to the sudden
decrease in mean stress. Figure 7. Variations of the moisture content of
As soon as the base of the mandrel passes from soil measured in the laboratory along the 0°, 45° and
the observation level during withdrawal, the pore 90° axes in relation to the installed PVD (surcharge
pressure experiences a sudden decrease attrib- loading = 50 kPa, consolidation pressure = 80 kPa).
utable to the vacuum created by passing of the
mandrel. The residual excess pore pressure that
exists once the mandrel has been completely with- As discussed by Ghandeharioon (2010), install-
drawn will drain over time as a result of radial and ing mandrel-driven PVDs affects both the hori-
vertical consolidation. These observations were zontal coefficient of permeability and the moisture
also used to validate the numerical simulation in content of the soil adjacent to the drain. Variations
Ghandeharioon et al. (2012). These variations of of moisture content are investigated in this research
pore water pressure reveal that when the vertical to determine the extent of smear zone, and also to
drain was fully installed the excess pore pressure evaluate the effect of in-situ effective stresses on
that built up decreased as the radial distance to the the size of the smear zone. Figure 7 illustrates the
PVD increased, but when the mandrel was com- results of laboratory measurements of the mois-
pletely withdrawn the residual pore pressure was ture content of 54 samples cored from various
larger further away from the PVD. locations along three different directions in the test
Distribution patterns for excess pore pressure corresponding to a surcharge pressure of 50 kPa,
with the radial distance from the center of the after consolidation was 90% complete. r, rm(θ) and
cavity predicted with the developed elliptical CET rsmear are the radial position measured from the
and conventional cylindrical CET, and measured center of the cell, equiv alent radius of the mandrel
in the laboratory are compared in Figure 6. It is and radius of the smear zone, respectively.
evident that the developed elliptical CET evaluates Based on the variation of the measured moisture
the pore pressure during mandrel installation more contents and by implementing the concept devel-
accurately than the conventional cylindrical CET. oped by Sathananthan and Indraratna (2006), it is
found that the radius of the smear zone normalized
by the equivalent radius of the mandrel is 2.55 in
that test. The normalized radius of the smear zone
is 2.65 and 2.85 for tests where the surcharge load-
ing applied to the surface of soil during installa-
tion was 32.5 kPa and 20 kPa, respectively. These
results verify the concept of an elliptical smear
zone identified by Ghandeharioon et al. (2010) for
the cases where drains were driven by rectangular
mandrels. Furthermore, these results indicate that
for a given type of soft soil subjected to a particular
rate of mandrel installation, the size of smear zone
decreases as the in-situ effective stresses increase.
Figure 6. Distribution patterns predicted for excess
pore pressure with radial distance using the elliptical CET
and cylindrical CET along the major axis of the mandrel 5 CONCLUSIONS
0.26 m below the soil surface, and measured when base of
the drain shoe passed that horizontal plane, with a sur- The disturbance and pore water pressure in soft sat-
charge loading = 50 kPa. urated clays were analyzed during the installation
116
of mandrel-driven prefabricated vertical drains. Cao, L.F., Teh, C.I., & Chang, M.F. (2001). Undrained
A new elliptical cavity expansion theory for an cavity expansion in modified Cam clay I: Theoretical
undrained condition was developed by incorpo- analysis, Géotechnique, 51(4): 232–334.
rating the modified Cam clay parameters to study Ghandeharioon, A. 2010. Analytical and numerical study
of soil disturbance associated with the installation of
the installation of PVDs in soft clay deposits. As mandrel-driven prefabricated vertical drains, Ph.D.
a mandrel is driven into the soil, the internal pres- Thesis, School of Civil, Mining & Environmen-
sure in the cavity that is expanding, and the cor- tal Engineering, Faculty of Engineering, Univ. of
responding stresses and excess pore pressure in the Wollongong, Wollongong, Australia.
soft soil were predicted. The plastic shear strain Ghandeharioon, A. 2012. Large-scale laboratory assess-
normalized by the rigidity index of the soil was ment of smear effects in soft soils stabilized by prefab-
used to characterize the disturbed soil surround- ricated vertical drains. Proceedings of 65th Canadian
ing the mandrel-driven PVDs. Based on this con- Geotechnical Conference: GeoManitoba—Building on
cept, the boundaries of critical zone, smear zone, the past, Canadian Geotechnical Society, Canada.
Ghandeharioon, A., Indraratna, B., & Rujikiatkamjorn, C.
marginally disturbed zone, plastic zone, and elastic 2010. Analysis of soil disturbance associated with man-
zone were identified. drel-driven prefabricated vertical drains using an ellip-
In this research the installation of PVDs was tical cavity expansion theory, International Journal of
studied in the laboratory under the field conditions Geomechanics, 10(2): 53–64.
using a large-scale consolidometer and a mandrel- Ghandeharioon, A., Indraratna, B., & Rujikiatkamjorn, C.
driving machine capable of working at realistic 2012. Laboratory and finite element investigation of
rates. Variations of pore water pressure during the soil disturbance associated with the installation of
installation of a PVD and withdrawal of the man- mandrel-driven prefabricated vertical drains, Journal
drel were monitored by pore pressure transducers. of Geotechnical and Geoenvironmental Engineering,
138(3): 295–308.
Subsequently, the pattern of distribution calculated Gibson, R.E., & Anderson, W.F. 1961. In-situ measure-
for excess pore pressure by the developed theory ment of soil properties with the pressuremeter. Civil
was verified by the measurements performed in the Engineering Public Works Reviews, 56, 615–618.
laboratory. Hird, C.C., & Moseley, V.J. 2000. Model study of seepage
Using the results of moisture content tests on in smear zones around vertical drains in layered soil,
samples, which in relation to the installed PVD Géotechnique, 50(1): 89–97.
were cored from various locations and directions, Indraratna, B., & Redana, I.W. 1998. Laboratory deter-
the extent of smear zone was measured in the mination of smear zone due to vertical drain instal-
large-scale consolidometer in each test performed lation. Journal of Geotechnical and Geoenvironmental
Engineering, 124(2): 180–184.
under a different surcharge loading. The results Sathananthan, I., & Indraratna, B. 2006. Laboratory
verified the concept of an elliptical smear zone evaluation of smear zone and correlation between
around drains that was identified in previous stud- permeability and moisture content, Journal of Geo-
ies for cases where PVDs were driven by common technical and Geoenvironmental Engineering, 132(7):
field mandrels. These results also showed that for 942–945.
a given type of soft soil subjected to a particu- Sharma, J.S., & Xiao, D. 2000. Characterisation of
lar rate of installation, the extent of smear zone a smear zone around vertical drains by large-scale
around the mandrel decreases as the in-situ effec- laboratory tests, Canadian Geotechnical Journal, 37:
tive stresses increase. 1265–1271.
Vesic, A.S. 1972. Expansion of cavities in infinite soil
mass. Journal of the Soil Mechanics and Foundations
Division, ASCE, 98, 265–290.
ACKNOWLEDGEMENTS Yu, H.S., & Houlsby, G.T. 1991. Finite cavity expansion
in dilatant soil: Loading analysis. Geotechnique, 41,
This research was supported by a scholarship from 173–183.
the Cooperative Research Center (CRC Australia) Yu, H.S., & Mitchell, J.K. 1996. Analysis of cone resist-
for Railway Innovation awarded to the author for ance: A review of methods. The University of Newcas-
his Ph.D. studies at University of Wollongong, tle, Australia, Report No. 142.09.1996.
Australia. Their support is acknowledged.
REFERENCES
117
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
ABSTRACT: In many geotechnical applications like pile installation processes or soil improvement
large deformations of the surrounding soil occur. These large deformations in combination with the com-
plex material behaviour of the soil lead to numerical boundary value problems which are often difficult to
solve. These difficulties are related to large mesh distortions and numerical problems due to complex con-
tact conditions. One possibility to overcome these difficulties is to use special numerical techniques which
are especially invented for large deformation simulations. In this work, the Coupled Eulerian-Lagrangian
method (CEL) is identified to be well suited for such boundary value problems involving large distortions
of the surrounding soil. Therefore, the numerical technique is presented and its suitability is shown in dif-
ferent geotechnical applications. The screwed pile installation and the deep vibration compaction process
are investigated using the CEL method.
118
which remains stationary enables the motion of
the material to be traced. Materials can move
freely through an Eulerian mesh, which remains
undeformed.
mechanical problems involving large deformations In this equation T is the objective Jaumann stress
has been approved e.g. by Qiu et al. (2009, 2010), rate, D the strain rate and M a fourth order tensor,
Qiu & Grabe (2011), Henke et al. (2010), Bienen which depends on the actual Cauchy stress T, the
et al. (2011), Pucker & Grabe (2012), Pichler et al. void ratio e and the intergranular strain δ. One
(2012). advantage of the constitutive model is the consid-
eration of the void ratio. Thus, compaction proc-
esses can be described accurately.
2.3 Contact formulation
Contact between Eulerian and Lagrangian struc-
tures is enforced using a general contact formu- 3 SIMULATION OF SCREWED PILE
lation that is based on a penalty method. The INSTALLATION
algorithm does not enforce contact between the
Lagrangian elements and the Eulerian elements. The installation process of piles influences the
The Lagrangian elements can move through surrounding soil depending on the installation
the Eulerian mesh without resistance until they method. The piles can be driven, jacked, vibrated
encounter an Eulerian element filled with mate- or drilled into the ground. The effects inside the
rial (EVF ≠ 0), see Figure 2. The penalty contact soil are still not completely investigated. There-
method is less strict compared to the kinematic con- fore, most methods to predict the pile behaviour
tact method used in the Lagrangian approach. It or its bearing capacity are based on empirical
approximates hard pressure-overclosure behavior. approaches.
This method allows small penetration of the The influences of the installation process on
Eulerian material into the Lagrangian domain. the surrounding soil have been investigated for
119
driven, jacked and vibrated piles by Mahutka has a diameter about 51 cm and the drill rod about
(2007), Henke (2008), Henke & Grabe (2009) in 32 cm. The diameter of the auger elements relate
numerical studies. In this section, the numerical to the displacement body.
simulation of the installation process of drilled The soil is modeled as an Eulerian area. The
piles is presented. Using this simulation, new shape is cylindrical with a diameter about 16 m
insight can be gained in understanding effects and a height of 20 m. Thereby, the distance
in the soil nearby installed piles. The installation between the boundaries and the drilling tool is
of a full displacement pile with lost bit (FDP) is more than 15 D (15 times of the pile diameter
simulated. Full displacement piles push the soil D), so that no influences of the boundary con-
sideways without excavating soil. The installation ditions on the results are expected. Above the
process is divided into six steps schematically pre- soil, a 2 m void area is provided, so that the soil
sented in Figure 3. can avoid into this free space during the drilling
In the first step the drilling tool is taken into process. The whole Eulerian area is discretized
position. During the second and the third step, the with approx. 163,000 eight-noded elements with
drilling tool is drilled to the purposed depth. Dur- reduced integration. The discretized model is
ing the drilling process, the soil is transported to shown in Figure 5.
the displacement body where the soil is compacted. The drilling tool penetrates the soil velocity
The reinforcement basket is installed in step four. controlled. Assuming, that the response of the soil
The drilling tool is removed in step five and the pile model does not depend on the velocity, the pene-
is concreted simultaneously. The concrete hardens tration velocity is taken to 1 m/s. The ratio between
in step six and the installation is completed. The the penetration velocity and the rotation velocity
presented numerical study simulates only the steps should be approx. 10. Therefore the rotational
two and three. A detail of the drilling tool is illus- velocity is chosen to 10 U/s. The reaction forces
trated in Figure 4. and moments are recorded during the simulation.
Figure 3. Schematic illustration of the installation process of full displacement piles (Busch et al., 2010).
120
3.2 Results
3.2.1 Changes of the stress state
The stress state around the drilling tool is signifi-
cantly influenced, see Figure 6. The transport of
the soil to the displacement body causes a compac-
tion of the soil at the height of the displacement
body. After the displacement body passes through,
a gap between the soil and the drill rod occurs
and the soil can relax. Therefore, the horizontal
stresses, illustrated in Figure 6, decrease nearby the
drilling tool. The massive compaction of the soil
in combination with the changing cross-section of
the drilling tool lead to a reduction of the horizon-
Figure 4. Scheme of the FDP drilling tool invented by tal stresses nearby the drilling tool.
BAUER (Busch et al., 2010) (left); numerical discretiza-
tion of the drilling tool (right). 3.2.2 Changes of the soil’s bulk density
The hypoplastic constitutive model is able to predict
changes of the soil’s bulk density, see section 2.4.
These changes caused by the drilling process are
shown in Figure 7. The influence of two different
velocity ratios between the penetration velocity vz
and the rotation velocity vr are investigated. Nearby
the drilling tool, the soil is loosened up, caused by the
changing cross-section of the drilling tool. Between a
distance of 1 D to 3 D, the soil is strongly densified.
3.3 Conclusion
The CEL-method is able to reproduce the soil’s
behaviour during a drilling process. The changes
of the stress state and of the bulk density can be
investigated. Therefore, this method can be used
to gain deeper insights into the soil behaviour and
the influencing installation parameters, such as
Figure 5. Discretized model with void area and mesh the penetration and rotation velocity as well as the
of the soil in detail. geometric properties of the drilling tool.
121
Figure 6. Horizontal stresses in the soil before (left) and after the drilling process (right).
Figure 7. Bulk density of the soil along horizontal paths at depths of 1 m, 3 m and 4.5 m after completion of the 6 m
drilling process for different drilling ratios.
The deep vibration compaction process is sche- the self-weight of the vibrator and vibration of the
matically depicted in Figure 8. First, the vibrator compactor. It is possible to reach depths of up to
is installed up to the wished penetration depth 60 m with this technique. Afterwards, the vibrator
by the help of a drilling fluid as it can be seen in is pulled stepwise in combination with compaction
Figure 8. The penetration is achieved only due to of the surrounding soil. The steps are between 0.3
122
Figure 8. Schematic description of the deep vibration compaction process after “Keller Grundbau”.
and 1.0 m and the vibration time at one penetra- allow material transport into this region during
tion depth lies between 30 and 90 s, see Witt (2009). the analyses. The vibrator itself has a length of
The vibration energy leads to shear waves in the 2.5 m and a circular cross-section with a diameter
soil such that the grain structure is rearranged. of 30 cm. At the beginning of the deep vibration
Therefore, cyclic shearing acts on the soil such compaction simulation the vibrator is modeled
that the void ratio is reduced and the soil is com- wished-in-place at a penetration depth of 10 m.
pacted. Due to this compaction soil is transferred The vibrator is vertically hinged to a tube with the
into the compacted region such that settlements at same cross-section. The vibrator and the stay tube
the ground surface become evident. To cope with are modeled linear elastic with the parameters of
these settlements the funnel is refilled with material steel.
(Fellin, 2000). To compact a large area the com- In all simulations the soil continuum is discretized
paction process is redone in discrete distances such using eight-noded-Eulerian elements whereas the
that a narrow grid can be improved. The vibra- vibrator components are modeled with Lagrang-
tion frequencies of typical vibrators lie between 30 ian elements (eight-noded-hexaeder elements with
and 60 Hz such that the maximum deflection of reduced integration). Contact between vibrator
the vibrator lies between 3 and 21 mm at the toe. and soil is considered using Coulomb’s friction
This comes along with horizontal forces between law. As a simplification the contact between soil
150 and 700 kN (Witt, 2009). Fellin (2000) points and vibrator is considered to be frictionless.
out that the soil around the vibrator is compacted All simulations are done using the hypoplastic
up to a dense packing in distances between 0.6 m constitutive model with soil parameters for loose
and 1.75 m. In greater distances it is not possible to Mai-Liao sand.
receive a dense packed sand. As loading history first a K0-stress state is
applied on the soil continuum. Afterwards, the
simulation of the deep-vibration compaction proc-
4.2 Numerical modelling
ess is carried out for a total duration of 3 s. The
A three-dimensional coupled Eulerian-Lagrangian vibrator’s movement is discretized both displace-
model is used to simulate the deep vibration com- ment and force-controlled. In the displacement
paction process, see Figure 9. The soil continuum controlled simulations the toe of the vibrator
has a diameter of 30 m and a height of 27 m. At oscillates at a certain frequency with a predefined
the beginning of the simulation the first two meters radius around the centre of gravity of the vibra-
of the soil continuum are modeled material-free to tor whereas in the force-controlled simulations the
123
Figure 9. Three-dimensional numerical model to simulate the deep-vibration compaction process.
force acting at the toe of the vibrator is predefined Table 2. Parameter combinations investigated regard-
and the frequency is considered to be f = 30 Hz. ing the deep vibration compaction process in sand.
The top point of the vibrator is connected to the
stay tube in all simulations. Vibration Oscillation Vibration
frequency f amplitude uH force Fdyn
All varied parameters in this numerical study Name (Hz) (mm) (kN)
are summarized in Table 2.
f30u4 30 4 –
f30u8 30 8 –
4.3 Results f30u12 30 12 –
During the displacement-controlled simulations f30u16 30 16 –
the movement of the vibrator is forced along a f30u20 30 20 –
circular path around the vibrator’s centre of grav- f40u12 40 12 –
ity. In contrast during the force-controlled simu- f50u12 50 12 –
lation the movement of the vibrator depends on f60u12 60 12 –
f30F150 30 – 150
the acting forces and the soil stiffness such that the
f30F300 30 – 300
deflection does not follow an ideal circular path. f30F500 30 – 500
In Figure 10 the deflection of the vibrator’s toe is f30F700 30 – 700
depicted as a plot in the x-y-plane. This line repre-
sents 3 s of vibration.
At the beginning of the deep vibration com-
paction process the soil is in a loose state. The the soil the stiffness increases and the amplitude of
surrounding soil is very compactable and has a the vibrator therefore is reduced. The drift of the
relatively low stiffness such that after start of the vibrator shown in Figure 10 results out of the local
vibration process large deformation amplitudes differences in soil stiffness due to the compaction
can be reached. With increasing densification of of the sand.
124
Figure 10. Deformation of the vibrator in x-y-plane Figure 11. Dependency of maximum oscillation ampli-
during the deep vibration compaction process, force- tude and vibration force—Comparison between numeri-
controlled simulation, Fdyn = 150 kN, f = 30 Hz. cal simulation and values found in literature.
Figure 12. Normalized force and horizontal displacement in time domain during the deep vibration compaction
process, force-controlled simulation, Fdyn = 150 kN, f = 30 Hz.
Furthermore, a phase shift between the direc- frequency the area of influence where a dense soil
tion of the vibration force and the direction of state is reached increases significantly. In contrast,
vibrator movement can be seen, see Figure 11. the oscillation amplitude, at least in the investi-
The vibration force is shifted with an angle ψ to gated range does not significantly influence the
the normalized horizontal deformation. The angle ψ void ratio distribution around the vibrator.
depends on the soil damping and the eigenfrequency To judge the quality of the numerical simula-
of the system. The higher the damping the closer this tions the calculated deflections of the vibrator in
angle is to 90° (Fellin, 2000). In the present case for a dependency of the vibration forces are compared
vibration force of Fdyn = 150 kN and a frequency of to results found in literature (Fellin, 2000), see
f = 30 Hz the phase shift is ψ = 113°. Figure 13.
Furthermore, the influence of the main parame- Regarding this graph, it can be seen that the
ters vibration frequency and oscillation amplitude numerically received results fit well into the data
on the compaction process is of main interest. In given by various companies and literature. Only
Figure 12 the results of displacement controlled the simulation with a vibration force of 700 kN
simulations with varying frequency and oscillation seems to overestimate the oscillation amplitude.
amplitude respectively are shown. This is caused due to the high force compared to
It is evident that mainly the frequency has an the comparatively low soil stiffness which leads to
impact on the void ratio distribution around non-uniform movement of the vibrator which was
the vibrator. It can be seen that with decreasing not judged to be realistic.
125
Figure 13. Top: Influence of the vibration frequency on the compaction success, Bottom: Influence of the oscillation
amplitude on the void ratio around the vibrator—displacement-controlled simulations; results along a horizontal path
at the height of the centre.
A more detailed discussion regarding the numer- dry granular soil material using a hypoplastic con-
ical simulation of deep vibration compaction proc- stitutive model.
esses using the Coupled Eulerian-Lagrangian In case of screwed piles the CEL method is used
method can be found in Henke et al. (2011) and to simulate the drilling process. The CEL-method
Henke et al. (2012). is able to reproduce the soil’s behaviour during this
process. The changes of the stress state and of the
bulk density can be investigated. Therefore, this
5 CONCLUSION method can be used to gain deeper insights into
the soil behaviour and the influencing installation
The simulation of large deformations of soil often parameters, such as the penetration and rotation
cause numerical boundary value problems in many velocity as well as the geometric properties of the
geotechnical applications like pile installation drilling tool.
processes or soil improvement. These large defor- Another geotechnical application to simulate
mations in combination with the complex material with the CEL method is the deep vibration com-
behaviour of the soil are often difficult to solve paction. It can be shown that the process of soil
numerically. The Coupled Eulerian–Lagrangian compaction can be simulated realistically. Effects
method (CEL) is well suited to model complex such as a phase-shift between the direction of
geotechnical installation processes, like screwed the vibrator force and the direction of motion
pile installation and deep vibration compaction, in of the vibrator and the size of the area of influ-
126
ence around the vibrator as observed in reality can Henke S., Hamann T. & Grabe J. 2011. Coupled
be simulated. Furthermore the numerical model Eulerian-Lagrangian Simulation of the Deep
is used to investigate the influence of the main Vibration Compaction Process as a Plastodynamic
parameters vibration frequency and oscillation Problem. Proc. of EURODYN 2011, Leuven.
Henke, S., Qiu G. & Grabe J. 2010. A coupled eulerian-
amplitude on the compaction process. lagrangian approach to solve geotechnical problems
At the moment research is done to enhance the involving large deformations. Proc. of 7th European
existing constitutive model to 2 or 3-phases such Conference on Numerical Methods in Geotechni-
that also the effect of pore pressure in dynamic cal Engineering (NUMGE) in Trondheim/Norway:
analyses can be taken into account. First results 233–238.
of this research developing an explicit formula- Herle, I. 1997. Hypoplastizität und Granulometrie
tion including pore water pressure are presented in einfacher Korngerüste. Heft 142 Institut für Bod-
Pichler et al. (2012). enmechanik und Felsmechanik der Universität
Fridericana in Karlsruhe.
Mahutka, K.-P. 2007. Zur Verdichtung von rolligen
Böden infolge dynamischer Pfahleinbringung und
REFERENCES durch Oberflächenrüttler, Dissertation, Veröffentlic-
hungen des Instituts für Geotechnik und Baubetrieb
Bienen B., Henke S. & Pucker T. 2011. Numerical study of der TU Hamburg-Harburg, Heft 15.
the bearing behaviour of circular footings penetrating Pichler, T., Pucker T., Hamann T., Henke S. & Qiu G.
into sand. Proc. of 13th International Conference of 2012. High-performance abaqus simulations in soil
International Association for Computer Methods and mechanics reloaded—chances and frontiers. Proc.
Advances in Geomechanics (IACMAG): 939–944. of International Simulia Community Conference in
Busch, P., Grabe, J., Gerressen, F.W. & Ulrich, G. 2010. Providense, Rhode Island/USA: 237–266.
Use of displacement piles for reinforcement of exist- Pucker, T. & Grabe J. 2012. Numerical simulation of the
ing piles, Proceedings of DFI and EFFC 11th Int. installation process of full displacement piles. Com-
Conf. in the DFI series, Geotechnical Challenges in puters and Geotechnics, 45: 93–106, DOI: 10.1016/j.
Urban Regeneration in London/UK: 113–119. compgeo.2012.05.006.
Dassault Systèmes 2010. Abaqus User Manual Version Qiu, G. & Grabe J. 2011. Explicit modeling of cone and
6.10. strip footing penetration under drained and undrained
Fellin W. 2000. Rütteldruckverdichtung als plastody- conditions using a visco-hypoplastic model. Geotech-
namisches Problem. Dissertation, Institute of Geo- nik 34(3): 205–271.
technics and Tunneling, University of Innsbruck. Qiu, G., Henke S. & Grabe J. 2009. Applications of
Advances in Geotechnical Engineering and Tunneling, coupled eulerian lagrangian method to geotechnical
Heft 2, 2000. problems with large deformations. Proc of SIMULIA
Henke, S. 2008. Herstellungseinflüsse aus Pfahlrammung Customer Conference 2009 in London: 420–435.
im Kaimauerbau, Dissertation, Veröffentlichungen Qiu, G., S. Henke, & J. Grabe 2010. Application of a cou-
des Instituts für Geotechnik und Baubetrieb der TU pled eulerian-lagrangian approch on geomechanical
Hamburg-Harburg, Heft 18. problems involving large deformation. Computers and
Henke, S., & Grabe, J. 2009. Numerical modeling of pile Geotechnics, DOI:10.1016/j.compgeo.2010.09.002.
installation, Proc. of 17th Int. Conf. on Soil Mechanics vonWolffersdorff, P.-A. 1996. A hypoplastic relation for
and Foundation Engineering (ICSMFE): 1321–1324. granular materials with a predefined limit state sur-
Henke S., Hamann T. & Grabe J. 2012. Numerische face. Mechanics of Frictional and Cohesive Materials,
Untersuchungen zur Bodenverdichtung mittels 1: 251–271.
Rütteldruckverfahren. 2. Symposium “Baugrund- Witt K.J. 2009. Grundbau-Taschenbuch, Teil 2: Geotech-
verbesserung in der Geotechnik”, TU Wien, zur nische Verfahren. Berlin, Ernst & Sohn Verlag.
Veröffentlichung akzeptiert.
127
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
ABSTRACT: There is a need for an improved understanding of rate effects over a wide range of strain
rates in order to improve the modelling and analysis of installation effects. Using triaxial testing on recon-
stituted kaolin over a wide strain rate range this paper examines the impact of strain rate on the aspects
of soil response which are important in the analysis of installation effects. It is demonstrated that shear
strength, small strain stiffness and the elastic shear strain threshold are rate dependent. The implications
of this for a common stiffness degradation model are analysed and an improved model proposed.
128
of test methods from cone penetrometer testing Table 1. Properties of the speswhite kaolin used.
to large scale shear boxes (Steenfelt 1993, Lehane
et al 2009). Property Value
As previously mentioned, small strain behaviour
Plastic limit, wP (%) 32.5
is important to FEA. Shibuya et al (1996) showed Liquid limit, wL (%) 65.0
that the initial small strain stiffness, Emax, is rate Plasticity index, PI (%) 32.5
independent and constant for the range of strain Clay fraction (%) 80
rates investigated (0.6 to 84%/hr). It was also found Activity (%) 40.6
that this linearity continued up to a strain defined Specific surface area (m2/g)* 36.7
as the elastic strain threshold, εEL, which was itself Permeability (mm/s)# 1.17 × 10−6
rate dependent, increasing with increasing strain cv (m2/year)+ 23.52
rate. These findings were confirmed by Mukabi & MCSL 0.9
Tatsuoka (1999). λ 0.101
Lo Presti et al (1996) found that soil stiffness is N 2.678
rate independent below 0.001% shear strain, and
increases with strain rate beyond this point. To *Determined from methylene blue spot testing;
allow comparison of relative stiffnesses at differ- #
Determined at an effective stress of 300 kPa; +Determined
ent strains and strain rates, Lo Presti et al (1996) for a 100 kPa stress increment.
defined the strain rate coefficient, α.
There are numerous geotechnical applications
where rate effects are an important consideration, the additional benefit of significantly reducing
including free falling penetrometer tests where consolidation times.
velocities can reach up to 12 m/s and Statnamic pile
tests where the pile is displaced at rates of the order 2.1 Testing apparatus
of 1 m/s (Chow & Airey 2011). In these cases, rate
The tests were carried out in a GDS advanced elec-
effects cause the capacities derived from the tests to
tromechanical dynamic triaxial rig specially modi-
exceed static values; an issue which is usually dealt
fied to carry out high speed monotonic tests. The rig
with by the inclusion of damping co-efficients in
is capable of axial displacement rates of 100 mm/s,
the analyses (Brown 2009). These damping co-
and during high speed testing it is controlled by
efficients are not always uniform throughout each
a GDS digital control system capable of control-
test, but may vary with strain level. This highlights
ling the axial displacement within a time interval
the importance of understanding the strain level
of 0.1 milliseconds. Both the back and cell pres-
dependence of rate effects in order to improve the
sures were provided by GDS pressure controllers.
accuracy of dynamic testing and installation effects
Lubricated end platens of a similar design to those
in the field, as well as for use in FEA.
proposed by Rowe & Barden (1964) were used in
order to minimise the inhomogeneity caused by
end restraint conditions. As these can introduce
2 MATERIAL AND TESTING PROCEDURES
errors into the measurement of small strains using
external methods, these were measured using
The tests were carried out on reconstituted
Hall effect transducers, two axial and one radial,
speswhite kaolin, the properties of which are
mounted directly on the sample providing a resolu-
shown in Table 1.
tion of 1 × 10−6% strain. Pore pressures were moni-
The samples were first prepared as slurry with
tored using a mid-height pore pressure transducer
a moisture content of 120% using de-aired, de-
mounted on the surface of the sample.
ionised water before being one dimensionally con-
solidated to an effective stress of 180 kPa for three
2.2 Testing programme
days. These were then trimmed to 200 mm length
and 100 mm diameter to create triaxial samples. The testing programme consisted of triaxial tests
Once installed in the triaxial apparatus, the sample at shear strain rates from 0.333 to 60,000%/hr in
was saturated to an effective stress of 50 kPa at a order to investigate strain rate effects over as large a
back pressure of 300 kPa and then re-consolidated range as possible. These were carried out at a com-
to an effective stress of 300 kPa to restore isotropic paratively low effective stress of 300 kPa as previ-
conditions. Sample drainage was facilitated by ous studies have shown that greater rate effects are
using vertical filter paper drains on the surface of observed at higher moisture contents (Bea 1982,
the sample, connected to both the top and bottom Chow & Airey 2011). Throughout the testing pro-
drainage valves. These were required as the use of gramme, the samples were allowed to drain through
lubricated end platens in the testing meant that the filter paper drains in order to allow rate effects
conventional drainage was not possible, and had due to consolidation to be investigated.
129
3 RESULTS AND DISCUSSION velocity associated with chosen reference rate and
V0 is the point after which time for consolida-
3.1 Observed rate effects and their modelling tion effects are negligible. This process was also
repeated for the measured rate effects at selected
The rate effects observed at the various strain rates
strains to identify the influence of strain level on
are shown in Figure 2, with a shear strain rate of
rate effects as shown in Figure 3. The parameters
100%/hr taken as the reference rate. In order to
obtained from the curve fitting process are shown
allow comparison with other studies using dif-
in Table 2.
fering materials and sample sizes, the strain rates
The fitting parameters at peak strength (which
have been converted in the normalised dimension-
occurs at varying strain levels) show that the rate
less velocities used by Randolph & Hope (2004) as
effect per log cycle was found to be 22.5% which
shown in Equation 1.
is higher than previous studies. This highlights the
need to develop a framework to predict rate effects
vd
V= (1) based on the current state and properties of the
cv material in question.
130
Table 2. Parameters used to fit the model by Randolph & Hope (2004) to the data at each
shear strain level.
ΔG (γ )
α (γ ) = (3)
)) G (γ ,γ refe )
Δ (log(γ ))
131
(Figs. 8 and 9). This was done for the slowest
(0.333%/hr) and fastest (60000%/hr) tests.
This showed that at the low strain rate, the cur-
rent HS small model matched the measured shear
moduli with relatively good accuracy, with a devia-
tion of less than 12%. However, for the high strain
rate test, the increase in εEL and the more rapid
subsequent degradation meant that this variation
increased to up to 71%. Clearly, this margin of
error is significant.
It is proposed that this can be better modelled
using the form shown in Equation 6.
G 1
= (5)
G0 ⎛ γ ⎞
1 + α HHS ⎜
⎝ γ 0.0 7 ⎟⎠
132
G 1 γ 0.7,dyn = γ 0.7 × 1.5 γ 0.2 (7)
= β
(6)
G0 ⎛ γ ⎞
1 + α HHS ⎜ β = 0.043 log( γ ) + 1.13 (8)
⎟
⎝ γ 0.7,dyn ⎠
where γ is the shear strain rate in%/hr.
where αHS = 0.384 as in the existing model. γ0.7,dyn
is the shear strain at which G/G0 = 0.7 in the fast 4 CONCLUSIONS
(dynamic) test, which has the effect of moving the
G/G0 curve to the right to intercept the measured Based on triaxial tests on reconstituted kaolin at
data. Similarly, the new power, β, allows the steep- strain rates over more than 5 orders of magnitude,
ness of the rate of degradation after the elastic it has been shown that rate effects have numerous
shear strain threshold to be increased. The pro- effects on the response of the soil, ranging from
posed model is shown alongside the current HS increased deviatoric strength to changes in small
small model in Figure 8 and Figure 9. As can be strain behaviour. These effects have also been
seen it closely follows the current model at slow shown to be strain level dependent.
strain rates, while better approximating the data at
faster rates, reducing the variance from the data to a. Peak strength rate effects in the partially drained
11%. For the soil tested, γ0.7,dyn and β were found domain due to time for consolidation have
to vary according to the following relationships been shown to be greater than undrained rate
(Equation 7 and Equation 8) such that the pro- effects.
posed model resolves into the existing model at low b. The undrained rate effects observed have been
strain rates. (Figs. 10 and 11). shown to be higher than those reported by
other studies at 22.5% per log cycle for peak
strength.
c. The importance of strain level in relation to rate
effects has been highlighted, with rate effects
increasing with strain when partially drained
and reducing with strain when undrained.
d. The applied strain rate was found to have a sig-
nificant impact on the small strain response,
with the elastic shear strain threshold increasing
with rate and G0 appearing to reduce with rate.
e. The need for a new model to accurately rep-
resent shear modulus degradation was high-
lighted, and a potential form for achieving this
described.
It is hoped that the results and discussion in
this paper will be of use in the FEA of installation
effects and in their analysis more generally.
Figure 10. Graph of γ0.7, dyn/γ0.7 against shear strain rate.
ACKNOWLEDGEMENTS
REFERENCES
133
Brown, M.J. & Powell, J.J.M. 2013. Comparison of rapid Plaxis bv. 2011. Material models manual 2011. Delft:
load test analysis techniques in clay soils. ASCE Jour- Plaxis.
nal of Geotechnical & Geoenvironmental Engineering. Quinn, T.A.C. & Brown, M.J. 2011. Effect of strain rate
Available online 21/03/12. on isotropically consolidated kaolin over a wide range
Chow, S.H. & Airey, D.W. 2011. Rate effects in free fall- of strain rates in the triaxial apparatus. Proc. Int.
ing penetrometer tests. Proc. Int. Symp. on Defor- Symp. on Deformation Characteristics of Geomateri-
mation Characteristics of Geomaterials, Seoul, 1–3 als, Seoul, 1–3 September 2011.
September 2011. Randolph, M.F. & Hope, S. 2004. Effect of cone velocity
Krieg, S. & Goldscheider, M. 1998. Bodenviskotät und on cone resistance and excess pore pressures. Proc. Int.
ihr Einfluβ auf das Tragverhalten von Pfählen. Bau- Symp. Eng. Practice and Performance of Soft Deposits:
technik 75: 806–820. Ernst und Sohn. 147–152.
Kulhawy, F.H. & Mayne, P.W. 1990. Manual on estimating Rowe, P.W. & Barden, L. 1964. Importance of free ends
soil properties for foundation design. Report EL-6800, in triaxial testing. ASCE journal of the soil mechanics
Electric Power Research Institute, Pala Alto. and foundations division 90 (SM1): 1–27.
Lehane, B.N., O’Loughlin, C.D., Gaudin, C. & Shibuya, S., Mitachi, T., Hosomi, A. & Hwang, S.C.
Randolph, M.F. 2009. Rate effect on penetrometer 1996. Strain rate effects on stress-strain behavior as
resistance in kaolin. Geotechnique 59: 41–52. observed in monotonic and cyclic triaxial tests. ASCE
Lo Presti, D.C.F., Jamiolkowski, M., Pallara, O. & GSP: Measuring and modelling time dependent soil
Cavallaro, A. 1996. Rate and creep effect on the stiff- behavior 61: 214–227.
ness of soils. ASCE GSP 61: 166–180. Steenfelt, J.S. 1993. Sliding resistance for foundations on
Mukabi, J.N. & Tatsuoka, F. 1999. Influence of reconsol- clay till. Predictive soil mechanics: 664–684. London:
idation stress history and strain rate on the behavior Thomas Telford.
of kaolin over a wide range of strain. Geotechnics for
developing Africa: 365–377. Rotterdam: Balkema.
134
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
ABSTRACT: The penetration resistance of freshly deposited silt is measured. Therefore, a miniature
ball-cone is designed with high accuracy at the applied scale. This paper presents the operation charac-
teristics of the newly designed ball-cone and presents the first results. The first results indicate that subtle
differences in penetration resistance are resolved.
2 EXPERIMENTAL
135
material. The dimensionless number on the x-axis The standard TU Delft 2 axis actuator, which
is a Peclet-like number and is used to express the has originally been designed for the geotechnical
variation form drained to undrained response. centrifuge, has been used to push the ball-cone into
Peclet numbers are used as well for assessing the sediment. This actuator has sufficient linear
drained and undrained erosion of sediment beds stroke for 100 mm penetration and the penetra-
(Winterwerp & Van Kesteren, 2004). tion rate has been varied between 0.05 mm/s and
The dimensions of the ball-cone are chosen 0.3 mm/s.
such that the full flow conditions are reached as
early as possible, in order to capture the resistance
2.3 Calibration procedure
in the first 100 mm of the bed, whilst maintain-
ing high sensitivity for low shear strengths. At the Calibration of the ball cone penetrometer is carried
same time scaling considerations require that the out on a regular basis using a calibration rig com-
minimum diameter of the ball should be large prising of a lever system. Figure 3 shows a schema-
enough to consider continuum conditions around tization of this calibration set up. The ratio of the
the probe. Hence, the ratio of the ball diameter Dball length between the hinge and loading point (L1) and
and the largest mean diameter of a silty sediment the hinge and the ball cone (L2) is 1:5. Care is taken
(D50 ∼ = 60 μm) is chosen to be Dball/D50 ≥ 200. that the ball cone is loaded vertically at all times.
