You are on page 1of 94

Bulletin of Earthquake Engineering (2005) 3:141234 Springer 2005

DOI 10.1007/s10518-005-1241-3

A Study of Piles during Earthquakes: Issues

of Design and Analysis

Kagawa University, 2217-20 Hayashi-cho, Takamatsu 761-0396 Japan. University of British
Columbia, Vancouver, Canada
(Tel: 81-87-864-2170; Fax: 81-87-864-2188; E-mail:

Received 18 January 2005; accepted 20 January 2005

Abstract. The seismic response of pile foundations is a very complex process involving iner-
tial interaction between structure and pile foundation, kinematic interaction between piles
and soils, seismically induced pore-water pressures (PWP) and the non-linear response of
soils to strong earthquake motions. In contrast, very simple pseudo-static methods are used
in engineering practice to determine response parameters for design. These methods neglect
several of the factors cited above that can strongly affect pile response. Also soilpile interac-
tion is modelled using either linear or non-linear springs in a Winkler computational model
for pile response. The reliability of this constitutive model has been questioned. In the case
of pile groups, the Winkler model for analysis of a single pile is adjusted in various ways
by empirical factors to yield a computational model for group response. Can the results of
such a simplied analysis be adequate for design in all situations?
The lecture will present a critical evaluation of general engineering practice for estimating
the response of pile foundations in liqueable and non-liqueable soils during earthquakes.
The evaluation is part of a major research study on the seismic design of pile foundations
sponsored by a Japanese construction company with interests in performance based design
and the seismic response of piles in reclaimed land. The evaluation of practice is based
on results from eld tests, centrifuge tests on model piles and comprehensive non-linear
dynamic analyses of pile foundations consisting of both single piles and pile groups.
Studies of particular aspects of pilesoil interaction were made. Piles in layered lique-
able soils were analysed in detail as case histories show that these conditions increase the
seismic demand on pile foundations. These studies demonstrate the importance of kine-
matic interaction, usually neglected in simple pseudo-static methods. Recent developments
in designing piles to resist lateral spreading of the ground after liquefaction are presented. A
comprehensive study of the evaluation of pile cap stiffness coefcients was undertaken and a
reliable method of selecting the single value stiffnesses demanded by mainstream commercial
structural software was developed. Some other important ndings from the study are: the
relative effects of inertial and kinematic interactions between foundation and soil on acceler-
ation and displacement spectra of the super-structure; a method for estimating whether iner-
tial interaction is likely to be important or not in a given situation and so when a structure
may be treated as a xed based structure for estimating inertial loads; the occurrence of large
kinematic moments when a liqueed layer or naturally occurring soft layer is sandwiched
between two hard layers; and the role of rotational stiffness in controlling pile head displace-
ments, especially in liqueable soils. The lecture concludes with some recommendations for

practice that recognize that design, especially preliminary design, will always be based on
simplied procedures.

Key words: case histories of pile failures, centrifuge testing of piles, dynamic effective stress
analysis, pile cap stiffnesses, piles in liqueable soils, pseudo-static analysis of pile groups,
seismic response of pile groups

1. Introduction
In 1999, on appointment to the Anabuki Chair of Foundation Geodynam-
ics at Kagawa University, I was given the opportunity to make a compre-
hensive study of pile foundations in liqueable and non-liqueable soils.
The study was supported by Anabuki Komuten, a major construction rm
in Western Japan. The rm constructs many buildings on reclaimed land
susceptible to liquefaction and use pile foundations extensively. Therefore
the research was focussed on the behaviour of pile foundations during
earthquakes in both liqueable and non-liqueable soils. The research study
focussed on three aspects of the pile foundation design process: determin-
ing the moments, shears and displacements for design, characterizing the
pile cap stiffness matrix so that the effects of the pile foundation on the
response of the superstructure could be included adequately in a computa-
tional model of the superstructure and assessing the reliability of the sim-
plied methods used in practice for the seismic analysis of pile foundations.
The seismic response of pile foundations to strong earthquake shaking
is a very complex process that is controlled by inertial interaction between
superstructure and pile foundation, kinematic interaction between founda-
tion soils and piles, and the non-linear stressstrain behaviour of soils. At
some sites, high seismically induced pore-water pressures (PWP) or lique-
faction add to the complexity. Over the years, a series of simple analytical
procedures have evolved for solving the various design problems associated
with pile foundations subjected to earthquake motions. These procedures
are simple because they model pilesoil interaction by Winkler springs and
ignore one or more of the signicant factors listed above that control seis-
mic response.
Simplied procedures are an essential feature of seismic design and are
here to stay. In the case of the seismic design of structures, simplied pro-
cedures are the tools for code regulated designs. However, in this case, the
simplied procedures are continually re-validated and improved over time
by tests on large scale model structures on shake tables, analysis of fail-
ures during earthquakes and by parametric studies using sophisticated non-
linear nite element analyses. The improved procedures appear in updated
building codes. In the structural environment, simplied methods are a
convenience for design and more sophisticated, conrmatory analyses can
be conducted on the nal design, if considered necessary. In the case of
performance based design, the proposed design is checked by a non-linear

push-over or non-linear dynamic analysis. The kind of quantitative con-

rmatory data on performance associated with the simplied methods of
structural design is not available to support the simplied methods for the
seismic analysis of pile foundations. Therefore an overall assessment of the
reliability of the simplied methods was made a component of the research
The basis of the simplied pseudo-static analysis widely used for the
seismic analysis of pile foundations was established 60 years ago and has
changed little in the last 30 years. In recent years some excellent research
has been conducted on model pile foundations in centrifuge tests, on
full-scale piles in tests on large shaking tables and in the eld. Some of this
work will be described in later sections of the paper. So far this work has
had limited impact on general practice.
The purpose of this paper is to present some of the more important
ndings from the comprehensive study of the seismic behaviour of pile
foundations in liqueable and non-liqueable soils. It is hoped that the
overview may provide a better understanding of the interactions of pile
foundations with the superstructure and the foundation soils. This may
lead to a better appreciation of the merits and limitations of approximate
methods for the design of pile foundations and the characterization of
the actions of pile foundations on structures in computational models for
structural response analysis.

2. Evolution of simplied procedures

2.1. Single pile analysis: elastic springs
The origins of the simplied methods that are in wide use today can be
traced back to a discussion by Chang of a paper by Feagin on lateral pile
loading tests. The paper and all discussions appear together in the 1937
Transactions of the American Society of Civil Engineers (Feagin, 1937).
The tests were conducted to provide design data for the design of the pile
foundations for Lock and Dam 26 on the Mississippi River at Altona,
Illinois. These tests were the rst major lateral load tests and generated
great interest and discussion. They still make very interesting reading today.
Six concrete monoliths were constructed on groups of timber and concrete
piles driven to an average penetration of 9.0 m. The heads of the piles
were xed by embedding them 0.6 m into the concrete. Lateral loading was
applied by inserting a jack between two monoliths and jacking them apart.
Chang (1937) presented a method for the analysis of Feagins test data in
which the interactions between soil and pile were represented by elastic
springs and the spring constant was assumed to be constant with depth.
He developed solutions for deections, moments and shears in the piles.

Chang (1937) appears to have been the rst to introduce the concept of
critical pile length. He recognized that the upper layers of soil contributed
most to resistance against lateral loads and suggested that the constant soil
modulus, Es , should be taken as the value at one third of the modulus at
the critical depth, in effect recommending a characteristic modulus. Using a
modulus backgured from the test data, which inherently included non-lin-
ear effects of the loading, he showed that his model of pilesoil interaction
could simulate the results of Feagins tests satisfactorily.
This method, designated Changs method, is still widely used today and
appears in the latest Japanese codes with a yield cut-off. The soil modulus
is given in Japanese codes as a function of the standard penetration resis-
tance, N. Computer programs readily allow a variable Es corresponding to
the distribution of N values. Liqueable layers are assigned either zero or
a very low modulus in practice.

2.2. Single pile analysis: non-linear springs

The next major development was to replace the elastic springs by non-
linear supports described by specied curves called py curves (Matlock,
1970; Reese et al., 1974). Here p is the resistance per unit length of pile
for a deection y at a particular depth H . The curves are based on the
results of static lateral loading of piles with diameters of about 0.6 m. Mod-
ied versions of these curves based on slow cyclic loading tests were devel-
oped to take into account the degradation in resistance with the many slow
cycles of loading characteristic of sea wave loadings. Both types of py
curves were adopted by the American Petroleum Institute for the analysis
of offshore pile foundations (API, 1993). Analyses using these curves repre-
sent the level of current practice for the analysis of single piles under lateral
loading in many applications, especially in North America.
The general equation for a py curve in sand is

p = Apu tanh[kHy/Apu ] (1)

in which k is the modulus of subgrade reaction, H is the depth, y is the lat-

eral deection, A is a constant depending on whether the loading is static
or slow cyclic and pu is the ultimate lateral resistance.
The k values recommended by Reese et al. (1974) for different soil con-
ditions are shown in Figure 1. These stiffnesses are considered to be rep-
resentative of low strain response. Another set of k values, recommended
by Terzaghi (1955), is shown in Figure 1 also. These are much softer, and
have been recommended for analysis of bridge pile foundations in two sets
of design recommendations; one to the California Department of Transpor-
tation (CALTRANS) by the Applied Technology Council (ATC, 1996) and

Figure 1. Moduli of subgrade reaction for py curves.

the other jointly by the Multidisciplinary Center for Earthquake Engineer-

ing Research and ATC (MCEER/ATC, 2003) in guidelines for the seismic
design of federal highway bridges.
The Terzaghi moduli are considered to reect the stiffness associated
with pile head displacements of about 2.5 cm (Martin, private communi-
cation). The MCEER/ATC guidelines were developed under the National
Cooperative Highway Research Program in the United States for the Fed-
eral Highway Administration. These two sets of subgrade moduli differ on
average by a factor of 4. It is surprising to nd both sets still recommended
for general practice despite years of lateral load testing of full-scale piles in
the eld and model piles in shake table and centrifuge tests.

2.3. Reliability of the py model

How reliable is the py method for single piles? Murchison and ONeill
(1984) carried out an extensive evaluation of the reliability of py based
analyses in cohesionless soils. They investigated four different py curves,

including those recommended by API. Their study included 24 full-scale

tests on piles in cohesionless soils; 14 static tests and 10 slow cyclic tests.
The site conditions varied from very loose clayey sands to very dense
sands. They found that the API curves, although the best of the set, gave
poor predictions with large errors. As a result of their study, Murchison
and ONeill (1984) concluded that the API curves were not adequate for
the analysis of static or slow cyclic loading tests. They stated that It is
likely that other factors, not included in the py formulation are opera-
tive and this observation suggests that further study into the fundamental
mechanisms of lateral pilesoil interaction in cohesionless soils is war-
ranted. One obviously missing fundamental mechanism is shear transfer
between the springs.
Gazioglu and ONeill (1984) carried out a similar evaluation of the
py relationships proposed for use in cohesive soils with similar results.
After studying 30 full-scale tests in clayey soils; 21 static and 9 slow cyclic
tests, they concluded that the condence in predicting deections and
moments is unfortunately rather poor.
These assessments of the reliability of the py model, which are based
on 54 full-scale eld tests, should be matters of concern to the profession.
API (1993), in recommendations for design, draws attention to what it calls
limitations in the ability to predict single pile soilpile interaction behav-
iour .
Apart from the concerns about the computational model expressed by
Murchison and ONeill (1984) and Gazioglu and ONeill (1984), there are
serious omissions from this analysis that may have a signicant effect on
the reliability of the analysis in some situations in a seismic environment.
The base shear and moment applied to the pile head are calculated on
the assumption that the structure is on a rigid base. Therefore any effect
that the pile foundation may have on the dynamic response of the super-
structure and hence on the base shear and moment is not considered. Iner-
tial interaction is not included fully. The effects of kinematic interaction
are also neglected. This interaction arises from the soil pressures gener-
ated against the pile to ensure that the seismic displacements of soil and
pile are compatible at points of contact along the pile. The corresponding
changes in curvature induced in the pile can lead to signicant moments.
These kinematic moments should not be neglected routinely.
Eurocode 8, Part 5 (ECN, 2003) recognizes the importance of kine-
matic interaction. It directs that the bending moments due to kinematic
interaction be computed for important structures in regions of moderate
to high seismicity, when the ground prole contains consecutive layers of
sharply differing stiffness. These ground conditions are met in liqueable
soils. Some very important cases are, if a non-liqueable layer overlies the
liqueable layer or if there is a layer of soft clay between two much stiffer

layers. Results from the analyses of pile foundations in these types of sites,
which will be presented later, clearly demonstrate the importance of kine-
matic moments in these conditions.
The response of the foundation soils to strong seismic shaking is
non-linear. The soil moduli degrade with strain, resulting in a reduction
in soil stiffness and therefore greater displacements under load. The simpli-
ed analysis is non-linear but it takes into account only the non-linearity
induced by the inertial loads from the superstructure. The very signicant
degradation in stiffness caused by the seismic ground motions is neglected
in the calculation of pile response.
Centrifuge test data by Wilson (1998) illustrates the uncertainty associ-
ated with pseudo-static analysis. He reports on three centrifuge tests on a
medium dense sand with Dr = 55%. These tests were part of a study of pile
response in liqueable sands but in these tests the PWP developed during
shaking were so low as to have no signicant effect on pile response. The
PWP, as a percentage of effective overburden pressure, were 20%, 11% and
4%. The piles experienced high inertial loads so that pilesoil interaction
was non-linear. Wilson found that the p-ordinate of the py curves had to
be multiplied by 2 in order for the results of pseudo-static analysis to pro-
vide a good match to the recorded data.
There is no generally accepted way for taking the effects of seismically
induced PWP into account in the simplied method. In the case of lique-
faction, it has been suggested that the p-ordinates of the py curves be
reduced by factors in the range 0.30.1. In the case of surface liquefaction,
the pile has been assumed sometimes to be free standing over the liqueed
depth. In Japanese practice using Changs elastic method, the lateral stiff-
ness is set to zero or a low value in the liqueed layers. The emerging state
of practice for liqueable soils will be discussed in more detail in a later

3. Pseudo-static seismic analysis of pile groups

3.1. Methods of analysis
The springs, which were developed to model single pilesoil interaction,
are not applicable directly to pile groups because the over-lapping displace-
ments elds of piles in the group affect the individual pile stiffnesses. A
solution for this difculty was developed by Poulos (1971). He used interac-
tion factors to quantify the incremental deformation of one pile due to the
presence of a similarly loaded neighbouring pile. Randolph (1981) greatly
simplied the use of interaction factors by developing simple equations for
the interaction between xed-headed piles, which is generally the most rel-
evant interaction factor in engineering practice. For a while, the state of

best practice was to use non-linear py curves for the individual piles in
the group and describe the interaction between the piles by elastic interac-
tion factors.
The advent of centrifuge testing showed the limitations of the elastic
interaction factors and an elastic approach in general. Barton (1982) was
the rst to document the unequal distribution of loads between rows of
piles in sand. She found that, for pairs of piles spaced at two diameters, the
front pile carried 60% of the applied load. Two different approaches were
developed for inferring group performance from analysis of a single pile.
The simplest approach is based on the concept of group efciency. Pinto
et al. (1997) conducted centrifuge static model tests of 3 3 pile groups in
sands with relative densities ranging from 17% to over 90%. They found
that the group efciency for free headed piles, at a centre to centre spacing
(s) of three times the diameter (d) or s/d = 3, was about 0.73, at displace-
ments of 75 mm. For s/d = 5, the efciency was about 0.9. In applications,
the efciency is often taken in the 0.70.8 range.
Rollins et al. (1998) conducted a full-scale test on a 3 3 pile group in
clay. The piles were closed-end steel pipes with a wall thickness of 9.5 mm
and were driven to a depth of 9.1 m. A single pile was also driven about
1.8 m from the group. The efciency of the group was found to be in the
range 0.40.5. This is considerably smaller than the values quoted by Pinto
et al. (1997) for sand.
The second approach to the analysis of group response uses p-multipli-
ers. The p-multiplier is a reduction factor that is applied to the p-term in
the py curve for a single pile to simulate the behaviour of piles in the
group. For the 3 3 group, Pinto et al. (1997) found the multipliers to be
0.8, 0.45 and 0.3 for the leading, middle and trailing rows respectively of
piles in sand with a relative density of 55%. Brown and Bollman (1996)
recommend the factors in Table I for design. The factors are representa-
tive of signicant pile head displacements of the order of 10% of the pile
Rollins et al. (1998) backgured p-multipliers from their full-scale test in
clay to be 0.6, 0.38 and 0.43. These multipliers are at the low end of the
p-multipliers obtained from other available full-scale tests and the third

Table I. p-Multipliers for pile group design Brown and Boll-

man (1996).

Row spacing Front row 2nd Row 3rd and More rows

3D 0.8 0.45 0.35

4D 0.9 0.65 0.55
5D 1.0 0.85 0.75

row shows a higher multiplier than the second row. Clearly p-multipliers
and group efciencies are quite dependent on site conditions, soil types and
the details of stratication. A very signicant nding from the Rollins test
is that the bending moments in piles in the group were 50100% higher
than in the single pile and that the location of maximum moment was

3.2. Reliability of pseudo-static pile group analysis

The pile group analysis is based on the py model for a single pile. How-
ever to approximate group action additional corrections are made to the
py curves by p-multipliers.
The multipliers were determined on the basis of static tests and the
assumption is made that dynamic group action is similar to static action.
Elastic response analyses show that this is not the case. The static analysis
of a pile group cannot be more reliable than the analysis of a single pile
and is likely to be less.
The API (1993) recommendations for design do present a method for
the analysis of single piles but do not recommend a specic method for
group analysis. It suggests that the designer use two or more appropri-
ate methods with upper and lower bounds on soil properties. Then, armed
with an appreciation of the uncertainties involved, the designer is advised
to use his judgment for the structural design of the foundation. However
because of the cited uncertainties with proposed methods of analysis, it
is not at all certain what may be considered appropriate analyses for pile
groups or that the results of all the analyses will provide upper and lower
bounds of group response.

3.3. Dynamic response analysis using py curves

The seismic analysis of single piles can be improved potentially by aban-
doning the pseudo-static approach and incorporating the py model of
pilesoil interaction into a dynamic formulation of the problem. This
allows some of the important factors previously ignored in the pseudo-
static method to be included in the analysis. The py model has been
incorporated into several dynamic analysis programs. Examples are SPASM
(Matlock et al., 1978), NONSPS (Kagawa and Kraft, 1980) and PAR
(PMB Engineering, 1988). Yet another version is incorporated in the pro-
gram PILE-PY (Thavaraj, 2001). The Thavaraj model is shown in Figure 2.
In this model, the near eld springs are the non-linear py curves. They
are treated as backbone curves and, after rst loading, all subsequent load-
ing paths are assumed to follow the extended Masing (1926) criteria in
loading and unloading proposed by Finn et al. (1977). This stressstrain

Figure 2. Dynamic Winkler computational non-linear model for pile analysis.

model automatically includes the hysteretic damping of the soil. The vis-
cous dashpot represents radiation damping which is specied by the damp-
ing coefcient proposed by Gazetas et al. (1993).
If a secant modulus approach is used to model non-linear response, then
a viscous damper is required to model hysteretic material damping. The
damping coefcient, cm , is given by

cm = 2s khsecant / (2)

where s is the damping ratio, khsecant is the secant modulus and is the
circular frequency. This model is used later to analyse the response of pile
foundations in centrifuge tests.

3.4. Pile head stiffnesses

A very important issue also is how to determine the pile head or pile
cap stiffness matrices. Mainstream commercial structural software does not
allow direct coupled modelling of pile foundations in non-linear soil. Typ-
ically single valued springs are used to represent the stiffness coefcients.
However the stiffnesses are time dependent. At any time, the stiffness
depends on the intensity of shaking because of the non-linear response of
soils. For specied design motions, how should a designer select an appro-
priate stiffness? One simple approach that has been used to dene lateral
stiffness is to compute the load to give a prescribed lateral displacement by
the simplied static method of analysis. Apart from the somewhat arbitrary

displacement criterion, this top down analysis neglects the stiffness changes
caused by kinematic interaction and seismic ground motions. The evalu-
ation of pile cap stiffnesses is considered later in connection with bridge

4. Simplied 3-D non-linear dynamic effective stress analysis

4.1. Basis of method
Dynamic non-linear nite element analysis of a pile group foundation in
the time domain, using the full three-dimensional wave equations, is not
common practice in engineering at present because of the time needed for
the computations. Seismic response analysis is usually conducted assum-
ing that the input motions are horizontally polarized shear waves propagat-
ing vertically. Therefore the important parameters controlling the response
of pile foundations are the shear stresses on vertical and horizontal planes
and the normal stresses in the direction of shaking. A much faster method
of computation can be developed by focussing the analysis on these key
parameters only.
A brief outline of a simplied method based on the important param-
eters cited above is given here. For details, the reader is referred to Wu
and Finn (1997a,b). The basic assumptions of the simplied 3D analysis
are illustrated in Figure 3.
Under vertically propagating shear waves the soil undergoes primarily
shearing deformations in xy plane except in the area near the pile where
extensive compressive deformations develop in the direction of shaking.
The compressive deformations also generate shearing deformations in yz
plane. Therefore, the assumptions are made that dynamic response is gov-
erned by the shear waves in the xy and yz planes and compression waves
in the direction of shaking, y. Deformations in the vertical direction and
normal to the direction of shaking are neglected. A nite element code
PILE-3D was developed to incorporate the dynamic soilpile-structure
interaction theory described previously. An 8-node brick element is used
to represent soil and a 2-node beam element is used to simulate the piles,
as shown in Figure 3. The global dynamic equilibrium equation in matrix
form is written as

[M]{v} + [C]{v} + [K]{v} = [M]{I }vo (t) (3)

in which vo (t) is the base acceleration, {I }, is a unit column vector, and v,

v and {v} are the relative nodal acceleration, velocity and displacement at
time, t, and [M], [C] and [K] are the mass, damping and stiffness matrices

Figure 3. Quasi-3D model of pilesoil response (Finn, 1999).

The loss of energy due to radiation damping is modelled using the

method proposed by Gazetas et al. (1993) in which a velocity propor-
tional damping force Fd per unit length is applied along the pile. Direct
step-by-step integration is employed in PILE-3D to solve the equations of
motion in Equation (3).
The equivalent linear constitutive model with strain dependent shear
modulus and damping, which is widely used in earthquake engineering
practice, was adopted to model the non-linear behaviour of the soil. This
is a robust constitutive model based on conventional engineering material
properties. These characteristics make it ideal for parametric studies. The
strain dependence relations for shear modulus (G) and damping ratio (D),
developed by Seed and Idriss (1970), and shown in Figure 4, were used
in the analyses described later. Since the equations of motion are formu-
lated in the time domain, the modulus and damping can be updated con-
tinuously during earthquake shaking to maintain compatibility with current
shear strain levels throughout the analysis. In the usual frequency domain
formulation, the properties are updated at the end of each complete seismic
analysis. A yield condition is incorporated that is consistent with the shear

Figure 4. Strain dependence of moduli and damping after (Seed and Idriss, 1970).

Figure 5. Rocking of a pile foundation.

strength of the soil and no tension is allowed to develop between the soil
and the pile, allowing gapping to occur, when appropriate.
In the motion of pile groups, rocking and translation are coupled. The
rocking stiffness develops due to the resistance of the piles to vertical
movement as shown partially in Figure 5. The rocking stiffness must be
updated continuously during the analysis to reect any changes in soil
properties. The simplied horizontal model cannot do this directly since it
is unidirectional. The difculty is overcome by switching the mode of calcu-
lation back and forth between horizontal and vertical. In the vertical mode
at any time t, the program calculates the rocking stiffness corresponding
to the current strain eld and hence the current degraded properties of the
soil. It then switches back to the horizontal analysis and after updating the
rocking stiffness proceeds to calculate response for the next time interval.

Wu and Finn (1997a,b) give a full description of this method and

present results of validation studies that show that the method gives quite
accurate results for excitation due to horizontally polarized shear waves
propagating vertically.
An effective stress version of this program, PILE-3D-EFF, incorporates
the Martin et al. (1975) PWP generation model but uses the simpler 2-con-
stant version of the volume change component proposed by (Byrne, 1991),
rather than the original 4-constant version. The effective stress analysis has
been validated by Finn et al. (1999) and Finn and Thavaraj (2001) using
data from centrifuge tests conducted on the geotechnical centrifuge at the
University of California at Davis (Wilson, 1998).
The effective stress program was used to develop an overview of pile
foundation response in liqueable and non-liqueable soils during earth-
quakes and to test the consequences of the assumptions underlying the
simplied pseudo-static procedures used in engineering practice. It is also
hoped that the review may demonstrate how a dynamic continuum analysis
can provide designers with a global picture of the response of pile founda-
tions in different site conditions, including liquefaction.