Next a stepwise loading cycle is applied varying
within the range of 0–100 N. At the end of a load-
2.2 Mechanical setup
ing cycle the lever was removed again and a gently
Following the requirements above the final design tap against the shaft is given in order to check for
for the penetrometer used in our experiments remaining friction between the tube and the con-
is equipped with a 12.0 mm diameter ball. As nection rod. This turned out to be negligible.
opposed to the design of Randolph & Hope (2004) Calibration results show a linear relation between
the ball resistance isn’t measured directly above the the output voltage and the applied mass during the
ball but the load is transmitted by a rod to a meas- loading cycle and some small hysteresis during the
uring bridge on top of the probe above the water subsequent unloading cycle (Fig. 4). A linear fit
table. The rod (ø = 3.0 mm) is guided in a tube
(ø = 4.0 mm), which is not connected to the ball, to
prevent pick up of shaft resistance on the rod and
assure that only the ball resistance is measured. The
normal load in the rod is converted in a bending
moment of the sensing element in the measurement
bridge. This design is more sensitive for registering
small loads and not susceptible for temperature
fluctuations. The design is optimized for loads up Figure 3. Schematization of calibration set-up. Ratio
to 150 N with accuracy better than 0.01 N. The L1:L2 = 1:5.
dimensions of the ball-cone are shown in Figure 2.
136
them with the solutions for the more classical
T-bar. The lower and upper bound solution follow
from using either a Tresca (upper bound) or Von
Mises (lower bound) yield criterion. Their solu-
tions lead to a non-dimensional bearing capacity
factor that varies in the range of 11.80–15.54 as the
ball interface condition varies from fully smooth
to fully rough. The corresponding values for a Von
Mises yield criterion are about 8% lower. Corre-
sponding lower bound solutions range from 10.98
for a smooth interface to 15.10 for a rough inter-
face as can be seen in Figure 6 (Randolph et al.,
2000). Here, only the theoretical solutions for pen-
etration resistance in isotropic Tresca soil will be
considered. The interface friction coefficient α is
assumed to be 0.3 (Low et al., 2011), resulting in
Figure 5. Grain size distribution of used silt, a bearing factor of 12.8. Now the net cone resist-
D50 = 18 μm. ance can directly be diverted to the measured shear
strength.
137
Figure 7. Net upward force due to buoyancy measured
during immersing in water.
Figure 9. Results of penetration tests after friction
reduction between the inner rod and outer. All test al
performed on the same sample.
138
Jacobs, W., 2011. Sand-mud erosion from a soil mechani- Randolph, M.F., Martin, C.M. and Hu, Y., 2000.
cal perspective, Ph.D. thesis, Delft University of Limiting resistance of a spherical penetrometer in
Technology. cohesive material. Géotechnique, 50(5): 573–582.
Low, H.E., Randolph, M.F., Lunne, T., Andersen, K.H. Roberts, J., Jepsen, R. and Gotthard, D., 1998. Effects of
and Sjursen, M.A., 2011. Effect of soil characteristics particle size and bulk density on erosion of quartz par-
on relative values of piezocone, T-bar and ball pen- ticles. Journal of Hydraulic Engineering, 124: 1261.
etration resitances. Géotechnique, 61(8): 13. Winterwerp, J. and Van Kesteren, W., 2004. Introduction
Randolph, M.F. and Hope, S., 2004. Effect of cone to the physics of cohesive sediment in the marine envi-
velocity on cone resistance and excess pore pressures, ronment. Elsevier Science Ltd.
Int. Symp. on Engineering Practice and Performance Winterwerp, J.C., van Kesteren, W.G.M., van Prooijen, B.
of Soft Deposits, Osaka, Japan, pp. 147–152. and Jacobs, W., 2012. A conceptual framework for
Randolph, M.F. and House, A.R., 2001. The complemen- shear-flow induced erosion of soft cohesive sediment
tary roles of physical and computational modelling. beds. Journal of Geophysical Research—Oceans.
IJPMG—International Journal of Physical Modelling
in Geotechnics, 1: 01–08.
139
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
ABSTRACT: During the construction of the 148 m long Øvre Sund Bridge on soft sensitive clay in
Drammen, in the south-eastern part of Norway, the stability of the river banks was a challenge. The
installation of displacement piles strongly compresses the adjoining soils and leads to build up of excess
pore water pressure. This temporary build up of excess pore water pressure, coupled with the sensitivity of
the clay soil, causes the soil to lose a good fraction of its shear strength in the short term. Dissipation of
the excess pore pressure generated during pile installations may allow pore pressure to rise in the vicinity
of the river bank slope, thus leading to failure of the slope, even where the soil has not been remoulded.
A strict construction control mechanism during the construction of the Øvre Sund Bridge in Drammen
was planned. The follow up of the development of excess pore pressure during piling activity was one of
the measures taken. Based on stability analyses of the slope along the river banks a criterion for excess
pore pressure margins due to pile installations was established. This paper discusses the result of this strict
pore pressure control for the piling activity. The result shows a systematic monitoring of the development
of excess pore pressure during pile installations in sensitive clay helps to observe the safety margins of
the slopes continuously during pile installation, saves waiting time due to excess pore water pressure and
avoids extra cost of the project due to the build up of excess pore pressure related to pile installations.
140
includes measures taken before piling started as 3.1 Before the pile driving started
part of site preparation, during piling activity and
In order to improve the safety factor of the river
after piling. These measures taken are discussed in
banks slope during pile installation, the river
section 3.
banks top terrain was unloaded by excavating the
top soil. This unloading improved the safety factor
of the riverbanks slope before starting the instal-
2 SITE CONDITION
lation of piles by 10–20%. Figure 2 shows the site
preparation which is mainly unloading the top soil
The Øvre Sund Bridge construction site is located
on the Grønland side of the river bank.
in the city of Drammen in the south-eastern part
of Norway about 50 km from Oslo, across the river
Drammen. The general soil condition on both sides 3.2 During pile driving
of the river and in the river bed is relatively similar. The installation of displacement piles strongly
The site condition consists of mainly three differ- compresses the adjoining soils and leads to build
ent soil layers. The thin top layer of silty sandy soil up of excess pore water pressure. The dissipation
with some gravel is overlying a relatively thick layer of the excess pore pressure generated during pile
of soft sensitive clay. The thick soft sensitive clay installations may allow pore pressure to rise in the
is overlying dense moraine. The depth to the dense vicinity of the river bank slope, thus leading to fail-
moraine layer is dipping in the direction of the ure of the slope, even where the soil has not been
Hamborgstrøm side of the river (Fig. 1). The depth remoulded. The development of excess pore pres-
to the bottom dense moraine layer varies from about sure during pile installation for the construction of
10 m on the upstream side of Grønland to about Øvre Sund Bridge was a major concern for the sta-
25 m on the downstream of the Hamborgstrøm side bility of the river banks. To reduce this build up of
of the river. The clay layer is normally consolidated pore water pressure due to pile installation proper
on both sides of the river and has an undrained selection of pile type, special installation technique
shear strength of 10–15 kN/m2 at the top of the on the riverbanks and strict monitoring of the
layer and increases with depth to 20–30 kN/m2 at build up of pore water pressure were carried out.
the bottom of the layer. The remolded undrained The foundation for axis 5 pillar, in the river, which
shear strength is in the order of 1 kN/m2, the unit is close to the river bank on the Hambergstrøm side
weight is in the order of 19–20 kN/m3 and the sen- of the river (Fig. 1) was designed using 813 mm
sitivity is 6–35. The transition between the soft clay diameter open ended steel pipe piles to reduce the
layer and dense moraine layer consists of stratified effect of soil displacement during pile installation.
sand and gravel. Clogging of these open ended pipe piles during
installation was a main concern which will affect
the main intended purpose to reduce the build up
3 MEASURES TAKEN TO CONTROL of excess pore pressure. The development of pore
THE STABILITY OF THE RIVER BANKS water pressure during pile installation in axis 5 was
also followed by piezometers.
The geotechnical evaluation of the river bank’s Vertical drains (Mebradrain) were also installed
slope stability shows that the safety factor is mar- on the outside wall of the open ended steel pipe
ginal. Any additional loading on the river banks piles (Fig. 3). A technique commonly adopted in
and construction activities which will increase the Scandinavia to reduce the effect of soil displace-
pore pressure in the soil will adversely affect the sta- ment during installation by preboring a slightly
bility of the slopes. This implies that there is a need
to take different counter measures in design and
during construction of the abutments foundation.
The bridge abutments foundation was therefore
designed in such a way that the load will be trans-
ferred deep down to the dense moraine layer. Pre-
cast concrete piles with dimension 270 × 270 mm
were chosen for the abutments foundation. Fur-
thermore different actions were taken to control
these marginally stable river bank slopes. These
actions include measures before piling started as
part of site preparation, during piling activity
and after piling completed. Based on economical Figure 2. Unloading of the top terrain on the Grønland
aspects and practical considerations of the river side of the river bank to improve the stability situation
system the following measures were taken. before piling.
141
water pressure value, alarming boundary, varies
for each piezometer according to their correspond-
ing locations in the ground. The excess pore water
development due to pile driving activity in the
Grønland side of the riverbank is discussed more
in detail in section 4.
142
The development of pore water pressure was Table 1. Elevation of the tip of the piezometers in the
monitored by piezometers at four locations in each Grønland side of the river bank.
side of the river bank and at three locations in the
river bed. Because of the relative proximity to the Station Piezometers Elevation, masl
river bank, pile driving activities in axes G4, G5, GP4 G-71306 −4.9
G6 and river front foundations (Figs. 1 and 5) were G-71606 −9.9
critical for the stability of the river bank slope. G-71706 −13.1
Figure 5 shows the locations of piezometer sta-
tions in the Grønland side of the river bank. GP3 G-53108 −10.0
Table 1 shows the elevation at which the tips of G-52808 −14.0
the eight piezometers where placed in the ground. GP2 G-53008 −9.0
The piezometers at station GP4 were installed two G-52908 −14.0
years before the startup of the construction of the
bridge. The two years measurement shows that the GP1 G-52508 −7.0
pore water pressure in the soil is close to hydro-
static and follows both the seasonal as well as tidal
variation of the river water level. The piezometers
at stations GP1, GP2 and GP3 were installed two
weeks before the piling activity started. The tips
of the piezometers in these three stations were
located along the critical failure plane where a
change in shear strength of the soil has a signifi-
cant effect on the stability of the river bank slope.
Figure 6 shows the follow up of the development
of excess pore water pressure during pile driving
for the riverfront foundation near station GP4.
Pile driving on land started near the river bank.
This piling activity near piezometer stations GP3
and GP4 was a very good chance to document
recorded data for the specific site condition. Based
on these data a pore pressure chart was developed.
Figure 6. Station GP4 with three installed piezometers
In the chart the maximum build up of excess pore in the Grønland side of the river and driving pile G184.
water pressure vs. distance from the pile can be Note that the river was frozen due to cold temperature
correlated. in January.
143
Figure 8. Measured excess pore water pressure build up
at piezometer G-71606 station GP4 during driving pile
Figure 7. Relative location and length of Pile G182 and G182 and G184.
Pile G184 installed into the ground near station GP4.
increases were much less pronounced. In Figure 8 depth in the soft clay layer where the piezometer
the reference numbers are: number 1—before pil- tip is located. Based on the recorded excess pore
ing activity, 2—after driving pile G184 about 12 m water pressure at stations GP2, GP3 and GP4 an
into the ground, 3—after driving pile G184 about excess pore water pressure chart was developed.
26 m into the ground, 4—after driving pile G184 Figure 9 shows the development trend of the antic-
and before driving pile G182, 5—after driving pile ipated excess pore water pressure in the 10–15 m
G184 and driving pile G182 about 12 m into the thick soft clay layer due to driving 270 × 270 mm
ground, 6—after driving pile G184 and driving pile concrete pile.
G182 about 24 m into the ground and 7—one day
after both piles were driven.
Furthermore the recorded excess pore water 5 CONTROL OF EXCESS PORE WATER
pressure for the deepest piezometer at station GP4 DEVELOPMENT DURING PILING
confirms that the piezometer tip is located just
on top of the dense moraine layer which consists Based on stability analysis critical failure planes of
of stratified sand and gravel. As is clearly seen the river bank slopes were identified. Further analy-
in Table 2 the recorded excesses pore water pres- ses were made to determine the maximum allowable
sure depends not only on the horizontal distance excess pore water pressure in the slope beyond which
between the pile and the piezometer but also on the the safety factor drops below the acceptable margin.
144
It was also observed that during pile driving the
excess pore pressure developments were measured
within approximately 13 m from the center of the
pile. No significant changes in pore water pres-
sure were observed beyond this horizontal distance
from the pile center immediately after driving. Sig-
nificantly high pore water pressure increases were
not recorded by deep piezometers when the piles
were driven into the underlying dense moraine.
The development of excess pore water pressure
was successfully controlled by limiting the mini-
mum distance between piles to be driven per day.
The pile driving schedule prepared based on mini-
mum distance between piles to be driven per day
was very effective. It was possible to keep the devel-
oped excess pore water pressure within the allow-
able range, continuously control the safety margin
of the river bank slopes and save the client from
Figure 10. The location of piezometer tips and the waiting time cost. Total lengths of about 2980 m of
critical failure planes. L1is the sandy, silty soil with some
gravel, L2 is the soft sensitive silty and sandy clay layer
pile length were driven into the ground in 28 days
and L3 is the dense moraine layer. which was about 106 m pile length per day.
ACKNOWLEDGEMENT
Figure 10 shows the critical failure planes and the
locations of the piezometer tips at GP1, GP2 and The authors would like to thank the Norwegian
GP3. The established alarming boundary zones for Public Roads Administration, NPRA, for provid-
the excess pore water pressure at station GP1, GP2 ing resources for this publication.
and GP3 were 10, 50 and 60 kPa respectively. When
the build up of excess pore water pressure along
the critical failure plane at station GP3, GP2, and REFERENCES
GP1 reaches a critical value of 60 kPa, 50 kPa and
10 kPa respectively piling activity has to stop until Aas, G. 1975. Skred som følge av peleramming i bløt leire.
it dissipates. The stability analyses also show that NGI Publication 110. 49–54 (In Norwegian).
a localized build up of excess pore water pressure Eigenbord, K.D. & Issigonis, T. 1996. Pore-water
more than 60 kPa can be tolerated if there is no pressures in soft to firm clay during driving of piles into
substantial change on the adjacent stations along underlying dense sand, Can. Geotech. J. 33, 209–218.
the critical failure plane. Based on the excess pore Fleming, K., Weltman, A., Randolph, M. & Elson, K.
2008. Piling Engineering, Taylor & Francis, New York.
water pressure development trend line (Fig. 9) a Johansen, S. & Finstad, J.A. 2009. Øvre Sund bru
detailed schedule for pile driving was prepared. A alarmgren-se poretrykksmålere Grønland, RIG
minimum distance between piles of 8 m for piles to 026-Rev A. (In Norwegian).
be driven per day was planned (Tvedt 2009). Based Kirkebø, S. 2006. Forslag til sikkerhetsfilosofi, RIG 001.
on the measured excess pore water pressure the (In Norwegian).
safety margins of the river bank slopes were con- Nr. 115513-4, Rv 283 HP: 400, Parsell Øvre Sund bru,
tinuously checked during construction. Geoteknisk rapport, 1. februar 2007 (In Norwegian).
Tvedt, G. & Tefera, T. 2009. Øvre Sund bru erfaringer
med poretrykk og bæreevne under peleramming på
Grøn-land, Notat. (In Norwegian).
6 CONCLUSION Tefera, T. Tvedt, G. & Oset, F. 2011. Excess pore pressure
during pile driving in soft sensitive clay. Proceedings
As was anticipated during driving of piles into the of 15th European Conference on Soil Mechanics and
soft clay layer excess pore water pressures were gen- Geotechnical Engineering, Athens. 1285–1290.
erated at both the shallow and deep piezometers.
145
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
D.J. White
Shell EMI Offshore Engineering, University of Western Australia, Australia
ABSTRACT: Seabed pipelines exemplify the type of geotechnical challenge addressed by the GEO
INSTALL project. Pipeline laying involves complex soil-structure interaction. The seabed undergoes
large deformations that are accompanied by changes in strength. Pipeline design requires assessment of
the available pipe-soil interaction forces after installation, and the potential mobility and scour of the
surrounding soil. These all depend on the installation process. This paper summarises recent research into
the geotechnical aspects of pipeline installation, including numerical and physical modelling as well as
field observations.
1 PIPELINE GEOTECHNICS
‘Rigid’ pipelines are made from steel tubes, typi-
cally 0.1–1.5 m in diameter, with layers of internal
1.1 Offshore pipeline networks
and external coating for corrosion, abrasion and
Pipelines form the vital transport network of off- impact protection, thermal insulation, and to add
shore hydrocarbon developments. They carry the weight for improved stability. ‘Flexible’ pipelines
hydrocarbon product and other fluids between are composites of metal and polymer, and range
wells and in-field processing facilities, and also to in diameter from 0.1–0.5 m. ‘Risers’ are separate
shore. A pipeline network is accompanied by vari- tubes or continuations of a pipeline that provide
ous forms of infrastructure to support, connect and the connection to the floating or fixed structure at
stabilize the pipelines, as illustrated in Figure 1. the ocean surface.
146
As offshore developments extend into deeper is a rather niche area of research and practice—to
waters located further from shore, pipelines and a wider geotechnical audience.
risers represent an increasingly important part The uncertainties on which GEO INSTALL
of the development infrastructure. Typical costs has focused, associated with soil-structure inter-
of gas export pipelines on the North-West Shelf of action and the resulting large deformations and
Australia now exceed $4 million per km, a signifi- changes in soil properties, are shown for pipelines
cant proportion of which is for measures to stabi- in Figure 2.
lise the pipe on the seabed. Stability is the major The laying process for a seabed pipeline is
focus of geotechnical design for pipelines, both dynamic. The pipe is either stored in short sections
under the action of hydrodynamic loading from and welded together as part of the lay process, or
waves and currents, and due to expansion and con- (for smaller diameter pipes) it is stored on a reel
traction arising from the elevated temperature and and ‘unwound’ as the vessel sails along the route.
pressure of the contents. In regions where over-trawling is not a hazard,
Even if long export pipelines are not required, pipelines are usually laid directly on the seabed
because the product is exported via tankers from without burial.
a floating facility, a large network of infield pipe- Depending on the lay method and soil conditions,
lines may be needed. BP’s Greater Plutonio devel- the process of pipe embedment may be drained,
opment provides a well-documented case study undrained or partially-drained. Motions of the ves-
(Jayson et al., 2008). This project is located 165 km sel and hydrodynamic loading on the hanging pipe
north-west of Luanda off the coast of Angola in section lead to slow dynamic motions between the
a water depth of 1400 m. A Floating Production, pipe and the seabed. The pipe is not simply lowered
Storage and Offloading (FPSO) vessel is moored statically onto the seabed. The seabed is disturbed
adjacent to a top-tensioned riser tower, which is a by these dynamic motions, and in soft soils heave
column of vertical pipes, or risers that carry prod- and remoulding occur around the pipe.
ucts between the FPSO and the seabed. The risers After laying, there is typically a period of sev-
are maintained in tension, to prevent buckling, eral months before the pipeline enters operation.
by a buoyancy unit at the top. The wells are dis- During this time, any excess pore pressure gen-
tributed over a wide area, and are each supported erated during the lay process dissipates and the
on subsea foundations. Connecting the wells are surrounding soil changes in strength. It is this
more than 100 km of pipelines. These are laid post-equalisation consolidated strength—which
directly on the seabed and are subject to signifi- varies spatially around the pipe—that controls the
cant expansion loads during cycles of startup and subsequent resistance of the pipe to lateral and
shutdown, as the temperature and internal pres- axial movements.
sure change. These issues will be familiar to piling engineers.
Pile design methods for axial capacity in clay soils
1.2 Pipeline installation effects
The GEO INSTALL project, which this confer-
ence forms the closing event of, has focused on a
research area that is extremely relevant to the anal-
ysis of seabed pipelines. The research activities in
GEO INSTALL mirror some of the research efforts
underway in our own group at the University of
Western Australia (UWA). Quoting from the GEO
INSTALL programme outline:
147
are often couched in terms of the post-installation change due to seabed mobility. In shallow water
strength or lateral effective stress on the pile, (<150 m depth) the action of waves and tides can
accounting for changes in strength from installa- lead to sediment transport at the seabed, and scour.
tion and equalization (e.g. Randolph et al. 1979, Depending on the metocean and seabed conditions,
Jardine et al. 2010). as well as the installed (or ‘as-laid’) embedment of
the pipeline, this seabed mobility may lead to net
erosion or burial around the pipe (Figure 2).
1.3 Pipeline design for global expansions
Installation of a pipeline on the seabed causes an
During operation, the internal temperature and increase in the local sediment transport capacity due
pressure cause a pipeline to expand. This expan- to the change in flow conditions around the pipe-
sion is opposed by the axial resistance between the line. A consequence of the increase of local sedi-
pipe and the seabed. Excessive compressive force ment transport capacity is the scour of sediments
may lead to buckling of the pipeline, depending away from an area surrounding the pipeline. Once
on the soil resistance and the pipe weight. Once a local scour is initiated at a cross-sectional location,
buckle forms, the axial force drops significantly as it can spread longitudinally along the pipeline to
a length of pipe feeds axially into the buckle. Exces- form a free span. The pipeline may then sag into the
sive feed-in—due to low axial resistance—leads to scour hole as the result of the structural deflections
high bending strains within the buckled section. and extra penetration into the soil at the span shoul-
In deep water, where there is no requirement ders (which attract increased load from the span-
to bury the pipe for protection from trawling, on- ning lengths of pipeline). As the pipeline sags into
bottom buckling in the horizontal plane is now a a scour hole, it deflects more flow over the top of
widely adopted solution to accommodate temper- the pipeline. This leads to a reduction of sediment
ature and pressure-induced expansions. However, transport capacity directly below the pipeline. The
controlled lateral buckling requires very careful scour hole is then backfilled with sediments and this
design in order to ensure that buckles form as induces the natural self-burial of the pipeline. Once
planned, and that the fatigue response within the the pipeline is buried or sags into the scour hole, its
buckle is acceptable throughout the operating life stability is enhanced through hydrodynamic shield-
of the flowline (Bruton et al. 2007, 2008). Thermal ing and the increase of lateral soil resistance.
cycles can also lead to the accumulation of axial Accurate modelling of the installation process
movement, which is termed pipeline walking (Carr and the resulting as-laid embedment is an important
et al. 2006). The walking arises from asymmetry in input to assessments of seabed mobility processes.
the heat-up and shutdown processes, or from other These processes are illustrated in Section 6.
asymmetries in the pipeline such as a high riser ten-
sion at one end or a seabed slope. Over many ther-
mal cycles, walking can lead to significant global 1.5 Soil mechanics context
displacement of the pipeline. Walking is not a limit
The stress levels relevant to pipeline geotechnics are
state for the pipeline itself, but can lead to failure at
significantly lower than found in conventional foun-
the mid-line or end connections.
dation engineering. The bearing pressure imposed
Buckling and walking must be assessed during
by pipeline is typically 0.5–5 kPa, and the soil in the
design, in order to control the stresses within the
shallowest 0.5 m of the seabed is generally very soft
pipeline and the movements relative to any asso-
(if fine-grained) or mobile (if coarse-grained).
ciated structures. Structural analyses of pipelines
Pipeline laying, and any subsequent movements
incorporate interaction forces between the pipe and
across the seabed, lead to large strains in the sur-
the seabed in the vertical, axial and lateral direc-
rounding soil. The consequent remoulding and
tions. The conventional approach is to construct
reconsolidation of the seabed soil leads to changes
independent models of the force-displacement
in topography and strength. Site investigation
response in each direction—in a manner analo-
technology is advancing rapidly, to provide bet-
gous to the ‘t-z’ and ‘p-y’ models used for axial and
ter tools to quantify these near-surface seabed
lateral pile-soil analysis. The pipe-soil resistance
characteristics.
in each of these directions depends on the pipe-
Two important aspects of the constitutive
line embedment and the surrounding soil strength.
response of fine-grained seabed soils, which affect
Correct modelling of installation effects is required
the interaction forces on seabed pipelines, are:
to assess these forces accurately.
1. The curvature of the effective stress failure enve-
lope at low stresses.
1.4 Pipeline design for hydrodynamic stability
2. The generation and dissipation of excess pore
During the period between installation (laying) and pressure during and in between episodes of
operation, the embedment of the pipeline may also shearing to very high strains.
148
This first effect is illustrated by the two types during the initial packet of cycles is partly due to
of data shown in Figure 3. The red lines are effec- the generation of positive excess pore pressure
tive stress paths determined from load and pore in this lightly over-consolidated material. As this
pressure data gathered by the ‘SMARTPIPE’—an positive pore pressure dissipates and the effective
instrumented model pipe that operates in situ, at the stress rises back to the geostatic state the mate-
seabed. As the pipe was slowly cycled axially, alter- rial densifies and the subsequent undrained shear
nating shear stresses were mobilized, tracing out the strength is higher.
failure envelope. The cross-shaped markers are the After just three episodes of full remoulding and
results from model tests performed in a large tank reconsolidation, the current remoulded strength is
of the same soil, reconstituted at NGI’s laboratory. comparable to the original intact strength. These
The second effect is best demonstrated by cyclic results illustrate the contrasting effects of remould-
penetrometer tests with periods of reconsolida- ing and reconsolidation that are shown schemati-
tion between each cyclic episode, such as the T-bar cally around a pipeline in Figure 2.
test shown in Figure 4. The reduction in strength This behaviour is best explored in a critical state-
type framework, since this provides an explicit
link between moisture content (which reduces as
positive pore pressures dissipate) and undrained
strength. White & Hodder (2010) describe a critical
state-type model that produces the back-analysis
indicated by the line ‘Model’ in Figure 4.
149
Figure 5. Typical undrained static vertical penetration
response (Randolph & Gourvenec 2011).
Figure 6. Pipeline laying notation (Randolph & White
2008b).
1998a, b), which is described in Section 4.1. The
variation in normalized vertical load V/su-invertD
with embedment, w/D fits a power law (where V is
the vertical pipe-soil load, su-invert is the in situ und-
rained shear strength at the pipe invert level, and D
is the pipe diameter).
However, assessment of as-laid pipe embed-
ment is not simply a matter of equating the pipe
submerged weight, W′, with the bearing capacity,
V, to determine the embedment, w. For a pipeline
being laid from a vessel, the maximum value of V
within the touchdown zone is greater than the sub-
merged pipe weight W′ due to the catenary effect.
This stress concentration can be estimated from
the lay tension, pipe bending stiffness and the sea-
bed response (Fig. 6) (Pesce et al. 1998, Lenci &
Callegari 2005, Randolph & White 2008b, Palmer
2009).
Parametric solutions for the static lay condi-
Figure 7. Effect of seabed stiffness on the static pipe-
tions have been presented by Randolph & White soil contact stress through touchdown zone (Westgate
(2008b). For most practical values of lay tension, et al. 2013).
analytical solutions (Lenci & Callegari 2005) and
numerical analysis using OrcaFlex (Orcina 2008)
all converge to unique design lines. The touchdown interaction in pipeline laying: the stiffness of
stress concentration, flay, can be expressed approxi- the seabed affects the loads applied to the struc-
mately as: ture. The range of stiffnesses, k = V/w, shown in
Figure 7 is wider than the usual uncertainty faced
Vmax ⎛ EIk ⎞
0.25 in bearing capacity calculations. This is because
f lay = ≈ 0.6 + 0.4 ⎜ (1) the lay process introduces additional uncertainty
W′ ⎝ T0 ⎟⎠ in the operative seabed strength, which is discussed
in Section 3.
where Vmax is the maximum vertical pipe-soil
load and the other parameters are as shown in
Figure 6. 3 PIPELINE INSTALLATION DYNAMICS
The stiffness of the seabed affects both the max-
imum stress concentration and also the length of While a pipeline is being laid, the vessel is sub-
the elevated contact stress through the touchdown jected to waves and the hanging length of pipe also
zone. Example profiles of stress concentration are attracts hydrodynamic load in the water column.
shown in Figure 7 for a range of seabed stiffnesses. These effects cause the pipe to oscillate at the
This illustrates the influence of soil-structure seabed, softening the soil and entraining water.
150
Also, in rough sea states, transient reductions in 4 SOIL DISTURBANCE FROM PIPE
lay tension can occur which increase the vertical LAYING
pipe-soil load (Westgate et al. 2009, Westgate et
al. 2010a). 4.1 Numerical modelling technique
The combination of both vertical and horizontal
Assessments of the degree of soil remoulding
pipe-soil loading reduces the available vertical
during pipeline laying can be made using Large
bearing capacity of the soil, as shown through
Deformation Finite Element (LDFE) analysis.
theoretical plasticity analysis for planar contacts
Our group at UWA has made extensive use of the
(Green 1954) and also shallowly-embedded pipes
‘RITSS’ form of LDFE analysis, which was set
(Randolph & White 2008a). Lastly, horizontal
out by Hu & Randolph (1998a, 1998b) and has
oscillations can displace the seabed soil laterally,
recently been coupled with the commercial soft-
creating a trench by lowering the seabed elevation
ware ABAQUS.
adjacent to the pipe, as shown via experimental
The RITSS scheme is illustrated in Figure 9. In
simulations in a geotechnical centrifuge (Cheuk &
this method, large displacements of the pipe are
White 2011) and via field observations (Lund 2000,
divided into many small steps in which the pipe
Westgate et al. 2009).
is moved to a very small incremental distance
During pipe laying, remotely operated vehi-
and small strain Lagrangian calculations are per-
cles are used to survey the touchdown zone dur-
formed. Stresses and other material parameters
ing pipelaying. Video observations show that the
are then recovered from the integration points to
amplitude of pipe oscillations are typically a frac-
the nodes following the superconvergent patch
tion of a pipe diameter, with a period comparable
recovery technique (SPR, Zienkiewicz & Zhu
to the wave period (Westgate et al. 2009). Struc-
1992).
tural modelling of the vessel response and the pipe
At the end of each step, the displaced soil
catenary can be used to estimate these movements
boundary is updated and the whole soil domain
a priori (Westgate et al. 2010a).
is remeshed. After that, the stresses and other
The amplitude of the pipe oscillations in the
parameters are interpolated from the old nodes to
touchdown zone increases with the seastate.
the integration points of the new mesh. With the
Westgate et al. (2010a) back-analysed the observed
interpolated stresses as initial conditions for the
embedment of a pipeline installed across uniform
next step, another small strain step is performed.
soil conditions but in a range of seastates. These
This process is repeated until the pipe is moved the
data show a strong correlation between seastate—
required distance. Mesh generation, small strain
and thus vessel motion and pipe oscillations in the
calculations and post processing of the data are
touchdown zone—and embedment (Fig. 8).
performed using ABAQUS. Python, the in-built
Experimental and numerical simulations have
scripting language of ABAQUS, is used to auto-
been performed to explore these dynamic installa-
tion effects. The critical questions are:
1. How does the installation process alter the
strength of the soil surrounding the pipeline?
2. What is the consequent effect on the as-laid
embedment and the breakout resistance?
Figure 8. Effect of metocean conditions on as-laid pipe Figure 9. RITSS FE analysis scheme (after Wang et al.
embedment (Westgate et al. 2010a). 2010).
151
mate the process. The whole process is executed 4.2 Numerical simulations of pipeline laying
by a main Fortran program that calls different
This numerical technique has been used to explore
subroutines and Python scripts to carry out the
the pipe-seabed interaction during laying. An
entire task.
example analysis is shown in Figure 10. The in
Within each small strain analysis, the soil is
situ soil strength profile was su0 = (sum + kz) with
modelled as an elastic perfectly-plastic material,
sum = 0.75 kPa, k = 1.6 kPa/m, and the soil sof-
with failure according to the Tresca yield criterion.
tened according to Equation (2) with parameters
The particular value of the Tresca shear strength
ξ95 = 10, δrem = 0.25, μ = 0. The pipe was held under
for each element within the mesh is calculated
a constant vertical load of V/sumD = 3.7 and oscil-
when each small strain analysis is initialised. This
lated laterally with amplitude of +/−0.05 D.
calculation accounts for the influences on the soil
These parameters were chosen to represent a
of the cumulative strain (i.e. the level of distur-
centrifuge model test reported by Cheuk & White
bance or remoulding) strength, as well as the cur-
(2011), which is discussed in Section 4.3. The
rent strain rate.
amplitude of horizontal oscillation is comparable
The cumulative strain induced in a soil element
to the video observations of pipe laying reported
by any previous pipe movement will cause soften-
by Westgate et al. (2009). The resulting pipe embed-
ing of that soil and a decrease in the operative shear
ment and the measured horizontal resistance agree
strength—akin to the behaviour shown during the
well with the centrifuge model test results.
initial cyclic episode shown in Figure 4.
The calculated patterns of lateral resistance
The effects of strain rate on the shear strength
and embedment and the local soil remoulding are
of soil have also been studied extensively by many
shown in Figure 10. Even small lateral motions
researchers (e.g. Casagrande & Wilson 1951,
of just +/−0.05 D lead to a remoulded zone that
Graham et al. 1983, Biscontin & Pestana 2001,
extends by al-most one pipe diameter to each side,
Lunne et al. 2006, Low et al. 2008).
and the pipe it-self rests on fully remoulded soil.
To combine these softening and strain rate
The horizontal resistance initially rises through
effects within an undrained analysis, the simple
the first few cycles, as the pipe embedment and
Tresca soil model is used. After each analysis step,
soil contact width increase. With further cycles the
the original shear strength of the soil at every node,
effect of soil softening on either side of the pipe
su0, is modified to an updated value, su, according
dominates, as plastic strain accumulates. This leads
to the following formula (Einav & Randolph 2005,
to a reduction in the normalized lateral resistance
Zhou & Randolph 2007, 2009, Chatterjee et al.
in the later cycles. The vertical embedment after
2010, 2012a):
20 cycles is significantly greater than the 0.1 D
penetration achieved under purely vertical loading
⎡ ⎛ ⎛ ( ε1 − ε3 ) v p / D ⎞ ⎞ ⎤ prior to the lateral cycles.
s u = ⎢1 + μ log ⎜ max ⎜1, ⎥
⎢⎣ ⎝ ⎝ δ D γ ref ⎟⎠ ⎟⎠ ⎥⎦ The pipe embedment has almost stabilized after
30 cycles. Any additional cycles do not signifi-
e + ( − rem ) e
× ⎣⎡δ rem −3 ξ ξ95
⎤⎦ s u 0 (2) cantly deepen the pipe, because the soil influenced
by the lateral motions is already fully softened,
and the vertical bearing capacity is provided by
The first part of the relationship captures the
deeper soil that is less influenced by the motions.
increase in strength due to the strain rate effect and
the second part reflects a strength decrease due to
strain softening. Δε1 and Δε3 are the major and
minor principal strains, respectively, resulting from
a displacement increment, δ. vp is the pipe velocity
and μ is the rate of strength increase per decade
of strain rate, typically in the range of 0.05–0.2
(Biscontin & Pestana 2001, Lunne & Andersen
2007). γ ref is the reference strain rate taken as 1%
per hour or 3 × 10−6 per second. Softening is mod-
elled as an exponential decay function of the cumu-
lative absolute plastic shear strain ξ. δrem is the ratio
of fully remoulded strength to the initial strength
and is essentially the inverse of sensitivity, St. The
parameter ξ95 indicates the relative ductility of soil
and is the value of ξ at which the soil strength has
reduced by 95% of the way from the initial to the Figure 10. LDFE simulation of soil strength after
fully remoulded values. dynamic pipe laying (after Wang et al. 2009).
152
This stabilization is a useful concept for the assess-
ment of dynamic embedment in practice, since it
narrows down the input parameters necessary for
a simplified analysis of the lay process. It may not
be necessary to specify precisely the number of
cycles that will be imposed during the lay process
if that number is sufficient for the embedment to
approach the steady value.
153
Figure 13. Disturbed soil after (a) monotonic
penetration and (b) dynamic embedment of a model pipe
(Westgate et al. 2013).
154
embedment calculation method. The pipelines Krost et al. (2011) and Gourvenec & White
ranged from 0.33 to 0.63 m in diameter, with (2010) present numerical solutions for consolida-
submerged pipe weights between 0.22 and 0.95 tion around pipelines ‘wished-in-place’ in elastic
kN/m. The sediments were fine-grained sediments soil. The dimensionless time factors, T = cvt/D2,
with undrained shear strength gradients between for 50% and 90% dissipation of the average excess
4 and 38 kPa/m (and negligible strengths at the pore pressure around the pipe surface are T50 ∼ 0.1
mudline). and T90 ∼ 2 respectively. Subsequent solutions,
Estimates of pipe embedment were made for using the Modified Cam Clay constitutive model,
each site using the pipeline properties, soil con- have considered the effect of the installation proc-
ditions, and model parameters representative of ess (Chatterjee et al. 2012b).
the site-specific lay conditions. Three calculations
of the embedment process were performed for
5.2 Lateral breakout resistance: Numerical results
each site, representing the range in sea state (wave
height) measured during laying, and therefore the Theoretical solutions for the combined vertical-
amplitude of oscillations in the touchdown zone. horizontal bearing capacity of a pipeline embed-
The ranges of calculated embedment values are ded in uniform undrained soil are well-established
compared to histograms of as-laid embedment for (Randolph & White 2008a, Merifield et al. 2008,
each site in Figure 14. In all cases, the agreement Martin & White 2012). However, the dissipation of
between the calculated and observed embedment is installation-induced excess pore pressure changes
excellent, capturing both the average embedment the strength of the surrounding soil, with a conse-
(when using the best estimate parameters) and quent enhancement of the V-H bearing capacity.
the range of embedment (when using the extreme Numerical simulations of pipe penetration
minima and maxima parameters). and lateral breakout in Modified Cam Clay soil
A less accurate but simpler approach to assess are reported by Chatterjee et al. (2013). The pur-
as-laid pipeline embedment on fine-grained soils, is pose of this study was to investigate the potential
to assume that the soil is fully remoulded and apply enhancement of lateral breakout resistance due to
this strength within the vertical bearing capacity consolidation hardening of the surrounding soil.
equation (Westgate et al. 2010b; Westgate et al. Two types of analyses were performed: undrained
2012). This approach is often sufficiently accurate vertical penetration followed by either unconsoli-
for typical lay conditions, but probably only suc- dated or consolidated undrained lateral breakout.
cessful due to compensating errors: the reduction After the initial penetration (modelled using the
in bearing capacity under V-H loading is ignored, LDFE approach—Section 4.1), the pipe was either
but the loss of strength is probably over-estimated immediately broken out laterally (the unconsoli-
through the assumption of full remoulding. dated case), or full pore pressure dissipation was
permitted before breakout (the consolidated case).
The unconsolidated and consolidated fail-
ure envelopes for an undrained embedment of
5 POST-LAYING BREAKOUT RESISTANCE
w/D = 0.1 and 0.5 are compared in Figure 15.
5.1 Equalisation
The seabed softening that results from the dynamic
movements during laying of a pipeline is princi-
pally due to positive excess pore pressure being
generated as the soil is sheared. After laying, this
excess pore pressure, and any excess pore pressure
generated simply in response to the static pipeline
self-weight, will dissipate.
During this dissipation period, the effective
stress around the pipe-soil interface will rise and the
strength of the surrounding soil will also increase.
These ‘set-up’ processes will cause the available
axial and lateral pipe-soil resistance to rise.
There is typically a period of several months
between the laying of a pipeline and the initial
operation. It is therefore the consolidated axial and
lateral resistance that is usually relevant to design, Figure 15. Unconsolidated and consolidated undrained
but some assessment of the pore pressure dissipa- yield envelopes for w/D = 0.1 and 0.5 (Chatterjee et al.
tion rate is required to confirm this. 2013).
155
The initial undrained strength was uniform with installation (Gourvenec & White 2013). Details of
depth and the soil was normally consolidated. a subset of these tests are set out in Table 1. The
After consolidation, the failure envelopes have purpose of these tests was to explore the changes
expanded by approximately 50% for purely verti- in lateral breakout resistance that can result from
cal loading, and the horizontal load at the apex installation effects in soft clay, associated with dis-
point (where H is maximum) is approximately turbance and reconsolidation. The model seabed
doubled. The changes in strength in the surround- was normally—consolidated kaolin clay.
ing soil, and the additional embedment that arises In all three cases, the initial embedment was
during the consolidation process, are shown in w/D = 0.5 and immediately after reaching this
Figure 16 for the w/D = 0.5 case. Clearly the post- embedment the simulated pipe weight was reduced
installation consolidation process has a significant to V/D = 3 kPa. Two alternative ‘laying’ methods
influence on the lateral breakout resistance, and were considered: undrained monotonic vertical
‘set-up’ effects of this magnitude should not be penetration and a specified pattern of oscillations
neglected. that remould the surrounding soil. The adopted
However, it should be noted that in these analy- ‘zig-zag’ pattern of lateral movement during pen-
ses the simulated pipe weight was maintained at the etration (visible in Fig. 17) mimics the disturbance
unconsolidated undrained vertical bearing capac- and remoulding associated with the real lay proc-
ity during the dissipation process, thus maximiz- ess, and the results from case RemCU are therefore
ing the potential strength gain. In practice, most the most representative of field conditions.
pipelines will be ‘overloaded’ on laying, because For the monotonic installation method, two
the self-weight will be lower than the maximum cases involving full (90%) and negligible dissipa-
vertical load applied in the touchdown zone (by at tion were performed (cases UU and CU). The
least a factor of flay, Section 2.2). The changes in same pipe weight was maintained during breakout,
V-H bearing capacity in this example are therefore
upper bounds for set-up in normally consolidated
Modified Cam Clay. Table 1. Lateral breakout comparison cases.
Figure 16. Contours of ratio of consolidated to original Figure 17. Pipe invert trajectories during lateral
shear strength after pore pressure dissipation (Chatterjee breakout for different installation and ‘set-up’ cases
et al. 2013). (Gourvenec & White 2013).
156
and the pipe was able to rise or fall to maintain that observed in case UU. This gain in resistance
this load. is less brittle, reflecting a wider zone of strength-
The displacement paths during lateral breakout ened soil. The remoulding during ‘laying’ leads
are shown in Figure 17. In all cases, the trajectory to a larger zone of elevated pore pressure and a
is approximately the same with the pipe moving correspondingly higher gain in undrained strength
upwards at a slope of approximately 10°. The after consolidation. The outcome—which is ini-
mobilised lateral resistance is significantly differ- tially counter-intuitive—is that more softening
ent in each case. during laying leads to more net hardening after
The UU case involves no consolidation effects so consolidation.
the original undrained strength might be assumed to
apply throughout, notwithstanding some softening
5.4 Summary of post-laying breakout observations
due to the accumulated strain. In this case, the
penetration and breakout resistance should be pre- These complementary simulations of pipeline
dictable using the in situ su profile and theoretical laying and breakout, using numerical modelling
solutions, such as failure envelopes based on plas- (Section 5.2) and centrifuge testing (Section 5.3)
ticity limit analysis (Randolph & White 2008a, show a strong installation effect. This effect results
Martin & White 2012). The line marked LA in from post-laying consolidation, and is enhanced
Figure 18 shows this calculated breakout resistance, if the lay process involves remoulding. Three key
ignoring heave (i.e. assuming a horizontal sea- conclusions emerge, which provide some answers
bed beside the pipe) and using the in situ strength to the questions raised in Section 3 for pipelines
profile. There is indeed good agreement with the laid on soft clay seabeds:
UU result, with the ∼10% under-prediction being
1. Pipeline installation can significantly enhance
attributable to the neglect of heave.
the strength of the surrounding soil due to con-
The more realistic CU case shows an 80%
solidation under the pipe self-weight.
increase in peak breakout resistance, which is com-
2. The dynamic pipe movement during installa-
parable to the LDFE results shown in Figure 15.
tion creates additional shear-induced excess
This gain in capacity is highly brittle, which is
pore pressure in the soil surround the pipeline.
also consistent with the small size of the zone of
Although this remoulding weakens the soil in
enhanced strength shown in Figure 16. However,
the short term, the subsequent strength gain
even beyond this initial peak, the CU case still
during consolidation is greater than following
shows higher resistance than the UU case, as the
monotonic installation.
strengthened soil continues to be mobilised and
3. The subsequent breakout resistance is signifi-
pushed ahead of the pipe.
cantly enhanced (‘set-up’) by these changes
The most realistic case—RemCU—shows an
in strength, and suitable modifications to the
even greater breakout resistance, more than double
operative strength are required when using the-
oretical calculations for breakout resistance.
It should however be noted that in these tests
the soil was initially normally consolidated. This
state provides the greatest tendency to contract
and harden in response to shearing. At a higher
OCR these ‘set-up’ effects will be smaller, and the
effects of remoulding and reconsolidation may
balance, meaning that use of the in situ strength
may be appropriate in assessments of breakout
resistance.
157
When pipelines are installed on mobile soils,
there is generally a risk of global instability—i.e.
gross movement of the pipeline—under hydro-
dynamic action. The design must ensure that any
such movements are tolerable.
Current design practice to assess global pipe-
line stability generally neglects the mobility of the
seabed. The analysis instead focusses on simulat-
ing fluid-pipe and pipe-seabed interaction without Figure 19. General view of UWA large O-tube facility.
including the fluid-seabed interaction that leads to
the scour and erosion processes that were intro-
duced in Section 1.4 (e.g. Zeitoun et al. 2008).
Significant research efforts have been devoted
to pipeline on-bottom stability and seabed mobil-
ity at UWA over the past three years, supported
by the ARC1 and industry partners Woodside and
Chevron through the STABLEpipe Joint Indus-
try Project. An innovative Large O-Tube (LOT)
testing facility has been established to simulate
ocean-seabed interactions at large scale (Cheng et Figure 20. Images from a large O-tube test showing
al. 2010, An et al. 2011, Jas et al. 2013). The LOT scouring and settlement of a pipeline on a mobile seabed,
followed by breakout.
is a recirculating water tunnel that can generate
cyclonic and tidal flow conditions that cannot be
modelled in traditional wave flumes of similar shallowly embedded, a scour tunnel opens beneath
dimensions (Fig. 19). the pipe, widening into a scour hole that grows
Precise control of the inline propeller allows longitudinally. As the pipe sinks into this hole is
maximum wave and current conditions of approx- becomes more stable, because of the shielding
imately 3.0 m/s with a peak period of 13 s and from the flow and the increased soil resistance that
4.0 m/s respectively. Irregular flow can also be gen- results from the trench.
erated, and typical 100-year return period tropical If the pipe is initially too deeply embedded for
storm conditions for Australia’s NWS are com- a scour tunnel to open up, then this process is
fortably within the performance envelope of the delayed or may not occur. Consequently, a pipeline
facility. that is installed at a deeper embedment can be less
The facility is also equipped with an instrumen- stable than a pipeline that is initially resting on the
ted model pipe mounted on a sophisticated actuator seabed at zero embedment.
system to record the applied horizontal and vertical Figure 20 shows a sequence of images from a
forces. The pipe is also equipped with a network of large O-tube test in which an irregular storm was
surface pressure cells to record the hydrodynamic imposed on a pipe initially at a very shallow embed-
load around the pipe circumference. The actuator ment. The images show this sequence of (i) tunnel
system prevents model pipe movements in unreal- scour, (ii) propagation of the tunnel (away from
istic degrees of freedom such as rolling. The feed- the window), (iii) settlement of the model pipe and
back system can provide neutral horizontal control, then (iv) breakout at the height of the storm. In
allowing the pipe free lateral movement in response other tests, different responses have been observed
to the natural balance between hydrodynamic load- for different installed embedments.
ing and soil resistance.
Model tests in the large O-tube have high-
lighted the influence of pipeline installation—and 7 CONCLUSIONS
the resulting as-laid embedment—on the subse-
quent on-bottom stability. Some of the results are Seabed pipelines are their own foundation. The
counter-intuitive, and show the reverse trends rela- geometry of this foundation—the depth and
tive to conventional design practice. width—is set by the installation process, which is
For example, on mobile sandy soils the gen- subject to uncertainty even until the day of installa-
eral response is for the seabed to begin scouring tion when the prevailing weather determines the lay
before the storm is sufficiently intense to dislodge vessel motions and the consequent oscillations as the
a pipe resting on the seabed surface. If the pipe is pipeline touches down on the seabed.
1
Australian Research Council Linkage Project LP0989936.
158
The installation of seabed pipelines exemplifies the at large scale, to provide data to underpin these
particular class of geotechnical engineering that the new approaches.
GEO INSTALL project has been focused on. Pipeline
installation causes large seabed deformations and
changes the properties of the surrounding soil. ACKNOWLEDGEMENTS
These effects alter the subsequent pipe-soil interac-
tion behaviour. This work forms part of the activities of the Cen-
This paper has summarised recent research tre for Offshore Foundation Systems (COFS),
efforts that have involved numerical analysis, exper- currently supported as a node of the Australian
imental testing and field observations. The over- Research Council Centre of Excellence for Geo-
arching aim has been to create an understanding of technical Science and Engineering. Support from
installation effects to allow more reliable estimates the ARC Future Fellowships program and Shell is
of the subsequent in-service behaviour to be made. acknowledged.
Large deformation finite element analysis is a The pipe-soil interaction studies described here
mature research tool that is increasingly being used have been performed in collaboration with many
in industry practice. The particular LDFE method colleagues at UWA, including Mark Randolph,
used at UWA provides a flexible basis for incorpo- Susan Gourvenec, Dong Wang, Zack Westgate,
rating the constitutive behaviour that is most rele- Santiram Chatterjee and Matt Hodder. The O-tube
vant to the problem being simulated. Softening and team includes Liang Cheng, Hongwei An, Cheng-
rate effects have been superimposed on a conven- cai Luo and Scott Draper.
tional Tresca model. Modified Cam Clay has been Industry support from BP (Andy Hill), Subsea 7
used to capture consolidation and drainage effects. (Paul Brunning), Woodside (Andrew Pearce, Nino
An equally important research tool is well- Fogliani) and the SAFEBUCK and STABLEpipe
controlled physical modelling. Examples from JIPs is also acknowledged.
both the UWA geotechnical centrifuge facilities
and the new O-tube flume have been presented.
Where soil-structure interaction involves large REFERENCES
deformations and many episodes of loading or
displacement, physical modelling remains essential An, H., Luo, C., Cheng, L. White, D. and Brown, T. 2011.
to uncover effects that could not be reliably fore- Introduction to the O-tube. Proc. Sixth International
seen using only numerical modelling. In offshore Conference on Asian and Pacific Coasts. Hong Kong,
engineering, physical modelling often provides the China.
only basis for gathering validation data for compu- Aubeny, C.P., Shi, H. & Murff, J.D. 2005. Collapse load
tations and design calculations, due to the limita- for a cylinder embedded in trench in cohesive soil.
tions of scale and location associated with offshore International Journal of Geomechanics, 5(4): 320–325.
Bruton, D., Carr, M. & White, D.J. 2007. The influence
infrastructure. of pipe-soil interaction on lateral buckling and walk-
For pipelines on fine-grained soils, an impor- ing of pipelines: the SAFEBUCK JIP. Proc. 6th Int.
tant theme is the contrasting effects on soil Conf. on Offshore Site Investigation and Geotechnics,
strength of remoulding and reconsolidation. London. 133–150.
Classical solutions for undrained penetration Bruton D.A.S., White D.J., Carr M.C. & Cheuk, C.Y.
and lateral breakout resistance have recently been 2008. Pipe-soil interaction during lateral buckling
derived, using traditional plasticity limit analysis and pipeline walking: the SAFEBUCK JIP. Proc. Off-
and complementary numerical tools. Field and shore Technology Conference, Houston, USA. Paper
laboratory testing has shown that the use of these OTC19589.
Biscontin, G. & Pestana, J.M. 2001. Influence of periph-
solutions requires careful selection of the opera- eral velocity on vane shear strength of an artificial
tive strength. During dynamic pipe laying on soft clay. ASTM Geotech. Test. J. 24(4): 423–429.
soils, the strength is reduced due to remoulding, Carr, M.C., Sinclair, F., & Bruton, D.A.S. 2006. Pipeline
but during subsequent breakout it can be higher walking—understanding the field layout challenges,
due to reconsolidation. and analytical solutions developed for the SAFE-
In coarse-grained soils the installation process BUCK JIP. Proc. Offshore Tech. Conf., Houston,
and the resulting embedment affect the propensity Paper OTC17945.
for seabed mobility around the pipe and subse- Casagrande, A. & Wilson, S.D. 1951. Effect of rate of
quent scour and self-burial. This cross-disciplinary loading on the strength of clays and shales at constant
water content. Géotechnique 2(3): 251–263.
topic is an important area of current research. It Chatterjee, S., Randolph, M.F., White, D.J. & Wang, D.
offers the potential for a more realistic methodol- 2010. Large deformation finite element analysis of
ogy for assessing pipeline stability on mobile sea- vertical penetration of pipelines into the seabed. Proc.
beds. UWA’s O-tube is a novel tool that allows full 2nd Int. Symp. on Frontiers in Offshore Geotechnics.
ocean-pipeline-seabed interaction to be simulated Perth. 785–790.
159
Chatterjee S., Randolph M.F. & White D.J. 2012a. The Jayson, D., Delaporte, P., Albert, J-P, Prevost, M.E.,
effects of penetration rate and strain softening on the Bruton, D. & Sinclair, F. 2008. Greater Plutonio
vertical penetration resistance of seabed pipelines. project—Subsea flowline design and performance.
Géotechnique, 62(7):573–582. Offshore Pipeline Technology Conf., Amsterdam.
Chatterjee S., Yan, Y., Randolph M.F. & White D.J. Krost K., Gourvenec S.M. & White D.J. 2011. Consoli-
2012b. Elastoplastic consolidation beneath shallowly dation around partially-embedded submarine pipe-
embedded offshore pipelines. Géotechnique Letters, lines. Géotechnique, 61(2): 167–173.
2(2):73–79. Lenci, S. & Callegari., M. 2005. Simple analytical models
Chatterjee S., White D.J. & Randolph M.F. 2013. Cou- for the J-lay problem. Acta Mechanica, 178:23–39.
pled consolidation analysis of pipe-soil interactions. Low, H.E., Randolph, M.F., DeJong, J.T. & Yafrate, N.J.
Canadian Geotechnical Journal, in review. 2008. Variable rate full-flow penetration tests in intact
Cheuk, C.Y. & White D.J. 2011. Modelling the dynamic and remoulded soil. Proc. 3rd Int. Conf. on Geotech.
embedment of seabed pipelines. Géotechnique. 61(1): Geophys. Site Characterization, Taipei, Taiwan
39–57. 1087–1092.
Cheng, L. White, D.J., Brown, T. An, H. and Luo, C. Lund, K.M. 2000. Effect of increase in pipeline soil pene-
2010. O-tube Pipeline Stability Project, Development tration from installation. Proc. of ETCE/OMAE 2000
of the Large O-tube Facility, Commissioning Report. Joint Conference; Energy of the New Millennium Paper
UWA Report GEO: 10500 v2. 115 pp. OMAE2000/ PIPE-5047.
De Catania S., Breen J., Gaudin C. & White D.J., 2010. Lunne, T., Berre, T., Andersen, K.H., Strandvik, S. &
Development of multiple-axis actuator control system Sjursen, M. 2006. Effects of sample disturbance and
Proc. Int. Conf. on Physical Modelling in Geotechnics. consolidation procedures on measured shear strength
Zurich, Switzerland, 325–330. of soft marine Norwegian clays. Can. Geotech. J.
Dingle, H.R.C., White, D.J., & Gaudin, C. 2008. Mecha- 43(7): 726–750.
nisms of pipe embedment and lateral breakout on soft Lunne, T. and Andersen, K.H. 2007. Soft clay shear
clay. Canadian Geotechnical Journal, 45(5): 636–652. strength parameters for deepwater geotechnical
Einav, I., & Randolph, M.F. 2005. Combining upper design. Proc. 6th Int. Conf. Offshore Site Investiga-
bound and strain path methods for evaluating pene- tion and Geotechnics: Confronting New Challenges and
tration resistance. International Journal for Numerical Sharing Knowledge, SUT, London, Vol. 1: 151–176.
Methods in Engineering, 63(14): 1991–2016. Martin C.M. & White D.J. 2012. Limit analysis of the
Gaudin C., White D.J., Boylan N., Breen J., Brown, T. & undrained capacity of offshore pipelines. Géotech-
De Catania S. 2010. Development of a miniature high nique, 62(9):847–863.
speed wireless data acquisition system for geotechni- Merifield, R.S., White, D.J. & Randolph, M.F. 2008a.
cal centrifuges Proc. Int. Conf. on Physical Modelling Analysis of the undrained breakout resistance of
in Geotechnics. Zurich, Switzerland, 229–234. partially embedded pipelines. Géotechnique 58(6),
Gourvenec S.M. & White D.J. 2010. Elastic solutions for 461–470.
consolidation around seabed pipelines. Proc. Offshore Merifield R., White D.J. & Randolph M.F. 2009. The
Technology Conference, Houston. Paper 20554. effect of surface heave on the response of partially-
Gourvenec S.M. & White D.J. 2013. Softening and con- embedded pipelines on clay. ASCE J. Geotechnical &
solidation around seabed pipelines: centrifuge model- Geoenvironmental Engineering, 135(6):819–829
ling. In review. Murff, J.D., Wagner, D.A. & Randolph, M.F. 1989.
Graham, J., Crooks, J.H.A. & Bell, A.L. 1983. Time Pipe penetration in cohesive soil. Géotechnique 39(2),
effects on the stress-strain behaviour of natural soft 213–229.
clays. Géotechnique 33(3): 327–340. Orcina. 2008. OrcaFlex software, Orcina Ltd, Ulverston,
Green, A.P. 1954. The plastic yielding of metal junctions UK.
due to combined shear and pressure, J. Mechanics and Palmer, A. 2009. Touchdown indentation of the seabed,
Physics of Solids, Vol. 2, pp. 197–211. Applied Ocean Research, Vol. 30, No. 3. pp. 235–238.
Hill A., White D.J., Bruton D.A.S., Langford T., Meyer V., Pesce, C.P., Aranha, J.A.P. and Martins, C.A. 1998. The
Jewell R. & Ballard J-C. 2012. A new framework for soil rigidity effect in the touchdown boundary layer of
axial pipe-soil interaction illustrated by a range of a catenary riser: Static problem, Proc. 8th Int. Offshore
marine clay datasets. Proc. Int. Conf. on Offshore Site and Polar Engineering Conf., Montreal, pp. 207–213.
Investigation and Geotechnics. SUT, London. Randolph, M.F., Carter, J.P., & Wroth, C.P. 1979. Driven
Hu, Y. & Randolph, M.F. 1998a. A practical numerical piles in clay—the effects of installation and subse-
approach for large deformation problems in soil. Int. quent consolidation. Géotechnique Vol. 29, No. 4,
J. Numerical and Analytical Meth. Geomech. 22(5): 361–393.
327–350. Randolph, M.F. and White, D.J. 2008a. Upper bound
Hu, Y. & Randolph, M.F. 1998b. H-adaptive FE analy- yield en-velopes for pipelines at shallow embedment in
sis of elastoplastic non-homogeneous soil with large clay. Géotechnique, 58(4):297–301.
deformation. Computers and Geotechnics. 23(1–2):61–83. Randolph, M.F. and White, D.J. 2008b. Pipeline embed-
Jardine, R.J., Chow, F.C., Overy, R. & Standing, J.R. ment in deep water: processes and quantitative
2010. ICP design methods for diven piles in sands and assessment. Proc. Offshore Technology Conference,
clays. Thomas Telford. Houston, USA. Paper OTC19128-PP.
Jas E., O’Brien D., Fricke R., Gillen A., Cheng L. & Randolph M.F. & Gourvenec S.M. 2011. Offshore
White D.J. 2013. Pipeline stability revisited. Journal of Geotechnical Engineering. ISBN 978-0415477444.
Pipeline Engineering. in press. 560 pages. Taylor & Francis.
160
Wang D., White D.J. & Randolph M.F. 2009. Numerical Westgate Z.J., White D.J. & Randolph M.F. 2013.
simulations of dynamic embedment during pipelaying Modelling the embedment process during offshore
on soft clay. Proc. Conf. on Offshore Mechanics and Arc- pipe laying on fine-grained soils. Canadian Geotechni-
tic Engineering, Honolulu. Paper OMAE2009-79199. cal Journal, in press.
Wang, D, White, D.J. & Randolph, M.F. 2010. Large White, D.J. & Hodder, M. 2010. A simple model for the
deformation finite element analysis of pipe penetra- effect on soil strength of remoulding and reconsolida-
tion and large-amplitude lateral displacement. Cana- tion. Canadian Geotechnical Journal, 47: 821–826.
dian Geotechnical Journal, 47(8): 842–856. Zeitoun H.O., Tornes K., Cumming, G., & Branković, M.
Westgate, Z.J., White, D.J. and Randolph, M.F. 2009. 2008. Pipeline stability—state of the art. Proc. 27th
Video ob-servations of dynamic embedment dur- Int. Conf. on Offshore Mechanics and Arctic Engineer-
ing pipelaying on soft clay. Proc. Conf. on Offshore ing, OMAE2008, Estoril, Portugal.
Mechanics and Arctic Engineering, Honolulu. Paper Zhou, H. & Randolph, M.F. 2007. Computational tech-
OMAE2009-79814. niques and shear band development for cylindrical
Westgate, Z.J., Randolph, M.F., White D.J. & Li, S. and spherical penetrometers in strain-softening clay.
2010a. The effect of seastate on as-laid pipeline Int. J. Geomechanics 7(4): 287–295.
embedment: a case study, Applied Ocean Research, Zhou, H. & Randolph, M.F. 2009. Penetration resistance
Vol. 32, pp. 321–331. of cylindrical and spherical penetrometers in rate-
Westgate, Z.J., White, D.J., Randolph, M.F. & Brunning, P. dependent and strain-softening clay. Géotechnique.
2010b. Pipeline laying and embedment in soft fine- 59(2):79–86.
grained soils: field observations and numerical simula- Zienkiewicz, O.C. & Zhu, J.Z. 1992. The superconver-
tions, Proc. Offshore Tech. Conf., Paper OTC20407. gent patch recovery and a posterior error estimates.
Westgate, Z., White, D.J. & Randolph, M.F. 2012. Field Part 1: The recovery technique. International
observations of pipeline embedment in carbonate Journal of Numerical Methods in Engineering, 33(7):
sediments. Géotechnique, 62(9):787–798. 1331–1364.
161
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
ABSTRACT: During installation of offshore anchors and foundations the soil undergoes large defor-
mations. Effects of installation such as penetration resistance, punch-through, changes in contact stresses
and strengths should be accounted for in the design and planning. Typically such effects have been
assessed by semi empirical methods and engineering judgement skills obtained from field and model
testing. Numerical methods for modelling large deformation and penetration problems in geomateri-
als are now available, and although research into new developments and improvements is still ongoing,
these methods can already be used for some design problems. This paper presents and briefly reviews two
modelling approaches: the Arbitrary Lagrangian-Eulerian method and the Coupled Eulerian-Lagrangian
method of Abaqus/Explicit. The CEL method is found to be the most promising of the two and this
method is used to model the penetration of a spudcan foundation. Excellent agreement is obtained with
the penetration resistance and failure mechanism measured in a centrifuge test.
162
Among the techniques proposed used in whereas the soil is discretised with the Eulerian
geotechnical engineering are: Updated Lagrangian elements. The effect of different FE discretisations
(UL), Material Point Method (MPM) and Smooth will be considered in section 3.3.
Particle Hydrodynamics (SPH). The different but The interaction between the pile and the soil is
complementary features of the Lagrangian and modelled as frictionless (smooth).
Eulerian frame suggest that it would be compu- If not otherwise specified, all analyses are
tationally beneficial to combine these two so as conducted with a constant penetration rate of
to strengthen their advantages and to avoid their 0.1 m/s.
disadvantages. This idea has led to the develop- Generally the authors advocate a stepwise mod-
ment of three approaches that apply both the elling approach from a simple 2D/Axisymmetric
Lagrangian and Eulerian formulation: Arbitrary FE-model which preferably can be validated
Lagrangian-Eulerian (ALE), Remeshing and Inter- against known analytical results towards a full 3D
polating Technique with Small Strain (RITSS) and FE-model with the full complexity of soil layering
Coupled Eulerian-Lagrangian (CEL). etc. Table 1 summarises the different LDFE analy-
This paper mainly focuses on application of the ses performed in this section. It’s worth noting that
ALE and CEL methods which are available in the the CEL method implemented in Abaqus/Explicit
commercial finite element program Abaqus. can only be used with the three-dimensional brick
elements.
3 LDFE MODELLING APPROACH
3.2 Effect of boundary
3.1 General description
An ALE axisymmetric FE-model with a radius of
A 5 m long cylindrical “anchor pile” with diameter 10 m (equivalent to 25D) from the centre line and
D = 40 cm and a tapered tip, which is penetrating a depth of 10 m is established to study the bound-
into a weightless uniform clay domain, is mod- ary effect. Figure 1 plots the radial displacements
elled using both the ALE and the CEL methods calculated at two different depths of 4 m and 5 m
in Abaqus/Explicit. During the penetration proc-
ess the clay is assumed to be undrained with an
undrained shear strength su of 5 kPa and rigidity Table 1. Summary of different LDFE analyses of
index G/su of 200. The clay material behaviour is anchor pile.
described by the Tresca constitutive model.
Both axisymmetric and full three-dimensional FE-models Modelling Different effects
FE-models are considered. The objective is to (Abaqus/Explicit) techniques studied
compare the two methods and to present and Axisymmetric ALE Boundary
discuss some modelling issues common to the Axisymmetric ALE Discretisation
LDFE analysis. Owing to axisymmetry, only a Three-dimensional ALE Penetration rate
15°-domain of the 3D problem is modelled so that Three-dimensional CEL CEL versus ALE
the computational time is optimized without caus-
ing excessive skewing of the elements close to axis
of symmetry.
In all the FE-models fully fixity is applied at the
base of the soil domain and roller conditions at its
vertical sides. At the axisymmetric line while the
roller conditions are used in the CEL FE-model,
a zipper-technique (Cudmani & Sturm 2006) is
employed in the ALE FE-model. This technique is
necessary to improve contact between pile and sur-
rounding soil during large penetration. For the 3D
FE-models, the symmetric boundary conditions
are in addition imposed on the two planes of sym-
metry by constraining the two in-plane rotations.
In the ALE FE-models the anchor pile is mod-
elled by analytical rigid elements. The soil domain
is discretised with Lagrangian elements except for
the soil within a depth of 7.5 m and a radius of
3 m from the axisymmetric line where the ALE
elements are applied. In the CEL FE-model, the Figure 1. Effect of boundary on LDFE analysis of
pile is modelled by Lagrangian rigid solid elements anchor pile.
163
versus the radius of the FE-model at the end of soils given by Liyanapathirana (2009) and Yu et al.
the anchor pile installation, i.e. the tip at 5 m deep. (2000). From the results presented in Figure 3, it is
It can be seen that from a radius of 5 m (12.5 D) seen that the coarser meshes give faster run times
the effect of the side boundary is considered to but also less accurate solutions. The mesh density
be insignificant with respect to the soil responses. of 0.1 D below the pile tip is considered to be suffi-
Therefore, the radius of 5 m will be adopted for all cient fine in order to satisfy both the accuracy and
following simulations. the efficiency of the simulations. Thus, it will be
used in the following 3D FE-models.
3.3 Effect of discretisation
3.4 Effect of penetration rate
The effect of discretisation is investigated using an
axisymmetric FE-model with the ALE method. In most real installation problems the rate of pen-
A total of six mesh densities with element size etration is so slow that dynamic effects can be
ranging from 0.05 D to 0.3 D as presented in disregarded and the problem can be regarded as
Figure 2 are generated and run with the same being static. Solving a static problem with Abaqus/
constant penetration velocity of 0.1 m/s. Figure 3 Explicit requires choosing a penetration velocity
plots the calculated end bearing pressures which all that is artificial but which does not introduce iner-
agree fairly well with the analytical solution for the tia effects of significance for the solution. Slower
bearing capacity of a cone penetration in cohesive velocity, however, requires longer computation
time. The aim is to find the highest penetration
rate that gives acceptable accuracy, i.e. gives insig-
nificant inertia.
An ALE 3D FE-model shown in Figure 4(a)
is created. Its geometric dimensions and discreti-
sation are decided based on the calculated results
discussed in sections 3.2 and 3.3. Three installa-
tion rates of 1, 0.75 and 0.5 m/s are considered.
Figure 3 reveals that the corresponding computa-
tion time required for a penetration depth of 5 m is
approximately 59, 74 and 108 minutes, respectively.
Figure 5 plots the calculated end bearing pressures
versus the penetration depth. It is shown that all
Figure 3. Computation times (hours) and calculated Figure 4. Geometry and discretisation of ALE 3D FE-
end bearing pressures for six different discretisations. model (a) and CEL 3D FE-model (b).
164
pile installation problem as used with the ALE FE-
model is re-analysed using the CEL approach. The
comparison results are discussed in next Section.
165
the applicability of the Abaqus CEL method for 85% of the strengths given in Table 2 are applied to
geotechnical large deformation problems. the CEL 3D FE-model.
This section focuses on applying the CEL The contact between the soil and the spudcan
method in Abaqus/Explicit to back-analyse foundation is modelled as smooth. The numeri-
the centrifuge test E2UU-II-T5 carried out by cal penetration rate of 1 m/s is adopted, which
Hossain & Randolph (2010a). In this laboratory is assumed slow enough such that the calculated
test, a spudcan of prototype diameter 6 m was results are not affected by inertia effects.
installed into a stiff clay (Clay 1), with a thick-
ness of 4.5 m, overlaying a soft clay (Clay 2). 4.2 Results
Figure 6 shows a CEL 3D FE-model of the centri-
fuge test and its discretisation. Figure 7 presents a comparison of the numerically
The undrained shear strengths and the rigidity predicted bearing pressure against the centrifuge
indexes of the two clay layers are adopted from the result. The penetration depth d is defined as zero
T-bar test given by Hossain & Randolph (2010a). from the level at which the maximum contact area
The soil behaviour is described by the Tresca consti- of the spudcan is at the original soil surface. Hence
tutive model of which the parameters are detailed in the normalised penetration depth d/D starts from
Table 2. In order to account for the effect of strain negative value. It can be seen from the figure that
softening due to the soil gradually being remoulded the CEL FE-model predicts very well not only
as it flows around the spudcan and the effect of strain the magnitude of the punch-through load (peak
rate, Randolph et al. (2008) suggested to reduce the load) but also the depth where the punch-through
undrained strength by about 15%. Therefore, only failure occurs (i.e. stage A). An excellent agreement
between the numerical and experimental spudcan
responses is observed until the normalised depth
d/D = 1. For penetration depth beyond d/D = 1
the two spudcan responses gradually diverge from
each other. This mismatch could be explained by
the fact that when the spudcan is displaced closer
to the container base at 11.5 m (equivalent to
d/D ≈ 1.92), the bottom boundary effect become
more and more prominent. The bottom bound-
ary in the CEL FE-model is, however, taken deep
enough (d/D = 6.67) so that the spudcan response
is not affected.
Figure 8 shows the deformed soil layers and flow
mechanisms depicted by the velocity fields at three
installation stages: stage A (d/D = 0.07), stage B
(d/D = 0.75) and stage C (d/D = 1.6). The numeri-
cal results are also compared with the centrifuge
Value
166
the two clay layers. Potential punching failure sur-
faces are observed and depicted by the dark red
solid lines. It is recalled that Stage A corresponds
to the peak spudcan resistance (see Fig. 7). In con-
sequence of the punch-through failure, the subse-
quence penetration leads to a significant reduction
of the calculated bearing capacity of the spudcan
foundation. When continuing to displace the spud-
can deeper down till Stage B at which the spud-
can base is at the original stiff/soft clay interface,
the stiff clay plug trapped beneath the spudcan is
gradually reformed in an inverted truncated cone
as shown in Figure 8(b). The underlying soft clay
is correspondingly forced to squeeze outward
because of the inward sidewalls of the inverted
truncated cone containing the stiff clay. The stiff
clay wall above the spudcan edge is seen to remain
standing thanks to its high cohesion. Figure 8(c)
shows that at Stage C, a complete deep penetration
mechanism is revealed. Very good agreement in
terms of the soil back-flow pattern and the cavity
depth are observed from the CEL FE-model and
the centrifuge test model results. It is also found
from Figure 8(c) that at this Stage the numerical
simulation predicts a localize failure mode while
it is a more diffuse type of failure observed in the
laboratory test. The difference of the failure modes
can be attribute to the fact that since the penetra-
tion depth is very close to the container base of the
centrifuge test, the measurement results, especially
beneath the spudcan base, are most likely affected
by the boundary condition. The observation
clarifies further the divergence of the two bearing
pressure results obtained beyond the normalised
penetration depth d/D = 1, which is presented in
Figure 7.
5 CONCLUSIONS
167
rate, the ALE simulation runs slightly faster and Hossain, M.S. & Randolph, M.F. 2010a. Deep-
predicts more accurately the pile bearing pressure penetrating spudcan foundations on layered clays:
than the CEL one. However, the ALE method in Centrifuge tests. Géotechnique 60(3): 157–170.
Abaqus does not support the boundaries between Hossain, M.S. & Randolph, M.F. 2010b. Deep-
penetrating spudcan foundations on layered clays:
different materials as they are considered as non- Numerical analysis. Géotechnique 60(3): 171–184.
adaptive mesh (or Lagrangian) boundaries. There- Hu, Y. & Randolph, M.F. 1998. A practical numerical
fore, in order to back-analyze the centrifuge test approach for large deformation problems in soil. Int J
E2UU-II-T5 of a spudcan foundation penetrating Numer Anal Methods Geomech 22(5): 327–350.
into stiff clay overlying soft clay, which is a rather Liyanapathirana, D.S. 2009. Arbitrary Lagrangian
complex boundary value problem in terms of struc- Eulerian based finite element analysis of cone pen-
tural geometry and layered soils, the CEL method etration in soft clay. Computers and Geotechnics 36:
has been applied instead. Through the compari- 851–860.
son results it was found that except for some later Menzies, D. & Roper, R. 2008. Comparison of jackup rig
spudcan penetration methods in clay. Proc Offshore
stages in the centrifuge test where the measured Technology Conference OTC-19545.
results were most likely affected by the container Qiu, G. & Henke, S. 2011. Controlled installation of
base boundary, the numerical predictions were in spudcan foundations on loose sand overlying weak
excellent agreement with the experimental observa- clay. Marine Structures 24(4): 528–550.
tions. In fact, the CEL 3D FE-model was able to Qiu, G., Henke, S. & Grabe, J. 2011. Applications of a
not only calculate the spudcan foundation resist- Coupled Eulerian-Lagrangian approach on geotechni-
ance but also predict and capture well the failure cal problems involving large deformations. Computers
mechanisms experimentally observed during the and Geotechnics 38(1): 30–39.
penetration process. Randolph, M.F., Wang, D., Zhou, H., Hossain, M.S., &
Hu, Y. 2008. Large deformation finite element analy-
sis for offshore applications. Proc 12th Int Conf of Int
Association for Computer Methods and Advances in
ACKNOWLEDGEMENT Geomechanics 3307–3318. Goa, India.
Tho, K.K., Leung, C.F., Chow, Y.K. &
The research has been performed as a part of Swaddiwudhipong, S. 2012. Eulerian finite-element
“GEO-INSTALL” (Modelling Installation Effects technique for analysis of jack-up spudcan penetration.
in Geotechnical Engineering) project. The work Int J Geomech 12(1): 64–73.
is partially funded by the European Community Wang, C.X. & Carter, J.P. 2002. Deep penetration of
through the program “Marie Curie Industry- strip and circular footings into layered clays. Int J
Geomech 2(2): 205–232.
Academia Partnership and Pathways”, under Yu, H.S., Herrmann, L.R. & Boulanger, R.W. 2000.
Contract No PIAP-GA-2009-230638 and partially Analysis of steady state cone penetration in clay.
supported by the Norwegian Research Council. J Geotech Geoenviron Eng ASCE 126(7): 594–605.
All these supports are gratefully acknowledged. Yu, L., Hu, Y., Liu, J., Randolph, M.F. & Kong, X. 2012.