4.2. Elastic analyses by PILE-3D

Before going on to non-linear comparative analyses, it may be of interest
to present an example involving the elastic analysis of pile groups. The ef-
ciencies of square pile groups from 2 2 up to 6 6 were computed using
PILE-3D. The efciency of a group is dened as the ratio of the average
load per pile in the group to the load in a single pile to cause unit dis-
placement. The piles were of concrete, 0.36 m in diameter and 7.2 m long.
Pile spacing was three diameters centre to centre. Youngs modulus of the
concrete was 22,000 MPa. The soil modulus varied parabolically from 0.0
at the surface to 230 MPa at a depth of 10 m. The efciencies are shown
in Figure 6. They are comparable to those calculated by Randolph for a
linear distribution of modulus with depth (Fleming et al., 1992).

5. Non-linear response analyses of piles: pile-3D and py methods

5.1. Centrifuge tests
PILE-3D and the dynamic py model were used to analyse the seis-
mic response of a single pile in centrifuge tests conducted at the Califor-
nia Institute of Technology by Gohl (Finn and Gohl, 1987; Gohl, 1991).
Figure 7 shows the instrumented soilpile-structure system used in the test.
The system was subjected to a nominal centrifuge acceleration of 60 g. The
foundation soil is a loose sand with a void ratio e = 0.78 and a relative

Figure 6. Calculated elastic efciencies of several pile groups.

Figure 7. The layout of the centrifuge test for a single pile (Finn, 1999).

density of 38%. Two tests will be discussed here. In the rst test, the
input motions had a peak acceleration of 0.158 g, which gave moderately
strong shaking and generated signicant non-linear response. The horizon-
tal input acceleration record for the second test was 0.04 g. This motion
level was chosen to ensure essentially elastic response. The distribution of

Figure 8. Finite element mesh for analysis of single pile.

Figure 9. Distribution of shear moduli around pile at a depth of 2.1 m at time

12.58 sec.

shear moduli with depth was measured prior to shaking, while the centri-
fuge was in ight, using bender elements. Therefore, accurate initial shear
moduli of the foundation soil were available for use in the analyses.

5.2. PILE-3D analysis

Figure 8 shows the nite element mesh used in the PILE-3D analysis. The
dynamic shear strains, generated by the seismic shaking, cause reductions
in shear moduli and increased damping. The reduction in moduli is illus-
trated in Figure 9 by the distribution of shear moduli at a depth of 2.1 m
in the soil around the pile at a time T = 12.58 sec. The distribution of shear
moduli and damping are both time-and space-dependent. This dependence
results in a corresponding time dependence of the global stiffnesses of the
pile head.
Dynamic impedances as a function of time were computed using the
time- and space-dependent non-linear shear moduli. Harmonic loads were

Figure 10. Time histories of pile stiffnesses.

applied at the pile head, and the resulting equations were solved to obtain
the complex valued pile impedances. The impedances were evaluated at the
surface of the sand. The time histories of the stiffnesses (the real part of
the impedances) are shown in Figure 10.
The dynamic stiffnesses of the single pile during the specied input
motions decreased dramatically as the level of shaking increased (Fig-
ure 10). The stiffnesses experienced their lowest values between about 10
and 14 sec, when the maximum accelerations occurred at the pile head. It
can be seen that the lateral stiffness component Kvv decreased more than
the rotational stiffness K or the coupled lateral-rotational stiffness Kv .
The equivalent damping coefcients increased with the level of shaking.
The time histories of stiffnesses in Figure 10 show clearly the difcul-
ties in selecting a single spring value to represent the lateral or rotational
stiffness of a pile foundation. To make a valid selection, one would need
to know which segment of the ground motion was most critical in control-
ling the seismic response of the structure. A spring based on the minimum
lateral stiffness would represent the mobilized stiffness during the period of
very strong shaking and would be more critical for longer period modes or
structures. Clearly, the time variation in stiffness and damping provides a
useful guide to the selection of the discrete springs and dashpots required
by commercial structural analysis software. In a later section, a quantita-
tive procedure for getting the best estimate of the single valued spring to
represent the time histories will be developed.
The computed and measured moment distributions along the pile at the
instant of peak pile head deection are shown in Figure 11. The moments
computed by PILE-3D agree quite well with the measured moments.

Figure 11. Comparison of measured bending moments with moments from different
analyses (Finn, 1999).

Results from py analyses presented in the next section are also shown in
Figure 11.

5.3. Analysis using API py curves

The distribution of bending moments was also determined using the py
curves prescribed by the American Petroleum Institute (API, 1993). These
py curves are dened by the equation,

p = 0.9pu tanh [(kHy)/(0.9pu )] (4)

where pu is the ultimate bearing capacity at depth H , k is the initial

modulus of subgrade reaction, y is the lateral deection, and H is the
depth under consideration. These curves are the ones specied for slow
cyclic loading by API and are softer than the standard curves, but as will
be seen even these are too stiff for the analysis of strong shaking. For
the low acceleration test, with a peak acceleration of 0.04 g, the response

Figure 12. Measured and computed pile moments for near elastic response using API

was essentially elastic and the tangent moduli stiffnesses of API would
be expected to apply. A tangent stiffness, k = 15, 000 kN/m3 , is required
to give the very good approximation to the measured moments shown in
Figure 12. This value is about 10% higher than the value associated with
a Dr = 38% in Figure 1 (k = 13, 500 kN/m3 ).
The strong shaking test with peak input acceleration of 0.158 g was
analysed using the initial tangent modulus back gured from the elastic
test. The resulting moment distribution is shown in Figure 11 above. The
peak bending moment is about 50% too large. The analysis was repeated
using a modulus k = 2 500 kN/m3 . This is signicantly less than Terzaghis
secant modulus for Dr = 38% from Figure 1 for a sand above the water
table. The prediction of peak moment is much better now but the moments
in the bottom half of the pile are greatly over-predicted. The initial tangent
moduli in the API curves seem clearly to be too stiff for the analysis of
strong shaking. The Terzaghi moduli seem more appropriate, although in
this case, they too seem to be a little too stiff. However another assump-
tion regarding the initial stiffness, the assumption that it is constant with
depth, can be shown to have a signicant affect on response.
Both the API and Terzaghi moduli of subgrade reaction are assumed to
be constant with depth. The analysis was repeated following the suggestion
of Gazetas and Dobry (1984) that the k value should vary with depth H
and the Youngs modulus Emax at depth H with kH = Emax and is a con-
stant. Thavaraj (2001) found the range in to be 0.91.3.

The shear wave velocity distribution in the centrifuge test was measured
during ight and, therefore, accurate values of the shear modulus, G, were
available. Emax was estimated for two values of Poissons ratio, = 0.2 and
0.3. The resulting moment distributions are shown in Figure 13. In these
analyses = 1.0. This approach to estimating the initial stiffness, gives a
better approximation to the measured bending moments, when using the
standard API py curves. For = 0.2 and = 1.0, both the maximum
moment and the entire moment distribution are very good.
The results of these analyses demonstrate the potential unreliability of
a dynamic computational model based on standard py curves for seis-
mic response analysis. The analyses suggest that the standard py curves
developed on the basis of data from static and slow cyclic loading tests
have limitations for seismic response analyses. The API initial stiffnesses
appear to be too high for non-linear response analysis and the Terzaghi
initial stiffnesses too soft for essentially elastic response. The suggestion of
Gazetas and Dobry (1984) that the initial stiffness be considered to vary
with depth and to be proportional to the soil modulus seems to make
a major improvement in the response predictions, especially if the stiff-
nesses are estimated from measured in situ moduli. On important projects,
it is likely that shear wave velocity data will be available for estimat-
ing moduli from seismic cone tests or cross-hole or down-hole seismic

Figure 13. Comparison of computed and measured moments when kH = Emax .


Figure 14. Simplied pile-superstructure model (after Abghari and Chai, 1995).

6. Approximate seismic analysis of a pile group

6.1. Simplied pile soil model
Abghari and Chai (1995) of CALTRANS proposed a simplied pilesoil
model (SPSM) of a bridge-foundation system that facilitates taking non-
linear soil response and inertial interaction into account. The model is
shown in Figure 14. The soilpile interaction is represented by py curves.
The pile group foundation is represented by a single pile with the pile
head xed against rotation. The pile carries a superstructure comprised of
a lumped mass on an elastic support. The mass corresponds to the total
mass on the pile group divided by the number of piles in the group. The
basic assumption of the model is that the moments, shears and deections
of the single pile are representative for piles in the group. The xed base
period of the SDOF superstructure system is taken to be the rst mode fre-
quency of interest, assuming the bridge to be on xed supports.
This is a very attractive computational model for a pile group which can
be used with any constitutive model of soilpile interaction. Therefore the
basic assumption of the model was checked by comparing the results from
the model with results from a full pile group analysis by PILE-3D, using
the same constitutive model in both types of analysis.
Analyses were conducted for 2 2 and 4 4 pile groups in a homoge-
neous cohesive deposit. Pile spacing was to two diameters, centre to centre.
Relevant data on soil and piles are given in Figure 14. The rst 20 sec of
the earthquake accelerations recorded at Grifth Park in Los Angeles dur-
ing the 1971 San Fernando earthquake was used as input motion.

Figure 15. Comparison of pile moment proles.

Figure 16. Comparison of pile shear proles.

Figure 15 shows a comparison of bending moment proles for the two

different pile groups and the single pile. In the case of the pile groups, the
results are reported for the corner pile.
The bending moment proles from the (2 2) and (4 4) analyses do
not deviate much from the moment prole of the single pile analysis. The
difference in the peak moment is less than 8%. The shear force proles in
Figure 16 show similar behaviour. It is evident from Figures 15 and 16

Figure 17. Comparison of pile shear proles.

that the group effect does not appear to be a signicant factor as far as
moments and shears are concerned for the different pile foundation-super-
structure systems analysed here. The distribution of deections along the
pile show more variation, as shown in Figure 17. Group effects appear to
be different in the dynamic environment from those quantied for the static
environment. One should be cautious about relying too readily on group
effect parameters obtained from static tests or static analyses.

6.2. Effects of pile cap rotation

The key assumption of the SPSM model is that the pile cap is main-
tained xed against rocking. The previous evaluation analyses were con-
ducted with the assumption that the pile cap was fully restrained against
rotation. The analyses were repeated without preventing any rotation of
the pile caps of the groups that is permitted by the axial forces in the
piles. Proles of maximum bending moments and shear forces are shown
in Figures 18 and 19.
In this case there is very poor agreement between results of analyses
based on the representative single pile and the full group analyses, even
though the foundation soils are treated identically for all three systems.
The distributions of rotations for the three pile-superstructure systems are
shown in Figure 20. The three rotation patterns are very different and are

Figure 18. Comparison of pile moment proles with pile group cap rotation.

Figure 19. Comparison of shear force proles with pile group cap rotation.

probably responsible for the poor representation of group response param-

eters by the single pile model.

6.3. Another look at the single pile concept

The time histories of lateral stiffnesses of the single pile and pile groups
were calculated using PILE-3D. The minimum pile stiffnesses reached dur-
ing earthquake shaking are given in Table II. The pile group efciency,

Figure 20. Comparison of rotation proles along piles.

Table II. Lateral stiffnesses of single pile and pile groups.

Type of pile foundation Minimum lateral Group effect, Kgroup/(n*Ksingle)

stiffness (MN/m) at minimum stiffnesses

Single pile 35
(2 2) Group pile 106 0.76
(4 4) Group pile 228 0.41

*n is the total number of piles in the pile group.

calculated as a ratio of minimum group stiffness over the minimum sin-

gle pile stiffness multiplied by the number of piles in the pile group is also
shown. The efciency is 0.76 for the 2 2 group and 0.41 for the 4 4
The fundamental frequencies of the three pile foundation systems were
determined corresponding to the initial stiffnesses and the minimum stiff-
nesses that occur during the time of strongest shaking. The frequencies
are shown in Table III. Although the group effects on foundation stiff-
nesses of the pile groups were signicant, the global system frequencies of
the pile group systems were not signicantly different from the frequency
of the single pile system under either elastic or non-linear response. The
maximum difference is about 15%. The similarity in frequencies of the
different foundation models may be responsible for the similarity in out-
puts. The similarity in frequencies is due to the fact that the global system

Table III. First mode frequency of the superstructure-pile founda-

tion system.