Numerical study of spudcan penetration in loose sand
overlying clay. Computers and Geotechnics 46: 1–12.
REFERENCES
168
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
ABSTRACT: Monopiles are being used more extensively for offshore wind turbine foundation. Research
into the behaviour of these large open-ended piles generally do not examine the installation effect on
lateral capacity. This paper presents the results of model pile load tests on monopiles in the geotechni-
cal centrifuge. The effect of monotonic pile installation on the subsequent cyclic lateral capacity were
investigated. In order to study these effects, a novel actuator was developed which simultaneously allows
to install the piles in-flight and is able to load the piles laterally without interrupting the test. A series of
initial tests on open-ended monopiles, which show the effectiveness of the setup, are presented. These
initial tests investigated the effect of 1 ⋅ g and N ⋅ g pile installation on the subsequent lateral capacity in
two way cycling for two different initial densities. The preliminary results indicate that N ⋅ g installation
has a small positive effect, which decays with the number of cycles, on the lateral capacity of short stiff
monopiles. Hence, for open-ended model piles the installation effects are negligible. At the present stage,
given the accuracy of the experimental setup, no conclusive results for the initial stiffness response were
gathered. However, tentatively, the conclusions for the lateral capacity also hold for the initial stiffness.
The latter needs to be confirmed in future experiments, where the setup is optimized for the measurement
of the small strain response of the monopile and a large number of loading cycles. These conclusions
are for tests on open-ended model piles only. If for similitude a close-ended model pile will be used the
pile installation stage could be omitted in order to better approach the negligible installation effects as
presented in this paper.
169
piles are significant and should be incorporated in 2.2 Mechanical features
the tests. Unfortunately, the latter is complicated
The aim for the new design of the actuator is to
by the scaling conditions. Especially, in cases where
allow for the simulation of the installation stage,
a closed-ended pile is used instead of an open-
whilst at the same time offering possibility to apply
ended pile, in order to properly scale the lateral
a lateral load on the pile head after installation
stiffness, this could have an over pronounced effect
has fin-ished. The load application of the mecha-
on the resulting lateral capacity. Hence, the effect
nism on the free pile head is designed such that no
of the pile installation on the subsequent lateral
bending moment is transferred to the pile head. To
pile response is still not completely understood.
accomplish this a transversal rod through the top
Current paper presents novel results of the
of the model pile which fits in the template of the
effect of the installation of an open-ended pile
actuator acts as a hinge at the pile head. The pin
in sand on the subsequent cyclic lateral capacity.
only falls into place in the template after the full
A series of geotechnical centrifuge tests were per-
pile installation stage has been accomplished. The
formed to investigate the effect of pile installation
actuator is illustrated in Figure 2. This new actua-
at low stress level (1 ⋅ g) and at large stress level in
tor allows for 106 mm vertical penetration of the
the geotechnical centrifuge (N ⋅ g). The cyclic lat-
model pile and 55 mm lateral displacement, which
eral load tests have in both cases been executed at
is about two times the outer pile diameter (Do is
N ⋅ g. These tests require a novel actuator, capable
27.36 mm). The displacement of the pile head is
of both pile installation and lateral loading with-
measured by registering the absolute position of
out fixing the pile head and without interruption
the actuator. At present no load cell is applied in
of the test.
between the actuator and the pile head to maxi-
mize the penetration length.
The template in the actuator which catches the
2 EXPERIMENTAL SETUP pile is instrumented with strain gauges to meas-
ure the lateral load on the pile head. The loca-
2.1 Geotechnical centrifuge tion of the strain gauges is chosen such that the
The geotechnical centrifuge at Delft University load measurement is insensitive for vertical loads
of Technology is a small beam centrifuge with
a radius of 1.22 m and is equipped with all the
necessary data acquisition and camera facili-
ties for contemporary centrifuge testing (Fig. 1).
The flight computer (ITX computer board with
National Instruments PCI-6221 data acquisition
card), strain gauge amplifiers (custom made 16
channels, 1x-10000x gain, <100 kHz) and motor
control (3 channels DC-servo) are all hosted on
the centrifuge. The communication to the com-
puter is by means of a wireless network link or
wired over the slip rings. The network camera has
its own set of slip rings.
170
and bending moments in the horizontal and ver-
tical plane. After construction of the frame it has
been calibrated for static loads in the laboratory.
However, the complete system as constructed in
the centrifuge, is more extensively calibrated at 1 ⋅ g
and N ⋅ g in order to deal with internal variation and
hysteresis of the stress and strain in the mechanism
resulting from cyclic loading at 1 ⋅ g and N ⋅ g. The
cyclic loading in compression has been calibrated
with a mechanical spring, due to the remaining hys-
teresis in the mechanism, the calibration of the test
data has been done with a typical hysteresis loop
as presented in Figure 3. Practical difficulties only
allowed the frame to be calibrated in compression.
However, the linearity of the static calibration of the
mechanism in tension and compression give confi-
dence to use this cyclic calibration also for tension.
Besides hysteresis also the stiffness of the actuator
is taken into account for the correction of displace-
ment which is measured on the actuator spindle.
The brass model pile, depicted in Figure 4,
is designed to properly scale the lateral bending
stiffness (Alderlieste 2011) and to prevent plugging
during installation (large Do/D50). The maximum
height of the roughness profile (Rz) of the pile Figure 4. Model pile.
is 1.54 μm and the mean deviation of the rough-
ness profile (Ra) is 0.37 μm. The low normalized
roughness (Rn) of 0.007 implies a smooth inter-
face and dilative behavior of the interface shear
band is expected to be low (Paikowsky et al. 1995,
Garnier & König 1998). The strongbox with actua-
tor and relevant physical dimension are depicted in
Figure 5. Dimensions such as eccentricity (e, 0.71 ⋅
Do), distance to vertical boundary (B1, 3.85 ⋅ Do),
distance between bottom and pile tip (B2, 1.3 ⋅ Do),
embedded pile length (Lemb, 5 ⋅ Do) and installed
pile length (A1, 3.9 ⋅ Do) are kept constant during
the experiments.
3 TEST PROCEDURE
171
Table 1. Test execution sequences for different installa- where in each cycle the pile head displacement is act-
tion conditions. ing in two opposing directions, has been applied.
1 ⋅ g installation 48 ⋅ g installation
3.3 Test program
Monotonically jack the Spin up of the centrifuge
pile 106 mm at 0.5 mm/s to 48 ⋅ g The test program was designed such that the dif-
Retract the actuator Monotonically jack the pile ference in lateral cyclic response between 1 ⋅ g and
106 mm at 0.5 mm/s N ⋅ g installation of an open-ended pile can be
Spin up of the centrifuge Retract the actuator identified. The initial density of the sample has
to 48 ⋅ g been varied to gain more insight into the influ-
ence of the initial soil state on the installation
Cyclical loading with a displacement amplitude of 1 mm, effect in subsequent lateral loading. Each test is
a rate of 0.23 mm/s and 200 cycles. performed in a newly prepared sample. Given the
very large pile diameters of the monopiles where
Table 2. Test details.
plugging is not likely to occur, at the present stage
plugging of the open-ended pile should not occur
Test code Installation Id Cycles in the model test either. Preliminary tests show
that even at higher acceleration levels no plugging
T01-80-1 ⋅ g 1⋅g 60 ± 2% 200 did occur.
T02-80-N ⋅ g 48 ⋅ g 60 ± 2% 200
T03-60-N ⋅ g 48 ⋅ g 80 ± 2% 200
T04-60-1 ⋅ g 1⋅g 80 ± 2% 200 4 RESULTS
172
correction of the displacement data for the stiff-
ness of the actuator, the corrected displacement
which is actually applied on the pile head reduces
as the load increases.
In order to better understand the effect of
installation on the lateral pile response the full 1st,
2nd, and 150th loading loops have been presented
in detail in Figures 8–10. Each Figure shows the
same cycle number for each test as function of
the initial relative density and the pile installation
method (1 ⋅ g and 48 ⋅ g). All these cycles have been
corrected for stiffness of the loading mechanism
and hysteresis. For ease of comparison the initial
offset at the beginning of the 150th cycle has been
zeroed.
The results indicate that at the start of the test,
i.e. in the first load cycle (Fig. 8) the differences Figure 9. Cycle 2 for tests presented in Table 2.
between 1 ⋅ g installation and 48 ⋅ g are large. The
negative load at the start of the cycle, for the 48 ⋅ g
test, most probably originate from pre-loading
of the actuator as a result of inclined pile instal-
lation during in flight installation. Although the
Figure 7. First 20 peak loads (black) and displacements maximum loads are quite similar for all tests, the
(gray) for tests presented in Table 2. stiffness of the pile-soil at the end of the primary
displacement do show a higher stiffness for the
piles installed at 48 ⋅ g. Load reversal in backward
direction shows a higher lateral resistance in case
of 48 ⋅ g installation. The latter is still visible for
backward loading direction in the second cycle,
as presented in Figure 9. Figure 10 shows a very
consist response for both 1 ⋅ g and 48 ⋅ g installa-
tion, i.e. the differences in response resulting from
pile installation decay as function of cycle number.
However, the differences between the tests per-
formed at different initial densities remain. Also,
a progressive stiffening of the pile-soil response
becomes apparent when the second cycle within
one test is compared to the 150th cycle.
Typical for all the cycles is the shape of the
loop. It seems that the passive earth pressure in
Figure 8. Cycle 1 for tests presented in Table 2. loading is being mobilized gradually whereas
173
the active earth pressure decreases more rapidly API 2007. Recommended Practice for Planning, Design
after reversal of displacement and thus loading and Constructing Fixed Offshore Platforms—Work-
direction. ing Stress Design. American Petroleum Institute.
ERRATA AND SUPPLEMENT 3, AUGUST
2007.
5 CONCLUSIONS Barton, Y., Fin, W., Pary, R., & Ikuo, T. 1983. Lateral pile
response and p-y curves from centrifuge tests. Offshore
Technology Conference Paper number OTC 4502.
A series of model pile tests on open-ended mono- Bienen, B., Dührkop, J., Grabe, J., Randolph, M., &
pile in medium dense and dense sand have been White, D. 2011. Response of piles with wings to monot-
performed. The effect of pile installation at 1 ⋅ g and onic and cyclic lateral loading in sand. Journal of Geotech-
at 48 ⋅ g (in flight during the centrifuge test) on the nical and Geoenvironmental Engineering 138(3), 364–375.
subsequent cyclic lateral capacity at elevated stress Brant, L. & Ling, H. 2007. Centrifuge modeling of piles
level (48 ⋅ g) in sand have been studied for two dif- subjected to lateral loads. In H.I. Ling, L. Callisto,
ferent initial densities (Id = 60 ± 2% and 80 ± 2%). D. Leshchinsky, J. Koseki, and G.M.L. Gladwell (Eds.),
Similar to previous model pile tests the results Soil Stress-Strain Behavior: Measurement, Modeling
clearly show a progressive stiffening of the lateral and Analysis, Volume 146 of Solid Mechanics and
Its Applications, pp. 895–907. Springer Netherlands.
pile response as function of the number of load 10.1007/978-1-4020-6146-2_68.
cycles. The N ⋅ g installation has a small positive Byrne, B., Leblanc, C., & Houlsby, G. 2010. Response of
effect, which decays with the number of load cycles, stiff piles in sand to long-term cyclic lateral loading.
on the lateral capacity of short stiff monopiles. Gotechnique 60(2), 79–90.
Hence, for open-ended model piles installation Clauss, G., Lehmann, E., & Östergaard, C. 1988.
effects are negligible. At the present stage, given the Meerestechnische Konstruktionen. Springer.
accuracy of the experimental setup, no conclusive Craig, W. 1985. Installation studies for model piles.
results for the initial stiffness response are gathered. Publication of: Balkema (AA).
However, tentatively, the conclusions for the lateral Cuellar, P., Georgi, S., Baeßler, M., & Rücker, W. 2012.
On the quasistatic granular convective flow and sand
capacity also hold for the initial stiffness. The latter densification around pile foundations under cyclic lat-
needs to be confirmed in future experiments, where eral loading. Granular Matter, 11–25.
the setup is optimized for the measurement of the Dijkstra, J. 2009. On the Modelling of Pile Installation.
small strain response of the monopile and a large Ph. D. thesis, Technische Universiteit Delft. ISBN:
number of loading cycles. 9789085704324.
These conclusions are for tests on open-ended DNV 2011. Offshore standard dnv-os-j101—design of
model piles only. If for similitude a closed-ended offshore wind turbine structures.
model pile will be used the pile installation stage Dührkop, J. & Grabe, J. 2008. Monopilegrundungen
could be omitted in order to better approach the neg- von offshore-windenergieanlagen–zum einfluss
einer veränderlichen zyklischen lastangriffsrichtung.
ligible installation effects as presented in this paper. Bautechnik 85(5), 317–321.
Dyson, G. & Randolph, M. 2001. Monotonic lateral load-
ing of piles in calcareous sand. Journal of Geotechnical
ACKNOWLEDGEMENTS and Geoenvironmental Engineering 127, 346–352.
Garnier, J. & König, D. 1998. Scale effects in piles and
The first Author of this paper acknowledges the nails loading tests in sand. In Centrifuge 98.
received financial support from Deltares for pre- GL 2005. Guideline for the certification of offshore wind
paring the manuscript. turbines.
Klinkvort, R., Leth, C., & Hededal, O. 2010. Centrifuge
modelling of a laterally cyclic loaded pile. Physical
REFERENCES Modelling in Geotechnics, 959–964.
Li, Z., Haigh, S.K., & Bolton, M.D. 2010. Centrifuge
Achmus, M., Abdel-Rahman, K., & Kuo, Y. 2007. modelling of mono-pile under cyclic lateral loads.
Numerical modelling of large diameter steel piles under 7th International Conference on Physical Modelling in
monotonic and cyclic horizontal loading. In Tenth Geotechnics 2, 965–970.
International Symposium on Numerical Models in Geo- Murchison, J. & O’Neill, M. 1983. An Evaluation of P-y
mechanics, pp. 453–459. Taylor & Francis London. Relationships in Sands. Report (American Petroleum
Alderlieste, E. 2011. Experimental modelling of lateral Institute). University of Houston-University Park.
loads on large diameter mono-pile foundations in Oldham, D. 1985. Experiments with lateral loading of
sand. M.sc. thesis, Delft University of Technology. single piles in sand. Publication of: Balkema (AA).
Alderlieste, E.A., Dijkstra, J., & van Tol, A.F. 2011. Paikowsky, S., Player, C., & Connors, P. 1995. A dual
Experimental investigation into pile diameter effects interface apparatus for testing unrestricted friction of
of laterally loaded mono-piles. ASME Conference soil along solid surfaces. ASTM Geotechnical Testing
Proceedings 2011(44397), 985–990. Journal 18(2), 168–193.
174
Peng, J., Clarke, B., & Rouainia, M. 2011. Increasing Vattenfall 2008. Kriegers flak offshore wind farm-design
the resistance of piles subject to cyclic lateral load- basis foundations. Technical report, Vattenfall Vind-
ing. Journal of Geotechnical and Geoenvironmental kracft AB.
Engineering 137(10), 977–982. White, D.J. & Lehane, B. 2004. Friction fatigue on dis-
Reese, L., Cox, W., & Koop, F. 1974. Analysis of laterally placement piles in sand. Geotechnique 54, 645–658.
loaded piles in sand. Offshore Technology Conference
OTC 2080, 473–486.
Rosquoet, F., Thorel, L., Garnier, J., & Canepa, Y. 2007.
Lateral cyclic loading of sand-installed piles. Soils and
Foundations 47(5), 821–832.
175
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
P.A. Vermeer
Deltares, Delft, The Netherlands
Institute of Geotechnics, Stuttgart, Germany
ABSTRACT: In a wide spectrum of geotechnical applications, materials undergo large deformations and/
or large displacements. On modeling these problems with a Lagrangian finite element method, the mesh
can become too distorted and re-meshing is essential. In the past decades, considerable efforts have been
made to adopt what is called meshfree methods to mitigate the problems related to mesh distortion. One
of these methods is the Material Point Method (MPM) that represents the continuum field as Lagrangian
material points (particles), which can move through the fixed background of an Eulerian mesh.
The objective of this paper is to formulate and validate a coupled FEM-MPM approach for the numer-
ical simulation of large deformation of a membrane containing soil e.g. dumping of a geocontainer. In
this approach the membrane is discretised by a surface mesh with accurate computation of the membrane
stresses. This membrane mesh is free to move through a 3D mesh of non-structured tetrahedral elements.
Furthermore, the proposed approach is applied successfully to model geotextile of a geocontainer being
released from a split barge. Frictional contact is defined between the geotextile and the barge. On the other
side, rough contact is assumed between the geotextile and the Mohr-Coulomb soil type inside.
176
the 1960s. Sulsky & Schreyer (1993) adopted the rately over the membrane using surface discretisa-
PIC for solid mechanics and gave it later the name tion. Due to deformation, this mesh would move
Material Point Method (MPM). MPM has been through the 3D mesh producing stresses in local
successfully used to model several problems related plane. Then, the in-plane stresses are integrated
to geotechnical applications, such as the discharge accurately to have the membrane forces acting on
of granular material from a silo (Wieckowski et al. the membrane nodes. Finally, the computed forces
1999) and pile hammering (Jassim et al. 2011). are mapped on the original tetrahedral mesh, where
York (1997) developed MPM for the modeling momentum equation is solved, by the location of
of thin membranes, in which a 2D membrane is the considered membrane node.
discretised by a collection of material points on its As in classical finite elements, MPM formu-
surface. lation starts with the weak formulation of the
By definition, the membrane is a thin-walled momentum equation as formulated in Section 3.
structure that has stiffness in its plane and no stiff- Subsequently spatial and time integration form
ness in bending. In addition, the stress over thick- the general solution scheme of the method. Sec-
ness is assumed to be constant. The condition of tion 4 details the representation of the membrane
constant stress through the thickness of the mem- using a new approach. Furthermore application of
brane is automatically enforced by using a single the new approach on a real problem of a releas-
layer of material points as shown in Figure 1. In this ing geocontainer from barge is investigated in the
approach, standard MPM procedure is adopted Section 5. Finally, the conclusion and outlook is
except that the stresses of the membrane material summarized in the last section 6.
points are computed on local coordinate (ξ, η, ς)
where a plane-stress condition is satisfied, rather
than compute it on global (x, y, z) coordinate. 3 BRIEF REVIEW OF MPM
Fernandes (2004) shows that linear shape func-
tions may result in unphysical stress oscillations In MPM the continuum body is discretised by
when a material point crosses from one element to Lagrangian material points as shown in Figure 2.
another. Furthermore, the accuracy of the integra- The momentum equation is solved on the back-
tion depends mainly on the considered number of ground Eu-lerian mesh which provides a conven-
material points inside the element. As an improve- ient means of calculating discrete derivatives.
ment, (Zhang et al. 2011) proposed a hybrid finite
element-material point (FEMP) method for mod-
eling reinforced concrete being subjected to impact 3.1 Spatial discretisation
loading. The essential of this method is to intro- The conservation of linear momentum is given by
duce a hybrid bar element into MPM, where the
nodal variables are updated from background u ∇ + g (1)
nodal displacements of the grid and the stresses
are updated on the element. where σ (x, t) is the Cauchy stress tensor at posi-
Similar to the above FEM/MPM combination tion x and time t, (x, t) is the mass density, g (x, t)
scheme, the membrane can be introduced accu- is the gravitational acceleration vector, u (x, t) is
rately in a new computational framework that is the displacement with the superposed dot denoting
valid for low order tetrahedral mesh. To have such differentiation with time.
a scheme, the integration should be done accu-
Figure 1. 3D membrane with its MPM representation. Figure 2. Continuum body represented by MPM.
177
By taking the virtual displacement δu as test np
function for a domain of volume V surrounded
by boundary S, the weak form of the momentum
mi ∑ mpN (xp ) (7)
p =1
equation can be written as
Referring to Equation 4, the internal force vec-
∫V δ ∫V δδεε dV ∫V δ u
T T T
dV ggdV tor is given by,
+ ∫S δ uT tdS (2)
t np
where np is the number of material points and where Δt is the current time increment, at and
mp is the mass of material point p at location xp. at Δt are the nodal velocities at time t and t + Δt,
However, in practice, the lumped mass matrix is respectively. The incremental nodal displacement is
preferred over the consistent mass matrix. This obtained by integrating the nodal velocity by the
simplifies the computations at the expense of Euler-backward rule (Jassim et al. 2011)
introducing a slight amount of numerical dissipa- Δt Δt
tion (Burgess et al. 1992). The lumped mass matrix at tat (11)
Ml is the diagonal matrix. The structure of the
lumped mass matrix is as follows The positions of the particles are subsequently
updated from
⎡ m1 0 0 ⎤
Δt Δt
⎢0 m … 0 ⎥ xpt xpt + N p at (12)
Ml = ⎢ ⎥
2
(6)
⎢ ⎥
⎢ ⎥ where xpt and xpt Δt are the particle positions at time
⎣0 0 … mn ⎦ t and (t + Δt) respectively.
For the present MPM solution procedure, a
where n is the total number of degrees of freedom slightly different algorithm has been adopted for
which could increase or decrease within the calcu- updating the particles velocity following Sulsky
lation as particles move between elements. How- et al. (1995).
ever, each component of the diagonal term (mi) in By solving the equation of motion for the nodes,
Equation 6 being the corresponding row sum of the elements deform and the material points in the
the consistent matrix is performed as interior of the element move in proportion to the
178
motion of the nodes, based on the nodal basis presented for a rectangular element with the rein-
functions. The position of the material points is forcement of a bar/membrane element embedded
updated using a single-valued continuous velocity through the width as shown in Figure 3. In the ini-
field and, hence the interpenetration of material is tial configuration of this figure, an external force is
precluded. This automatic feature of the algorithm applied on top. As a result, the membrane particles
allows simulations of no-slip contact between dif- would follow the deformations kinematically pro-
ferent bodies without the need for special interface ducing in-line strain εξ which in turns produces the
tracking and contact algorithms. stress σξ. Accordingly, this stress is integrated along
After getting the grid node velocities, the strain the length to get the membrane forces acting on
increment Δεp of the material point p at time opposite direction to the deformation. Lastly, the
(t + Δt) is calculated from, internal membrane force is transferred to the original
mesh by weighting taking into account the location
Δ Δt
Δ εpt Δt B ( x p )aet (13) of the computational nodes relative to the membrane
nodes.
where aet Δt is the vector of nodal velocities of
the element e which the material point belongs 4.1 Membrane formulation with accurate
to. Based on the calculated strain increment, the integration
stress increment Δσ pt Δt at each material point
In the new membrane representation, the surface of
is updated by applying the appropriate constitu-
the membrane is discretised by 2D triangular mesh
tive model. The constitutive models are applied at
as shown in Figure 4. The mesh connectivity is main-
material points and this allows easy evaluation and
tained during the simulation where the membrane
tracking of history-dependent variables. It also
nodes are free to move through the 3D mesh. The
allows computations with more than one material,
displacement of these nodes is described by the solu-
since each material point retains its material prop-
tion of the momentum equation on the 3D compu-
erties throughout the computation, for example in
tational mesh, but the increment in membrane strain
the modeling of soil/membrane interaction.
and stress is based on the deformation of the 2D
As the mass of the material point mp is kept
triangular mesh. The membrane strains and stresses
constant during the calculation, the density of the
are calculated at the integration point as shown in
material is implicitly updated by updating the vol-
Figure 4. As long as 3-noded triangular element is
ume of the material point through
adopted in this scheme, the location of the integra-
Δt t Δt tion point could be anywhere inside the triangle and
Vpt Vpt ( + Δ v p ) (14) does not need to correspond to the Gaussian point.
In standard MPM integration approach the
where Δεv,p is the incremental volumetric strain of internal force vector is calculated using the strain-
the material point p computed from the incremen- displacement matrix of the 3D element and the
tal strain components (Δεv = Δεx + Δεy + Δεz) in integration is performed by summing over material
(x, y, z) coordinates. point volumes (Equation 8). In the new approach,
At the end of time (t + Δt) all the material point the strain-displacement matrix is based on the 2D
variables are updated and a new cycle is begun linear triangular element and integration point is
using the information carried by the material used to obtain the internal force vector on the 2D
points to initialise nodal values on the grid. Note mesh. The internal force is then mapped from the 2D
that at this stage, a new grid can be defined since grid nodes to the 3D grid nodes where the momen-
all the state variables are carried by the material
points. In practice however, the most efficient way
is to keep the original grid.
4 MPM REPRESENTATION
OF A MEMBRANE
179
(Equation 17) and the derivates of the shape func-
tions. At this stage, the matrix β is transformed to
local coordinate βl using the transformation ten-
sor T which is defined by the directional cosines
as given by (Bathe 1996). The tensor T should be
adjusted to satisfy the membrane theory, i.e. the
out-of-plane stresses are zeros, which is done by
eliminating rows three, five and six. The reduced
rotation tensor Tl then takes the form,
T
Figure 4. Membrane discretisation using 2D triangular ⎡ tx2ξ tx2η 2txξ txη ⎤
element. ⎢ ⎥
⎢ ty2ξ ty2η 2tyξ tyη ⎥
⎢ ⎥
l ⎢ tzξ ⎥
2
tum Equation 4 is solved. The mapping is based on tz2η 2tzξ tzη
the shape functions of the 3D mesh and not on the T ⎢ ⎥ (18)
gradient of the 3D shape functions. Hence, there is ⎢txξ tyξ tx tyη t ξ tyη + txηtyξ ⎥
⎢ ⎥
no change in the sign of the material points contri- ⎢tyξ t ξ tyηtzη tyξ t η + tyηt ξ ⎥
bution to the internal force vector when the point ⎢t t tzηt tzξ t η + tzηtxξ ⎥⎦
crosses an element boundary in the 3D mesh. Apart ⎣ zξ ξ η
from the membrane, the other materials are mod-
eled using the standard MPM procedure. where txξ is the directional cosine between x and ξ.
The general formulation for 2D elements in 3D The local strain-displacement matrix βl is given by
space with three translational degrees-of-freedom
per node were used to model the membrane. The l l
local coordinate system of an element is shown in 4. (19)
The displacement field v is approximated in terms
of interpolation functions H and nodal displace- Subsequently, the incremental local strain Δεt+Δt
ments d, v = H d. Hence, according to Bathe (1996) of the membrane element is calculated from
the x-component of the nodal displacements is
Δε l t +ΔΔt t Δt
Δ β l am with ε l [εξξ εηη εξη ]T (20)
nn nn
ζ
x(ξ ,η,ζ ) ∑ Hi xi ∑ hi Hi nxi
2 i =1
(15) t Δt is the nodal velocities vector of the
where a m
i =1
membrane element m. This vector is updated con-
where, Hi is the interpolation functions of node i tinuously by finite element interpolation of the 3D
represented at location (ξ, η, ς), nn is the number nodal velocity a t Δt .
of nodes per element, h is the membrane thick- Afterward, the membrane constitutive model
ness, and nx is the x-component of the unit normal is applied to compute the incremental local stress
n. For specific application of 3-noded elements tensor Δσl at the integration point of the triangular
with constant membrane thickness h, Equation 15 element,
reduces to,
l t +Δ
Δ Δt
Δσ Δε l ,t , (21)
nn
ζ hhn
x(ξ ,η,ζ ) ∑ Hi xi + 2 x (16)
where D is the constitutive tensor given for linear
i =1
elastic material by,
The Jacobian matrix J which involves the deriv-
atives of global coordinates (x, y, z) with respect to ⎡( − v ) v 0 ⎤
the local coordinates (ξ, η, ς) can be constructed D=
E ⎢ v ( − v) 0 ⎥⎥ (22)
as follows ( + v )( − v ) ⎢
⎢⎣ 0 0 ( − v ) ⎥⎦
∂ ∂
= Jiij where iand j = 1, 2, 3 (17) The total stresses tensor at the integration point
∂ξi ∂x j
σl is then updated using the increment given by
Equation 21
The global strain-displacement matrix β of the
membrane element can be formed in the usual
σ l t Δt σ l t σ l ,t Δt
(23)
way of using the inverse of the Jacobian matrix
180
The membrane force F2D is calculated on the
membrane mesh, using standard finite element
integration,
F 2D = ∫ A β l ,T σ l h dA
d (24)
nm
f i3 D ∑ i
m
f m2 D (25)
m =1
5 APPLICATION: RELEASING
GEOCONTAINER FROM BARGE
181
lifted up by the barge. On the contrary, the center
part of the geocontainer will move downward as it
loses the support underneath. Hence, the deforma-
tion pattern of the soil will take a circular shape,
whilst its center does not have any deformation as
shown in Figure 7 for an angle of opening θ = 14°.
As the barge opening increases, tensile forces in
the geotextile (TGT) increase as the membrane keeps
the soil together. Eventually, these forces pull the
soil away from the barge which in turn decreases
the frictional forces between the geotextile and the
barge. As a result, the geocontainer will lose con-
tact gradually with increasing angle of opening.
The top left part of Figure 9 shows the vertical dis-
placement of the very bottom point as a function
of the opening angle. The last mark in this curve
represents the state where there is not equilibrium Figure 8. Principal stress directions showing the arch-
state possible anymore which is corresponds to ing effect.
θ = 14°. For this state, the total displacement of the
soil is already shown in Figure 7.
The stress field inside the soil is redistributed
during loading; showing the effect of arching as
illustrated in Figure 8. During this process, the
horizontal force (Fh) increases gradually and simul-
taneously with the geotextile tensile forces (FGT) as
shown in Figure 9. Holding quasi-static equilib-
rium, the difference between these curves increases
with loading, however, this is due to the loss con-
tact during the releasing shown in Figure 6. Since
the frictional force (μ Fn) and the normal force to
the barge (Fn) decrease, the term (Fh − TGT ) would
increase to insure equilibrium state.
The last part of this analysis is done to simulate
the geocontainer beyond quasi-static equilibrium
state. Accordingly, the opening angle increase fur-
ther by opening the barge continuously. Similar to
the previous quasi-static solution phase, the open-
ing is done slowly to reduce the dynamic effects. As
182
Allowing water to penetrate into the geocontainer
and including the effect of water drag would be
more attractive for real geocontainer applications.
Currently, modeling free water with MPM is under
development.
ACKNOWLEDGMENT
REFERENCES
183
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
D.A. Kort
Norwegian Geotechnical Institute, Oslo, Norway
S. Raymackers
GeoSea NV, Zwijndrecht, Belgium
H. Hofstede
GustoMSC B.V., Schiedam, The Netherlands
V. Meyer
Norwegian Geotechnical Institute, Oslo, Norway
ABSTRACT: Leg penetration assessments were performed for two self-elevating units with tubular legs
operating at 21 locations within an offshore wind farm. The assessments were performed on the basis
of available CPT data, experience-based soil parameters and the calculation methods recommended by
SNAME T&RB 5-5A. In general the soil conditions are dense sand over stiff clay with strong variations
of the sand layer thickness across the wind farm. Initially at some locations the assessment indicated a
significant risk of punch-through, which could potentially lead to problems with insufficient leg length
and create difficulties for leg extraction after installation of the wind turbines. To reduce the uncertain-
ties of the leg penetration assessments, preload trials were performed at three critical locations. The three
preload trials provided a useful basis to revise the input soil parameters and thereby enhance the leg pen-
etration predictions. No indications of large penetrations or rapid penetrations were encountered during
the offshore field campaign.
1 INTRODUCTION
184
are presented. The methodology of the leg penetra- 2.2 Available geotechnical data
tion assessments was then used to predict leg pen-
The available geotechnical data consisted of one
etrations at other locations at the Thornton Bank
CPTU at each windturbine location. Sampling
wind farm. The third part of this paper compares
was performed at selected locations and index and
predicted leg penetrations to field measurements
advanced laboratory testing (triaxial testing and
from 21 different locations.
DSS testing) were performed on samples from the
underlying clay layer.
2 BASIS FOR THE LEG PENETRATION
ASSESSMENTS 2.3 Dredging activities
Dredging activities had taken place at the windtur-
Basis for the leg penetration assessments include
bine locations after the site investigations, resulting
the leg specifications and the available geotechnical
in updated reference seabed levels.
data. Relevant data are briefly summarised below.
Neptune Vagant
185
Figure 4. Interpreted soil profile at Trial 2 location. Figure 6. Predicted and measured penetrations at Trial 1.
Figure 5. Interpreted soil profile at Trial 3 location. Figure 7. Predicted and measured penetrations at Trial 2.
relative density (Dr) interpreted from the CPT conical shape of the leg tip, penetration resistance
data, and simplified profiles used in the penetra- is only built-up when the soil becomes in contact
tion analyses. The profiles are related to the water with the end bearing plate of the pile (Figure 2).
depth in TAW (Tweede Algemene Waterpassing). The applied foundation load is composed of the
The leg penetration depths presented are buoyant weight of the leg and the load applied to
related to the leg tip. However, due to the open the leg.
186
strength profile for the clay layers. As the friction
angles are assessed from correlations with relative
density, assessment of the relative density is also
required. The significant soil parameters for the
bearing capacity analyses include relative density,
friction angles and undrained shear strength pro-
files; soil parameter assessment is discussed below.
4.1.1 Assessment of relative density
A best estimate of the relative density was assessed
from the corrected CPT data using the method of
Baldi et al. (1986). The CPT data was corrected
for in-situ stress relief due to dredging following
the procedure for normalising the cone resistance
given by Gregg (2007), which includes some minor
updates for interpretation of silt-sand mixtures
and clay-silt mixtures compared to the procedure
developed by Robertson & Wride (1998).
Although the Robertson & Wride (1998) proce-
dure for normalising cone resistance was originally
developed for liquefaction analyses, the method is
widely supported and frequently used world-wide.
The stress-corrected cone resistance due to dredging
was significantly lower than the original cone resist-
Figure 8. Predicted and measured penetrations at Trial 3. ance in the upper metre but the effect on the calcu-
lated relative density appeared to be insignificant.
To account for a possible loosening of the sea-
The results of the jacking trials could be con-
bed due to dredging and other effects (such as tidal
sidered as an additional soil investigation on the
current), a maximum relative density of 80% was
basis of which the soil input parameters could be
assumed in the upper 0.5 metre.
revised. Because none of the jacking trials resulted
in rapid penetrations and the measured leg penetra- 4.1.2 Assessment of friction angles
tions appeared to be relatively low compared to the In the present method the relevant friction angles to
predictions, it was obvious that the assumed fric- be considered for very dense sand are peak friction
tion angles of the upper sand could be increased, angles, φ′p, and constant volume friction angles,
as presented in Section 4. φ′cv, from drained triaxial testing in sand. Drained
peak and constant volume friction angles are con-
sidered applicable for the following reasons:
4 METHODOLOGY OF THE LEG
– The process of leg penetration may involve very
PENETRATION ANALYSES
large shear strains (say 100–1000%) and shear
strain rates in the soil underneath the leg tip.
The leg penetration analyses have been performed
Experimental data show that the density of very
in accordance with the industry standard guidelines
dense sand around a penetrating pile tip will
published by SNAME (2008) for spudcan penetra-
decrease during penetration to a certain criti-
tion analyses. The method is based on conventional
cal density state (Dijkstra, 2009). In that case
bearing capacity analyses. Punch-through—dense
the sand underneath the pile tip will behave in a
sand over soft clay—has been analysed using the
drained manner.
method described in the commentary of SNAME.
– Peak values of the drained friction angles are
The leg penetration analyses consisted of two
associated with shear strains of order of mag-
main activities:
nitude 1% and constant volume drained fric-
1. Assessment of soil parameters. tion angles are applicable from say 10% shear
2. Leg penetration calculations. strain and more. Hence, peak values of the
drained friction angles may be applicable in
the outcropping zone of the failure mechanism
4.1 Assessment of soil parameters
whereas constant volume friction angles may
Soil parameters needed for the leg penetration be applicable underneath the leg tip. This state-
analyses include unit weight, equivalent friction ment is supported by the simplified FE analysis
angle for the sand layers and undrained shear in Figures 9–11.
187
– Appendix E of ISO 19905 (2012) proposes use
of critical state friction angles (constant volume
friction angles) increasing by up to 5% with
increasing relative density on the basis of back-
analyses of field penetration records and cen-
trifuge tests.
The weighting between peak and constant vol-
ume friction angles is based on a simplified back
analysis of the field measurement from Trial 2
(Section 4) using Plaxis (2008).
Figure 9 presents a detail of an FE mesh show-
ing the most relevant geometry and soil parameter
input and the calculated principal stresses. The
model is based on a 3.5 m diameter pile with a rigid
base at 2.1 m depth, which is a simplified represen-
tation of Trial 2 at 2.85 m.
The FE result in Figure 10 confirms that rela-
tively large shear strains below the leg tip may be
expected, and thus applicability of constant vol-
Figure 9. Model detail of a simplified bearing capacity ume friction angles, and that relatively small shear
FEM analysis of a tubular leg. strains outside the leg tip may be expected, and
thus applicability of peak volume friction angles.
Figure 11 shows three load-displacement
curves:
– Full end bearing: φ′ = 40.4º is taken underneath
and outside the leg tip
– Reduced end bearing: φ′ = 34.0º is taken under-
neath the leg tip and φ′ = 40.4º outside the leg
tip
– Equivalent friction angle: φ′ = 37.5º is taken
underneath and outside the leg tip.
Figure 11 shows that the reduced end bearing
capacity can be assessed with equivalent friction
angles as defined in Figure 9. Based on the simpli-
fied model it is suggested to assess the equivalent
friction angle by taking 45% of the friction angle
underneath the leg tip and 55% of the friction
angle away from the leg tip.
Figure 10. Typical shear strain contour diagram at
failure.
Low estimate and high estimate friction angles
used in the calculation are therefore based on a
weighted combination of peak friction angles and
constant volume friction angles:
φ ′ LE φ ′ cv , LE + 0.55 ⋅ φ ′ cv , HE (1)
φ ′ HE φ ′ cv , HE + 0.55 ⋅ φ ′ p BE (2)
188
from shallower penetrations). On the other hand, 4.2 Leg penetration calculations
the high estimate parameters are based on con-
The leg penetration calculations were performed
stant volume friction angles for in situ consolida-
for the characteristic foundation loads given in
tion stress conditions underneath the leg tip and
Table 1. Low and high estimated characteristic soil
on peak friction angles for in situ stress conditions
strength parameters have been selected to evaluate
outside the leg. This refers to the case where the
the range in leg penetration for a given location.
soil underneath the leg tip is experiencing low
The punch-through—dense sand over soft
consolidation pressure prior to loading to fail-
clay—analyses require input of a load spread fac-
ure and the soil outside the leg tip is experiencing
tor (Fig. 13).
small shear strains (no disturbance from shallower
A load spread factor of n = 2.5 is used, which
penetrations).
seems to be justified by the FEM analyses pre-
Figure 12 provides guidance on the assessment
sented in Figure 10. A load spread of 1:2.5 is out-
of the various friction angles from the relative
side the 1:3 to 1:5 range recommended in SNAME
density, which are based on the empirical database
T&RB 5-5A and ISO 19905-1, but can be justified
from Andersen & Schjetne (2012).
for these locations on the basis that:
4.1.3 Assessment of undrained shear strength 1. Trial installations have been undertaken at these
in clay locations on which the analyses have been sub-
The undrained shear strength for triaxial com- sequently calibrated.
pression in clay was assessed from the CPT data 2. A load spread of 1:3 to 1:5 has been reported,
using suC = (qt − σv0)/Nkt as presented by e.g., under certain conditions, to under-predict peak
Lunne et al. (1997). Nkt ranging from 15 to 20 punch-through loads for both centrifuge tests
was concluded for the purpose of the leg penetra- (Teh et al., 2010; Lee, 2009) and from back-
tion assessments at the Thornton Bank site on analysis of field data (Baglioni et al. 1982).
the basis of a study of the available laboratory Note, however, that this is not always the case.
data and literature. 3. A load-spread of 1:2.5 is within the range sug-
The deep clay layer was described as highly plas- gested in DNV Classification Note 30.4 (DNV,
tic with plasticity index Ip > 100% and on the basis 1992) for spudcan penetration analysis (1:2 to
of Bjerrum (1973) the undrained shear strength of 1:3).
the clay is assumed isotropic. Hence, the average
As the bearing capacity formulae and empirical
undrained shear strength accounting for strength
relationships given in by SNAME (2008) have been
anisotropy is suavg/suC = (suC+suDSS+suE)/(3 suC) = 1.0.
derived specifically for spudcans, the contribution
Here is suC = undrained shear strength measured
in triaxial compression; suDSS = undrained shear
strength measured in direct simple shear; suE = und-
rained shear strength measured in triaxial exten-
sion (Bjerrum 1973).
Figure 12. Assessment of friction angles from relative Figure 13. Load spread method according to SNAME
density. (2008).
189
of the leg shaft friction capacity has been added to field campaign, predictions of leg extractions have
the calculated penetration resistance. The leg shaft not been included in this paper.
friction capacity has been determined using
5 PENETRATIONS DURING WINDFARM
Fs ∫ s Odz (3) INSTALLATION
190
ACKNOWLEDGEMENT
REFERENCES
191
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
M. Lloret-Cabot
Centre for Geotechnical and Materials Modelling, The University of Newcastle, Newcastle, Australia
Department of Geoscience and Engineering, Delft University of Technology, Delft, The Netherlands
ABSTRACT: Artificial sand islands were constructed in the Canadian Beaufort Sea for use as
hydrocarbon exploration platforms in the 1970s and 1980s. For some of these islands, extensive Cone
Penetration Test (CPT) data are available for characterising the hydraulically placed sand during and after
the construction process. Tarsiut P-45 was the first island using the ‘Molikpaq’ concept, which consisted
of a mobile arctic caisson system to provide the temporary structure for the exploitation. Two main sand
fills were constructed: (a) a sandfill berm on which the caisson system was founded; and (b) the body of
the island structure (island core). This paper presents an investigation of the variability of the sand in the
berm in terms of the vertical and horizontal scales of fluctuation. This geo-statistical investigation is car-
ried out using CPT data from the berm before and after the founding of the caisson system, and sets the
basis for a preliminary discussion on the potential soil variability changes caused by the installation and
infilling of the caisson structure when placed on the berm.
192
3.1 Scale of fluctuation
Based on the concept of the variance function
discussed in Vanmarcke (1977), Wickremesinghe &
Campanella (1993) proposed an alternative strategy
to estimate θv which has since been used in a number
of studies (Wong 2004, Hicks & Onisiphorou
2005, Lloret-Cabot et al. 2012). Their strategy is
very convenient for analysing the vertical soil vari-
ability, not only because the method can be eas-
ily automated (Wickremesinghe 1989) but also
Figure 1. Side view of Tarsiut P-45 (Wong 2004). because many data points are typically available
in the vertical direction of a CPT. In addition, in
the current study, it has been considered useful to
also include results using an alternative method to
islands was sand, which was typically dredged
allow for comparison. Both approaches are out-
from local borrow pits.
lined in the following paragraphs.
The island of Tarsiut P-45 was constructed
during 1984 in a water depth of 25.5 m. The
3.1.1 Approach A
berm was constructed from the seabed up to
Approach A assumes that the available data are
19.5 m below Mean Sea Level (MSL) using a
statistically homogeneous (or stationary). The nec-
bottom dumped placement method. The caisson
essary requirements for a function to be statistically
was then positioned on the constructed berm
homogeneous are: (a) constant mean and constant
and the core was filled-in with sand using a pipe-
standard deviation; (b) an autocorrelation func-
line placement technique. The sand specifica-
tion independent of the location and dependant
tion for the berm and core was for a D50 greater
only on the separation distance (or lag distance,
than 300 μm and a silt content less than 4%
τ). For geotechnical engineering purposes, this is
(Wong 2004). Figure 1 illustrates a cross-section
equivalent to saying that the mean and standard
through Tarsiut P-45 in which the two sand fills
deviation do not spatially vary and that the cor-
are shown.
relation between property values at two different
locations is only a function of their separation dis-
tance. A constant mean can be obtained after de-
3 STATISTICAL EVALUATION
trending the data as illustrated in Figure 2b, where
the identified linear trend of qc with depth (Fig. 2a)
Two analyses are included in this study. In the
has been removed from the data. The approach of
first analysis, the cone tip resistances qc from
removing a linear trend of qc with depth from the
four CPTs are used to estimate the vertical and
horizontal scales of fluctuation of the berm prior
to set-down of the caisson. The second analysis
uses six CPTs for the calculation of the scales of
fluctuation of the berm after core infilling. All
the CPT data comprise tip resistance readings
at 2 cm intervals. In each analysis two comple-
mentary methods are used to determine the verti-
cal scale of fluctuation θv. The first of these two
methods will be referred to as Approach A, and
uses the strategy proposed by Wickremesinghe &
Campanella (1993). The second method will be
referred to as Approach B and estimates the ver-
tical scale of fluctuation by best fitting the the-
oretical correlation model to the experimental
correlation model. Due to the limited data avail-
able, Approach B is the only technique used for
the estimation of the horizontal scale of fluctua-
tion θh. A brief description of each approach is
given below, further details can be found else-
where (Wickremesinghe & Campanella 1993,
Baecher & Christian 2003, Wackernagel 2003, Figure 2. Example of a CPT profile: (a) original data,
Fenton & Griffiths 2008). (b) detrended data, (c) normalised data.
193
data, provides a useful approximation that has Note that various alternative expressions for
been applied in a number of prior studies (Uzielli the correlation model are also possible (Fenton &
et al. 2005, Hicks and Onisiphorou 2005, Lloret Griffiths 2008) but, for simplicity, only the expo-
et al. 2012). A constant standard deviation and the nential correlation model is used in this paper, i.e.
condition (b) stated above, are likely to be reached
if the data are extracted from the same soil layer, as ⎧ −2 τ ⎫
approximate uniform fluctuations are likely to be ρ ( τ ) = exp ⎨ ⎬ (3)
observed if the analysed data are contained within ⎩ θ ⎭
the same soil type layer (Phoon & Kulhawy 1999).
Note that this is a reasonable assumption for the where θ is the scale of fluctuation in the vertical or
case studied here, as all the berm sand was dredged horizontal direction. The experimental correlation
from the same location and deposited in the same function is given by
manner (Wong 2004).
n − j +1
Once the linear depth trend of the soil layer 1
ρˆ (jΔτ ) = ∑ (X i − μˆ )(X i + j − μˆ ) (4)
investigated has been removed, the next stage in σˆ (n − j) i =1
Approach A is to normalise the detrended data as
shown in Figure 2c. Then, the normalised data are where μ̂ and σ̂ are the estimated mean and stand-
considered in pairs (n = 2) and the moving average ard deviation from the in situ CPT data. Note
of the data is calculated with an averaging length that for this estimator it is desirable that the data
equal to the spacing between data points (τn = 2). The be equispaced (Fenton & Griffiths 2008) at a
variance of this series is then calculated (σ2n = 2). spacing Δτ.
This procedure is repeated for the case of triplets
n = 3, with the corresponding variance (σ2n = 3) and a
spacing average (τn = 3) being equal to twice the spac- 4 RESULTS
ing between the original data points. The process
can be further extended until the total number of This section presents the results obtained for the
data used in the averaging are n = N. A consequence estimation of θv and θh, before and after core infill-
of spatial averaging is that the variance tends to ing. The results are presented in two parts. The
decrease with increasing n, due to the cancelling first part describes the soil variability of the sand
out of fluctuation as a result of spatial averaging. in the berm before caisson installation. It includes
For each n, the variance function can be estimated the estimation of the vertical scale of fluctua-
from (Wickremesinghe & Campanella 1993): tion (using Approaches A and B) followed by the
results obtained for the horizontal scale of fluctua-
σ 2n tion (using Approach B). The second part of the
Γ ( τ) ≈ (1)
σ2 analysis includes the results for the soil variability
in the berm after core infilling. All results are sum-
marised in Tables 1 and 2.
where σ2 is the variance for the original data. On the
other hand, for large τ the variance function Γ(τ)
can be estimated as (Vanmarcke 1977, Vanmarcke 4.1 Variability of the sand in the berm
1984): before caisson installation
194
Table 2. Estimation of the scales of fluctuation in
Tarsiut P-45 berm after caisson set down and subsequent
core infilling.
CPT05 0.89 0.88 – Figure 3. Plan view of CPT locations for berm before
CPT06 0.62 0.61 – caisson installation.
CPT07 0.29 0.26 –
CPT08 0.42 0.44 –
CPT09 0.32 0.30 –
CPT10 0.66 0.63 –
Mean, m 0.53 0.52 18.1
Variance, m2 0.05 0.05 –
*
Values using Approach A. **Values using Approach B.
195
Figure 7. Plan view of CPT locations for berm after
caisson infilling.
196
the results from Tables 1 and 2. Furthermore, REFERENCES
when looking at the results for the horizontal scale
of fluctuation, a significantly larger value of θh is Baecher, G.B. & Christian, J.T. 2003. Reliability and
obtained following Stage 2 of the construction. In statistics in geotechnical engineering. John Wiley &
this second analysis, the estimated horizontal scale Sons Inc.
of fluctuation gives a value of 18.1 m as illustrated Fenton, G.A. & Griffiths, D.V. 2008. Risk assessment
in geotechnical engineering. John Wiley & Sons,
in Figure 10 and the degree of anisotropy of the New Jersey, USA.
heterogeneity ξ is once again approximately 35. Griffiths, D.V. & Fenton, G.A. 2001. Bearing capacity
The increases in the estimated scales of fluc- of spatially random soil: the undrained clay Prandtl
tuation seem to suggest that the installation, and problem revisited. Géotechnique 51(4): 351–359.
subsequent infilling of the caisson system, lead to Hicks, M.A. & Onisiphorou, C. 2005. Stochastic evalua-
less variability across the site. However, it has been tion of static liquefaction in a predominantly dilative
already highlighted that the values of the hori- sand fill. Géotechnique 55(2): 123–133.
zontal scale of fluctuation should be treated with Hicks, M.A. & Samy, K. 2002. Influence of heterogeneity
caution due to the limited data. In fact, the results on undrained clay slope stability. Quarterly Journal of
Engineering Geology and Hydrogeology 35(1): 41–49.
from the horizontal scale of fluctuation when using Hicks, M.A. & Smith, I.M. 1988. Class A prediction
Approach B are expected to be influenced by the of Arctic caisson performance. Géotechnique 38(4):
distance between CPTs, and this influence has not 589–612.
been studied in here. Thus, a deeper investigation Hicks, M.A. & Spencer, W.A. 2010. Influence of het-
involving a larger number of CPTs with smaller erogeneity on the reliability and failure of a long 3D
horizontal spacing is desirable. slope. Computers and Geotechnics 37(7–8): 948–955.
Lloret, M., Hicks, M.A. & Wong, S.Y. 2012 Soil char-
acterisation of an artificial island accounting for
5 CONCLUSIONS soil heterogeneity. GeoCongress 2012, R.D. Hryciw,
A. Athanasopoulos-Zekkos & N. Yesiller (Eds.), San
The vertical and horizontal scales of fluctuation have Francisco, 2816–2825.
been estimated for the berm of Tarsiut P-45 before Lloret-Cabot, M., Hicks, M.A. & Eijnden, A.P. van den.
and after caisson set-down. Two methods have been 2012. Investigation of the reduction in uncertainty
used for the computations of the vertical scale of due to soil variability when conditioning a random
field using Kriging. Géotechnique letters 2: 123–127.
fluctuation and these mainly give consistent results. Nuttall, J.D. 2011. Parallel implementation and applica-
This suggests that both approaches provide good esti- tion of the random finite element method. PhD thesis,
mates of θv. The average value for the vertical scale of , UK.
fluctuation of the berm before caisson installation is Phoon, K-K., & Kulhawy, F.H. 1999. Characterization
0.35 m when using Approach A and 0.38 m if using of geotechnical variability, Canadian Geotechnical
Approach B. On the other hand, the average θv of Journal 36(4): 612–624.
the berm after core infilling when using Approach A Spencer, W.A. 2007. Parallel stochastic and finite element
is 0.53 m and 0.52 m when using Approach B. modelling of clay slope stability in 3D. PhD thesis,
The estimated horizontal scale of fluctuation for , UK.
Uzielli, M., Vannucchi, G. & Phoon, K-K. 2005.
the berm prior installation of the caisson system is Random field characterisation of stress-normalised
12.7 m whereas θh for the berm after core infilling is cone penetration testing parameters. Géotechnique
18.1 m. The degree of anisotropy of the heterogene- 55(1): 3–20.
ity is therefore approximately 35 for both analyses. Vanmarcke, E.H. 1977. Probabilistic modeling of soil
The increase in the estimated scales of fluctuation profiles, Journal of the Geotechnical Engineering Div.
suggests that the installation, and subsequent infill- ASCE 103(11): 1227–1246.
ing of the caisson system, lead to a more uniform Vanmarcke, E.H. 1984. Random Fields: Analysis
material. The computed results, however, should be and Synthesis. The MIT Press, Cambridge,
treated with caution due to the limited data consid- Massachusetts.
Wackernagel, H. 2003. Multivariate geostatistics: An
ered here. Further studies would be desirable for introduction with applications. Springer, Germany.
more conclusive results, particularly with regard to Wickremesinghe, D.S. 1989. Statistical characterization
the horizontal scale of fluctuation. In this investi- of soil profiles using in situ tests. PhD thesis, University
gation θh may have been overestimated due to the of British Columbia, Canada.
spacing between CPTs being too large. Wickremesinghe, D.S. & Campanella, R.G. 1993. Scale
of fluctuation as a descriptor of soil variability.
Proc. Conf. Probabilistic Methods in Geotechnical
ACKNOWLEDGMENTS Engineering, Canberra, 233–239.
Wong, S.Y. 2004. Stochastic characterisation and
The authors gratefully acknowledge the support reliability of saturated soils. PhD thesis, University of
provided by the EU FP7 programme under the Manchester, UK.
project: ‘Geo-Install’ (PIAP-GA-2009-230638).
197
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
P. Becker
Kempfert und Partner Geotechnik, Hamburg, Germany, previous University of Strathclyde, Glasgow, Scotland, UK
M. Karstunen
Chalmers University of Technology, Göteborg, Sweden
University of Strathclyde, Glasgow, Scotland, UK
ABSTRACT: The deformation behaviour of deep mixed columns in anisotropic soft soil is a three-
dimensional problem which has to be considered adequately in numerical modelling. However, in litera-
ture simplifications are often to be found due to computational costs of fully coupled 3D analyses by
either modifying geometry of the problem or material parameters in 2D-plainstrain or -axisymmetric
conditions. This paper uses an enhanced 2D volume averaging technique for numerical modelling of deep
mixed columns. The method enables mapping the 3D problem in two dimensions, and yet modelling the
two constituents (column and soft soil) appropriately within a homogenized material. The performance
of the technique is demonstrated by considering unit cell simulations of floating columns in soft soil, in
which the results from 2D finite element simulations utilizing volume averaging technique are compared
against conventional and fully 3D coupled finite element analyses. It is shown that the quality of the
match is very good but depending on the number of columns and stiffness ratio between columns and
soft soil.
198
2 VOLUME AVERAGING TECHNIQUE local equilibrium between the soil and the column
material in each integration point can be formu-
2.1 Introduction and fundamental assumptions lated with the following equilibrium conditions,
which assure that there is no stress discontinu-
The basic idea of the volume averaging technique
ity between soil and column material in terms of
is to model the periodic system as a homogenous
radial and shear stress, see also Figure 2:
material instead of modelling columns and natural
soil separately (Fig. 1). The principles adopted by
σ eq
x σ sx = σ cx (2a)
Vogler & Karstunen (2007, 2009) are based on the
ideas of Schweiger & Pande (1986), further refined
by Lee & Pande (1998). The formulation has been σ eq
z σ sz = σ cz (2b)
extended to three dimensions and a new solution
routine has been developed to cope with the highly τ eq
xy τ sxy τ cxy (2c)
non-linear constitutive models.
Within the volume averaging technique a peri- τ eq τ syz = τ cyz (2d)
yz
odic distribution of the columns in the natural soil
is assumed. Furthermore perfect bonding, in other
words no slip between natural soil and columns, is Furthermore, perfect bonding between the col-
assumed. The method allows for adopting any elas- umns and the soft soil is assumed, and hence no
to-plastic constitutive model to the two constitu- slip is permitted between the two materials. This
ents: natural and improved soil. Local equilibrium can be achieved with the following kinematic con-
between soil and column as well as compatibility ditions, see also Figure 3:
and validity of the constitutive relations are satis-
fied through stress/strain redistribution within a ε eq
y ε sy = ε cy (3a)
sub-iterating procedure.
γ zx
eq
γ zx
s
= γ zx
c
(3b)
2.2 Equivalent material stiffness matrix
The constitutive equations for the constituents
Homogenisation is carried out by determin- can be described in terms of effective stress incre-
ing the strain increment and the stress increment ments as:
in the homogenised equivalent material according
to the following averaging rules:
(σ s )′ Ds ε s (4a)
σ eq
Ωs σ + Ωc σ
s c
(1a)
ε eq Ωs ε s + Ωc ε c (1b)
199
(σ c )′ Dc ε c (4b)
200
Table 1. Initial values for state parameters and K0. integration scheme. Because both constituents
exhibit highly non-linear behaviour and an implicit
Layer e0 OCR K0 α0 x0 integration scheme is used, a sub-iteration scheme
was necessary: If the initially predicted internal
Soft clay 2.9 1.25 0.56 0.458 10
strain distribution between soft clay and improved
columns leads to an violation of the equilibrium
conditions (Eq. 2), the strains between this two
Table 2. Conventional soil constants.
materials are redistributed in an iterative scheme
Layer γ (kN/m3) κ ν′ λ M until equilibrium is satisfied (more details in
Vogler & Karstunen, 2007).
Soft clay 15 0.02 0.15 1.15 1.2
201
Figure 8. Shear stresses in column, soil and VAT.
202
column and soft soil, i.e. no differential displace-
ments between column and soft soil.
Perfect bonding is achieved for the presented
variations. Figure 10 shows the axial strains with
a perfect match for 10 kPa loading conditions.
There is a small overestimation of axial strains
of VAT with increasing vertical load which is still
acceptable. Whereas the vertical strains in column
and soil are showing perfect bonding with the dis-
crete simulation of floating columns.
ACKNOWLEDGEMENTS
203
Karstunen, M., Krenn, H., Wheeler, S.J. Koskinen, M. & Vogler, U. & Karstunen, M. 2007. Numerical model-
Zentar, R. 2005. Effect of anisotropy and destructu- ling of deep mixed columns with volume averaging
ration on the behaviour of Murro test embankment. technique. In Pande, G.N. and Pietruszcak, S. (eds.),
ASCE International Journal of Geomechanics, Vol. 5, Proc. of the 10th intern. Symp. on Numerical Mod-
No. 2, 87–97. els in Geomechanics (NUMOG X), Rhodes, Greece,
Koskinen, M. & Karstunen, M. 2004. The effect of, 25–27 April 2007, London: A.A Balkema, 495–503.
structure on the compressibility of Finnish clays. Vogler, U. & Karstunen, M. 2009. Application of volume
SGF Report 3:2004, Proc. of XIV Nordic Geotechni- averaging technique in numerical modeling of deep
cal Meeting 19–21 May 2004, Ystad, Sweden, Vol. 1, mixing. In Karstunen, M. & Leoni, M. (eds), Geo-
A-11–A-22 Linköping: Swedish Geotechnical Society. technics of Soft Soils – Focus on Ground Improve-
Lee, J.-S. & Pande, G.N. 1998. Analysis of stone-column ment, London: Taylor & Francis Group, 189–195.
reinforced foundations. Int. J. for Numer. Anal. Meth. Wheeler, S.J., Näätänen A., Karstunen M. &
Geomech., Vol. 22, 1001–1020. Lojander M. 2003. An anisotropic elasto-plastic
Schweiger, H.F. & Pande, G.N. 1986. Numerical Analysis model for natural soft clays. Canadian Geotechnical
of stone column supported foundation. Computers Journal 40(2):403–418.
and Geotechnics, 347–372.
204
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
J. Castro
University of Cantabria, Santander, Spain
W. Wehr
Keller Holding GmbH, Offenbach, Germany
ABSTRACT: Several theoretical procedures to estimate the soil improvement produced by installation of
vibro-stone columns are described. Particularly, finite element model and analytical solutions of a cylindrical
cavity expansion were compared with results from an actual field test which was performed in silty sand and
clayey soil treated with a column group. The results show that after dissipation of pore pressure the instal-
lation effects produce considerable improvement due to a large increase of the horizontal effective stress
and due to densification process of sand. The load settlements response of the tested column group has
been analyzed and compared with theoretical estimation of the improvement with and without installation
effects, and with Priebe’s analytical solution. It is observed that the column group installation effects have an
important influence that should be evaluated with more advanced modelling or directly with in situ testing.
205
measurements (Watts et al. 2001, Kirsch 2006, Gäb induced into the ground, hypoplastic model has
et al. 2007, Castro 2008), but related to a specific been considered, which fits more accurately to the
cases and hence, cannot be generalized in a straight- sand behavior. In general, results obtained from
forward manner (Castro & Karstunen 2010). models are in accordance with the experiences on
On the other hand, there have been various actual observed effects, stressing the importance
attempts with different approaches for theoreti- of the compaction time in order to achieve better
cal modelling (Kirsch 2006, Elshazly et al. 2007, results. It is also important to emphasize the model
Guetif et al. 2007, Castro & Karstunen 2010, performed by Arnold et al. (2008), which confirms
Castro et al. 2012). In general, field measurements the existence of a zone with low densification
and theoretical approaches have shown that instal- immediately close to the vibrator, a second zone
lation of vibro-stone columns in saturated clays with best compaction results that extends from
cause its remolding close to the vibrator, and pro- approximately 0.5 to 3 m from vibrator axis, and
duces increase of pore pressures and horizontal a third zone at a distance of more than 3 m, where
stress. However, after a relative short period of soil compaction does not occur.
consolidation, soil tends to gain in stiffness and Mentioned findings are directly related to the
additionally provides greater confinement to the vibrocompaction process and may be applied
columns. Whereas, in intermediate soils with vari- to vibro-stone column installation for predomi-
able fines content such as silty sands or sandy silts, nantly sandy soils. For predominantly fine grained
vibrations could also achieve a significant den- soils the densification is insignificant using
sification by rearranging the soil particles into a vibrocompaction.
denser state. This would be the case of soils in the
marginal compactable zone shown in Figure 1.
2.2 Cavity expansion
In the following sections some theoretical pro-
cedures that are considered as reasonably suitable As it was mentioned, if vibro-stone column are
to evaluate the vibro-stone column installation performed in saturated clays the vibration energy
effects will be described and compared with field has a negligible effects on the densification of the
trial measurements obtained during column natural soil, but effects caused by the displacement
installation. of soil due to column installation can be evaluated
by the application of the cylindrical cavity expan-
sion theory. As columns are built in a reasonably
short period the undrained conditions should be
2 MODELLING PROCEDURES
considered even for relatively permeable soils, e.g.
sandy silts. Due to difficulties to determine the
2.1 Vibro equipment performance
internal cavity pressure produced by the vibrator
A rigorous modelling should consider both, the it is more suitable to simulate column expansion
actions imposed by vibrator and the soil response. considering radial displacement (Castro 2008).
Due to the complex interaction between vibrator Egan et al. (2008) summarized the development
and soil, densification modelling of vibro tech- of analytical solutions in three aspects: soil behav-
niques is still a challenging task (Heibrock et al. ior, consideration of finite or infinite medium,
2006). However, various attempts have been done and the influence of the initial radius of the cav-
by several authors (Fellin 2000, Cudmani et al. ity. Concerning mentioned aspects there are sev-
2003, Arnold et al. 2008, Arnold & Herle 2009). eral solutions proposed by various authors, e.g.
Ranges of vibrators basic characteristics, and ones Vesić (1972), Carter et al. (1986), Yu & Houlsby
usually considered for modelling are shown in (1991), Yu (2000). For complex soil models numer-
Table 1. ical analysis very frequently is selected as more
Usually dynamic effects are substituted by adequate, but large strains should be enabled and
quasi-static loading; one-dimensional or 2D considered.
axisymmetric conditions are often adopted, as In any case the key aspects to be analyzed will
well as dashpot-spring and full numerical models. be the rapid increase of pore pressure and its dis-
Furthermore, in order to simulate the shear waves sipation over time, as well as the final stress state in
terms of the at rest earth coefficient K0.
206
beyond ρF stresses are in the elastic zone. Thus, the pore pressure increments take place only in the
radial stress pF at point ρF represents fully plastic plastic zone. In the same way, if we take into
behavior. For a cohesive and frictional soil in an account the equilibrium equation of cylindrical
infinite medium and assuming Mohr Coulomb cavity expansion and adopting zero pore pressure
behavior Baguelin et al. (1978) proposed the in the elastic zone as a boundary condition, it can
expression (1) to determine the plastic radius, be obtained the equation 6 to determine de excess
based on the assumption of no volume change and pore pressure distribution as function of ρ.
considering total stresses:
⎛ρ ⎞
a 2 (p0 + c ⋅ cot φ ) ⋅ sin φ Δu 2c u ln ⋅ ⎜ F ⎟
= (1) ⎝ ρ⎠ (6)
ρF2 G
where a is the radius of the cavity; p0 is the pre- 2.3.2 Influence of soil constitutive model
existence horizontal stress; and G is the shear In the previous sections the elastic perfectly-plastic
modulus. φ is the angle of internal friction and c models have been applied in order to highlight
the cohesion of the soil. the possibilities of available and relatively simple
It can be noted that for purely cohesive clay with analytical solutions to reach a simple approach
φ = 0, or adopting Tresca criterion, equation 1 turns of most important effects of vibro-stone columns
into equation 2, which is the same solution devel- installation. In fact, saturated clays with low sensi-
oped by Randolph & Wroth (1979) for large strains: tivity may be well modelled with these approaches.
With hardening plastic behavior a first enhance-
a2 cu ment of the modelling could be made consider-
= (2)
ρF2 G ing the stress state modification due to plastic
strains and stress dependency of stiffness. This
where cu is the undrained shear strength. model is suitable for the application of both, sandy
Since plastic radius is known, and combin- and clayey soils subjected to a cavity expansion,
ing the equilibrium equation of cylindrical cavity although calculations usually have to be done with
expansion with Mohr Coulomb failure criterion, numerical model.
the solution for distribution of major principal On the other hand, Castro & Karstunen (2010)
stress σr in the radial direction within the plastic presented a numerical modelling of Bothkennar
zone can be obtained for purely cohesive soil with clay based on S-CLAY1 and S-CLAY1S, which
equation 3, and for cohesive-frictional soils with are Cam clay-type models that consider anisotropy
equation 4, as a function of radial distance ρ. and destructuration. Results agree with practical
experience, showing a great remolded zone close
ρF to vibrator. Authors recommend for practical pur-
σ r = pF + c u ⋅ ln (3) poses a reduction of 15%–20% of initial undrained
ρ strength for standard columns grid.
where pF = p0 + cu
2.4 Back-calculation procedures
1− K a
⎛ ρ2 ⎞ 2 As it was stated by contractors experience and
σ r = (pF + c ⋅ cotφ ) ⋅ ⎜ F2 ⎟ − c ⋅ cot φ (4) several field measurements, during vibro-stone
⎝ρ ⎠ columns installation a certain heave on the
surface could appear (Egan et al. 2008), and is
where pF = p0 ⋅ (1 + sinφ) + c ⋅ cotφ; and Ka = tan2(45- more important for closer distances between
φ/2). columns. It confirms that soil stress state var-
Whereas in the elastic zone the well-known ies, and from the heave measurement data it is
solution can be used to determine radial stress possible to perform back-calculation of final soil
distribution: characteristics.
The real scale load test on treated soil might be
ρF2 very useful for back analysis regarding the addi-
σr p0 (pF − p0 ) ⋅ (5) tional stiffness of the soil. This procedure can be
ρ2
used to estimate the installation effects of column
groups, e.g. Kirsch (2006) proposed the evaluation
2.3.1 Pore pressure of an enhancement zone placed around column
Considering that excess pore pressures are caused group by means of stiffness variation and its com-
only by the variation of mean total stresses, all parison with measured settlements.
207
3 FIELD TEST piezometer and its comparison with the theoretical
predictions that are exposed in next chapter.
3.1 Field measurements Maximum values of 65 kPa and 110 kPa were
measured at depths of 10 m and 16 m respectively,
In order to evaluate the installation effects of
and at radial distance of 1.8 m. At a radial dis-
vibro-stone columns executed with dry bottom
tance of 5 m were observed peaks values of 50 kPa
feed method in a profile of intermediate soil i.e.
and 68 kPa, probably related to the presence of
silty sand and clayey silt, field test consisting of
sand layer that is able to transmit the generated
a group of 13 vibro-stone columns, 20 m to 25 m
vibrations.
long and with diameter between 0.90 and 1.00 m
Figure 4 shows that after column installation
was carried out. Instrumentation campaign was
the excess pore pressure is rapidly dissipated and
composed by 12 piezometers distributed along
after 5 hours (300 minutes) almost no excess pore
4 lines and located at depths of 6 m, 10 m and
pressure was detected. Moreover, the same Figure
16 m each. To control vertical displacements and
shows that consolidation rate estimated with finite
stresses 1 extensometer was extended up to depth
element method reasonably agrees with the meas-
of 40 m. Figure 2b shows a section and plan view
ured consolidation rates.
of the field test. In order to compare measurements
with theoretical procedures, the pore pressure was
monitored during the installation of one column
3.2 Modelling of field measurements
(see column A, Fig. 2b).
Piezometer radial distances from the axis of the Finite element modelling with Plaxis v8 code was
column A can be seen in Figure 2b (distance of performed and results were compared with analyti-
1.25 m for piezometer Pz3, 1.8 m for piezometer cal solutions presented in chapter 2.2. Modelling
Pz4, whereas both piezometer Pz1 and Pz2 were set comprises an axisymetric model of 4 soil layers
at radial distance of 5 m). with Mohr-Coulomb behavior and extrafine mesh
The first evaluation step of the effects of col- of 15-noded elements close to the column axis. The
umns installation can be seen in Figure 2a. Typical whole geometry is indicated in Figure 2c.
cone penetration resistance between columns Column expansion was modeled by a prescribed
increase 8 to 14 times comparing with tip resist- displacement. According to Carter et al. (1979) the
ance before the treatment. From the combination expansion from a finite radius can be related to the ide-
of friction ratio FR% and tip resistance it can be alized expansion on an infinite medium where radius
observed that soil most probably experienced a starts from 0, by the relationship r2f,fin − r20,fin = r2c – 0.
certain improvement due to densification accord- Therefore, adopting initial radius r0,fin of 0.1 m and
ing to the marginal compactable zone shown in actual column radius rc of 0.55 m, the magnitude of
Figure 1. expansion to be considered is rf,fin − r0,fin = 0.46 m.
Regarding excess pore pressure, in Figure 3 To enable the soil free movements roller bound-
are presented both, the peak values measured by aries were assumed on all sides. Soil properties are
Figure 2. Field test characteristics, (a) cone penetration resistance after and before the treatment, (b) plan view and
section of field test and (c) geometry of finite element modelling.
208
decrease with depth, reaching approximate radial
distances of 4 m, 3 m and 2.5 m at 6 m, 10 m and
16 m depths respectively. It is also in accordance
with the increase of undrained shear strength cu,
while the maximum values of Δu increase with
depth. At a radial distance close to 1.75 m the pore
pressures are almost equalized at the three ana-
lyzed depths (6 m, 10 m and 16 m), and the Δu
values that fits best with theoretical procedures are
close to this zone, near normalized radial distance
of 4 column diameters.
On the other side, Figure 6 shows the results
from finite element method for the distribution
of normalized effective mean stress p and normal-
ized coefficient of earth pressure K at the end of
consolidation. Therefore, it is supposed that the
stress state associated to increased K and p values
Figure 3. Predicted and measured peaks of excess pore will be suitable to estimate the stiffness generated
pressures. due to column installation. Although the increase
of the stiffness is dominated by the great increase
of the radial stress, the reduction of the circum-
ferential stress in the plastic zone has to be taken
into account.
Consequently, for the estimation of the enhanced
modulus of soil, instead of the coefficient of lateral
earth pressure K, the mean effective stress should
be considered. In general, a power law in the form
E = E0 (p’/p’0)m might be used. However, K is still
the best indicator to assess the extent to which the
soil stiffness increases due to columns installation.
Figure 6 also shows that the zone influenced by
K/K0 = 1.5–2 could be very suitable for the esti-
Figure 4. Measurements of excess pore pressure mation of the stiffness increase. This zone is in
dissipation. between 4 and 6 column radii at 16 m depth, while
at 6 m depth is between 10 and 14 column radii.
It is important to emphasize that in the analyzed
listed in the Table 2. The water table was situated case less difference between K/K0 and p/po was
at 1.20 m depth. Because the modelling is aimed to observed with the increase of the depth.
study the effects on the soil, the column was con-
sidered only as a void with infinite permeability.
3.3 Field load testing
It was adopted Plaxis K0-procedure corresponding
to normally consolidated soil. An equivalent load of 75 kPa was placed above
Calculations were performed in two phases, the field-test site shown in Figure 1b. Loading
with undrained conditions for the first phase and was made of 4 m height backfill material, widely
consolidation analysis for the second phase. The larger than the testing zone. In Figure 7 the load
“up-dated” option of Plaxis software was activated settlements behavior of a group composed of
to take into account large strains. 13 columns are indicated.
Figure 5 shows the distribution of initial excess Several theoretical estimations are compared
pore pressures Δu after cavity expansion, obtained with the actual measurements, taking into account
at the same depths of piezometers. It is observed the situation without improvement and situations
that results from finite elements modelling and the considering the improvement with and without
analytical solution (equation 6) are quite similar. any installation effects.
The undrained shear strength, adopted to estimate Thus, a settlements estimation with coefficient of
the plastic radius ρF, was determined combin- lateral earth pressure at rest K0 is compared with the
ing equations (1) and (2) and considering effec- case of improved lateral earth pressure K* = 1.75
tive stresses and effective soil parameters (c’, φ’). according to the modelling results shown in
Results show that the plastic radius ρF coincides Figure 6, and with the analytical solution proposed
with the influence zone of Δu, and its values by Priebe (1995) which represents the case of K = 1.
209
Table 2. Soil properties.
Depth γ’ c’ φ’ Eoed* kh
Layer (m) (kN/m3) (kPa) (o) (kPa) (m/s)
*confined modulus.
Figure 8 shows the accumulated compressibility Figure 8. Accumulated compressibility results of field
monitored by the incremental extensometer test.
installed in the middle of field test, together with
the theoretical estimation of improvement consid-
ering the same cases described in Figure 7. During Furthermore, as have been stated by Kirsch
the load testing the extensometer only provided (2006), the settlement reduction in terms of
data up to a maximum foundation load of 35 kPa. improvement factor “n” depends on the load
The reduction of settlements obtained with the level, which can be noted comparing results from
consideration of K* improved by cavity expansion Figures 7 and 8. It has to be stressed that the dif-
is about 20–30% greater than the case of improve- ference between actual measurements and theo-
ment without any installation effects (K0). The retical estimation could be attributed to the global
results closer to the actual measurements are those installation effects of the column group, possible
estimated according to Priebe (1995). densification effects and the global increase of
210
horizontal stresses. The evaluation of these global Elshazly, H., Elkasabgy, M. & Elleboudy A. 2008. Effect
effects should be made with more advanced model- of Inter-Column Spacing on Soil Stresses due to
ling or directly by means of in situ testing. Vibro-Installed Stone Columns: Interesting Findings.
Geotech Geol Eng 26:225–236. Springer Science—
Business Media.
4 CONCLUSIONS Fellin, W. 2000. Rütteldruckverdichtung als plastodyna-
misches Problem. Advances in Geotechnical Engineer-
ing and Tunnelling Vol 3.
Installation effects should be considered in the Gäb, M., Schweiger, H.F., Thurner, R., & Adam, D.
design of vibro-stone column treatments, and can 2007. Field trial to investigate the performance of a
be estimated according to cavity expansion theory floating stone column foundation. In Proceedings
with reasonable accuracy. The improvement induced of the 14th European Conference on Soil Mechan-
by individual column installation can be expressed ics and Geotechnical Engineering, Madrid, Spain,
by the increase of coefficient of earth pressure. For 24–27 September 2007. Millpress, Amsterdam, The
the evaluation of global effects induced by a column Netherlands. pp. 1311–1316.
group installation, advanced modelling should be Greenwood, D.A. & Kirsch, K. 1984. Specialist Ground
Treatment by Vibratory and Dynamic Methods.
performed to consider more realistic characteristics Piling and Ground Treatment. The institution of Civil
of vibro-stone column execution and its effects. Engineers: Tomas Telford, London.