Type of structure First mode frequency (Hz)

Initial Minimum

Single pile-superstructure 3.66 1.89

(2 2) Group pile-superstructure 3.45 1.82
(4 4) Group pile-superstructure 3.07 1.61

frequencies result from the combined stiffnesses of the super-structure and

pile foundation rather than the stiffness of the pile foundation alone. In
this study, the superstructure stiffness dominates the system frequency.
CALTRANS adjusts the results of the SPSM for group effects using a
group reduction factor. Since group effects need to be incorporated during
the dynamic analysis, when they affect the non-linear response, their use
after the event may not be most effective.
The results of the evaluation study raise the possibility that the SPSM
model for the analysis of pile foundations may be valid only, if the sys-
tem frequency of the representative pile-super-structure model and the fre-
quency of the full foundation-super-structure model are approximately the
same. This is likely to occur only when the stiffness of the support struc-
tures of the bridge dominates the system frequency. For this to happen, the
support structures must be much more exible than the pile foundation in
the degree of freedom under consideration. Parametric studies using many
earthquakes with different frequency contents would be desirable to explore
how group foundation-structure system frequencies compare with represen-
tative single pile-structure system frequencies for typical bridges and foun-
dation soils under strong shaking. The study would allow the reliability of
the representative pile concept to be evaluated more thoroughly.

7. Characterization of pile foundation for performance based design

7.1. Performance based design
Traditionally the objectives of seismic design codes for structures have been
to protect life safety under the extreme events envisioned by the codes and
to maintain serviceability under the smaller events with a greater proba-
bility of occurring during the life of the structure. During the 1989 Loma
Prieta earthquake in California, the life protection objective was met but
the level of damage was considered high for such a short duration, moder-
ate earthquake. The direct costs of repair or replacement of buildings and
in many cases huge indirect losses due to business interruption motivated

Figure 21. Performance levels for seismic design (modied from Shapiro et al., 2000).

the progressive structural engineers of California to critically review design

concepts and propose the idea of performance based design.
The concept underlying performance based design is to design for an
acceptable damage level specied by the owner. The range of performance
options is illustrated in Figure 21.
Note that along the capacity curve for the structure, potential perfor-
mance options are dened in terms of global displacement. The effec-
tiveness of a performance based design is assessed by an appropriate
non-linear analysis that establishes the demand on capacity in terms of
global displacement. The location of the calculated performance point on
the capacity curve relative to the desired performance point is a measure
of how design meets the design criteria.
A non-linear pushover analysis is commonly advocated but it is recog-
nized that in some cases a non-linear dynamic analysis may be required.
For an evaluation analysis to be valid, the structural model must include
all elements of the building-foundation-soil system that signicantly affect
the seismic demand and response of the structure. Hitherto the seismic
demand for code designed buildings was determined assuming that the
structure rested on a rigid base. The actions of the foundations were com-
pletely ignored. The demands of performance based design make it imper-
ative to include the effects of soils and foundations on seismic demand and
structural response.
The rst detailed examination of how to include the actions of the
soils-foundation system in code design provisions was presented in a report
by the Applied Technology Council (ATC-3, 1978) in which tentative pro-
visions for seismic regulations were advanced. This procedure was based on
the work of Veletsos and Wei (1971). Veletsos et al. (1988) presented an
updated review of these issues in a state of the art paper to the 9th World

Conference on Earthquake Engineering in Tokyo. They noted that a proper

analysis of the physical system of structure, foundation and foundation
soils was necessary to get reliable estimates of structural demand and seis-
mic response. They evaluated the relative contributions of kinematic and
inertial interactions for surface based structures and concluded that, for
these cases, inertial interaction had the greatest inuence. These studies
were based on analyses of a simple elastic structure on a rigid mat founda-
tion welded to the surface of a homogeneous, elastic half space. Today the
technology is available to consider deep, exible foundations and non-linear
soil response.
Investigations of structural performance during earthquake loading have
conrmed the importance of treating the structure, foundation and foun-
dation soils as a complete system. Wallace et al. (1990) analysed the seis-
mic response of two 10-storey buildings designed and built in the 1970s.
The buildings were instrumented to record strong motions. One building,
in Northern California, was analysed for the motions recorded during the
1984 Morgan Hill earthquake (Ms = 6.2). The second building, in Southern
California, was analysed for the motions recorded during the Whittier
earthquake (Ms = 5.9). Good correlation was achieved between computed
and recorded motions, when the exibility of the foundationsoil system
and cracked section properties were taken into account. Otherwise the cor-
relation was poor.
The authors also evaluated the response of shear wall buildings in Chile
that performed unexpectedly well during the 1985 Chilean earthquake. On
the basis of conventional rigid base analysis, these buildings had a ductil-
ity demand of 3 and should have suffered appreciable damage. However,
when the effects of foundation exibility were taken into account, the duc-
tility demand dropped to 2. The benecial effects of foundation exibility
were an important factor leading to the good performance.
Foundation exibilities in the cases discussed above were all based on
treating the soil response as elastic. More recent developments consider
the non-linear response of the soil to strong shaking. Furthermore allow-
ing uplift during rocking and permitting yielding of the foundation soils
is being advocated to reduce seismic demand on structures. These new
advances are important developments for performance based design of new
buildings but are even more important for developing cost effective retrot
strategies. Retrotting to meet current code demand levels can be prohibi-
tively expensive. The inclusion of foundation exibility gives a more real-
istic picture of where retrots are critical and can lead to lower seismic
demand. These benets result in more cost effective retrots.
Clearly to evaluate a performance based design or the capacity of ret-
rotted building requires a realistic computational model of the struc-
tural system. This, in turn, requires a way of characterizing the actions of

Figure 22. Three span box girder bridge on pile foundation.

foundations and supporting soils on structural response that is compatible

with available commercial computational software for structural analysis.

7.2. Stiffness of pile groups

The characterization of pile foundations will be presented in the context of
a bridge on pile foundations. A three span continuous box girder bridge
structure shown in Figure 22 was chosen for the numerical studies of pile
foundations in non-liqueable ground (Thavaraj, 2001). A rigid base ver-
sion of this bridge is used as an example in the guide to the seismic
design of bridges published by the American Association of State High-
way and Transportation Highway Ofcials (AASHTO, 1983). The sectional
and physical properties of the superstructure and the piers were taken from
(AASHTO, 1983).
Each pier is supported on a group of sixteen (4 4) concrete piles. The
diameter and length of each pile are 0.36 and 7.2 m respectively. The piles
are spaced at 0.90 m, centre to centre. The Youngs modulus and mass den-
sity of the piles are E = 22, 000 MPa and = 2.6 Mg/m3 respectively. As will
be shown later, the pile cap stiffnesses of this foundation are much greater
than the corresponding column stiffnesses. The large disparity in stiffnesses
was selected in order to facilitate a study of the effects of the ratio of col-
umn to foundation stiffnesses on the seismic response of the super-structure
by increasing the column stiffnesses only.
The soil beneath the foundation is assumed to be a non-linear hyster-
etic continuum with unit weight, = 18 kN/m3 , and Poissons ratio, =
0.35. The low strain shear modulus of the soil varies as the square root

Figure 23. Distribution of initial shear moduli at site of AASHTO (1983) bridge
and of effective shear moduli from a SHAKE free-eld analysis.

of the depth with values of zero at the surface and 213 MPa at 10 m
depth (Figure 23). The variations of shear moduli and damping ratios with
shear strain are those recommended by Seed and Idriss (1970) for sand.
The surface soil layer overlies a hard stratum at 10 m. For the PILE-
3D nite element mesh, the surface layer was divided into 10 sub-layers
of varying thicknesses. Sub-layer thicknesses decrease towards the surface
where soilpile interaction effects are stronger. 900 brick elements were
used to model the soil around the piles and 64 beam elements were used
to model the piles. The input acceleration record used in the study was the
rst 20 sec of the N-S component of the free eld accelerations recorded at
CSMIP Station No.89320 at Rio Dell, California during the April 25, 1992
Cape Mendocino Earthquake. The power spectral density of this accelera-
tion record shows that the predominant frequency of the record is approx-
imately 2.2 Hz.

7.3. Pile cap stiffnesses

The pile cap stiffnesses of the pile foundation shown in Figure 22 will be
determined for two different ratios of the column/foundation stiffness ratio,
7% and 50%. A PILE 3-D analysis is conducted rst and the spatially
varying time histories of modulus and damping are stored. Then an asso-
ciated program PILIMP calculates the time histories of dynamic pile head
impedances using the stored data. The dynamic impedances are calculated
at any desired frequency by applying a harmonic force of the same fre-
quency to the pile head and calculating the generalized forces for unit dis-
placements. In this paper the focus will be on the stiffnesses only as these
are the parameters of primary interest for current practice. However the

damping components are always obtained also during the analyses. The
stiffnesses are calculated rst without taking into account inertial interac-
tion between the superstructure and the pile foundation. This is the usual
condition in which stiffness is estimated either by static loading tests, static
analysis or by elastic formulae. The stiffnesses are calculated also tak-
ing the inertial effects of the super-structure into account. In this latter
case, both kinematic and inertial interactions are taken into account. Since
the entire pile group is being analysed, pilesoilpile interaction is auto-
matically taken into account under both linear and non-linear conditions.
Therefore the usual difcult problem of what interaction factors to use or
what group factor to apply is avoided.

Figure 24. Time history of lateral and cross-coupled stiffness under strong shaking.

Figure 25. Time history of rotational stiffness under strong shaking.


Time histories of lateral and cross-coupling stiffnesses are shown in

Figure 24; rotational stiffness in Figure 25. These stiffnesses, resulting from
kinematic interaction only, were calculated for the predominant frequency
of the input motions, f = 2.2 Hz. It is clearly not an easy matter to select
a single representative stiffness to characterize the discrete single valued
springs used in structural analysis programs to represent the effects of
the foundation. In the absence of a complete analysis, probably a good
approach to including the effects of soil non-linearity on stiffness is to get
the vertical distribution of effective moduli by a SHAKE (Schnabel et al.,
1972) analysis of the free eld and calculate the stiffnesses using these
moduli at the appropriate frequency. The constant stiffnesses calculated in
this way are shown also in Figures 24 and 25. However these are kine-
matic stiffnesses. Later it is shown that inertial interaction with the super-
structure may cause greater non-linear behaviour leading to substantially
reduced frequencies. A SHAKE analysis cannot capture this effect.

8. Seismic response of code bridge to transverse earthquake loading

8.1. Finite element model of the bridge structure
A three dimensional space frame model of the bridge is shown in Figure 26.
At the abutments, the deck is free to translate in the longitudinal direc-
tion but restrained in the transverse and vertical directions. Rotation of the
deck is allowed about all three axes. The space frame members are mod-
elled using 2-noded 3-D beam elements with twelve degrees of freedom,
six degrees at each end. The bridge deck was modelled using 13 beam ele-
ments and each pier was modelled by three beam elements. The cap beam
that connects the tops of adjacent piers was modelled using a single beam
element. The sectional and physical properties of the deck and the piers
are those provided in the AASHTO guide (1983). The pier foundation is
modelled using a set of time-dependent non-linear springs and dashpots
that simulate exactly the time histories of stiffnesses and damping from the
PILE-3D analyses.
The response of the bridge structure was analysed for different founda-
tion conditions to study the inuence of various approximations to foun-
dation stiffnesses and damping, using the computer program BRIDGE-NL
(Thavaraj, 2001). The free eld acceleration was used as the input acceler-
ation and the peak acceleration was set to 0.5 g.

8.2. Foundation conditions for analyses

The seismic response of the bridge to transverse earthquake loading was
analysed for the four different foundation conditions listed below.

Figure 26. Stick model of the bridge with the foundation springs and dashpots.

1. Rigid foundation and xed base condition is assumed

2. Flexible foundation with elastic stiffness and damping
3. Flexible foundation with kinematic time dependent stiffness and
4. Flexible foundation with stiffness and damping based on the SHAKE
effective moduli.
The fundamental transverse mode frequency of the computational model
of the bridge structure with a xed base was found to be 3.18 Hz. This
is the frequency quoted in the AASHTO-83 guide. This agreement in
fundamental frequencies indicates an acceptable structural model. In this
analysis, the lateral stiffness of the bridge pier is only 7% of the founda-
tion stiffness. For this extremely low stiffness ratio, the columns control the
fundamental frequency of the bridge and the inuence of the foundation
is negligible. Results from analyses in which the column/foundation stiff-
ness ratio is 50% will be presented here. The stiffness ratio was raised by
increasing the stiffnesses of the piers only, with no changes to the super-
structure. Normally much stiffer piers would imply a heavier superstructure
and therefore higher inertial forces.
For a 50% stiffness ratio, the xed base fundamental frequency of the
bridge is 5.82 Hz. When the stiffnesses associated with low strain initial
moduli are used, the fundamental frequency is 4.42 Hz, a 24% reduction
from the xed base frequency. With kinematic strain dependent stiffnesses,
the frequency reached to a minimum value of 3.97 Hz during strong shak-
ing, a 32% reduction from the xed base frequency. When the foundation
stiffnesses are based on effective shear moduli from a SHAKE analysis of
the free eld, the frequency was 4.18 Hz, a 28% change from the xed base

Figure 27. Time history of the frequency of rst transverse mode.

Figure 28. Effects of foundation exibility on deck displacements.

frequency. Figure 27 shows the variation with time in fundamental trans-

verse modal frequency for different foundation conditions.
The response of the bridge deck at Bent 2 (Node No. 5 in Figure 26)
was computed for two cases: the xed base case and a exible foundation
with kinematic time dependent stiffnesses. The effect of including the foun-
dation exibility is shown in Figure 28. There is a dramatic change in the
deck displacement during the strong shaking, when the foundation exibil-
ity is included in the model. The peak displacement increased from 7 to
17 mm.
An elastic analysis of the displacement response of the deck at Bent 2
was carried out also with stiffnesses calculated using the initial low strain
initial moduli of the site. The elastic response underestimates the displace-
ments based on the kinematic stiffnesses but is a big improvement over the
xed base solution.

Figure 29. Pile foundation with structure.

8.3. Inertial interaction of structure and pile

The time dependent stiffnesses used in the analyses described above were
computed without taking the inertial interaction of superstructure and
foundation into account. The primary effect of this interaction is to
increase the lateral pile displacements and cause greater strains in the soil.
This in turn leads to smaller moduli and increased damping. The preferred
method of capturing the effect of super-structure interaction is to consider
the bridge structure and the foundation as a fully coupled system in the
nite element analysis. However, such fully a coupled analysis is not pos-
sible with mainstream commercial structural software. The specialist soft-
ware that is available, however, requires signicant expertise to operate and
heavy demands on storage and time.
An approximate way of including the effect of super-structure interac-
tion is to use the model shown in Figure 29. In this model, the super-
structure is represented by a single degree of freedom (SDOF) system. The
mass of the SDOF system is assumed to be the portion of the super-struc-
ture mass carried by the foundation. The stiffness of the SDOF system is
selected so that the system has the period of the mode of interest of the
xed base bridge structure.
This approximate approach will be demonstrated by the analysis of
the centre pier at Bent 2. The fundamental transverse mode frequency of
the xed base model was found earlier to be 5.82 Hz. The static portion
of the mass carried by the centre pier is 370 Mg. The super-structure can

Figure 30. Effects of inertial interaction on lateral pile cap stiffness.

be represented by a SDOF system having a mass of 370 Mg at the same

height as the pier top and frequency 5.82 Hz. The corresponding stiffness
of the SDOF system is 495 MN/m.
A coupled soilpile-structure interaction analysis can be carried out
using PILE-3D by incorporating the SDOF model into the nite element
model of the pile foundation. The pile foundation stiffnesses derived from
this nite element model incorporate the effects of both inertial and kine-
matic interactions and are called total stiffnesses. The time histories of stiff-
nesses with and without the super-structure are shown in Figure 30. The
reduction in lateral stiffness is greater throughout the shaking, when the
inertial interaction is included. There is a similar reduction in the rotational
and cross-coupling stiffnesses.
When inertial interaction was included, the lateral stiffness reached a
minimum of 188 MN/m which is 22% of the initial value. When the iner-
tial interaction was not included, the minimum was 400 MN/m. Clearly in
this case, inertial interaction has a major effect on foundation stiffness. The
effects of inertial interaction may be underestimated somewhat in this case,
because the column stiffness of the AASHTO code bridge was increased
from 7% to 50% of the foundation stiffness without any increase in super-
structure mass. Such a stiffness ratio would normally be associated with a
heavier super-structure.
An eigen-value analysis of the complete bridge structure was carried
out, using the total foundation stiffnesses. The variation in rst mode
transverse frequency with time is shown in Figure 31. This gure also
shows the frequency variation for the case in which the inertial interaction
was not considered. The frequency reached a minimum of 3.62 Hz, when
the inertial interaction was included and 3.97 Hz, when the interaction was

Figure 31. Effects of inertial interaction on foundation frequency.

Figure 32. Effect of super-structure interaction on deck acceleration.

ignored. Figures. 32 and 33 show the effects of super-structure interaction

on the time histories of acceleration and displacement respectively. When
the super-structure interaction effect is included, it leads to greater acceler-
ations and displacements.
The results of the analyses for four different foundation conditions are
summarized in the acceleration and displacement spectra for transverse
vibrations of the bridge, shown in Figures 34 and 35 respectively. The dis-
placement spectra clearly show the importance of including inertial inter-
action, when calculating foundation stiffnesses in this case. The xed base
model for estimating response is inadequate. As the ratio of super-struc-
tural stiffness to foundation stiffness is reduced, the effect of inertial inter-
action on system frequency is reduced and kinematic stiffnesses become
adequate. Only for low stiffness ratios is the xed base model adequate.

Figure 33. Effect of super-structure interaction on deck displacement.

Figure 34. Spectral accelerations of AASHTO bridge for four different foundation

The role of the stiffness ratio in controlling system frequencies is demon-

strated in the next section.
For the example bridge, when effective moduli from a SHAKE analysis
of the free eld are used in an elastic analysis to obtain a discrete, foun-
dation stiffness for each degree of freedom, the corresponding system fre-
quencies lead to acceleration and displacement responses very close to the
responses from a PILE-3D non-linear analysis. This is true when the com-
plete pile foundation is included in the analysis. It or may not be true, if
the effective moduli are used to get the stiffness of a single pile and the
stiffness of the pile group is developed from this with the help of empiri-
cal factors for group effects. This result suggests that kinematic stiffnesses
may be obtained taking non-linear soil effects into account by an elastic
structural program that can model the pile group foundation, if the effec-
tive elastic moduli from a SHAKE analysis are used.

Figure 35. Spectral displacements of AASHTO bridge for four different approxima-
tions to foundation conditions.

9. Pile cap stiffnesses and system frequencies

This study has shown that the different approximations made in the eval-
uation of pile cap stiffnesses can lead to large differences in the estimated
pile cap stiffness matrix. It does not follow, however, that there will nec-
essarily be correspondingly large differences in the seismic responses of the
structure. Structural response depends on the system frequencies that result
from the coupling of foundation and structural stiffnesses. The impact that
pile cap stiffnesses have on system frequencies depends on the relative stiff-
nesses of the super-structure and the pile foundation. This effect can be
estimated by the period shift in the rst mode frequency.
The extent of the period change depends on the non-dimensional period
ratio, TP /TF , where TP is the system period for a xed base and TF is the
system period for a exible base. The ratio depends on many factors char-
acterizing the structural system but a dominant factor is the ratio of super-
structural stiffness to foundation stiffness in the direction of the degree of
freedom of interest. In the lateral mode of translation, the ratio is KPS /KLF ,
where KPS is the superstructural lateral stiffness and KLF is the lateral stiff-
ness of the pile foundation.
A parametric study was conducted on a simple two span bridge to
dene the dependence of period shift on relative super-structure/ pile cap
stiffness. The bridge is shown in Figure 36. It has a single column bent
supported by pile group foundations and two seated type abutments. The
bridge is similar to one studied by Lam and Martin (1986) and Imbsen
and Penzien (1986). A computational model of the bridge incorporating
translational and rotational stiffnesses of the pile foundations is shown
Figure 37. Two different pile groups were used to support the central pier,
a (2 2) and a (3 3). Three different soil conditions were considered;
in one case the shear modulus varied parabolically with depth and in the

Figure 36. Two span bridge used in parametric study.

Figure 37. Computational model of bridge.

other two cases the moduli were uniform but different. Three different
pile/soil modulus ratios, Ep /Es , were used. The foundation parameters for
the study are shown in Table IV.

Table IV. Foundation parameters.

Case No. of piles Gmax prole Siffness ratio, Spacing ratio, Depth ratio, Density Ratio,
Ep /Es S/D L/D p /s

1 (2 2) Uniform 100 5 15 0.7

2 (2 2) Uniform 1000 5 15 0.7
3 (2 2) Parabolic 312 5 15 0.7
4 (3 3) Uniform 100 4 15 0.7
5 (3 3) Uniform 1000 4 15 0.7
6 (3 3) Parabolic 312 4 15 0.7
Table V. Superstructure and pier parameters.

Parameter Values

Mass of superstructure, M (Mg) 500, 750, 1000

Youngs modulus of pier, E (MPa) 25000
Density of pier, (Mg/m3 ) 2.5
Height of pier, H (m) 5, 10, 15, 20
Diameter of pier, D (m) 1.25, 1.50, 1.75

Table VI. Calculated pile cap stiffnesses.

Case Lateral Cross-coupling Rotational

stiffness stiffness stiffness
(MN/m) (MNm/rad) (MN/rad)

1 704 364 2440

2 507 118 1244
3 273 99 996
4 1075 578 7375
5 858 177 4169
6 507 178 3412

Figure 38. Period shift for bridge-foundation system as a function of relative lateral
stiffnesses of super-structure and pile foundation.

The mass, height and diameter of the pier were also varied. The param-
eters characterizing the pier and the super-structure are given in Table V.

The calculated foundation stiffnesses for the six cases are shown in
Table VI.
The results are shown in Figure 38, where the non-dimensional period
ratio, TP /TF , is plotted against the non-dimensional stiffness ratio, KPS /KLF .

10. Role of ambient vibration measurements

There are two phases in the calculation of pile foundation stiffnesses, cal-
culating the stiffnesses at the initiation of shaking and tracking the evolu-
tion of these stiffnesses under the varying intensities of earthquake shaking
during the earthquake. Evolution requires a constitutive soil model that has
the demonstrated capability to model non-linear soil behaviour satisfacto-
rily for the problem under consideration. The initial stiffnesses depend only
on the initial distribution of moduli in the ground before the earthquake
and are calculated for elastic response. Since they represent the starting
point on the evolutionary path, it is important to get the best possible esti-
mate of the initial moduli.
Site investigations to characterize foundation soils vary widely in the
technology used and level of detail in dening the in situ soil properties.
Therefore, it is desirable, at least for critical structures, to check indepen-
dently the estimates of initial stiffnesses. Ambient vibration testing provides
a cost effective way to do this for structures being evaluated for retrot.
The utility of the method will be demonstrated by an example from engi-
neering practice.
The Queensborough Bridge over the Fraser River in Vancouver, Canada,
is over 800 m long with three major structures. The North Approach is
220 m long, the South Approach is 490 m and the Main Span over the Fra-
ser River is 204 m. The bridge was built in 1960. It is considered to be
a critical structure and was retrotted in the mid-90s. The Main Span is
shown in Figure 39. Abutment and river piers are shown in Figure 40,
together with the nite element pier model in which the initial founda-
tion stiffnesses are represented by two sets of springs. The initial foun-
dation stiffnesses were calculated for pier S1 , using soil data from a site

Figure 39. Main span of the Queensborough Bridge showing the river (S1 , N1 )
and the abutment piers (S2 , N2 ) (Ventura et al., 1994a).

Figure 40. Typical piers and a nite element pier model of Queensborough Bridge
(Ventura et al., 1994a).