Guetif, Z., Bouassida, M. & Debats, J.M. 2007.
Improved soft clay characteristics due to stone col-
REFERENCES umn installation. Computers and Geotechnics 34(2):
104–111.
Arnold, M., Herle, I. & Wehr, J. 2008. Comparison of Heibrock, G., Kebler, S. & Triantafyllidis, TH. 2006. On
vibrocompaction methods by numerical simulations. modelling vibro-compaction of dry sands. Numerical
In Karstunen et al. (eds), Geotechnics of Soft Soils: Modelling of Construction Processes in Geotechnical
Focus on Ground Improvement: University of Strath- Engineering for Urban Environment—Triantafyllidis
clyde, Glasgow. (ed). Taylor & Francis Group. London. ISBN 0 415
Arnold, M. & Herle, I. 2009. Modelling of vibro- 39748 0.
compaction using hipoplasticity with intergranu- Kirsch, F. 2006. Vibro stone column installation and
lar strains. Proceedings of the 17th International its effect on ground improvement. In Proceedings
Conference on Soil Mechanics and Geotechnical of Numerical Modelling of Construction Processes
Engineering—ICSMGE. in Geotechnical Engineering for Urban Environment,
Baguelin, F., Jezequel, J.F. & Shields, D.H. 1978. The Bochum, Germany, 23–24 March 2006. Taylor and
Pressuremeter and Foundation Engineering. TransTech Francis, London: 115–124.
Publications. Kirsch, K. & Kirsch, F. 2010. Ground Improvement by
Carter, J.P., Randolph, M.F. & Wroth, C.P. 1979. Stress Deep Vibratory Methods. New York: Spon Press.
and pore pressure changes in clay during and after Massarsch, K.R. 1991. Deep Soil compaction Using
the expansion of a cylindrical cavity. International Vibratory Probes in Deep Foundation Improvement.
Journal for Numerical and Analytical Methods in STP1089 ASTM.
Geomechanics. Priebe, H.J. 1995. Design of vibro replacement. Ground
Carter, J.P., Booker, J.R. & Yeung, S.K. 1986. Cavity Engineering 28(10):31–37.
expansion in cohesive frictional soils. Geotechnique Randolph, M.F. & Wroth, C.P. 1978. An analytical
36. No. 3:pp. 349–358. solution for the consolidation around a driven pile.
Castro, J. 2008. Análisis teórico de la consolidación y International Journal for Numerical and Analytical
deformación alrededor de columnas de grava. Ph.D. Methods in Geomechanics. Vol. 3: pp. 217–229.
thesis, University of Cantabria, Santander, Spain. Slocombe, B.C., Bell, A.L. & Baez, J.L. 2000. The den-
Castro, J. & Karstunem, M. 2010. Numerical simulations sification of granular soils using vibro methods.
of stone column installation. Canadian Geotechnical Geotechnique 50. No. 6. 715–725.
Journal 47(19):1127−1138. Sonderman, W. & Wehr, W. 2004. Deep vibro techniques.
Castro, J., Kamrat-Pietraszewska, D. & Karstunem, M. Ground Improvement. 2nd Edition Ed. Moseley &
2012. Numerical modelling of stone column installation Kirsch.
in Bothkennar clay. International Symposium on Ground Vesić, S.A. 1972. Expansion of Cavities in Infinite Soil
Improvement IS-GI Brussels; ISSMGE-TC 211. Mass. Journal of Soil Mechanics and Foundation
Cudmani, R.O., Osinov, V.A., Bühler, M.M. & Gudehus, G. Engineering Division. ASCE. 98(3):pp. 265–290.
2003. A model for de evaluation of liquefaction sus- Watts, K.S., Johnson, D., Wood, L.A. & Saadi, A.
ceptibility in layered soils due to earthquakes. 12th 2000. An instrumented trial of vibro ground treat-
Panamerican Conference on SMGE. Cambridge. ment supporting strip foundations in a variable fill.
Degen, W. 1997. Vibroflotation ground improvement. Géotechnique 50(6):699–708.
Altendorf, unpublished. Yu, H.S. & Houlsby, G.T. 1991. Finite cavity expansion
Egan, D., Scott, W. & McCabe, B. 2008. Installation in dilatant soils: loading analysis. Geotechnique. 42(2):
effects of vibro replacement stone columns in soft pp. 173–183.
clay. In Geotechnics of Soft Soils—Focus on Ground Yu, H.S. 2000. Cavity expansion methods in geomechan-
Improvement. Glasgow, 3–5 September 2008. ics, Dordrecht, Boston; Kluwer Academic Publishers.
211
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
J. Castro
University of Cantabria, Santander, Spain
M. Karstunen
Chalmers University of Technology, Gothenburg, Sweden
University of Strathclyde, Glasgow, UK
N. Sivasithamparam
University of Strathclyde, Glasgow, UK
Plaxis BV, Delft, The Netherlands
C. Sagaseta
University of Cantabria, Santander, Spain
ABSTRACT: The paper presents the results of numerical simulations studying the installation effects
of stone columns in a natural soft clay. Stone column installation is modelled as an undrained expansion
of a cylindrical cavity, using the finite element code PLAXIS that allows for large displacements. The
properties of the soft clay correspond to Bothkennar clay, a soft Carse clay from Scotland (UK). The
complexity of this material is simulated via two advanced recently developed constitutive formulations
able to account for the soil structure, namely S-CLAY1 and S-CLAY1S. Modified Cam Clay model is also
used for comparison purposes. The paper shows the new stress field and state parameters after column
installation and the subsequent consolidation process. This sets the basis for including installation effects
in studying the settlement reduction caused by stone columns.
212
about installation effects for column design, if the bonding and degradation of bonds, using an
assumptions made in the model are validated by intrinsic yield surface and a hardening law describ-
experimental measurements. Furthermore, few ing destructuration as a function of plastic straining.
attempts (Kirsch 2006, Guetif et al. 2007) had been The models have been implemented as User-defined
made in this field, using simple soil models. There- soil models in Plaxis. An implementation that uses
fore, the authors recently decided to study installa- an implicit integration scheme (Sivasithamparam
tion effects numerically using advance soil models 2012) has been used, instead of an explicit previous
to reproduce the behaviour of natural structured version in Castro & Karstunen (2010).
soft soils (Castro & Karstunen 2010). The results The values for S-CLAY1 model param-
were satisfactory, as they compared well with field eters (soil constants) and the initial state vari-
measurements (e.g. Roy et al. 1981, Kirsch 2006). ables for Bothkennar clay are listed in Tables 1
Here, a detailed analysis of those numerical results and 2, respectively. The additional parameters for
once the excess pore pressures have been dissipated is S-CLAY1S are detailed in Table 3. S-CLAY1 model is
presented. The new state of the soil that is obtained formulated to be a hierarchical model, which reduces
after column installation and full consolidation sets to MCC model by assuming μ = 0 and α0 = 0.
the basis of a future study of the influence that stone The geometry of the numerical model and
column installation has on the ground improvement, the finite element mesh is shown in Figure 1.
especially on the settlement reduction. Parametric studies were carried out to check how
wide the model should be to have a negligible influ-
ence of the outer boundary. Mesh sensitivity stud-
2 NUMERICAL MODEL ies were performed to confirm the accuracy of the
mesh. Calculations accounted for large displace-
The finite element code Plaxis v9 (Brinkgreve ments using the “updated mesh” option in Plaxis,
2008) was used to develop a numerical model of which uses an updated Lagrangian formulation
a reference problem to study installation effects of described by McMeeking & Rice (1975).
stone columns. The installation of only one stone Column installation is modelled as the expan-
column was considered, to simplify the problem sion of a cylindrical cavity, which is considered to
to an axisymmetric two-dimensional geometry. In occur in undrained conditions, because columns
order to consider a realistic situation, properties are usually installed in a short period of time. The
of Bothkennar clay were used for the soft soil. The expansion of the cavity is modelled as a prescribed
Bothkennar soft clay test site has been the subject of displacement from an initial radius, a0, to a final
a number of comprehensive studies (Géotechnique one, af. Values of a0 = 0.1 m and af = 0.41 m rep-
Symposium in print 1992). The soil at Bothkennar resent the installation of a column with a radius
consists of a firm to stiff silty clay crust about 1.0 m of rc = 0.4 m (Carter et al. 1979). After undrained
thick, which is underlain by about 19 m of soft clay.
The ground water level is 1.0 m below the ground
surface. Typically, in a structured soil, the in situ Table 1. S-CLAY1 parameters for Bothkennar clay.
water content is close to the liquid limit.
Depth γ
Stone columns have been applied in Bothkennar (m) (kN/m3) κ ν′ λ M μ β
clay (Watts et al. 2001, Serridge & Sarsby 2008)
or other Carse clays (Egan et al. 2008). For the 0–1 18.0 0.02 0.2 0.48 1.4 30 0.94
numerical model in this paper, a column length of 1–10 16.5 0.02 0.2 0.48 1.4 30 0.94
10 m is used. The untreated clay underneath is not
modelled, because the installation effects in this
part of the soil are not particularly significant and Table 2. S-CLAY1 initial state variables.
furthermore, modelling the tip of the column may
lead to some numerical instabilities. Depth POP
The behaviour of Bothkennar clay was modelled (m) e0 α0 OCR (kPa) K0
using two advanced constitutive models, namely
S-CLAY1 (Wheeler et al. 2003) and S-CLAY1S 0–1 1.1 0.539 – 30 1.35
(Karstunen et al. 2005). The Modified Cam Clay 1–10 2 0.539 1.5 – 0.544
model (MCC) (Roscoe et al. 1958) is also used for com-
parison purposes. S-CLAY1 is a Cam Clay type of
model with an inclined yield surface to model inherent Table 3. S-CLAY1S additional parameters.
anisotropy, and a rotational component of harden-
ing to model the development or erasure of fabric λi χ0 a b
anisotropy during plastic straining. The S-CLAY1S
0.18 5 11 0.2
model accounts, additionally, for interparticle
213
Figure 1. Model geometry and finite element mesh.
3 STRESS FIELD
214
on the soil model) where soil behaviour is always around 3 times the initial value of the undrained
elastic and hoop stresses decrease, (2) an area that shear strength and the radial stresses are between
is plastic during undrained expansion of the cav- 4 and 5. Randolph et al. (1979) predicted a slightly
ity but is not after consolidation, where vertical higher value because their analysis is in perfect
stresses change only slightly and (3) points that plane strain conditions while in the present analy-
are on the yield surface also after consolidation sis there are some vertical strains since the model
(closer than 4.5–6 column radii), where densifica- realistically considers the soil surface. That also
tion and the increase in mean effective stresses are causes subtle differences in the shape of the curves.
important. In the far field, the initial stresses of the present
Randolph et al. (1979) presented similar results analysis are lower because of the lower value of the
for normally consolidated Boston Blue clay using σ'z/cu relationship.
the MCC model and a similar numerical model The stress field described in Figure 2 is the basis
for driven piles. The results from this study with for an ongoing study on the influence that the instal-
the MCC model are compared with those results lation effects have on the settlement reduction. As the
(Fig. 3). The comparison shows that, although finite element mesh gets very distorted after column
there are differences in the OCR, the soil param- installation, it is advisable to input directly the stress
eters and the initial stresses used in the current field as the initial one in a new model. That requires
study, the values near the column/pile are very sim- curve fitting of the stress field and the modification
ilar and well correlated with the initial undrained of the soil model input parameters to allow for this
shear strength. The vertical and hoop stresses are particular initial stress field. Special care should be
taken to ensure that equilibrium is fulfilled:
∂σ ′ r σ ′ r − σ ′θ
+ =0 (1)
∂r r
a1
e a3 − (2)
r rc − a2
215
Figure 4. Void ratio after column installation and Figure 5. Size of the yield surface and mean effective
consolidation. stresses after column installation and consolidation.
The results for MCC and S-CLAY1 are quite sim- on the improvement as the column is better laterally
ilar but soil destructuration due to column installa- confined and (b) the increase in the mean effective
tion causes a greater densification of the soil. stresses, which may be positive if the yield surface is
The value of the void ratio is directly related to expanded and the soil hardens (Zone 3a) or may be
the changes in the mean effective stresses and the negative if the yield surface is not clearly expanded
mobilized soil stiffness. The mean effective stresses as the soil loses its overconsolidation and for a sub-
and the size of the yield surface, given by p′m, are sequent loading process, there is not an elastic region
plotted in Figure 5, where the distance to the col- anymore and the soil has not hardened either.
umn axis is in logarithmic scale to amplify the zone
of interest near the column. The different zones
previously distinguished are also visible here: (1) far 5 ANISOTROPY
from the column, the mean effective stresses does
not change, (2) the mean effective stress increases The advanced soil models S-CLAY1 and
but without expanding the yield surface because S-CLAY1S can be used to reproduce the changes
the overconsolidation ratio is OCR = 1.5, (3) the in soil fabric due to column installation. The incli-
current stress point is on the yield surface and (3a) nation of the yield surface changes, as shown in
the yield surface is notably expanded through strain Figure 6, where the components of the fabric
hardening, p′m increases. Note that the extension tensor, {αi}, are plotted. The two models predict
of these zones is slightly different depending on the almost identical changes in anisotropy.
soil model. The current mean effective stress, p′, do To help to visualize the changes in soil fabric,
not necessarily coincide with p′m when the point is Figure 7 shows the (αz − αθ, αr − αθ) vector in arbi-
on the yield surface because p′m is the mean effec- trary points. This vector changes from horizontal
tive stress at the apex of the yield surface on the direction for an initial vertical cross anisotropy
right (Wheeler et al. 2003). For MCC, the differ- towards a nearly vertical one for radial cross ani-
ences are noticeable but for S-CLAY1 type models, sotropy. Its length is equal to α0 = 0.539 for the ini-
they are very similar. The relative expansion of the tial situation and changes only slightly.
yield surface is similar for MCC and S-CLAY1, The components of the fabric tensor after col-
which explains the similar results for the void ratio. umn installation and consolidation (Figure 6) must
For S-CLAY1S, the intrinsic yield surface, p′mi, is be input as the initial values to study the influence
significantly expanded in Zone 3a. that those changes have on the settlement reduc-
For normal column spacings, the soil is in tion. That requires curve fitting of those compo-
Zone 3, but the densification or the strain harden- nents. By definition of the fabric tensor, the fitting
ing is only important in Zone 3a, i.e. for closely of those components must fulfilled that
spaced columns.
The changes in the stress field and the initial α r + αθ + α z = 3 (3)
state variables, if anisotropy and destructuration
are not considered, have an influence on the ground So, for the sake of simplicity, it seems sensible to
improvement through the following features: (a) the keep αθ constant and decrease αz the same amount
increase in the radial stress, which has a positive effect that αr increases.
216
case occurs for closely spaced columns. In any
case, the increase of the radial stress improves the
lateral confinement of the column. The initial ver-
tical cross anisotropy of the soil changes towards
a radial one when approaching the soil-column
interface.
ACKNOWLEDGEMENTS
217
Gäb, M., Schweiger, H.F., Thurner, R. & Adam, D. 2007. Priebe, H.J. 1995. Design of vibro replacement. Ground
Field trial to investigate the performance of a float- Engineering 28(10): 31–37.
ing stone column foundation. In Proc. of the 14th Randolph, M.F., Carter, J.P. & Wroth, C.P. 1979. Driven
European Conf. Soil Mech. Geotech. Eng., Madrid, piles in clay-the effects of installation and subsequent
24–27 September 2007. Millpress: Amsterdam, consolidation. Géotechnique 29(4): 361–393.
pp. 1311–1316. Roscoe, K.H., Schofield, A.N. & Wroth, C.P. 1958. On
Géotechnique Symposium in print. 1992. Bothkennar the yielding of soils. Géotechnique 8(1): 22–53.
soft clay test site: characterisation and lessons learned. Roy, M., Blanchet, R., Tavenas, F. & La Rochelle, P.
Géotechnique 42(2). 1981. Behaviour of a sensitive clay during pile driving.
Guetif, Z., Bouassida, M. & Debats, J.M. 2007. Improved Canadian Geotechnical Journal 18(2): 67–85.
soft clay characteristics due to stone column installa- Sivasithamparam, N. (2012). Development and imple-
tion. Computers and Geotechnics 34(2): 104–111. mentation of advanced soft soil models in finite
Karstunen, M., Krenn, H., Wheeler, S.J., Koskinen, M. & elements. PhD thesis, University of Strathclyde,
Zentar, R. 2005. Effect of anisotropy and destructu- Scotland.
ration on the behaviour of Murro test embankment. Slocombe, B.C., Bell, A.L. & Baez, J.I. 2000. The densifi-
ASCE International Journal of Geomechanics 5(2): cation of granular soil using Vibro methods. Géotech-
87–97. nique 50(6): 715–726.
Kirsch, F. 2004. Experimentelle und numerische Unter- Watts, K.S., Chown, R.C. & Serridge, C.J. 2001. Vibro
suchungen zum Tragverthalten von Rüttelstopfsäu- stone columns in soft clay: A trial to study the influ-
lengruppen. Dissertation, Technische Universität ence of column installation on foundation perform-
Braunschweig. ance. In Proc. of the 15th Int. Conf. Soil Mech. Geotech.
Kirsch, F. 2006. Vibro stone column installation and its Eng., Istanbul, 28–31 August 2001. Taylor and Francis:
effect on ground improvement. In Proc. of Numeri- London, pp. 1867–1870.
cal Modelling of Construction Processes in Geotech- Watts, K.S., Johnson, D., Wood, L.A. & Saadi, A. 2000.
nical Engineering for Urban Environment, Bochum, An instrumented trial of vibro ground treatment sup-
Germany, 23–24 March 2006. Taylor and Francis: porting strip foundations in a variable fill. Géotech-
London, pp. 115–124. nique 50(6): 699–708.
Lee, F.H., Juneja, A. & Tan, T.S. 2004. Stress and pore Weber, T.M., Plötze, M., Laue, J., Peschke, G. &
pressure changes due to sand compaction pile installa- Springman, S.M. 2010. Smear zone identification and
tion in soft clay. Géotechnique 54(1): 1–16. soil properties around stone columns constructed in-
Massarsch, K.R. & Fellenius, B.H. 2002. Vibratory com- flight in centrifuge model tests. Géotechnique 60(3):
paction of coarse-grained soils. Canadian Geotechni- 197–206.
cal Journal 39(3): 695–709. Wheeler, S.J., Näätänen, A., Karstunen, K. & Lojander, M.
McMeeking, R.M. & Rice, J.R. 1975. Finite-element for- 2003. An anisotropic elastoplastic model for soft clays.
mulations for problems of large elastic-plastic defor- Canadian Geotechnical Journal 40(2): 403–418.
mation. International Journal of Solids and Structures
11: 606–616.
218
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
ABSTRACT: The reparation and underpinning of existing structures and infrastructure, due to differ-
ent post-constructive pathologies, often needs the ground improvement of man-made fills and soft soils
in general. The increasing growth of deep soil mixing methods has evolved into the development of the
special Springsol® tool, which permits the installation of deep mixed soil-cement columns under existing
superstructures by the application of new procedures linked with a controlled opening system. Based on
the results of different recently executed field trial tests and projects, this paper describes some advantages
of the Springsol® tool used for the construction of deep mixed columns. The execution parameters and
methods of quality control during and after column installation are described and analyzed, taking into ac-
count installation effects and its influence on the geo-mechanical characteristics of the improved soil. Full
scale test results are presented and compared with results obtained by laboratory tests on core samples.
219
Part of the column with the maximum diam-
eter of 600 mm and part of the column with the
reduced diameter of approximately 300 mm can
be observed. It is important to emphasize that the
variation was executed in a programmed and con-
trolled way. The installation of both vertical and
inclined columns is possible.
⎛ω ω ⎞
Im = N ⎜ b + s ⎟ (1)
⎝ U b Us ⎠
220
predominantly cohesive and granular types of
soils. The cement used for the construction of
these columns in both field trial tests presented
was the Portland type CEM II/B-M 32.5 R. More
details on experiences based on various projects
performed in Spain with the Springsol® tool appli-
cation are given in Melentijevic et al. (2012).
Considering that the soil-cement columns are
stiffer than the surrounding soil, the stress is con-
centrated on them so that the main factor for the
design is the compressive strength. Unconfined
Compressive Tests (UCS), being the most com-
Figure 4. Pumping and automatic flow rate control monly used test for cement treated soil, were con-
equipment. ducted both on core samples and full scale tests,
observing the typical crushing failure for these
types of tests. Core samples (cylinders of 2″ diam-
eter with the relation diameter/height = 0.5) were
obtained in the laboratory by the coring of blocks
that were cut from excavated columns at the field
trial test site. The relationship between axial strain
at failure (εf) and UCS is shown in Figure 14, mostly
covering typical low values for cement treated col-
umns of εf < 1%. The stress-strain curves of the
cement stabilized soil are generally characterized by
high strength and low values of axial strain at fail-
ure, while the native soil is defined by small strength
and large strain at failure. The modulus of elasticity
(E50) is an important parameter for design and is
Figure 5. Spoil collection system, peristaltic bomb, mostly determined by the relationship with uncon-
spoil container. fined compressive strength. These correlations
obtained during the tests are given in Figure 13.
The compressive strength is taken at peak value and
automatically in order to meet the increasing the value of E50 as a secant stiffness modulus at a
or decreasing resistance of soil layers that could stress equal to 50% of the failure stress.
affect the penetration velocity. Triaxial compression tests (CU type—consoli-
– The monitoring by the continuous registering of dated undrained) have also been carried out on core
the execution working parameters in real time samples, presenting clear shear bands developed
(depth (m), penetration velocity (m/h), torque during failure. In triaxial compression tests, with
pressure (bar), rotational speed (rpm), flow rate the increase of the confining pressure the stress-
(L/min) and grouted volume (L/m)). strain relationships change from strain softening
– The system for the management of spoil (see to strain hardening. The typical stress-strain corre-
Fig. 5). The construction of Springsol® columns lations of the soil-cement columns are comparable
is characterized by the “clean” environment after to ones observed in the cases of overconsolidated
its execution due to the system of spoil collection clays or dense sands, considering the increase up
installed at the base of the mast of the drilling to the peak value and then decreasing to residual
rig, which is connected to the peristaltic pump value at the post peak state (see Figure 15).
that is drawing the spoil directly to the container.
Apart from the prevention of contamination of 4.1 Field trial test 1
the working platform with spoil, it allows indirect
control of homogeneity and geomechanical char- Field trial test 1 comprises 12 test columns of 5.0 m
acteristics of the treated or improved soil. length and 400 to 530 mm diameter, performed in
sandy clay. The Springsol® tool applied for the test
is shown on the left side of Figure 2 and represents
4 COLUMN GEOMATERIAL its basic form or mechanism. The consumption of
CHARACTERIZATION cement varied from 250 to 350 kg/m3 of the treated
soil, while the average blade rotation number was
Some of the field trials that were carried out 650, ranging from 350 to 1000, basically due to the
in Spain are presented in this section, both for penetration velocity that was influenced by the soil
221
density/consistency. The lower value of the blade test 1 for the laboratory testing (UCS and triaxial
rotation number corresponds to the upper part CU tests) is presented in Figure 8.
of the column and in general to a softer original
soil layer. The rotational speed of the blades dur-
4.2 Field trial test 2
ing penetration was maintained at 50 rpm, while
the mixing penetration blade velocity ranged from A total of 5 trial test columns were performed,
10 to 25 cm/min. The soil-cement columns were with a minimum length of 3.0 m and 600 mm
exposed after 28 days of their execution, allowing diameter. The ground was defined as silts (ML)
observation of the column shape, diameter, homo- and silty well graded sands (SW-SM), whose sieve
geneity, etc. (see Fig. 6). analysis chart is presented in Figure 9. The average
A few columns were completely extracted for blade rotation number varied from 250 to 750, the
full scale tests. The extracted columns were also lower value being representative to the first meter
cut in blocks in order to obtain core samples for of the column corresponding to soft materials and
laboratory testing by unconfined compression higher penetration rate. The average consump-
tests and triaxial tests. The full scale test and the tion of cement was 300 kg/m3. Trial columns were
failure cracks of the column after the test can be executed with the new modified and improved tool
observed in Figure 7. shown in the right photo of Figure 2. The rota-
Considering that core samples are most prob- tional speed of the blades during penetration var-
ably the principal source for verification of the ied from 55 to 110 rpm in accordance with the soil
properties of cement treated soils, and taking into density/consistency. The mixing penetration veloc-
account that coring is an invasive method that pro- ity ranged from 10 cm/min in a stiffer sandy layer
vokes micro fractures influencing the quality of to 45 cm/min in silts and automatic flow grout
results, less affected or better core samples were injection control was applied.
selected with the objective to provide more realistic The soil-cement columns were fully exposed after
results. The aspect of some core samples of field 28 days of construction to be verified geometrically
and geotechnically. Some of the columns were
extracted to be tested in the laboratory at full scale
in order to determine the unconfined compressive
222
strength and its corresponding elastic modulus 4.3 Results
(E50) after 28 days. The general appearance of the
The correlation of E50 and UCS varies between
excavated columns is given in Figure 10. This shows
50 and 1000, as reported by several published
the column with the uniform diameter of 600 mm
investigations (Bruce, 2001; CDIT, 2002; etc.) for
and the one with the controlled diameter variation
different soils, basically depending on the natural
achieved by the application of the system for open-
soil grain size. This relationship for the field trial
ing and closing blades.
tests is given in Figure 13, varying from 50 to 420
The UCS tests on columns of 600 mm diam-
for core samples and 90 to 380 for full scale tests.
eter were performed at full scale (with the rela-
The summary of values of failure axial strain for
tion of height to diameter greater than 2). The
both field trial tests is presented in Figure 14, with
cracks observed after failure of the columns are
the majority of strains at failure being low values
presented in Figure 11. The UCS and triaxial tests
(εf < 1%).
were also conducted on cored samples of the trial
The results of triaxial tests (CU) for the core
test columns. The appearance of some of the core
samples of field trial test 2 are shown in Figure 15.
samples before and after the failure during tests is
The undrained residual strength parameters are
presented in Figure 12.
5 to 40% lower than the peak values.
The correlation between the blade mixing
number and the UCS is presented in Figure 16,
with the proposed relationship taking into account
both field trial tests.
The quantity of cement in some of the con-
structed Springsol® columns during field trial test 2
Figure 12. The aspect of core samples before and after Figure 14. Correlation between axial failure strain and
UCS tests for field trial test 2. UCS for core samples and full scale tests.
223
Table 1. Parameters of Springsol® columns.
Diameter 400–700 mm
W/C 0.6–1.2
Perforation velocity 15–50 cm/min
Im min 350 rev/m
Ii 150–350 kg/m3
UCS 0.5–6.0 MPa
E50 (50–500) UCS
Shear strength 20–40% UCS
Bending strength 8–15% UCS
5 CONCLUSIONS
224
The relationship between axial failure strain and CDIT (Coastal Development Institute of Technol-
UCS obtained in presented tests covering typical ogy), Japan. 2002. The Deep Mixing Method,
low values of εf < 1% for cement treated columns. A.A. Balkema.
The stress-strain curve is characterized by high Laboratorio de Geotecnia. Cedex. 2011. Informe de lab-
oratorio para Grupo Rodio-Kronsa. Columnas suelo-
strength and low values of axial strain at failure. cemento para el proyecto de investigación de nuevas
The results of UCS tests on core samples, influ- técnicas y herramientas de soil-mixing.
enced by the disturbance produced during their Le Kouby, A., Bourgeois, E. & Rocher-Lacoste, F. 2010.
coring, are lower than values of UCS obtained on Subgrade improvement method for existing railway
full scale tests. It confirms that, for more repre- lines—an experimental and numerical study. EJGE
sentative results, the dimension of the test sample Vol. 15: 461–494.
is of major importance. Actually more tests are Melentijevic, S., Prieto, L. & Arcos, J.L. 2012. Aplica-
being performed in order to verify the influence of ciones de columnas suelo-cemento tipo Springsol®. 9º
test sample dimension, which can be of additional Simposio Nacional de Ingeniería Geotécnica. Ciment-
aciones y Excavaciones Profundas. Proc. Symp.,
importance in heterogeneous soils, very character- Sevilla, 17–19 October 2012: 175–189.
istic for the Iberian Peninsula. Rodriguez Abad, R. & Estaire Gepp, J. 2012. Determi-
nación mediante WD-XRF del contenido de cemento
en suelos inyectados y en mezclas de suelo-cemento. 9º
ACKNOWLEDGEMENTS Simposio Nacional de Ingeniería Geotécnica. Cimenta-
ciones y Excavaciones Profundas. Proc. Symp., Sevilla,
The authors are grateful to the personnel of 17–19 October 2012: 255–268.
Rodio-Kronsa that participated in the execution UNE-EN 15309:2007. Characterization of waste and
of the Springsol® columns, with special assist- soil. Determination of elemental composition by
X-ray fluorescence.
ance from Juan Manuel Dimas, Esteban Casado
and José Luis Arcos. The research program is
partially financially supported by CDTI con Ref:
IDI-20110066.
REFERENCES
225
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
ABSTRACT: Traditionally, the majority of numerical studies investigating stone column behaviour
have studied the problem under unit cell (axisymmetric) conditions, in which the granular columns tend
to be ‘wished-in-place’ (no installation effects). In this study, cylindrical cavity expansion is used to work
out post-installation lateral earth pressure coefficients (and hence a post-installation stress-regime in the
ground) arising due to the lateral expansion and subsequent soil remoulding caused by the vibrating
poker as columns are installed in a soft clay. Two sets of two-dimensional axisymmetric finite element
analyses have then been carried out using PLAXIS 2D to examine load-settlement behaviour, the first set
assuming the coefficient of earth pressure to be unaffected by column installation, while the second set
have been conducted using increased earth pressure coefficients based on the cavity expansion procedure.
The Hardening Soil Model (no viscous effects) has been used to model the behaviour of the granular
column material and the soft clay. Settlement improvement factors calculated using both approaches have
been compared to establish the effect of column installation. Predicted improvement factors have been
put into context by comparison with existing analytical settlement design approaches. The results indicate
that this approach can be used as a realistic means of accounting for column installation in conjunction
with unit cell analyses, with larger improvement factors predicted when installation (increased K) is taken
into account.
226
In this paper, PLAXIS 2D (Brinkgreve et al. process must start from a finite radius, the final
2010) is used to provide an indication of the possi- cavity radius, af, should be obtained by rearranging
ble effect of column construction on K values sur- Equation 2 (i.e. observing volume conservation),
rounding the columns, and in turn, the influence of where a0 is the initial cavity radius.
the new K values on the settlement improvement
factors for an infinite grid. Benchmark (Case A) af 2 a 2 = Rc 2 (2)
unit cell analyses were performed on columns that
were wished-in-place. Case B unit cell analyses Elshazly et al. (2006) have backfigured the post-
involved a modified initial K profile derived from installation coefficient of lateral earth pressure, K,
prior Cylindrical Cavity Expansion (CCE) analy- by matching measured load-settlement behaviour to
ses of single columns. The approach used in Case finite element analyses (by changing K values) car-
B is novel, as the prescribed displacements in the ried out using PLAXIS and found that K/K0 may
CCE stage would cause both excessive heave and range from 1.1 to 2.5, with best estimates of approx-
numerical problems (due to boundary proximity) imately 1.5. Guetif et al. (2007) have used CCE to
for the size of unit cell required to give typical Ac/A evaluate the improvement to the Young’s Modulus
ratios. of a soft clay as a result of column installation.
The improvement factors derived have been put Castro & Karstunen (2010) have applied the cavity
in the context of those derived from existing ana- expansion technique to the undrained installation
lytical settlement design methods based on the unit of a single stone column. The authors confirm that
cell approach (i.e. Priebe 1995, Castro & Sagaseta column installation generates excess pore pressures
2009, Pulko et al. 2011). and increases the horizontal stresses in the soil.
Although not entirely representative of the vibra- The numerical modelling carried out consists of
tory action of the poker and the progressive com- two distinct parts. The first part involves the use
paction of the columns from the base upwards, of CCE to establish post-installation lateral earth
CCE nevertheless provides a convenient means pressure coefficients generated by column installa-
of simulating the lateral expansion of a granular tion in a soft clay soil. The second part involves
column into the surrounding soil. In practical situ- carrying out two sets of wished-in-place unit cell
ations, column installation involves the ‘creation’ analyses under an applied load for (i) a soft clay
of a cavity in the soil, i.e. modelling column instal- assuming the clay properties to be unaffected by
lation using CCE should thus involve expand- column installation (Case A) and (ii) a soft clay
ing a cavity from a zero initial radius to the final with increased lateral earth pressure coefficients
radius of the column. However, numerical simu- calculated using the CCE analyses (Case B).
lations must begin with a finite cavity radius to Essentially, the analyses carried out in this paper
avoid the development of infinite circumferential are intended as a progression of some of the afore-
strain. Consequently, it must be ensured that the mentioned work by Guetif et al. (2007), Castro &
internal cavity pressure reaches the limit pressure Karstunen (2010), etc., albeit with a simpler soil
(plim), as defined by Gibson & Anderson (1961) model in the latter case.
in Equation 1 (p0 is the original in-situ horizontal
total stress, cu is the undrained shear strength of 3.1 Finite Element (FE) model parameters
the soil, E is the Young’s Modulus of the soil and ν
is its Poisson’s ratio). The Hardening Soil (HS) Model has been used to
model the behaviour of both the granular column
⎧ ⎡ E ⎤⎫ material and the soft clay soil. The HS model is a
plim = p0 + cu ⎨1+ ⎢ ⎥⎬ (1) hyperbolic elastoplastic model that takes account
⎩ ⎣ u 2c
c ( ) ⎦⎭ of the stress dependency of stiffness moduli but
does not account for viscous effects such as creep
Carter et al. (1979) have described the use of the or stress relaxation. The formulation of the model
cavity expansion technique in two different types has been described by Schanz et al. (1999). The
of elastoplastic soil, and report that doubling the clay has been modelled as an undrained material
cavity size is sufficient in most cases (the internal (PLAXIS Undrained A approach), with param-
cavity pressure will be within 6% of the ultimate eters defined in terms of effective strength and
limit pressure; any further expansion beyond this effective stiffness. The stone has been modelled as
will only cause further growth of the annular a highly permeable drained material (permeability,
region of yielded soil). Since the cavity expansion k = 1.7 m/day).
227
The soft clay properties (Table 1) are representa- boundary is located 30 m away from the axis of
tive of a simplified single layer profile (no crust) symmetry (to ensure that results are unaffected
loosely based on parameters for the Bothkennar by boundary proximity). For the unit cell radii
soft clay test site (e.g. Nash et al. 1992). A reference required for typical Ac/A ratios encountered in
cohesion of 1 kN/m2 has been used for numerical practice, the boundary would have been too close
stability. The power, m, dictating the stress depend- (leading to numerical problems). Roller bounda-
ency of stiffness moduli with depth has been set ries have been applied to all sides. The finite ele-
equal to 1 in order to simulate logarithmic com- ment mesh (Fig. 2), consisting of approximately
pression behaviour (e.g. Brinkgreve et al. 2010). The 5000 6-node triangular elements, has been refined
soil stiffnesses have been defined so that the ratio in the region surrounding the cavity (largest strains
of the oedometric/constrained moduli between the in this region).
granular material and the clay (commonly referred Simulating column installation in PLAXIS
to as the modular ratio Ec/Es) is approximately 20. 2D using a cavity expansion technique will
The ratio of λ (=Cc/ln10) to κ (=Cs/ln10) has been involve three phases after initial stress generation
fixed at approximately 7 (representative of Both- (initial stresses have been generated using the K0
kennar clay, e.g. Allman & Atkinson 1992) where procedure):
Cc and Cs are the compression and swelling indices
a. Install ‘dummy material’ (e.g. Guetif et al.
respectively.
(2007) suggest modelling the cylindrical hole
The parameters for the granular column mate-
created by the poker using a ‘dummy material’
rial have been chosen based on Killeen & McCabe
with a weak Young’s modulus) over the entire
(2010). The friction angle, ϕ′, has been set equal to
column length to a radial extent, a0.
45o (representative of bottom feed columns). The
b. Apply a prescribed displacement (undrained
dilatancy angle has been calculated as ψ = ϕ′ − 30o.
conditions) from the initial radius, a0, to a final
The oedometric modulus, Eoedref, was assumed equal
radius, af (Fig. 2). Three different a0 values of
to the secant modulus, E50ref and the unload-reload
0.10 m, 0.15 m and 0.20 m have been used for
modulus, Eurref, was calculated as Eurref = 3E50ref (the
which af values have been calculated accord-
default setting in PLAXIS). A complete listing of
ing to Equation 2 as af = 0.316 m, 0.335 m and
model parameters is given in Table 1.
0.361 m respectively.
c. This cavity expansion stage is followed by a
3.2 Installation (CCE) stage for Case B consolidation phase to allow excess pore pres-
sures to dissipate to establish the long-term
A column diameter (D) of 0.6 m (column radius,
stress changes in the ground caused by column
Rc = 0.3 m) has been adopted for this study, which
installation.
is typical of columns at soft cohesive soil sites, e.g.
Watts et al. (2000). The unit cell model is 5 m long This approach is described in detail by Castro &
(i.e. column length, L = 5 m) and the external far Karstunen (2010). The application of a prescribed
displacement in this manner was deemed to be the
best option for numerical stability purposes (Kirsch
Table 1. FE model parameters. 2006, Castro and Karstunen 2010), as opposed to
the application of a volume strain to an expand-
Clay (drained) Stone backfill (drained) ing soil cluster. Based on the increased horizontal
stresses (after the consolidation phase), it will be
γunsat (kN/m3) 16.5 19.0
possible to work out post-installation lateral earth
γsat (kN/m3) 16.5 19.0
pressure coefficients, K, as detailed in Equation 3,
kx (m/day) 1 × 10−4 1.7
ky (m/day) 6.9 × 10−5 1.7
einit 2.0 0.5
φ′ (o) 34 45
ψ (o) 0 15
cref (kPa) 1.0 1.0
K0nc 0.441 0.296
Eoedref (kPa) 3500 70000
E50ref (kPa) 4350 70000
Eurref (kPa) 21874 210000
m (power) 1.0 0.3
pref (kPa) 100 100
ν′ur 0.2 0.2
K0 0.441 0.296
OCR 1.5 –
Figure 2. Finite element mesh for CCE stage.