Figure 41. Sensitivity of computed transverse frequencies to soil spring stiffness (Ven-
tura et al., 1994b).

investigation conducted at pier S1 . Since soil data was not available for
piers S2 , N1 and N2 , the translational and rotational stiffnesses of S1 were
scaled in proportion to the areas and second area moments of inertia of
their bases to provide estimates of their stiffnesses.
A series of ambient vibration studies (AVS) of the Main Span were con-
ducted before the retrot work started by Ventura et al. (1994a,b). Mode
shapes and modal frequencies were determined from the ambient vibra-
tion data. The experimental and analytical rst three frequencies of trans-
verse vibration are shown in Figure 41. The sensitivity of these analytical
frequencies to variations in the initial moduli is also shown in Figure 41,
where variations in the moduli are expressed as multiples of the initial

Figure 42. Transverse mode between piers S19 and S23 and displacements at S21 (19)
(Ventura et al., 1994b).

values. For an accurate prediction of the rst transverse mode, the initial
moduli need to be doubled. But the rst modal frequency based on the ini-
tial moduli is quite good; it is 85% of the measured frequency.
For the higher frequencies the soil properties needed adjustment. The
relative insensitivity of all frequencies to very high multiples of the initial
soil moduli arises from the fact that the system frequencies depend not on
foundation stiffnesses alone but on the combined stiffnesses of the founda-
tion and super-structure, as discussed earlier.
The use of ambient vibration data to adjust the initial stiffnesses of the
foundations is based on the tacit assumption that the structural modelling
needs no adjustment. Ambient vibration can also provide valuable data on
potential displacement patterns of a bridge.
Additional lateral vibration measurements were taken at selected piers
to verify the frequencies of the lateral modes of vibration. The transverse
mode shape between piers S19 and S23 in the South Approach is shown
in Figure 42. Two piers showed signicant translation during the AVS. The
pile cap of pier S21 had a lateral translation that was 40% of the trans-
lation of the deck. The pile cap is supported by 25, 14 m long, untreated
timber piles that penetrate 14 m of peat to bed on ne sand. The low trans-
lational stiffness of the pile cap is consistent with the low stiffness of the
peat. The lateral motion at pier S21 is shown in Figure 42 and the coupling
of the different modes of displacement is evident.

11. Behaviour of pile in liqueable soils during earthquakes

11.1. General comments
Loose cohesionless sands and silts below the water table develop high
PWP or liquefy, during strong earthquake shaking, and as a result expe-
rience signicant degradation of strength and stiffness. The seismic design
of pile foundations in these liqueable soils poses very difcult problems
in analysis and design. The degradation of stiffness and strength during

liquefaction leads to substantial increases in pile cap displacements above

those for the non-liqueed case. Depending on when liquefaction occurs,
the pile foundation may undergo substantial shaking, while the soil is in a
fully liqueed state and soil stiffness is at a minimum. During this shak-
ing phase, the pile is prone to suffering severe cracking or even fracture.
After liquefaction, if the residual strength of the soil is less than the static
shear stresses caused by a sloping site or a free surface such as a river
bank, signicant lateral spreading or down slope displacements may occur.
The moving soil can exert damaging pressures against the piles, leading
to failure. Such failures were prevalent during the 1964 Niigata and the
1995 Kobe earthquakes. Lateral spreading is particularly damaging, if a
non-liqueed layer rides on top of the liqueed soil. It is only in the last
few years that the profession has begun to deal effectively with these two
critical design issues. The progress is due to developments in analysis and
ndings from centrifuge tests. The progress in analysis has allowed more
fundamental and comprehensive evaluations of case histories, leading to a
deeper understanding of the mechanics of soil-pile interaction in liqueable
soils and a greater appreciation of the associated design problems.

11.2. Case histories of pile foundations during earthquakes

During liquefaction, large ground displacements can take place on sloping
ground or towards an open face such as a river bank. Displacements from
lateral spreading during the Niigata earthquake are shown in Figure 43.
Displacements as large as 10 m occurred towards the Shinano River. Such

Figure 43. Displacements in Niigata during the 1964 earthquake (Hamada et al., 1986).

Figure 44. Damage to a pile by ground displacement, Niigata 1964.

displacements were very damaging to pile foundations and caused the fail-
ure of two major bridges.
Damage to a pile under a building in Niigata caused by about 2 m of
ground displacement is shown in Figure 44. The damage occurred at the
boundary between liqueed and non-liqueed soil.
The complete shearing of a pile supporting a warehouse on Port Island
by about 2 m of ground displacement during the 1995 Kobe earthquake is
shown in Figure 45. These piles were designed for vertical loads only and
could not carry large moments and shears. Their only function was to con-
trol settlement.
However piles can be designed to carry the moments and shears gen-
erated by earthquake shaking or post-liquefaction ground displacements.
Figure 46 shows a bridge on pile foundations. The foundation soils lique-
ed during the 1983 Nihon-Kai-Chubu earthquake. This led to a failure of
the approach embankments by lateral spreading but the pile foundations
survived without damage.

Figure 45. Shearing of a pile by ground displacement in Kobe earthquake, 1995 (Finn,

Figure 46. Bridge on undamaged pile foundations with failed approaches

Nihon-Kai-Chubu earthquake, 1983 (Finn, 1999).

A pile supporting a crane rail on Port Island, just offshore of Kobe City,
is shown in Figure 47.
The ground moved about 1.0 m in this location after liquefaction
occurred during the Kobe earthquake. The relative motion between the
ground and the pile is clearly evident in Figure 47. However the pile was
designed to carry signicant shears and moments from crane loading and
survived without damage.

11.3. Performance of CIDH piles in liqueable soils

Matsui and Oda (1996) evaluated the damage to the foundations of ve
major elevated expressways in the Kobe region, Japan, caused by the 1995
Kobe earthquake. They conned their study to cast-in-deep-hole (CIDH)

Figure 47. Undamaged pile supporting a crane rail in ground that moved more than
1.0 m (Finn, 1999).

Table VII. Classication of pile damage (Adapted from Matsui and Oda, 1996).

Damage type A-Severe B-Heavy C-Light D-None

Damage Many cracks with Many cracks with Some cracks Almost
description concrete separation concrete separation with separation no cracks
all over pile near pile top near top

Buckling of Many cracks Some cracks in

main around middle middle and
Pile shaft and lower end lower end
separation of pile of pile
Residual pile Totally Probably adequate Adequate Adequate
capacity inadequate vertical, partial capacity capacity
lateral capacity

reinforced concrete piles, as these comprised 80% of all foundation types.

The piles were all over 1 m in diameter. Damage was classied into the four
categories given in Table VII and estimates of the residual load resisting
capacities of piles in each damage class are also given.
The damage was assessed by direct observation of pile shafts, examina-
tion of cores taken from the piles and observations made using borehole

Figure 48. Lateral load test on 1 m diameter CDIH pile (Kimura et al., 1994).

television (BHTV) cameras. The BHTV system was very effective, even hair
cracks could be observed in the images.
Non-destructive methods such as velocity logging, impact wave and elec-
tromagnetic wave methods were also used.
Matsui and Oda (1996) found that pile damage correlated with sub-soil
conditions. Damage was largely conned to areas of liquefaction with and
without lateral ow of liqueed soil. Despite the extensive liquefaction and
the severe damage to the elevated super-structures, damage to the CIDH
piles was negligible. The most extensive damage was along the No. 5 Bay
Route of the Hanshin Expressway: 11% B, 37% C and 52% D. On the No.
3 Kobe route, the damage was 16% C and 84% D. There was no instance
of A category damage. Matsui and Oda (1996) explained cracking pattern
as follows. The cracks near the top of the pile are to be expected as this
is usually the location of maximum moment. The cracks lower down the
pile occur at the location of the second largest moment, at an interface
between soft/liqueed soils and a harder formation or where there is an
abrupt change in the density of reinforcement.
The comments on the residual capacity of the damaged piles in Table VII
were based on tests conducted on CDIH piles, 1 m in diameter, by the
Hanshin Expressway Public Corporation in 1993 (Kimura et al., 1994). The
tests involved single piles and a 3 3 pile group. Data from a typical load
test is shown in Figure 48. The piles showed cracking at around 10 cm of
displacement. At 40 cm displacement, the piles still retained sufcient lat-
eral capacity. A photo of the external condition at the head of one of
these piles corresponding to a displacement of 40 cm is shown in Figure 49.

Figure 49. Damage to 1 m diameter CDIH pile at 40cm displacement (Kimura et al.,

Clearly the CDIH piles behaved very well, more particularly as they
were designed for much less intense ground shaking than they experienced
during the Kobe earthquake.
The review of case histories has clearly demonstrated the design prob-
lems posed by pile foundations in liqueed soils. To cope with these prob-
lems it is essential to have a reliable method of calculating the effects of
earthquake shaking and post-liquefaction displacements on pile founda-
tions. Methods for analysis and design introduced tentatively into practice
since 1995 and of ndings from extensive research programs involving case
histories, full-scale tests and shake table and centrifuge tests in recent years
will be reviewed critically with the aim of presenting an integrated up to
date assessment of the state of the art.

12. Analysis of pile foundations in laterally spreading ground

12.1. General comments
In the case histories section, it was shown that large post-liquefaction
displacements can occur and that these can be very damaging to pile
foundations. These potential deformations can control design but they
are very difcult to predict reliably. In engineering practice, the
displacements at the top of the liqueed layer are often estimated by empir-
ical formulas based on eld data from past earthquakes. The rst pre-
dictor equation was developed in Japan by Hamada et al. (1986). Very
comprehensive predictor equations have been developed by Youd et al.
(1999) in the USA which are used in practice in North America. An
updated version of the Hamada equation has been adopted by the Japan

Figure 50. Distortion of pile foundation by lateral soil displacement.

Water Works Association (JWWA, 1997) based only on ground slope

and the thickness of the liqueed layer. Bardet et al. (2002) developed a
method for modelling post-liquefaction displacements on a regional basis.
In practice, the free eld displacements are assumed usually to vary lin-
early from top to bottom of the liqueed layer. The deformed shape of
a pile foundation caused by these displacements is shown in Figure 50.

12.2. Force analysis

A force based analysis is recommended in a number of Japanese design
codes for analysis of piles foundations in liqueed soils, undergoing lateral
ow (JRA, 1996; JWWA, 1997). The underlying concepts are rational and
simple. An non-liqueed surface layer, which is transported on the mov-
ing liqueed soil is assumed to apply passive pressure on the foundation.
A liqueed layer is assumed to apply a pressure less than the equivalent
hydrostatic pressure on the piles because of the internal ow resistance of
the liqueed sand. The transmitted lateral pressure was found to average
about 30% of the overburden pressure on the basis of back analysis of
case histories. The pressure distribution against the foundation for design
is shown in Figure 51. These pressure distributions are associated with the

Figure 51. Design pressures against piles in laterally owing liqueed soils (JRA,

potentially large deformations that foundations close to sea-walls might

experience. Data from the Kobe earthquake suggested that deformations
a 100 m or more from the quay wall approached zero. The above pres-
sures can be reduced by 50% for distances between 50 and 100 m from the
wall. The codes should be consulted before use. Only the essential outline is
given here to indicate what is available for solving this difcult problem. An
attempt to convert the distance criteria to displacement criteria to broaden
the area of application of the method was abandoned because the two
data bases available (Ishihara et al., 1996; Hamada et al., 1998) give sig-
nicantly different displacement/distance relationships. Brandenburg et al.
(2005) studied load transfer from the non-liqueed layer to piles in cen-
trifuge tests. Their results show that the load transfer is a function of dis-
placement. Large relative displacements are required to develop peak load
against the piles. This study conrms in a qualitative way the guidelines in
JRA (1996).
Dobry and Abdoun (2001) and Ramos et al. (1999) have studied the
behaviour of piles in laterally owing soils by centrifuge tests. The test set-
up is shown in Figure 52.
Typical test results for moments in the piles are given in Figure 53. In
order to simulate the moments, they adopted the two different pressure
distributions: inverted triangular and uniform for the liqueed soil. Adop-
tion of the inverted triangular distribution may have been inuenced by the
inverted triangular distribution of displacements in the liqueed soil.
However, when there is lateral restraint at the pile head, both distributions
seem to over-estimate the bending moments in the upper part of the pile.
Tarek and Wang (2000) studied the effects of lateral spreading of ground with
a upper lightly cemented layer on piles in centrifuge tests. They concluded

Figure 52. Centrifuge test on pile in owing soil (Dobry and Abdoun, 2001).

Figure 53. Computed and measured pile moments (Dobry and Abdoun, 2001).

Figure 54. A Winkler spring model for pile foundation analysis.

that the moments in the pile were dominated by the lateral pressures from
the cemented layer. They found the force model to be useful.

12.3. Displacement analysis

The rst step in the analysis is to estimate the post-liquefaction free eld
displacements using one of the empirical formulas or by an appropriate
method of analysis. These displacements are usually assumed to vary lin-
early from the top to the bottom of the upper liqueed layer. These dis-
placements are then applied to the free eld ends of the near eld springs
in the very general Winkler model shown in Figure 54 and a static analysis
is performed. Degraded py curves have usually been used for this kind of
The effects of lateral spreading on 1.5 m diameter CDIH piles support-
ing the structure shown in Figure 85 below were analysed as described
above. The free eld displacements at the surface were estimated to be
between 15 cm and 25 cm. The computed pile displacements, assuming that
the pile head is xed against rotation, are shown in Figure 55.
The resulting bending moments are shown in Figure 56. Note that the
maximum bending moment is near the interface between the liqueed and
non-liqueed layers.
13. Soil properties for displacement analysis
13.1. Japanese practice
In Japanese practice, the springs in the Winkler model are linearly elas-
ticplastic. The elastic soil stiffness is determined by semi-empirical code

Figure 55. Calculated pile displacements for specied ground ow (Finn, 1999).

Figure 56. Pile moments induced by eld displacements (Finn, 1999).


Figure 57. Japanese computational model for pile groups (JRA, 1996).

formulas related to the elastic modulus of the soil. This modulus is

deduced from plate loading tests or correlations with the SPT-N measure-
ments and therefore includes some non-linear effects. In some design of-
ces, the spring constant K is taken as zero in liqueed soil. The typical Jap-
anese computational model for pile groups is shown in Figure 57.
The JRA (1996) code for highway bridges recommends reductions in
the spring stiffness for use in liqueable soils that depend on the fac-
tor of safety, FL , against liquefaction. The reduction factors are given in
Table VIII. The resistance to liquefaction, RL , is determined by cyclic tri-
axial tests on undisturbed samples obtained by in situ freezing techniques.
This strength is modied depending on whether Type 1 or Type 2 motions
are used in design, by a factor cw . Then R = cw RL is the dynamic shear
strength ratio in Table VIII. The factor cw has a value of 1 for Type 1
motions. For Type 2 motions, cw = 1.0 for RL 0.1, cw = 3.3RL + 0.67 for
0.1 < RL 0.4 and cw = 2.0 for 0.4 < RL .
The code should be consulted for details of the 2 types of motions.
Generally Type 1 motions are the design motions before the Kobe earth-
quake. Type 2 motions were introduced to provide protection against
another earthquake like Kobe.

13.2. North american practice

There is no general consensus in North American practice on the appro-
priate modelling of the Winkler springs for post-liquefaction analysis. The
basis of most analyses is a degraded form of the API py curves or curves
Table VIII. Reduction coefcients for soil constants due to liquefaction (JRA, 1996).

Range of FL Depth from the present Dynamic shear strength ratio R

ground surface x (m)

R  0.3 0.3 < R

FL  1/3 0  x  10 0 1/3
10 < x  20 1/3 1/3
1/3 < FL  2/3 0  x  10 1/3 2/3
10 < x  20 2/3 2/3
2/3 < FL  1 0  x  10 1/3 1
10 < x  20 1 1

due to Reese et al. (1974). The practice is to multiply the py curves, by a

uniform degradation factor , called the p-multiplier, which ranges in value
from 0.3 to 0.1. This follows from the original work of Dobry et al. (1995).
They found that bending moments could be predicted adequately using a
Winkler model type of analysis, if the commonly used py curves were uni-
formly degraded by multiplying by a degradation factor that appeared
to diminish with increasing PWP to a value of 0.1 at 100% excess PWP.
Wilson et al. (2000) conrmed these results but showed that the p-multi-
plier for fully liqueed soil depended also on relative density, ranging in
value from 0.10.2 for sand at about 35% relative density and 0.250.35 for
a relative density of about 55%.
Wilson et al., (2000) also found that the resistance of the loose sand did
not pick up even at substantial strains but that the denser sand, after an
initial strain range in which it showed little strength, picked up strength
with increasing strain. This nding suggests that the good performance of
the degraded py curves which did not include an initial range of low or
zero strength, must be test specic and the p-multiplier may be expected to
vary from one design situation to another.
The very low initial strength range in the laboratory py curves fol-
lowed by a range of increasing strength is related to the dilatancy char-
acteristics of sand at low effective stresses. Similar behaviour is observed
in tests in which undrained monotonic loading is conducted on sand spec-
imens after cyclic loading to liquefaction. Typical examples of this phe-
nomenon from undrained torsional tests are shown in Figure 58 (Yasuda
et al. 1999). Vaid et al. (1996) showed that the strain range of very low
undrained resistance after liquefaction depends on the number of cycles of
stress reversal the sand experiences after liquefaction, before the undrained
monotonic loading is applied.

Figure 58. Post-liquefaction undrained stressstrain behaviour of sand (Yasuda et al.,


13.3. Findings from full-scale tests

Weaver et al. (2001) conducted full-scale cyclic loading tests in the eld on
a 0.6 m diameter cast-in steel-shell (CISS) pile in liqueed soil to assess
the accuracy of the py type of analysis. The test site is on Treasure
Island in San Francisco Bay which is the location of the National Geo-
technical Experimentation Site. Therefore soil conditions at the site are
very well known. Liquefaction was caused by blasting and the cyclic load-
ing was conducted using a high speed hydraulic actuator. The back g-
ured py curves for the liqueed sand differed signicantly in shape from
the standard py curves modied by the p-multiplier. The shape of the
back-calculated py curves are shown in Figure 59.
The standard py curves including the p-multiplier effect are also shown
for comparison. The two sets of py curves have distinctly different shapes
and give different estimates of soil resistance. The shapes of the Weaver
et al. (2001) curves are consistent with the post-liquefaction undrained
monotonic the test data from Yasuda et al. (1999) and Vaid and Thomas
The slope of the standard py curve is greatest at small displacements
and eventually decreases to zero at large displacements. The back calcu-
lated py curves show no resistance for a range of displacements between
20 and 50 mm. The soil resistance increased thereafter and was still increas-
ing after 150 mm.
The py curves from the full-scale eld test share some characteristics
with the py curves obtained by Wilson et al. (2000) from centrifuge tests.
The Wilson data for one cycle of loading at different depths are shown
in Figure 60. The hysteresis loops are very attenuated showing almost no
pressure being exerted against the pile. The standard py curves, with the
p-multiplier effect included, are shown for comparison.
Again the shapes are radically different. Liu and Dobry (1995) found
similar results and concluded that liqueed loose sands provide very lit-
tle resistance. In this case it may be reasonable to ignore the effects of

Figure 59. Comparison of standard py curves with curves back-gured from test
data at depths of (a) 0.2 m and (b) 2.3 m from a full-scale pile test (Weaver et al.,

Figure 60. Comparison of standard py curves with curves back-gured from centri-
fuge data (Wilson et al., 2000).

liqueed loose sand as far as pressure on the piles is concerned even in

the force analysis. This is consistent with the judgment of some Japanese
designers who assign zero stiffness to the elastic springs in their form of
Winkler displacement analysis as discussed earlier.

13.4. Lateral spreading in centrifuge tests

Brandenburg et al. (2001) conducted a very comprehensive series of tests to
determine the effects of various parameters on pile performance in laterally
spreading ground. Centrifuge tests on single piles and 2-pile groups were
conducted on the centrifuge at UC Davis. Pipe piles were used. The single
piles had prototype diameters of 0.36, 0.73 m, and 1.45 m. The piles in the
pile group were 0.73 m in diameter. The foundation soil prole sloped gen-
tly towards a channel at one end of the shear box as shown in Figure 61. It
consisted of a non-liqueable layer of clay, with a thin sand cover, under-
lain by a liqueable layer of sand with a relative density of 35% and a base
layer of dense sand at a relative density of 85%.
The responses of the piles to lateral spreading were analysed using a
non-linear Winkler model. The Matlock (1970) static py relation for soft
clay and the Reese et al. (1974) static py relation for sand were used to
represent the non-linear springs. A p-multiplier = 0.1 was used for fully
liqueed sand.
Three cases were considered: (1) original py curves for loose sand with
p = 0.1 and only the properties in the loose liqueed sand were degraded
for pore pressure effects: (2) original py curves for loose sand with = 0.1
and reductions in py stiffness and capacity of the dense sand due to PWP
in that layer; (3) the same as case (2) except that the standard py adjust-
ment factors to the static py curves for cyclic loading were made also. As
Brandenburg et al. (2001) point out these latter adjustments were developed

Figure 61. Centrifuge model test (Brandenburg et al., 2001).


for the large number of water wave generated stress cycles associated with
a major offshore storm and are probably not applicable to the far fewer
signicant stress cycles associated with earthquake shaking. Comparison of
measured and computed responses led to a number of important conclu-
sions. The three most important ones are quoted verbatim below.
The recorded responses of the three single piles and the one group of
two piles could be modelled within the range of parameter variations that
were studied, but all the responses could not be accurately modelled with
the same set of input parameters.
The parameter studies also showed that the standard adjustments to py
relations for cyclic loading would have resulted in substantial under-predic-
tion of lateral loads from the clay layer.
The calculated bending moments were more sensitive to the strength and
py parameters for the upper clay and sand cover layers, and less sensitive
to the p-multiplier assigned to the liqueed layer.
These ndings pose clear warnings for anyone contemplating analyses of
piles in laterally spreading soils using the standard North American py
curves. The crucial factors seem to be; the dominating role of the non-
liqueable layer, the inappropriateness of using the standard cyclic loading
reduction factors for earthquake shaking and the large uncertainty associ-
ated with the results of any analysis.
Some of the problems of arriving at a generally acceptable set of
Winkler non-linear py curves for analysis of post-liquefaction deforma-
tions in liqueable soils arise from the assumed form of the curves. If the
form is incompatible with the actual stressstrain behaviour of the soil,
problems in simulating the responses of different pile foundations with one
set of py curves is not surprising. The North American py curves are
concave downwards and this is not compatible with the post-liquefaction
undrained behaviour of liqueed sand under monotonic loading which is
concave upwards (Figure 58)

13.5. MCEER/ATC (2003): lateral spreading and bridge foundations

MCEER/ATC (2003) contains recommended guidelines for evaluation and
design of pile foundations for bridges in laterally spreading soil. The pro-
posed methodology is summarized in the ow chart in Figure 62.
The guidelines cover the same two ground conditions as the Japanese
codes; liqueable soils with and without a non-liqueable surface layer.
Liqueed soils are assigned a residual strength and treated as cohesive
soils. The friction angle of cohesionless soils is taken as 100 when the fac-
tor of safety against liquefaction is FS = 1.0 and is interpolated between
for 1.0 FS 1.5. The modulus of subgrade reaction is reduced in a
similar way, with a value equal to that of a soft clay at FS = 1.0. The

Figure 62. Flow chart for design of piles in spreading soil (after MCEER/ATC, 2003).

recommendations for analysis are presented in the context of single pile

response. If there is no non-liqueable layer at the surface, the likely result
is that the soil will ow around the pile and exert passive pressure on it. In
this case, the ground deformations are likely to be much greater then the
pile deformations as shown in Figure 63. If there is a non-liqueable layer
at the surface, this layer will either grip the foundation, forcing it to fol-
low the ground deformation, (Figure 64), or the soil will ow around the
In the rst case, if the pile cannot withstand structurally the imposed
ground displacements, then remedial action is necessary. One option is to
retrot the pile foundation by adding either active or passive piles.
The analysis to determine the number of piles is treated as an uncoupled
problem. The Newmark rigid block analysis (Newark, 1965) is suggested
for estimating ground displacements. The pinning action of the piles is
represented in the sliding block model by their shearing resistance. The
displacements can be plotted against the pinning forces as shown in Fig-
ure 65 (Martin, 2004). The pile displacements needed to mobilize differ-
ent amounts of shearing resistance are shown also in Figure 65. The

Figure 63. pilesoil relative displacements in fully liqueable soil (after MCEER/ATC,

Figure 64. pilesoil displacements with non-liqueable crust (after MCEER/ATC,


Figure 65. Determining compatible pile pinning forces (after Martin, 2004).

intersection of the two curves gives the equilibrium state for the pile foun-
dation. If the displacement meets the performance objective, the retrotted
foundation is satisfactory. The guidelines allow the designer to use more

Figure 66. Cross-section of Sardis Dam.