228
where σ′xx, σ′yy and σ′zz are the effective radial, ver- modelled. Field values of K/K0 are slightly greater
tical, and tangential (hoop) stresses respectively. than HS model values for r/Rc > 8, but the agree-
ment, for the most part, is relatively good for the
⎛ σ ′ + σ ′ zz ⎞ ‘Field 2’ data. Field values are below those pre-
K = ⎜ xx ⎟ (3) dicted by the HS model for r/Rc < 8 (Castro &
⎝ 2.σ ′ yy ⎠ Karstunen (2010) have suggested that this ‘dip’
close to the column may be due to remoulding and
The variation of K/K0 with r/Rc (r is the radial dynamic effects).
distance from the axis of symmetry) calculated by The numerical predictions by Castro &
the HS model is shown in Figures 3a and 3b. K/K0 Karstunen (2010) have been obtained using two
is independent of the amount of lateral expan- ‘user-defined’ anisotropic (the HS model is an iso-
sion and it has also been verified that the K/K0 tropic soil model) constitutive soil models imple-
values are independent of depth (K0 was set equal mented in PLAXIS 2D (L = 10 m, D = 0.8 m). The
to 1 − sin ϕ′ = 0.441 for the normally consolidated S-CLAY1 model, based on the Modified Cam Clay
clay considered in this study). (MCC) model, employs an inclined yield surface
The K/K0 values calculated using the HS model and a rotational component of hardening to model
have been compared to field values of K/K0 anisotropy of plastic behaviour. The S-CLAY1S
reported by Kirsch (2006) mostly beyond r = 5Rc model is an extension of the S-CLAY1 model
and numerical predictions of K/K0 presented by that also takes account of bonding and destructu-
Castro & Karstunen (2010) in Figures 3a and 3b ration (in addition to anisotropy). The computed
respectively. The first set of field data (‘Field 1’) values of K/K0 are lower for S-CLAY1S than for
pertains to 9 m long 0.8 m diameter stone columns S-CLAY1 owing to the destructuration caused by
installed in a silt layer (K0 = 0.91) while the sec- column installation (Castro & Karstunen, 2010).
ond set of data (‘Field 2’) was obtained for 6 m The K/K0 values obtained using the S-CLAY1
long columns installed in a sandy silt (K0 = 0.57); and HS models are in excellent agreement for r/
the latter set are more comparable to the situation Rc < 5 with deviation at higher r/Rc values. K/K0
values obtained using S-CLAY1S appear to be in
good agreement with the HS model for r/Rc > 7.
However, it is clear that the model predictions will
not (and should not) be in perfect agreement owing
to different assumptions inherent in their develop-
ments (in addition, the different geometries and
soil types employed in the different studies will
lead to different predictions). The purpose of the
comparisons carried out was simply to indicate
that the HS model K/K0 values are in reasonable
agreement with previous studies.
The K/K0 decay with r/Rc is approximated by a
stepwise reduction as shown in Figure 4; average
K/K0 values have been interpreted at intervals of
0.10 m representing concentric rings around the
axis of symmetry. Extrapolation has been used
to estimate K/K0 values for r/Rc < 1.15 (numeri-
cal discontinuities in this region lead to difficul-
ties in establishing accurate values, e.g. Carter
et al. (1979)). These values are used as input
parameters for the unit cell (Case B) models.
Also demarcated on Figure 4 is the radial extent
(r/Rc = 3.16) relevant to the largest unit cell mod-
elled (A/Ac = 10).
229
Figure 4. Adopted K/K0 profile.
4 RESULTS
230
ACKNOWLEDGEMENTS
REFERENCES
231
Guetif, Z., Bouassida, M. & Debats, J.M. 2007. Improved Priebe, H.J. 1976. Evaluation of the settlement reduction
soft clay characteristics due to stone column installa- of a foundation improved by Vibro-Replacement.
tion. Computers and Geotechnics 34(2): 104–111. Bautechnik 2: 160–162.
Killeen, M.M. & McCabe, B.A. 2010. A Numerical Priebe, H.J. 1995. The design of vibro replacement.
Study of Factors Affecting the Performance of Stone Ground Engineering 28(10): 31–37.
Columns Supporting Rigid Footings on Soft Clay. Pulko, B., Majes, B. & Logar, J. 2011. Geosynthetic-
Proceedings of the 7th European Conference on Numer- encased stone columns: Analytical calculation model.
ical Methods in Geotechnical Engineering, Trondheim, Geotextiles and Geomembranes 29(1): 29–39.
833–838. Schanz, T., Vermeer, P.A. & Bonnier, P.G. 1999. The
Kirsch, F. 2006. Vibro Stone Column Installation and hardening soil model: Formulation and verification.
its Effect on Ground Improvement. Proceedings of Beyond 2000 in Computational Geotechnics—Ten Years
the International Conference on Numerical Modelling of PLAXIS International, Amsterdam, 281–290.
of Construction Processes in Geotechnical Engineering Vesic, A.S. 1972. Expansion of cavities in infinite soil
for Urban Environment, Bochum, 115–124. masses. Journal of the Soil Mechanics and Foundations
McCabe, B.A., Nimmons, G.J. & Egan, D. 2009. A review Division 98(4): 265–290.
of field performance of stone columns in soft soils. Watts, K.S., Johnson, D., Wood, L.A. & Saadi, A.
Proceedings of the ICE—Geotechnical Engineering 2000. An instrumented trial of vibro ground treat-
162(6): 323–334. ment supporting strip foundations in a variable fill.
Nash, D.F.T., Powell, J.J.M. & Lloyd, I.M. 1992. Initial Géotechnique 50(6): 699–708.
investigations of the soft clay test site at Bothkennar.
Géotechnique 42(2): 163–181.
232
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
X.A.L. Stodieck
Federal Waterways Engineering and Research Institute, Karlsruhe, Germany
Norwegian University of Science and Technology, Trondheim, Norway
T. Benz
Norwegian University of Science and Technology, Trondheim, Norway
ABSTRACT: This paper is concerned with cement grout filtration during installation of pressure
grouted ground anchors in non-cohesive soils. It is aimed for a simple model that can predict the build up
of filter cake during grouting. A series of laboratory tests were carried out to investigate the time depend-
ent expulsion of water from cement grouts at different grouting pressures and water contents. Addition-
ally, permeability tests were performed on filter cake material. Different approaches to model the filtration
process of cement grouts are compared and applied in back calculation of the performed tests. Material
parameters such as permeability and consolidation coefficients are provided for the cement grouts tested.
Amethod to estimate the time required for filter cake formation is proposed.
233
3.15 kg/l the respective void ratio ranges from gives the discharge rate Q in relation to the filter
e = 1.56 to e = 1.28. cake’s current thickness Lc and its permeability kc
The concentration of cement grouts is usually
m
specified by the mass ratio w / z = mwc . Assuming kc σ
Q= A (2)
all voids are fully saturated, the relation between γ w Lc
e and w/z is
where A is the drainage area and σ the total stress
V m γ acting on top of the sample.
e= w = w c, (1) Assuming a constant concentration of cement
Vc mc γ w
particles within each phase, the calculation of filter
cake formation with time is straight forward. The
where mw and mc are the masses of water and
piston deformation during filtration δ, shown in
cement respectively.
Figure 1, can be calculated with the initial void ratio
The void ratio of the dry cement without com-
of the grout eg and the final filter cake void ratio ec:
paction corresponds to a water content between
w/z = 0.5 and w/z = 0.4. The highest bulk density eg − ec
which can be achieved by vibration ranges from δ = Lc . (3)
1.6–1.9 kg/l, which corresponds to e = 0.96 − 0.66 1 + ec
and w/z = 0.30 − 0.21 for full saturation.
Warner (2004) states that cement grout is a non- If water and cement particles are regarded
Newtonian fluid that is a viscous fluid for water incompressible, the volume change of the sample
contents above w/z = 0.6, but has paste consistency is equal to discharge rate of water. Dividing
for water contents below w/z = 0.4. In between it Equation (2) by A gives
flows to some extent but is also capable to store
some energy. dδ kc σ
= . (4)
Therefore, it is assumed that the particles start dt γ w Lc
to contact each other for water contents between
0.4 and 0.5, but might rearrange easily as only little Combining equations (3) and (4) and integrating
contact forces are established between the grains finally gives the filter cake thickness with time
and most stresses are transferred by pore pres-
sure. In this study it is however looked for a filter 2σ kc ⎛ ec ⎞
cake with a stable grain skeleton and consider- L2c = ⎜ ⎟ t. (5)
able strength and stiffness. For grouting pressures γ w ⎝ eg ec ⎠
between 5 kPa and 60 kPa McKinley (1993) found
filter cakes with water contents between 0.3 and The filtration process ends, when the piston
0.5, while Picandet et al. (2011) and Lee et al. (2012) reaches the filter cake at Lc,fin = h0 − δfin.
report filter cake water contents of around 0.3.
1.3 Terzaghi consolidation
1.2 Filter cake formation analysis Other authors used Terzaghi’s classical consolida-
tion theory to describe the filtration process of
The one dimensional filtration process of a suspen-
cement grouts (Lee et al. (2012), Kleyner & Krizek
sion can be modelled by a simple two-phase filtra-
(1995) and Picandet et al. (2011)) and they claim
tion approach. McKinley (1993) used this model
that the process could be modelled reasonably well
to describe the filtration process of cement grouts
with this approach.
and states that the formation of filter cake can be
Classical consolidation theory is based on the
considered as a distinct phase change from a liquid
assumption of homogeneous material properties
to a solid phase. The cement particles retained by
within the sample, constant over time. The dissipa-
the filter medium form a filter cake with constant
tion of excess pore pressures over sample height is
material properties while the suspension retains the
initial concentration. The build up of filter cake is du d 2u
defined by the amount of water expelled from the = cυ 2 , (6)
suspension. dt dz
The rate of water expulsion is limited by the
hydraulic resistance of the filter cake and filter where cv is the consolidation coefficient, related to
medium. McKinley (1993) found that for a suf- the permeability k and stiffness of the material
ficient large permeability of the filter medium, it
k
only influences the very first part of the filtration cυ = . (7)
process and can thus be disregarded. Darcy’s Law γ w mv
234
Figure 1. Filtration test set-up. Figure 2. Typical deformation and pore pressures
measured at the piston during filtration test with grout-
ing pressure = 1000 kpa and w/z = 0.5.
The compressibility coefficient mv is defined as
∈
mv = (8)
σ′
2 FILTRATION TESTS
2.1 Materials
Figure 3. Typical piston deformation compared to the
For all tests Norcem Standard Portland Cement two phase cake filtration solution and classical consolida-
with a specific grain density of 3.15 kg/l was used. tion theory.
The retarding agent (Sika® Retarder) was added to
prevent setting effects during the test procedure.
which only little further deformations takes place.
The pore pressure at top of the sample remains
2.2 Test procedure
constant in a fist phase until it suddenly drops at a
The experimental set-up for a filtration test is shown certain point in time (filtration time tfilt). This shows
in Figure 1. A 54 mm diameter steel cylinder is filled that in the first phase of the tests some parts of the
with cement mix up to a height of h0 = 80 mm and grout are still liquid and do not transfer effective
a piston is placed on top of the sample. stresses. The drop of pore pressures at tfilt indicates
The piston is loaded until the desired constant that cement particles are in contact and trans-
pressure is reached. Then a valve is opened at the bot- fer effective stresses from piston to the bottom.
tom of the test apparatus to allow drainage through Considering cake filtration theory this would be
a filter paper and porous stone. During the filtration the time when the piston reaches the top of the fil-
test water is drained out of the cement grout while ter cake material and all excess water is filtered out.
cement particles are retained by the filter medium. However, more water is expelled after the filtration
The piston deformation δ and pore pressure u are point which shows that after a first phase of filtra-
recorded at top of the sample during the test. tion the filter cake material consolidates further.
Tests with grouting pressures of 100, 200, 500, Figure 3 shows the deformation curve from the
1000 and 1400 kPa were carried out with water test results lying between the deformations calcu-
contents of w/z = 0.4, 0.45, 0.5 and 0.6. With each lated with Terzaghi’s consolidation theory and the
mixture, a minimum of two tests were conducted two-phase filtration model. The two approaches
for each grouting pressure. represent an upper and lower limit. The two
phase filtration model allows a clear distinction
between grout and filter cake while in consolida-
2.3 Results
tion approach the expulsion is associated with the
Results of a filtration test with w/z = 0.5 and a dissipation of excess pore pressure which happens
grouting pressure of 1000 kPa is shown in Figure 2. gradually. While in the filtration approach the pore
The piston deformation is approximately linear pressures on top would theoretically stay constant
with time until it declines at a certain point after until the piston reaches the filter cake, they decrease
235
gradually from the beginning when using the Based on the filtration model the coefficient of
consolidation approach. Although it seems that the permeability of the filter cake kc was determined
filtration approach describes the test results more by rearranging (5) to
accurately, using consolidation theory would be
very handy as it is readily available in common finite h2fiin γ w ⎛ eg ec ⎞
kc = . (10)
element programmes for geotechnical applications. 2σ t ⎜⎝ 1 + e ⎟⎠
fiin c
236
calculation are presented in Figure 5. The The filtration rate for different initial water con-
permeabilities calculated with the consolidation tents and grouting pressures is illustrated in Figure 6.
approach are in general higher than the values from Filtration rates clearly increase with grouting pressure,
filtration approach, as the mean value of the entire due to higher pressure gradients causing higher dis-
sample is taken into account while the filtration charge rates (Equation (2)). Although the test results
approach only considers the permeability of the show some scatter, the filtration rate decreases with
filter cake. The results form permeability tests are increasing initial water content. This finding is plau-
close to the results derived from the filtration tests sible, as with higher initial water content, more water
using two-phase filtration theory. Figure 5 shows must be drained to reach the final water content.
a clear dependency between permeability k and The theoretical filtration rate was back calcu-
final water content of the filter cake. This agrees lated for different initial water contents, using the
with the understanding of a void ratio dependent filtration model. Rearranging Equation (5) and
permeability of granular materials. Picandet et al. differentiation with respect to t gives
(2011) propose to describe the void ratio dependent
permeability of cement grouts with the function
dL
Lc ⎛ 2kk ⎞⎛ ec ⎞
= ⎜ c⎟⎜ ⎟σ. (19)
e 3 d t γ
⎝ w ⎠ ⎝ g ec ⎠
e
k Cx (15)
1+ e
In reasonable agreement with Figure (4), the back
which was originally introduced by Carman (1956), calculation was conducted with a constant, represent-
or with an logarithmic approach ative filter cake void ratio ec = 0.95, corresponding to
a water content of w/zc = 0.3. For w/zc = 0.3 the perme-
Δe ability kc = 7.5 × 10−8 m/s was then determined from
Δ (log10 k ) = (16) the correlation presented in Figure (5). The back cal-
Ck
culated filtration rates fit well with the test data.
suggested first by Taylor in 1948. The Carman
equation fitted for permeability tests on cement 3.2 Coefficient of consolidation
grout filter cake carried out by Picandet et al. (2011) (Consolidation Model)
is plotted in Figure 5. For the permeability tests Using Terzaghi’s consolidation theory the defor-
of this study a better fit was found with Taylor’s mation rate is defined by the coefficient of con-
logarithmic relation given in Equation (16). solidation cv and compressibility coefficient. The
compressibility coefficient mv was estimated with
Equation (8), assuming full consolidation was
3 BACK CALCULATION OF TEST achieved at the end of the test. The coefficient
RESULTS of consolidation cv was estimated from t50 at 50%
consolidation, assuming that the consolidation
3.1 Filtration rate (Filtration model) curve is linear with t . The theoretical solution of
Based on the two-phase filtration model, the filter
cake formation can be calculated directly from the
filtration rate, which is considered constant with
t . The rate of filter cake formation is defined
as the change of filter cake thickness Lc with t
and was determined from test results using the
filtration approach. Assumingg a linear relation
between deformation δ and t , the inclination of
the filtration curve was approximated with
dδ δ
≈ 50 . (17)
d t t50
237
Terzaghi’s consolidation theory at this point gives Figure 9 shows the degree of consolidation U = δ/
the time parameter T = cvt50 /h2 = 0.197 and therefore δfin of a typical filtration test compared to consoli-
the consolidation parameter is determined with dation curves for different values of cv. The consoli-
dation curve for the cement grout passes through
t50 different Terzaghian consolidation curves, which
cv = 0.197 . (20) means cv changes gradually with degree of consoli-
h2
dation U and therefore with effective stress σ′.
This method is a simplification of the ‘log(t)- The filtration tests were back calculated with a
method’ (Casagrande & Fadum 1940), which addi- modified Terzaghi consolidation approach, using
tionally includes the determination of an initial the finite difference method. The sample height was
and final deformation. For this study it was found divided in 30 elements and in each element the con-
adequate to assume δ0 = 0 and take the last meas- solidation coefficient cv was updated for each time
ured deformation as δfin. step. A linear interpolation between the initial cv,ini
The coefficients of consolidation cv versus and final cv,fin consolidation coefficient was used:
grouting pressure determined from this method
are plotted in Figure 7. An increase of cv with cv = cv,ini + (cv, fin
i cv,ini )U . (21)
increasing grouting pressure is shown, while no
clear correlation of initial water content and cv can Unfortunately the correlation presented in
be identified. A linear correlation of cv and grout- Figure 7 can not be used for this purpose as the
ing pressure was hence used for all test results: parameters were determined from t50 and represent
cv = 6.07 × 10−8 ⋅ σ + 7.79 × 10−6. Using this cor- the cv value at 50% consolidation for the given pres-
relation for cv, tests at different grouting pressures sure. The result of this back calculation is shown in
were back-calculated with Terzaghi’s consolidation Figure 9 (solid line).
theory. Results are presented in Figure 8. The cal-
culated consolidation curves fit reasonably well
with the test results.
The assumption of a constant cv value during the
test, as used in classical consolidation theory, might
not be appropriate for this material, as it exhibits large
changes in permeability and stiffness during the con-
solidation process. Both parameters, permeability km
and compressibility mv, decrease with decreasing void
ratio σ ′ or increasing effective stress cv. Looking at
Equation (7), cv could either decrease, increase or stay
constant with effective stress, dependent on which
parameter changes more. It is usually assumed that
the change of km and mv cancels each other out and
the assumption of a constant cv is appropriate. This
does not apply for all materials, and Abuel-Naga &
Pender (2012) proposea modified consolidation
approach with linear effective stress dependent cv. Figure 8. U versus t for w/z = 0.4, 0.45, 0.5 and 0.6
for different grouting pressures. Back calculation of test
results with Terzaghi consolidation theory (red line).
Figure 7. Consolidation parameter cv versus grouting Figure 9. Consolidation curve of filtration test in com-
pressure. Best fit for all values cv = 6.07 × 10−8σ + 7.79 × 10−6 parison to back calculation with constant cv and effective
with R2 = 0.976. stress dependent cv (cv,ini = 2.5e − 5; cv,fin = 8.0e − 5).
238
3.3 Estimation of filter cake thickness tents of the cement grouts used range between the
limit for liquid and plastic consistency and in this
In the two phase filtration model consolidation
case the two models represent an upper and lower
of the filter cake is not taken into account and the
limit to describe the filtration process.
extension of the final filter cake material is directly
Required parameters for the filtration model are
calculated from Equation (5). A clear distinction
the initial and final water contents and the perme-
between filter cake and unfiltered grout is made.
ability of the filter cake. The filter cake material
In contrast, no distinction is made between fil-
exhibits a void ratio dependent permeability kc
ter cake and grout in classical consolidation the-
between 3.81 × 10−8m/s and 1.97 × 10−7m/s. The val-
ory, as the whole sample consolidates gradually
ues calculated on the basis of filtration theory com-
with time. Material properties such as stiffness and
ply well with the directly measured values, which
permeability are regarded constant. However, if
confirms that the filtration model is applicable.
we assume that a stable contact of grains is estab-
Assuming a final water content of w/zfin = 0.3 and
lished at a certain void ratio, the expansion of filter
the corresponding permeability kc = 7.5 × 10−8m/s
cake could then be related to the degree of consoli-
from figure (5) the filtration rate could be back-
dation U.
calculated for all tested grouts.
The consolidation model requires the consoli-
δ
U= . (22) dation coefficient cv and compressibility mv of
δ ffin
i the material. The compressibility mv can be deter-
mined from the strain required to achieve the final
The consolidation ratio U at tfilt, at which pore water content and the applied grouting pressure.
pressures at top of the sample start to decrease, Values for cv were found between 6.86 × 10−6 m2/s
ranges between 0.75 to 0.95, which corresponds and 1.33 × 10−4 m2/s, based on Terzaghi’s consoli-
to water contents between 0.28 to 0.36. Theoreti- dation theory. The mean values of all tests can be
cally the deformation and therefore degree of con- described by a linear correlation between cv and
solidation at tfilt depends mainly on the initial water grouting pressure. With this correlation tests could
content of the grout, as this defines the amount of be back calculated using consolidation theory. To
deformation needed to reach a certain mean water determine the filtration point, it is proposed to use
content in the sample. In addition the applied the degree of consolidation U. By assuming a final
grouting pressure defines the compression of the water content w/zfin = 0.29, a mean water content
filter cake material. at the filtration time w/zfilt = 0.32 was found as
However, no significant correlation between maximum value to set limits to the determined U
initial water content, grouting pressure and U values.
was found. This could be due to variations of The determined calculation parameters for both
the test results and inaccuracies of the interpreta- models are dependent on the final water content
tion. Assuming that all tests result in a final water of the filter cake material which was found to vary
content of 0.29 as indicated in Figure 4, a mean between 0.25 and 0.35. No correlation between
water content at the filtration time of w/zfilt = 0.32 w/zfin and grouting pressure could be identified.
would give Ufilt values of 0.75, 0.83, 0.87 and 0.91 Therefore it is not possible to determine the void
for initial water contents of 0.75, 0.83, 0.5 and 0.6 ratio at which the grout actually acts as a solid
respectively. Although the compression of filter and deformation is stress dependent. It seems that
cake for different grouting pressures is ignored in the scatter of test results exceed the influence of
this approach, it fits in the range of test results. grouting pressure and initial water content for the
range of water contents and grouting pressures of
this study. Where no correlations could be identi-
4 SUMMARY fied, mean values of all tests were taken for back-
calculation of test results, which gave reasonable
This paper presents a series of filtration tests on results. These parameters might not be applicable
cement grouts with water contents ranging from for water contents and grouting pressures outside
0.4 to 0.6. The influence of initial water content the test range.
and grouting pressure was investigated and test The filtration model allows a straight-forward
results were back calculated with a two-dimen- determination of the filter cake thickness with
sional filtration model and classical consolidation time, but for an investigation of the filtration proc-
theory. ess in borehole geometry the use of consolidation
Both models could be used to back-calculate theory could be favourable, as it is already imple-
the laboratory tests. While the filtration model is mented in common geotechnical FEM software.
designed to describe filtration of solid particles When using consolidation theory it is not straight-
from a suspension, consolidation theory considers forward to define the filtration time, especially as
the expulsion of water from a solid. The water con- calculated deformations are inaccurate in the final
239
part of the curve. However, a rough estimation of Lee, S.-W., Kim, T.-S., Sim, B.-K., Kim, J.-S. & Lee, I.-M.
the minimum filtration time is possible. 2012. Effect of pressurized grouting on pullout resist-
ance and group efficiency of compression ground
anchor. Can. Geotech. J. Canadian Geotechnical
Journal 49(8), 939–953.
REFERENCES McKinley, J.D. 1993. Grouted ground anchors and the soil
mechanics aspects of cement grouting. Ph. D. thesis.
Abuel-Naga, H.M. & Pender, M.J. 2012. Modified terza- Picandet, V., Rangeard, D., Perrot, A., & Lecompte, T.
ghi consolidation curves with effective stress-dependent 2011. Permeability measurement of fresh cement
coefficient of consolidation. Gotechnique Letters 2(2), paste. Cement and Concrete Research 41(3), 330–338.
43–48. Taylor, D.W. 1948. Fundamentals of soil mechanics. New
Carman, P.C. 1956. Flow of gases through porous media. York: J. Wiley.
London: Butterworths Scientific Publications. Warner, J. 2004. Practical Handbook of Grouting—Soil,
Casagrande, A. & Fadum, R.E. 1940. Notes on soil testing Rock, and Structures. Norwich, NY: John Wiley &
for engineering purposes. Cambridge, Mass.: Harvard Sons.
University, Graduate School of Engineering.
Kleyner, I. & Krizek, R.J. 1995. Mathematical Model for
Bore-Injected Cement Grout Installations. Journal of
geotechnical engineering. 121(11), 782–788.
240
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
J. Wehr
Keller Holding GmbH, Offenbach, Germany
ABSTRACT: Piles, vibro concrete columns and rigid inclusions cannot be installed in liquid media like
water with cu = 0 kPa, because a cone shaped slope will form instead of a cylindrical body. In international
standards a limit of cu = 15 kPa is currently used as a lower boundary. Additionally the minimum center
to center distance between fresh concrete displacement piles without permanent casing is specified in
EN 12699 between 6–10 times the diameter depending on the cu-value. These limits should be applied
to all kinds of displacement piles like rigid inclusions in order to avoid damaging neighbouring columns
during installation.
However, recent world-wide site experiences reveal that these limits are not valid for granular columns like
vibro stone columns. The limit of the undrained cohesion should be reduced to cu ≈ 5 kPa.
1 INTRODUCTION Austria:
• Klagenfurt, VSC, metro-market, cu = 5–10 kPa
It is common knowledge that vibro stone columns (Marte et al., 2005)
and vibro concrete columns cannot be installed in
liquid media like water with cu = 0 kPa, because a Germany:
cone shaped slope will form instead of a column. • Lübeck-Herrenwyk, VSC, metal factory,
In the German recommendation (FGFS 1979) for cu = 11–26 kPa (Völzke, 2001)
soil improvement with depth vibrators, dated 1979, • Mering, VCC, railroad embankment,
limits of cu = 15–25 kPa are given. Since that time cu = 5–20 kPa (Borchert et al. 2004)
the lower limit value has been, and still is, used in • Rethen, VSC, sugar plant, cu = 12–18 kPa
all kinds of German and international standards. • Zossen, VCC, dam for ring road, cu = 4–8 kPa
However, more than 10 years of site experiences of (Zimmermann, 2003)
the Keller company around the world reveal that Malaysia:
this limit value is far too conservative and should
be reduced to cu = 5 kPa. • Shah Alam Expressway, VSC, (Raju/Hoffmann
The aim of this publication is to find out the limit 1996, Raju 1997)
value of the undrained cohesion where a column Kinrara Interchange, VSC, cu = 6–40 kPa
installation is no longer possible. Site experiences Sri Petaling Interchange, VSC, cu = 8–50 kPa
of Keller have been compiled and summarized and Shah Alam West, VSC, cu = 5–15 kPa
furthermore systematic model tests are presented • Putrajaya Boulevard,VSC, cu = 5–20 kPa
varying the cu-value. (Raju, 2002)
Poland:
2 PREVIOUS SITE EXPERIENCES • VSC, highway embankment, cu = 5 kPa
(Marte et al., 2005)
The cu-values of Keller sites have been com- In summary, the experience of all the above pre-
piled, where an installation of vibro stone col- vious sites show that the installation of vibro stone
umns (VSC) or vibro concrete columns (VCC) columns and vibro concrete columns is possible
with a depth vibrator was possible in soils with well below the old limit value of cu = 15–25 kPa,
cu < 15 kPa. Details may be found in the following and that the actual limit value should be approxi-
references. mately cu = 4–5 kPa.
241
3 MODEL TESTS AT KAISERSLAUTERN
UNIVERSITY
242
Table 2. Comparison of vibrators.
Reality Model
243
Marte, R. & Schuller, H. 2005. Verbesserung sehr weicher Raju, V.R. 2002. Vibro replacement for high earth
Seesedimente und Torfe durch Schottersäulen—zwei embankments and bridge abutment slopes in Putra-
Fallbeispiele, Bauingenieur, Volume 80, 430–440. jaya, Malaysia, International Conference on Ground
Perlea, V. 2000. Liquefaction of cohesive soils, soil Improvement Techniques, Malaysia, p. 607–614.
dynamics and liquefaction, ASCE geotechnical special Völzke, B. 2001. Böschungssicherung für die Sanierung
publication, no. 107, 58–76. des Metallhüttengeländes in Lübeck-Herrenwyk, 8.
Raju, V.R. & Hoffmann, G. 1996. Treatment of tin mine Darmstädter Geotechnik Kolloquium, 233–239.
tailings in Kuala Lumpur using vibro replacement. Proc. Zimmermann K.-U. 2003. Gründung von Verkehrswege-
12th SEAGC, May 1996, Kuala Lumpur. bauten in Feuchtgebieten mit organischen Böden
Raju, V.R. 1997. The behaviour of very soft cohesive soils geringer Scherfestigkeit, Mitteilungen des Instituts für
improved by vibro replacement, Ground Improvement Grundbau und Bodenmechanik, Universität Braunsch-
Conference, London. weig, Pfahlsymposium, Issue 71, 71–81.
244
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
ABSTRACT: Buried corrugated steel culverts, cushioned on top by a layer of granular material, are
sometimes used to protect infrastructure from rock falls. A crucial point in design of such structure is
the cushion material ability to absorb the kinetic energy from the rock block. This is evaluated by two
important parameters, i.e. the resulting impact load and penetration of the falling rock. In this article a
Discrete Element Model (DEM) is used in simulation of the impact from a falling rock block on such a
cushion layer. The simulation is compared to measurement data from a full-scale field test. Applicability
and potential of DEM to simulate rock fall impacts on cushion are assessed and discussed in the light
of measurements from full scale test and estimations using empirical relations proposed by Norwegian
guideline.
245
applied for rockfall impact analysis on e.g. embank- large scale odeometer has diameter 500 mm and a
ments (see Plassiard and Donzé, 2009). The experi- height of 400 mm. The tests were run by loading
mental study presented by Ebeltoft et al. (2006) to several load levels with unloading between each
and Ebeltoft and Larsen (2006) is used as a basis load increment. Figure 3 shows results of such
for the numerical study presented in this paper. two oedometer tests, where preloading to 25 kPa
was done prior to the presented data. During the
preloading as much as 10 mm of displacement was
measured. The two tests shows some differences,
2 BACKGROUND
however, one can estimate the oedometer stiffness
to be between 18 and 26 MPa for a stress increment
2.1 Experimental work
of 500 kPa loading from 50 kPa.
Ebeltoft and Larsen (2006) studied an alternative Figure 4 shows high speed–camera documenta-
protection system based on a buried steel structure. tion of the rock fall during impact.
Figure 2 shows an illustration of the experimen-
tal setup. A 15.0 m long structural MP200 arch
type metal culvert was installed. Backfill material
in the near zone consists of 8–16 mm gravel. The
outer zone consists of 23–64 mm size gravel. The
top 1.2 m of the fill made the so-called cushion
layer, which was constructed without compaction.
The experimental study used natural rock boul-
ders with different weights and drop heights. The
test discussed in this paper involves a 200 kg rock
dropped from a 10 m height.
Large scale oedometer tests were ran on the
gravel to find oedometer stiffness properties. The
246
2.2 Semi empirical approaches and Strack (1979). STAR-CCM+ uses a classical
mechanics method to model DEM, and is based on
The Norwegian practice for calculating penetra-
soft-particle formulation where particles are allowed
tion depth and maximum contact force (SVV,
to develop an overlap. The calculated contact force
2011) is adopted from the Swiss recommendations
is proportional to the overlap, as well as to the parti-
given by the “Bundesamt für Strassen” (ASTRA,
cle material and geometric properties (CD-adapco,
2008). Accordingly, Equation 1 and 2 are adopted
2012). In addition to the standard linear momen-
to estimate a maximum contact force, Pmax [kN],
tum, the formulation incorporates the angular
and penetration depth, t [m], respectively.
momentum conservation equation (equation 4).
R 02
Pmax = 2.8 R tan ϕ ⋅ Ek 0.6
M E 0.4 tan (1) ⎛ ⎡ ω p ⎤⎞
e d
(
Ip p ) ∑ ⎜ rc ⋅ ⎢ Fc Crr Fc ⋅ n c ⋅ ⎥⎟
dt contacts ⎜
⎝ ⎢⎣ ω p ⎥⎦⎟⎠
2Ek
t= (2) (4)
Pmax
where, Ip is the particle moment of inertia, ωp is the
where, R is the radius of falling boulder [m], e is particle angular momentum, rc is the vector from
the soil layer thickness [m]. ME is the soil modulus the particle center of gravity to the contact point,
(Modulus of subgrade reaction) [Pa], ϕ is the fric- Fc is the contact force vector, Crr is the Rolling Fric-
tion angle [°], and Ek is the energy of the falling tion Coefficient and nc is the is a particle surface
boulder during impact [J]. normal at the contact point. The contact force vec-
According to SSV (2011), ME can be estimated as: tor is controlled by the contact theory.
In the work presented in this article, the Hertz
Eoed Mindlin contact theory is used for the simula-
ME = (3) tions. The inputs to the simulations are then the
2
Static and Rolling Friction Coefficient (μ and
where, Eoed is the oedometer modulus. Crr respectively) and the Normal and Tangen-
If one assumes ME = 10 MPa, Ek = 19.6 kJ, tial Coefficients of Restitution (COR) (εn and εt
tanϕ = 1.0, R = 0.26 m, and e = 0.6–1.2 m, then respectively), both for particle-particle interac-
Pmax = 240–330 kN and t = 0.12–0.16 m. Figure 5 tions and particle-wall interactions. In addition to
gives the deceleration history for the 200 kg stone these properties, the elastic stiffness of the parti-
in the field test. From the figure one can estimate cles, E, Poisson’s ratio, ν and particle density, ρ,
the maximum force to be about 220 kN, when are analysis inputs. In this study spherical parti-
multiplying the deceleration with the weight of cles with normally distributed diameters of the
the stone. The measured penetration depth after particles were used.
impact was 0.20 m.
3 DEM SIMULATIONS
2.3 Discrete element method
In this work, the Star-CCM+ program v. 7.04 (CD- 3.1 Discrete element properties
adapco, 2012) is used for the dynamic discrete ele- The properties used in the simulations are as given
ment simulations. DEM was established by Cundall in table 1 and table 2. The spheres used in the sim-
ulations were normally distributed with a mean
diameter of 43 mm and a standard deviation of
9 mm. In addition, the minimum and maximum
diameters were set to 23 mm and 64 mm, respec-
tively. This gives a cumulative weight distribution
curve as presented in figure 6. The particles were
loosely packed with a random distribution that
gives a void ratio of about 0.9.
ρ [kg/m ]
3 E [GPa] ν
2700 50 0.3
Figure 5. Deceleration history for the 200 kg stone.
247
Table 2. DEM particle interaction
properties.
μ Crr εn εt
248
take at least 10 time steps for the particle to move
the full length of the radius. The three criteria are
given in equation 5 to 7 below.
π ⋅ Rmin
Δtt1 = (5)
vRayleigh
⎛ 5 2 ⋅π ⋅ ρ 1− ν 2 ⎞ R
Δt2 = 0.1 ⋅ 2.94 ⋅ ⎜ ⋅ ⎟ ⋅ (6)
⎝ 4 E ⎠ vimpact
5
R
Δt3 = 0.1 ⋅ (7) Figure 10. Penetration depth versus time simulated
v particlel with DEM.
249
− The test was done with a flexible culvert. The age will increase the demand for computational
idea with the flexible culvert is to reduce the resource.
impact load. The numerical study support such The DEM simulations should be coupled with
an effect. Hopefully further studies will support the finite element or the finite volume method
this hypothesis. for the simulation of the culvert and far field parts
− The shapes of the boulders in the tests were not of the model. This is not possible in the present
as spherical as in the DEM simulation. Degago version of the software. However, it is in the future
et al. (2008) gave attention to the shape effect plans of the software developers. For details on
on penetration depth and maximum peak load. such coupling one can see e.g. Rojek and Oñate
They found that spherical objects give larger (2008). Such analysis will potentially provide a bet-
impact forces and less penetration depths than ter documentation on the effects rock fall on flex-
pyramidal shaped objects. ible culverts and will potentially be used in design
− For peak deceleration it is believed that the hori- of rock shield structures.
zontal distance to the boundaries has no effect Finally back calculation of other drop heights
on the results. However, for final settlement and and boulder masses should be carried out. In the
the deceleration process a box of 2 m x 2 m is tests mass up to 2650 kg and drop heights of 20 m
small, both when it comes to reflections from the were used. Such large boulders with such high
boundaries and particles sliding/rolling along kinetic energy would require that a large zone
the boundaries. should be modeled with DEM, again increasing
calculation time.
The effect of particle properties in the DEM
simulation, when it comes to both shape and mate-
rial properties, such as stiffness etc. should explain ACKNOWLEDGEMENTS
some of the differences. Finally the DEM simula-
tion done in this study does not take into account Authors are grateful to acknowledge the Norwe-
particle crushing. gian Public Roads Administration for financial
support for the experimental studies. The authors
also wish to thank the International Centre for
6 CONCLUSIONS AND Geohazards (ICG) and the Research Council of
RECOMMENDATIONS Norway for supporting this research.
250
Imre, B, Räbsamen, S. & Springman, S.M. 2008. A coef- Pichler, B., Hellmich, C. & Mang, H.A. 2005. Impact of
ficient of restitution of rock materials. Computers & rocks onto gravel, design and evaluation of experi-
Geosciences, 34(4): 339–350. ments. International Journal of Impact Engineering
Jacquemound, J. 1999. Swiss Guideline for the Design 31(5): 559–578.
of Rockfall Protection Galleries: Background, Saftey Plassiard, J.-P. & Donz´e, F.-V., 2009. Rockfall Impact
concept and case histories, Joint Japan-Swiss Scientific Parameters on Embankments: A Discrete Element
Seminar on Impact Load by Rock Falls and Design of Method Analysis, Struct. Eng. Int., 19, 333–341.
Protection Structures, Kanazawa, Japan, 95–102. Rojek, J. & Oñate, E., 2008. Multiscale analysis using a
Kawahara, S. & Muro, T. 2006, Effects of Dry Density coupled discrete/finite element model. Interaction and
and Thickness of Sandy soil on Impact Response due Multiscale Mechanics, 1(1): 1–31.
to Rockfall, Journal of Terramechanics, 43: 329–340. SVV, 2011. Sikring av veger mot steinskred, Grunnlag for
Kishi, N., Ikeda, K., Mikami, H. & Takemoto, S. 1999, veiledning, VD rapport, No. 32.
A proposed New Design Procedure for RC Rock- Timoshenko, S. 1951. Theory of Elasticity. McGraw-Hill
Sheds, Joint Japan-Swiss Scientific Seminar on Impact Book Company.
Load by Rock Falls and Design of Protection Struc- Volkwein, A., Schellenberg, K., Labiouse, V., Agliardi, F.,
tures, 103–112. Berger, F., Bourrier, F., Dorren, L.K.A., Gerber, W. &
Labiouse, V., Descoueudres, E. & Montani, S. 1995, Jaboyedoff, M. Rockfall characterisation and struc-
Numerical Analysis of Rock Blocks Impact a soil Cush- tural protection—a review. Nat. Hazards Earth Syst.
ion, Numerical Models in Geomechnics-NUMOG V, Sci., 11, 2617–2651.
Pande & Pieetruszczak (eds), 645–650. Yoshida, H., Masuya, H. & Ihara, T. 1988, Experimen-
Masuya, H., Tanak, Y., Onda, S. & Ihara, T. 1999, tal Study of Impulsive Design Load for Rock Sheds,
Evaluation of Rock Falls on Slopes and Simulations IABSE Proceedings P-127/88, 61–74.
of the Motion of Rock Falls in Japan, Joint Japan-
Swiss Scientific Seminar on Impact Load by Rock
Falls and Design of Protection Structures, Kanazawa,
Japan, 21–28.
Montani, S.S., Labouse, V. & Descoeudres, F. 1999,
Action of Falling Blocks Impacting Rocksheds Cov-
ered with a Soil Cushion, Joint Japan-Swiss Scientific
Seminar on Impact Load by Rock Falls and Design of
Protection Structures, Kanazawa, Japan, 51–57.
251
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
ABSTRACT: The paper presents the results of field tests performed to study the effects of the installa-
tion technique, degree of plugging, cyclic loading and ageing on the shaft resistance developed on open-
ended piles in sand. Two instrumented model piles were jacked and driven into an artificially created
loose sand deposit in Blessington, Ireland. The results from these tests indicated that the equalized radial
effective stresses which are suggested to control the shaft capacity vary strongly with the degree of plug-
ging and number of load cycles experienced during installation. A comparison of jacked and driven
installations suggest similar radial stresses were developed provided the jacked pile had experienced a suf-
ficient number load cycles. The degree of plugging experienced during installation controlled the radial
stresses near the bottom of the pile, with closed-ended or plugged piles developing high stresses near
the pile base and exhibiting friction fatigue up the shaft, compared with open-ended coring piles which
exhibited relatively low stresses along the length of the pile shaft. A comparison with full scale 340 mm
diameter pipe piles driven into the dense sand in Blessington noted comparable radial stresses when the
pile was fully coring but exhibited a larger increase in radial stress near the pile toe as the pile became
plugged. Further tests on the effects of ageing show a pile shaft capacity increase of 260% over 220 days
after driving. Further research is underway in Blessington to investigate the mechanisms controlling this
ageing behaviour.
252
cal spreading and cyclic degradation (the primary where ΔLp is the change in plug length for a given
causes of friction fatigue) to vary principally with change in pile penetration, ΔL (IFR = 0 means fully
h/R, where R is the pile radius, for closed-ended plugged, IFR = 1 means fully coring). It was sug-
piles. gested that using qb instead of qc as a design input
Chow (1997) sought to extrapolate from the parameter in equation 3 (and using R instead of
closed-ended ICP tests to establish an approxi- R*) could provide a more logical extrapolation from
mate link which could be used for the design of closed-ended behaviour. White et al. (2005) pre-
open-ended piles. It was suggested that using a sented cavity expansion analyses which suggested
modified R* term, where R* is equal to the radius a weaker dependence on IFR than that proposed
of an equivalent closed-ended pile of the same by Gavin and Lehane (2003). The UWA-05 design
solid area, would result in a higher rate of friction method was developed by modifying the original
fatigue degradation which could be applicable for ICP equations slightly, by dropping the vertical
open-ended piles. The aforementioned research led effective stress function and using Ar,eff instead of
to the development of the Imperial College Pile R* to account for open-ended behaviour, as seen
design method (ICP-05) shown in equation 3: below:
σ′rc = 0.029 ⋅ qc ⋅ (σ′v0/Pa)0.13 ⋅ (h/R*)−0.38 (3) σ′rc = 0.03 ⋅ qc ⋅ (h/D)−0.5 ⋅ Ar,eff0.3 (6a)
where σ′v0 is the in-situ vertical effective stress Ar,eff = 1 – IFR (Di/D)2 (6b)
and Pa is atmospheric pressure (≈100 kPa). The
ICP-05 method includes an expression to predict where D is the external pile diameter and Di is the
the stress increase caused by dilation, Δσ′rd, based internal pile diameter.
on the work of Lehane and Jardine (1994): While both the UWA-05 and ICP-05 methods
offer comparably good performances in database
Δσ′rd = (4G/D) Δy (4) studies, Gavin et al. (2011) showed that when these
were used to design piles for a typical North Sea
where G is the operational shear modulus of the oil platform, each method gave significantly differ-
soil (which can be correlated with CPT qc) and Δy ent pile length requirements. Recognizing that the
is the radial displacement occurring during pile ICP-05 and UWA-05 methods for open-ended pile
loading (≈ 0.02 mm for lightly-rusted steel piles). design were fundamentally based on the results of
The Δσ′rd component is inversely proportional instrumented closed-ended model piles and which
to pile diameter and as such was considered to be were subsequently adapted and calibrated against
relatively small for field scale piles. The method databases of field scale open-ended piles, the
for predicting local shear stress, which was seen to UCD Geotechnical Research Group conducted a
be primarily controlled by the CPT qc and (h/R*) prolonged research campaign to further improve
components, was a major advancement over tra- understanding of open-ended pile behavior.
ditional design methods and has been shown to
offer significantly improved reliability in database
2 EXPERIMENTAL DETAILS
studies.
The extrapolation of the measurements of
2.1 UCD instrumented piles
radial stress on closed-ended piles to the design of
open-ended piles has caused much debate. Gavin Model open- and closed-ended piles were used
and Lehane (2003), Lehane et al. (2005), Igoe et al. in the tests described in this paper. The stainless
(2011) and others have suggested that using the steel open-ended pile had an external diameter of
R* parameter to extrapolate to open-ended pile 168 mm and a wall thickness (t) of 9 mm, giving a
behavior does not account for the complex con- D/t ratio typical for pipe piles. The lower instru-
ditions such as soil plugging experienced during mented section, 2 m in length, was constructed
the installation of an open-ended pile. Gavin and using the twin-walled technique, where two strain
Lehane (2003) suggested that an open-ended pile gauged steel pipes with slightly different diameter
would have a lower installation base resistance are joined (at the top) thus allowing separate meas-
(qb), which in turn would result in lower shaft urement of the internal and external shaft resist-
resistance. The qb was shown to depend on the qc ance and the base load developed on the annulus of
and the amount of plugging experienced during the pile toe itself. Three instrumented units which
installation as defined by the Incremental Filling housed the radial total stress sensors and pore
Ratio (IFR): pressure transducers were located at h/D = 1.5,
5.5 and 10.5. Spacer sections machined from the
IFR = ΔLp/ΔL (5) same pipe, which were 1.0 m in length and of equal
253
diameter, could be used to allow pile installation to of the pile have been described by Kirwan et al.
6 m. The stress sensors were TML (Tokyo Sokki (2012). For continuity with previous publications
Kenkyujo Co.) PDA-500 kPa, and had a rated the open—and closed-ended model piles are des-
capacity of 500 kPa. The same sensors were used ignated pile OE and CE respectively while the full
to measure pore pressure response (located diamet- scale pile is designated pile S5.
rically opposite the total stress sensors). A porous
ceramic disc was mounted flush with the outer pile
2.2 UCD test site
surface in front of each of the pore pressure sen-
sors. The location of the instrumentation, includ- The experimental pile tests were performed at the
ing the electrical resistance strain gauges are shown University College Dublin (UCD) sand test site
in Fig. 1. Full details of the design, construction which is situated in Blessington, 25 km southwest
and calibration of the model pile are contained in of Dublin city. The ground conditions comprise
Igoe et al. (2010). of glacially deposited, very dense, fine siliceous
The model closed-ended pile used in the field sand with a CPT qc resistance in the range of 15 to
tests was a 73 mm, stainless steel pile with radial 20 MPa, and small strain stiffness (G0) in the range
total stress and pore-water pressure sensors at the 75 to 150 MPa. Sand replacement tests indicated
same relative locations (h/D = 1.5, 5.5 and 10.5) as that the relative density of the in-situ sand deposit
the open-ended piles and a similar level of overall was between 90% and 100%. Particle size distri-
instrumentation. This instrumentation is described bution analyses performed on samples indicated
in detail by Igoe et al. (2010). that the mean particle size, D50, was approximately
In addition to the model piles, a larger scale, 0.15 mm. The equilibrium ground water table was
340 mm diameter, 7 m long, steel, open-ended approximately 13 m bgl. The natural water content
pile was instrumented with four radial total stress was relatively uniform at 10–12% above the water
sensors at relative locations (h/D = 1.5, 5.5, 10.5 table and was found not to vary seasonally.
and 17.5). The sensors used were the same make/ To allow for comparison of pile behaviour in
type as those used in the model piles only with a conditions where only the sand state varied, a
higher rated capacity of 3 MPa (to survive driving trench 2.5 m wide, 10 m long and 6 to 7 m depth
and denser soil conditions). Details of the design was excavated at the site. Immediately following
excavation the material was backfilled into the
trench at its natural moisture content (10–12%)
using an excavator bucket maintaining a drop-
height of between 1 and 2 m, with the intention
that this air pluviation technique would produce
a uniform loose sand deposit. Four CPT profiles
were performed at 2 m spacings along the trench.
The CPT qc profiles are compared with profiles
measured in the in-situ, dense sand in Fig. 2a. The
CPT profiles were consistent at all locations in
the trench, with qc decreasing from 2 MPa near the
ground surface to 1.0 MPa at depths below 1.5 m
bgl until the cone reached the base of the trench
at 6–7 m bgl. This compared to CPT qc values of
15–20 MPa in the dense in-situ soil over this depth
range. The friction ratios (Fr = ratio of CPT sleeve
friction fs to cone resistance qc) were similar at
1–2% in the loose and dense sand deposits. Profiles
of the small strain shear modulus, G0, determined
by Multi-channel Analysis of Surface Waves
(MASW) are shown in Fig. 2c. The G0 values in
the trench increased from 20 MPa near the ground
surface to 40 MPa at depth.
254
Figure 2. CPT profile at Blessington.
installed through initial starter holes, varying from lent to the short term equalized radial stresses (σ'rc),
0.85–1.9 m in depth. The first test (OE1) involved were measured between jacking strokes. The jacked
jacking the open-ended model pile into the sand, closed-ended model pile (CE) developed large σ'rs
using the CPT truck as reaction, in 100 mm jacking values near the pile toe during installation, and
strokes to a final depth of 5.9 m. The installation exhibited friction fatigue with much smaller values
was paused between each jacking stroke to moni- noted further up the pile shaft. Similar behaviour
tor the plug development and record the stationary was noted for the jacked OE model pile when it
radial stresses. The second test (OE2) involved driv- was experiencing full plugging during installation.
ing the pile to a depth of 4.1 m, using a specially fab- The driven OE model pile, which was virtually
ricated 25 kg hammer falling through a drop-height fully coring throughout installation (IFR > 80%),
of 0.5 m. The pile was jacked a further 1.45 m into exhibited very little friction fatigue, with low σ'rs
the ground using the CPT truck with 50 mm jack- values noted along the entire shaft. Compression
ing strokes. The third test involved the installation cyclic load tests performed on the CE and OE
of the 73 mm diameter closed-ended model pile jacked piles showed a significant reduction in σ'rs
(CE), which was installed using the CPT truck in over the first 20–50 load cycles, reducing to values
50 mm jacking strokes to a final depth of 3.55 m. similar to those measured on the driven model pile
At the end of installation for each model test (OE1, (at sensor locations remote from the pile toe). The
OE2 & CE) a series of cyclic tests were performed combined effect of plugging and cyclic loading on
to assess the effects of cyclic loading on the radial the model piles is shown in Fig. 3. It is evident that
stresses. A final pile test involved driving the full for closed-ended piles or fully plugged open-ended
scale pile S5 into the dense natural sand deposit piles (IFR = 0) that the σ'rs values near the pile toe
using a Junttan PM16 pile driving rig to a final (at h/D = 1.5) are significantly higher than further
depth of 7 m bgl. Driving was paused every 1 m to back up the shaft, even after 100 or more cycles (i.e.
assess the IFR and stationary radial stresses. friction fatigue is evident). This is related to high
stress mobilized in the vicinity of the pile toe and
the geometry of the steady flow around the pile tip.
3 EXPERIMENTAL RESULTS For a fully coring pile (IFR = 100%), much lower
stresses are noted at h/D = 1.5 and these are in line
During installation of the model piles, the station- with the minimum threshold values further up the
ary radial effective stresses (σ'rs), which were equiva- pile shaft (i.e. no friction fatigue is evident).
255
ICP-05, UWA-05 and UCD-11 design methods
compared to the radial stresses measured from
the full scale pile, S5 driven into Blessington. Due
to damage to two of the sensors during driving,
radial stress data is only available from sensors at
h/D = 1.5 and 5.5. The IFR during installation of
pile S5 reduced steadily from 95% over the first 1 m
of penetration to 45% at the end of installation. It
is evident that ICP-05 and UWA-05 methods do
not capture the full effects of plugging near the pile
toe. The UCD-11 method predicts a larger range
σ′rc,min = γ ⋅ qc (7b)
256
ington show that while recent CPT based design
methods provide significantly improved radial
stress predictions, they are still unable to capture
the full combined effects of installation plug-
ging, cyclic loading and ageing. Further research
is being undertaken at UCD using a new highly
instrumented full scale pile to reliably measure the
radial and shear stresses during installation, ageing
and load testing.
REFERENCES
257
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
ABSTRACT: In this paper we are going to summarize the procedures used to manage the operation
of a TBM (Tunnel Boring Machine) type EPB (Earth Pressure Balance) with continuous and “real time”
monitoring of working and parameters. To show the characteristics of this method we chose a relevant
example, Sagrada Familia Basilica designed by Antoni Gaudi, which is a UNESCO World Heritage Site.
Some control parameters have been followed in order to detect possible damage to the building structure.
During this time, Sener Ingenieria y Sistemas supported the project with direct monitoring and risk as-
sessment to the TBM, reaching an advanced level in order to detect anomalies in the TBM operation,
ground movements or building failure.
1 INTRODUCTION
258
pressure inside the excavation chamber. To regulate
the pressure and the amount of material inside, a
screw conveyor is used with two relevant param-
eters: the rotation speed and the gate aperture
percentage. Additional to these parameters is the
treatment of the excavated soil by foam and differ-
ent kinds of polymers.
When an EPM TBM is working, the operator
always tries to find a balance between the material
entering the chamber, the screw conveyor rotation
and the aperture of the gate at the end of the screw
conveyor. This state is very difficult to reach in
mixed soils with different densities and behaviors.
It is accepted that the safest way to operate an EPB
is with the chamber full of material, but sometimes
this is not possible. Think, for example, of a mixed
ground formed by sand and clay. The friction of
Figure 2. Typical cross section showing tunnel location the clay against the internal wall of the chamber
with respect to the temple. raises the temperature and it is necessary to cool it
down through injecting water that can wash away
the sand very fast, making a chaotic scenario of
densities inside the chamber.
treatment in the surface. During the execution of
A critical moment for the front support occurs
the pile wall the settlements detected in the Temple
when the TBM must stop. There are different
were around 2 mm.
causes. During normal operation for a single shield,
when the stroke of the hydraulic jacks reach the
1.3 The construction project end a new ring has to be installed. At this moment
the cutter head stops its rotation, the gate of the
This urban tunnel is part of the High Speed Railway
screw conveyor is closed and some excavated mate-
Line Madrid-Zaragoza-Barcelona-France border
rial remains in the chamber. To prevent unexpected
and it crosses Barcelona city. The total length of the
loss of pressure during the ring installation, the
track is 4849 m and the average cover around 28 m.
TBM has been equipped with a bentonite injection
The geometry design parameters are as it follows:
system which is prepared to actuate if the pressure
− Minimal free surface: 85 m2 drops below the reference level. The engineer takes
− Internal radius: 5.2 m care during this moment and, in the case of exces-
− Free height of the section: 7.36 m sive pressure loss, informs the operator to turn on
− Cover in considered section: 25.5 to 26.5 m the bentonite system. In this way, during stops it is
possible to maintain the pressure.
Taking account of the geotechnical properties of
Another important issue is the hyperbaric oper-
the excavated material, it was decided to perform
ation. Human intervention at the front is one of
the job using an EPB (Earth Pressure Balance)
the most dangerous and complex situations that
TBM (Tunnel Boring Machine) manufactured
can happen during tunnel driving. The different
by Herrenknecht, with an excavation diameter of
phases to form the bentonite “cake”, that gives the
11.55 m, 105 m length, 2300 ton weight and named
stability to the ground and allows work inside the
“Barcino”.
chamber, must be followed and monitored with
A very complex monitoring system for the struc-
special attention. Also, during long time mainte-
ture was installed. Level sensors were placed at dif-
nance stops the bentonite “cake” can modify the
ferent heights, and extension gauges, piezometers,
surrounding ground behaviour.
inclinometers and a robotic survey complete a huge
network that can detect any single movement.
2.2 Monitoring EPB operation
The complexity of the operations requires a big
2 EPB TBM WORKING OPERATION
control of parameters, in some cases around 200.
AND MONITORING
The monitoring system hereby proposed and
described is not only oriented to the operator or
2.1 EPB working operation
the Contractor. The system gives support to all
An EPB TBM was operated with continuous front parties implicated in the job from a global and
support using the excavated material to keep the dedicated point of view.
259
Figure 3. Scheme of monitoring and data acquisition
in real time.
260
filled with excavated material. Of course, the effect consequent formation of an air bubble. After con-
of the purge is not immediate; the shift monitoring veying the situation to the operator the vent valve
engineer calculates the amount of material needed starts to drain the air and the FER is corrected. Of
to refill the chamber and, in the case of the density course, this operation is not a matter of seconds; it
not recovering in a reasonable stroke length, the takes at least, in the case shown in the figure, eight
engineer informs the EPB operator. rings (ring 976 to ring 984) excavated to recover the
Taking into account that the scales are not the right density. Eight rings means, in terms of dis-
same in Figure 5, we can check the importance tance, 32 m. This is one of the principal reasons for
of a good soil treatment in terms of operational this dedicated survey. Without noticing this situa-
safety. It is represented by the density measured tion the distance could have been 2–3 times more,
in the top of the chamber together with the FER causing an enormous risk of suffering a “blow-
(Foam Expansion Rate). FER is a parameter that out” in the ground.
gives a good idea about the quality of the foam
used at each moment. The value comes from the
3.3 Monitoring densities
expression:
A good tool for the monitoring engineer is to com-
V compressed air pare the densities between the different levels in the
E =
FER (1) chamber. If we pay attention to Figure 6, which is
V foam solution representing the density at two levels of the cham-
ber, it is possible to recognize an anomaly after ring
where: 978; the density at the medium level remains around
the same value without significant variations and
− Vcompressed air indicates the volume of compressed
the density in the top of chamber (P1-P2P3) goes
air injected
down. This is a signal that something is happen-
− Vfoam solution indicates the total volume of foam
ing at the face; a soil change, or, in the TBM, a
obtained.
mechanical problem or a bad operation.
A normal working ratio of FER is between 8
and 20, but these values must be adapted to the
3.4 Monitoring bentonite injection
geological conditions every time. For example, a
saturated soil needs dry foam which will be indi- Another parameter to take account of to control the
cated with a high FER value. On the other hand, a presence of an air bubble in the top of the chamber
lower expansion rate is normal for wet foams. is the bentonite injection through the shield. The
The dashed line represents the optimal density function of this injection is to fill the gap gener-
for operation with minimum risks. It is clearly ated by the cutter head. This gap could be suscep-
demonstrated that, when the FER increases, the tible to filling with air, causing a risk just behind
densities in the top of the chamber decrease with the cutter head and before the installation of the
261
concrete ring. It is a volume without any apparent the ground composition and TBM characteristics.
support and means a big risk of settlements in the For the Sagrada Familia crossing, a theoretical
surface. The injection of bentonite must create a weight of 368 tons extracted per ring excavated
pressure balance with the material inside the cham- was calculated.
ber; in fact, if the top part of the chamber is empty Of course there is an interval of admittance in
the bentonite without pressure will displace to the the weight. The upper alarm limit was established
front filling the empty space. In Figure 7 we can see as 405 tons and the lower alarm limit as 331 tons.
a good example. These two limits are represented in Figure 8.
Using the same ring interval as before and Coming back to ring 980, the density in the top of
comparing the cubic meters of bentonite injected the chamber goes down and the extracted amount
through the shield with the density in the top of of material shares this tendency (this is also possi-
the chamber, we can observe that when the density ble to detect two rings before). The reason for this
in the top of the chamber decreases, the amount behavior is the necessity to refill the chamber with
of bentonite injected rises dramatically to 4.6 m3, material after the wrong treatment occurred in ring
ring 980, before going down to around 1.0 m3 when 978 (FER rises to 15). The TBM operator receives
the density reaches again the optimal balance. In a from the engineer advice that an air bubble is in
good operation the injection is reduced to a mini- progress and immediately two actions take place:
mum volume, which is needed to fill the gap and
− The volume of bentonite is increased to help
give a certain pressure to it. From ring 981 to 987
during the refilling of the chamber with exca-
the volume of bentonite is more or less constant
vated material.
around 1 m3 and the density is maintained at over
− The extracted material reduces its amount oper-
1.4 t/m3 with a good margin of reaction. In ring
ating over the screw conveyor. In this case the
987 it comes back to the same situation. The FER
rotation speed is reduced in order to keep the
rises very fast (Figure 5), the density drops off
material in the chamber. See Figure 9.
and the volume of bentonite injected is doubled.
This example focuses on a few rings demonstrat- The fast detection of an air bubble progress
ing that these three parameters are directly linked makes changes in the density possible in only three
and that a good interpretation can solve dangerous rings of the advance using a combination of oper-
situations. ations in the TBM:
− Using the vent valve to drain the air retained in
3.5 Monitoring the extraction of material the top of the chamber
The next parameter which focuses the attention of − Filling the empty space with bentonite to com-
the engineer in charge is the amount of material pensate for the absence of material
extracted from the chamber. For each part of the − Controlling the extracted material with the oper-
tunnel a theoretical quantity is calculated based on ation of the screw conveyor.
Figure 7. Graphical representation of bentonite vol- Figure 8. Graphical representation of the material
ume injection. extracted from the chamber.
262
tions, sand with clay layers and saturated in water,
we can conclude that this kind of monitoring is an
essential tool to avoid the associated risks with this
kind of job.
4 CONCLUSIONS
263
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
ABSTRACT: Suction Embedded PLate Anchors (SEPLAs) are a relatively new type of anchorage in
deep waters. The SEPLA is inserted vertically into clayey seabed and then pulled to rotate until it becomes
nearly perpendicular to the loading inclination. The keying response has been studied by means of centri-
fuge tests and large deformation finite element (LDFE) analyses. However, the predicted ultimate losses
of embedment were lower than the experimental results in most scenarios. The effect of anchor rough-
ness on the keying process is then investigated using LDFE approach with contact algorithm. A friction
coefficient of 0.3 in typical normally consolidated kaolin clays is determined by comparing the numerical
results with three groups of experimental data measured. The loss of embedment during keying depends
on the combined influences of anchor roughness, anchor thickness and loading eccentricity. When the
anchor thickness ratio is less than 0.7 and the loading eccentricity ratio not larger than 0.5, the anchor
roughness needs to be considered.
1 INTRODUCTION
264
The embedment loss of SEPLA was firstly ses using the RITSS, the keying process is divided
explored through in-situ and centrifuge model tests into typically dozens to a hundred small steps. In
(Wilde et al. 2001, O’Loughlin et al. 2006, Gaudin each step, the translation and rotation of SEPLA
et al. 2008, Gaudin et al. 2010). The disadvantage must be small enough to avoid severe mesh distor-
of the in-situ test is that the evolved inclination and tion around the SEPLA. Then the deformed soil is
embedment depth of SEPLAs cannot be measured re-meshed and updated Lagrangian calculation is
in a straightforward and accurate way. In the centri- performed for each step, followed by field variables
fuge tests using the advanced Particle Image Veloci- (e.g. stresses and material properties) mapped to
metry (PIV) technique (White et al. 2003), the soil the new mesh at the commencement of next step.
flow mechanism around the keying anchor was cap- Here the RITSS strategy is coupled with com-
tured and the trajectory of the anchor was tracked mercial package ABAQUS/Standard (SIMULIA,
continuously and reliably (O’Loughlin et al. 2006). 2010). The flow chart of ABAQUS-based RITSS
To implement the PIV technique in the centrifuge, a and mapping techniques are detailed in Randolph
transparent side is set for the chamber containing soil et al. (2008) and Wang et al. (2010, 2011).
sample, and both ends of the anchor are placed in The SEPLA’s installation is usually completed
contact with the sides of the chamber, i.e. the anchor within a few hours, therefore, the process can be
is under plane-strain condition. The pull-out resist- regarded as undrained due to low permeability of
ances measured in the PIV tests are not reliable due to clay. An elastic-perfectly plastic model with Tresca
the friction of the chamber sides. However, the posi- yield criterion was used to describe the soil, with
tions of anchor are quantified to demonstrate the his- Young’s modulus of 500su and Poisson’s ratio of
tories of the anchor inclination and embedment loss. 0.49. The undrained shear strength of clay su var-
Apart from the model tests, the keying processes ied linearly with depth as follows:
of strip and rectangular SEPLAs were investigated
by Song et al. (2009) and Wang et al. (2011) using a su = sum + kz (1)
large deformation finite element (LDFE) approach
named ‘remeshing and interpolation technique Where sum represents the undrained strength at
with small strain’ (RITSS). The RITSS approach is seabed surface, k is the strength gradient and z
based on frequent mesh regeneration to overcome denotes the soil depth. The anchor was idealised as
mesh distortion around the keying anchor. Two a rigid body since its stiffness is much higher than
simple equations were presented to predict the ulti- soil. The coefficient of earth pressure at rest was
mate loss of embedment through a large amount taken as 1.
of parametric studies varying the anchor geom- The schematic geometry of the anchor models
etry, soil strength profile and loading eccentricity used in the centrifuge is shown in Figure 1b. The
ratio. However, the SEPLAs modelled in these initial embedment depth of the SEPLA prior to
large deformation analyses were assumed as fully keying (Hi) must be sufficiently deep, Hi ≥ 3B, to
rough, i.e. no relative tangential movement between guarantee deep failure mechanism in the subse-
the anchor and ambient soil was permitted. The quent daily operation stage (Song et al. 2009). B
embedment loss tends to be underestimated, which represents the anchor width. The SEPLA is led to
is on the unsafe side for practical applications. rotate by a vertical or inclined pulling force applied
In this paper, the influence of anchor rough- at padeye. The distance from the padeye to the
ness on the keying response and embedment loss front face of plate is indicated as loading eccentric-
is studied using two-dimensional RITSS approach. ity, e. It was found in previous experimental and
An appropriate range of friction coefficient is sug- numerical researches that the loading eccentricity
gested by comparison with previous vertical pull- ratio, e/B, is a dominant factor for keying response
out tests using the PIV. The inclined pull-out tests (Gaudin et al. 2008, Wang et al. 2011). For practical
were abandoned since the chain-soil interaction SEPLAs, the padeye is connected to the fluke via
would introduce more complexity on determin- shanks. However, the shanks were replaced with a
ing the embedment loss. The focus of this paper slender shaft in most centrifuge tests using PIV, in
is limited to the roughness of anchor, which will order to lower the complexity of manufacturing
be pursued by offsetting the uncoupled factors. the anchor models. The shaft was ignored in the
The combined effect of anchor roughness, anchor numerical analyses, since the area and self-weight
thickness and loading eccentricity ratio on the ulti- of the shaft are marginal compared with those of
mate embedment loss is also studied. the fluke. This may lead to a slight overestima-
tion of the embedment loss. It was assumed that
no separation occurs between the SEPLA and soil
2 METHODOLOGY during the entire keying process. The assumption is
supported by the phenomena witnessed in almost
The RITSS approach can be regarded as a special all centrifuge tests (Gaudin et al. 2006, O’Loughlin
Arbitrary Lagrangian-Eulerian method. In analy- et al. 2006, Gaudin et al. 2008, Gaudin et al. 2010,
265
Song et al. 2009). This is due to the negative pore trifuge of the University of Western Australia, and
pressures (e.g. suction) generated on the back the soil samples were composed of kaolin clay. The
face of the SEPLA given the anchor embedment model anchors were inserted vertically into the soil
depth is sufficiently deep. This ‘bond’ mechanism, manually. The moderate soil disturbance induced
however, does not necessarily mean the soil can- by the insertion was almost fully recovered during
not slide frictionally along the anchor faces. The spinning to the high g-level for the pull-out test,
frictional soil-anchor interactions was simulated as therefore, the insertion process is not mimicked
Coulomb friction contact, but the frictional resist- in the LDFE analyses. Only vertical pull-out tests
ance is limited by a maximum shear stress, τmax, were replicated, with the prototype geometries of
which was defined as the product of the intact soil SEPLAs and soil properties listed in Table 1. In the
strength at anchor centre (suc) and friction coeffi- table, the thickness of SEPLAs and the submerged
cient α unit weight of soil are represented with t and γ ′,
respectively.
τmax = αsuc (2)
A rough contact state, in geotechnical sense, 3.1 Case A
is achieved as α = 1. Note that the anchor centre Three model tests were reported by O’Loughlin
does not coincide with fluke centre when a keying et al. (2006), with loading eccentricity ratio
flap is attached (see Figure 1). Previous centrifuge e/B = 0.17, 0.5 and 1. For e/B = 0.17, the meas-
tests concerning offshore installations of suction ured loss of embedment was as high as 2.2B, given
caissons and pipelines have indicated that under the initial embedment depth was Hi/B = 3. In addi-
static loads, the friction coefficient between steel tion, the soil was observed to be separated from
and kaolin clay ranges from 0.3 to 0.5 (Chen and the back face of the anchor before the rotation
Randolph, 2007, Dingle et al. 2008). In ABAQUS- was completed. This ‘separation’ mechanism is not
based LDFE analysis, the maximum shear strength allowed in practical applications. Hence the test
at soil-anchor interfaces was assumed to be con- with e/B = 0.17 was abandoned. For e/B = 0.5 and
stant in each small step, but it was updated at the 1, the experimental and numerical keying responses
end of each step according to the current embed- are compared in Figure 2, where Δz represents the
ment depth of anchor centre. The uncertainties are embedment loss of fluke centre. When e/B = 0.5,
then limited to the determination of the friction the numerical curve of α = 0.3 approach good
coefficient. agreement with the experimental curve until the
anchor inclination is reduced to 40° to the horizon-
3 COMPARISON WITH MODEL TESTS tal. The experimental ultimate loss of embedment
is reached at anchor inclination of 27° rather than
Two-dimensional LDFE analyses were carried out ∼0 obtained from LDFE calculation. The differ-
to reproduce the keying responses of strip anchors ence may be attributed to the experimental errors
in 3 groups of centrifuge tests by O’Loughlin et al. on loading inclination at the padeye and non-uni-
(2006), Gaudin et al. (2008) and Gaudin et al. formity of the soil sample. If the anchor is assumed
(2010), respectively. The 3 groups are ordered as as rough (α = 1), the embedment loss predicted is
Case A, B and C in Table 1. Within our knowl- much lower than the experimental and numerical
edge, these are all the published tests to measure result with α = 0.3. When the loading eccentricity
embedment loss of SEPLAs using the PIV tech- ratio is increased to e/B = 1, the predicted keying
nique. These tests were conducted in the drum cen- responses in terms of α = 0.3 and 1 are close, indi-
SE PLA Soil
266
At the start of vertical pull-out, V = 0, whereas
H = Ta and M = eTa. Ta is the vertical pulling force
at padeye. These initial values satisfy Eq. 3. The
ratio of rotation and upward sliding displace-
ment δβ/δu at the start of keying is then deter-
mined as:
δβ ∂f ∂M m ⎡⎛ Hmax ⎞ ⎤
n
= = ⎢⎜ ⎟ − 1⎥ (4)
δ u ∂f / ∂H en ⎢⎝ | H | ⎠ ⎥⎦
⎣
Figure 2. Comparison for Case A. It can be postulated from the preceding equa-
tion that the reduction of Hmax, resulting from a
decreased friction coefficient α as all other param-
cating that the effect of anchor roughness is mini- eters are maintained, leads to a smaller rotation-
mal and the keying process dependent mainly on sliding displacement ratio, indicating the rotational
the eccentricity ratio. The model anchor ceased to motion has slowed down with respect to soil-fluke
rotate at anchor inclination of 23° to the horizon- movement. On the other hand, when the influence
tal, which is also earlier than those predicted by the of loading eccentricity e is pursued from Eq. 4, it is
LDFE analyses. shown that a smaller e value corresponds to a more
The influence of the anchor roughness can be significant change rotation-sliding ratio. Both pos-
explained using the plasticity model developed by tulations are in agreement with observations in
Cassidy et al. (2012). The plasticity model incor- Figure 2.
porates combined loads into a yield envelop as
follows: 3.2 Case B
1/ p Gaudin et al. (2008) conducted 10 experimental
⎛ ν ⎞
q ⎡⎛ | M | ⎞ m ⎛ | H | ⎞ n ⎤ tests with load inclination varying from 30° to
f =⎜ − 1 + ⎢⎜ +⎜ ⎥ =0
⎝ νmax ⎟⎠ ⎢⎝ Mmax ⎟⎠ ⎝ Hmax ⎟⎠ ⎥ 120°, of which two were vertical pull-out cases with
⎣ ⎦ loading eccentricity ratio e/B = 0.25 and 1. The
(3) experimental and numerical results of the vertical
pull-out cases are compared in Figure 4. For lower
where V, H and M are equivalent normal, sliding eccentricity ratio e/B = 0.25, the numerical curve
and rotational loads at anchor centre and f defines representing α = 0.3 matches the experimental data
a plastic potential, of which the derivatives against very well, whilst the assumption of rough contact
equivalent loads represent the corresponding local tends to grossly underestimate the loss of embed-
normal (δw), sliding (δu) and rotational (δβ) dis- ment. As the loading eccentricity ratio is as large
placements (Figure 3), given an associated flow as e/B = 1, similar to that observed in Figure 2, the
rule is assumed. The fitting parameters within this keying responses are nearly independent of the
equation, p, q, m and n, are pre-decided at current friction coefficient.
depth of embedment.
267
3.3 Case C It is noteworthy that the strength gradient in
Case C is as high as k = 2.72 kPa/m, which is 3.86
The model anchors used in Case A and B are flap-
and 2.72 times of that in Case A and B, respectively.
less anchors. Gaudin et al. (2010) investigated the
An interesting observation is that the fitting fric-
activation of keying flap in centrifuge, with flap
tion coefficient α = 0.3 for Case A and B is 3 times
width of 0.5 or 0.375 times of the fluke width. It
of that for Case C. Therefore, it is deduced that
was observed in their tests that the flap remains
the maximum shear strength along the anchor-soil
aligned with the fluke until the fluke is nearly
interface is not proportional to the local intact soil
perpendicular to the pull-out force. This unex-
strength at the anchor centre, instead, the real fric-
pected phenomenon can be explained through the
tion coefficient reduces with the local soil strength.
moment applied on the keying flap (Wang et al.
For offshore normally consolidated clays, the typi-
2012). Therefore, the fluke and flap were simpli-
cal strength gradient is in the range 0.9–1.3 kPa/m
fied as a rigid body in the LDFE analyses, with the
and friction coefficient is suggested as 0.3 for pre-
anchor width being the sum of the fluke width,
dicting the keying of SEPLAs (see Figs. 2, 4). If
flap width and the width of gap between the fluke
the strength gradient is apparently higher than this
and flap. The intact soil strength suc in Eq. 2 is rela-
typical value, it is conservative to select a lower
tive to the centre of whole anchor. Gaudin et al.
value, say, α = 0.1.
(2010) reported two experimental keying curves
In the above comparisons (Figs. 2, 4, 5), the
both for B = 3.2 m (see Table 1). The numerical
LDFE analyses with friction coefficient of 0.3 or
keying responses are plotted in Figure 5, where the
0.1 predicts more reliable keying responses than
embedment loss refers to the fluke centre. In the
those with α = 1. As the loading eccentricity ratio
test with e/B = 0.625, the flap is activated as fluke is
e/B < 0.5, the embedment loss is significantly
rotated to ∼20° to the horizontal. The experimen-
underestimated if the friction coefficient is taken
tal curve is located in a narrow range determined
as 1. The influence of friction coefficient is mini-
by the numerical curves with α = 0.1 and 0.3. The
mal as the loading eccentricity ratio is increased to
anchor with α = 1 is predicted to rotate signifi-
e/B = 1.
cantly faster than the measured. For e/B = 0.391,
the experimental loss of embedment is at least
twice of the prediction in terms of α = 1. The fric-
tion coefficient α = 0.1 provides better agreement 4 FACTORS AFFECTING
with experimental curve than α = 0.3. EMBEDMENT LOSS
268
ratio and anchor thickness ratio. The friction effect
becomes dominant for loading eccentricity e/B ≤
0.5 or for anchor thickness t/B ≤ 0.07.
REFERENCES
269
Installation Effects in Geotechnical Engineering – Hicks et al. (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00041-4
Author index
271