Figure 67. Cross-section of Sardis Dam showing liqueable zones and treated zone.

sophisticated methods of analysis, if the benets justify the additional costs,

but it is difcult to see how this judgment can be made without actually
doing the analysis.

13.6. Coupled analysis of pile pinning

The rst coupled analysis of pile pinning in seismic design was conducted
during a seismic safety evaluation of Sardis dam in Mississippi (Finn et al.,
1999). A more complete report may be found in Finn et al. (1998). A
cross-section of the dam is shown in Figure 66. Areas of potential liq-
uefaction: parts of the core and the upstream slope, and a thin layer of
clayey silt under the upstream slope, are shown in Figure 67. The post-
liquefaction behaviour of the dam was evaluated using Tara-3 FL (Finn
and Yogendrakumar, 1988) which is based on a Lagrangian formulation for
large strain non-linear analysis. The deformed shape of the dam is shown
in Figure 68 for one choice of post-liquefaction residual strength in the
thin layer.
After detailed engineering and nancial studies of various remedial mea-
sures to stabilize the upstream slope, pile pinning emerged as the cheapest,

Figure 68. Deformed shape of Sardis Dam after liquefaction.

Figure 69. Piles pinning the upstream slope to the foundation.

most easily constructible and offering the best quality assurance. An eleva-
tion view of the pinning design is shown in Figure 69.
Initially a uniform distribution of piles over the remediation area was
analysed to obtain an overall view of how load would be shared among
the piles after liquefaction. An effective stress non-linear dynamic analysis
was rst carried out using the program TARA-3 (Finn et al., 1986). This
analysis gave time histories of moments, shears and deections. A static
analysis was then conducted using TARA-3 FL to determine the effect of
the static thrust of the upstream slope after liquefaction of the weak hor-
izontal layer. In this analysis the strengths and stiffnesses were gradually
reduced from pre-earthquake values to values consistent with the seismi-
cally induced PWP and the residual strengths of fully liqueed soils. The
distribution of peak static moments is shown in Figure 70.
A major portion of the load is carried by the line of piles that is clos-
est to the dam crest. The peak moment is nearly double that of the next
line of piles. This occurs because there is no pile cap to transmit load ef-
ciently between piles. Transmission of load depends on pile deformations.
Although it is possible to design for this load distribution, there is clearly
a risk of a chain reaction failure, if the piles in the rst group should fail.
Therefore the pile spacing was decreased for the rst three rows and the

Figure 70. Distribution of peak moments from thrust of upstream slope after lique-

Figure 71. Plan view of pinning piles at Sardis Dam (Finn et al. 1998).

piles in these rows were designed for the moments and shears in the lead-
ing row. A plan view of the nal layout of piles is shown in Figure 71.
The time history of dynamic moments in the piles in the leading row
nearest the crest of the dam is shown in Figure 72. The design moment for
the leading three rows of piles was the sum of the static moment and 67%
of the peak dynamic moment.
This case history is a good example of how useful a continuum anal-
ysis can be in giving a designer a detailed picture of how the structure
is likely to behave in different circumstances and to alert him to poten-
tially dangerous phenomena. Any continuum analysis with a reasonably
representative constitutive model can provide this function, at least qual-
itatively. To guarantee a measure of reliability in quantitative predictions,
the method of analysis should be validated by data from other than labo-
ratory element testing. Centrifuge testing provides the best environment for

Figure 72. Time history of dynamic moments in the leading row of piles (Finn et al.,

validation. These tests provide non-homogeneous stress, strain and PWP

elds that are representative of eld conditions. An example of such an
evaluation is given below.

14. Dynamic analysis of piles in liqueable soils: centrifuge tests

14.1. Description of tests
Dynamic centrifuge tests of pile supported structures in liqueable sand
were performed on the large centrifuge at University of California at Davis,
California (Wilson, 1998). These tests provided an opportunity to evalu-
ate the capability of the non-linear effective stress program PILE-3D-EFF
which is used in the next section to analyse the behaviour of large diameter
piles supporting a 14 storey building in liqueable reclaimed land. The typ-
ical arrangement of structures and instrumentation is shown in Figure 73.
Full details of the centrifuge tests can be found in Wilson et al. (1998). The
models to be analysed here are a structure supported by single piles and a
structure supported by the 2 2 pile group. Model tests were performed at
a centrifugal acceleration of 30 g.
The soil prole consists of two level layers of Nevada sand, each
approximately 10 m thick at prototype scale. Nevada sand is a uniformly
graded ne sand with a coefcient of uniformity of 1.5 and mean grain
size of 0.15 mm. Sand was air pluviated to relative densities of 7580%
in the lower layer and 55% in the upper layer. Prior to saturation, any
entrapped air was carefully removed. The container was then lled with
a hydroxy-propyl methyl-cellulose and water mixture under vacuum. The

Figure 73. Layout of models for centrifuge tests (Wilson, 1998).

Figure 74. Instrumented pile for single pile test (Wilson, 1998).

viscosity of this pore uid is about ten times greater than pure water to
ensure proper scaling. Saturation was conrmed by measuring the compres-
sive wave velocity from the top to the bottom.
The model dimensions and the arrangement of bending strain gauges
for the single pile and the 2 2 group are shown in Figures 74 and 75,
respectively. Model tests were performed at a centrifugal acceleration of
30 g.
The responses of the single pile and the 2 2 pile group to the Santa
Cruz acceleration record obtained during the 1989 Loma Prieta earth-
quake, scaled to 0.49 g, are described and analysed here.

Figure 75. Instrumented test piles and superstructure for 2 2 pile group (Wilson
et al. 1998).

Figure 76. Finite element mesh for single pile.

14.2. Effective stress dynamic analysis of piles-superstructure

The Davis tests were analysed using the effective stress version of PILE-3D,
the program PILE-3D EFF. The nite element mesh used in the analysis is
shown in Figure 76. The nite element model consists of 1649 nodes and
1200 soil elements. The upper sand layer which is 9.1 m thick was divided
into 11 layers and the lower sand layer which is 11.4 m thick was divided

into 9 layers. The single pile was modelled with 28 beam elements. 17 beam
elements were within the soil strata and 11 elements were used to model
the free standing length of the pile above the soil. The super-structure mass
was treated as a rigid body and its motion is represented by a concentrated
mass at the centre of gravity. A rigid beam element was used to connect the
superstructure to the pile head.
The small strain shear moduli, Gmax , were estimated using the formula
proposed by Seed and Idriss (1970).

Gmax = 21.7 kmax Pa m /Pa (5)

in which kmax is a constant which depends on the relative density of the

soil, m is the initial mean effective stress and Pa is the atmospheric pres-
sure. The program PILE-3D-EFF accounts for the changes in shear moduli
and damping ratios due to dynamic shear strains at the end of each time
increment. The shear strain dependencies of the shear modulus and damp-
ing ratio of the soil were dened by the curves suggested by Seed and Idriss
(1970) for sand. The friction angles of the upper and the lower layers were
taken as 35 and 40 , respectively.
Increments in seismic PWP were generated in each individual element
depending on the accumulated volumetric strain prevailing in that element
and the current increment in volumetric strain (Martin et al., 1975). The
moduli and shear strengths of the foundation soils were modied continu-
ously to account for the effects of the changing seismic PWP.
The Santa Cruz acceleration record from the 1989 Loma Prietc Earth-
quake was scaled to 0.49 g and used as input to the shake table. The base
accelerations of the model were measured at the east and west ends of the
base of the model container. (Wilson, 1998) showed that both accelerations
agreed very well. The base input acceleration is shown in Figure 77.

Figure 77. Input acceleration time history.


Figure 78. Comparison of measured and computed super-structure acceleration time


Figure 79. Response spectra of measured and computed superstructure accelerations.

14.3. Results of single pile and 2 2 group analyses

Figure 78 shows the measured and computed acceleration response of the
superstructure. There is generally good agreement between them, especially
in the time period of peak response.
Response spectra of the computed and measured accelerations at the
superstructure mass are shown in Figure 79. The agreement between them
is quite good.
Figure 80 shows comparisons between measured and computed PWP at
three different depths; 1.14, 4.56, and 6.78 m in the free eld. There is gen-
erally good agreement between the measured and computed PWP response.
Figure 81 shows the measured and computed bending moment time his-
tories at two different depths; 0.76 and 1.52 m. Generally there is a very
good agreement between the measured and computed time histories. Fig-
ure 82 shows the proles of measured and computed maximum bending
moments with depth. The comparison between measured and computed
moments is fairly good. The distributions of maximum computed and mea-
sured moments for the 2 2 pile group, shown in Figure 83, also agree
quite well.

Figure 80. Comparison of measured and computed PWP at three depths.

The capability of PILE-3D-EFF to simulate satisfactorily the full range

of response parameters in the centrifuge tests; accelerations, PWP and
bending moments, gives some credibility to the ndings from analyses of
foundation piles in the following section.

15. Analysis of piles under 14 storey building in liqueable soils

15.1. Description of analyses
In this section, the relative contributions of inertial and kinematic interac-
tions to the seismic response of the foundation piles of a 14 storey build-
ing will be discussed. The effect of a non-liqueable surface layer or dry
crust, which was found to be so important in the case of laterally spread-
ing ground, will be presented also.
Seismic analyses were conducted on a 1.5 m diameter cast-in-place rein-
forced concrete pile supporting a column of a 14 storey building. The soil

Figure 81. Comparison of measured and computed bending moments at two depths.

Figure 82. Comparison of measured and computed maximum moments along the pile.

Figure 83. Comparison of measured and computed maximum bending moments along
the piles in the 2 2 pile group.

conditions and pile are shown in Figure 84. Slightly idealized site condi-
tions are shown in Figure 85. The soils in the upper 10 m are expected to
liquefy or develop high-PWP during the design earthquake.
The mass mounted on the pile in Figure 85 represents the mass
equivalent of the static reaction force carried by the pile. The purpose in
placing the mass on the pile is to model approximately the inertial inter-
action between the super-structure and the pile foundation. It is mounted
on the pile head by a massless support with a stiffness that ensures a
period of vibration of 1.4 sec corresponding to the estimated fundamen-
tal period of the prototype structure. Dynamic effective stress analyses
of this system include both inertial and kinematic interactions and the
effects of high PWP and liquefaction. Analyses were also conducted with-
out including the mass of the superstructure. These latter analyses give
the kinematic deections and moments. In all these analyses, the non-lin-
earity of the soil and the effects of seismic PWP are taken into account
by adjusting the soil properties continuously for current PWP and shear
The peak acceleration of the input acceleration record is 0.25 g and is
amplied to 0.4 g at the surface. The surface accelerations become negligi-
ble after liquefaction has occurred.

Figure 84. Site in reclaimed land.

Figure 85. Model of soilpile-structure system.


Figure 86. Maximum moments along the pile; pile head free to rotate (Finn, 1999).

15.2. Results of analyses: liqueable surface layer

15.2.1. Analyses with inertial interaction
Moment distributions with and without liquefaction are shown in Fig-
ure 86 and Figure 87 for two pile head conditions; free to rotate and xed
against rotation respectively.
Liquefaction results in a major redistribution of maximum moments in
the pile with depth. Without liquefaction, the maximum moment in the
free headed pile occurs at a depth of 3 pile diameters. The maximum pile
head moment is the maximum base moment from the structure. If the site
is liqueable, the dynamic response of the system to the input motions is
quite different. As a result, there is a 50% drop in the base moment of the
superstructure and a 300% increase in bending moment at the boundary
between the stiff and soft soils.
When the pile head is xed against rotation, maximum moment occurs
at the pile head with or without liquefaction. The base moment from the
structure does not contribute to the pile head moment in this case. The
pile head moment results from the combined actions of the base shear from
the structure and kinematic interaction between pile and soil, while taking
into account the changes in soil properties due to seismic motions in the
foundation soils. If a pseudo-static analysis were performed, the same base
shear and moment would have been applied to the pile head in both cases
because they is usually calculated, assuming the structure is on a rigid base.
Pile displacements and moments, at the instant of maximum pile head
displacement, are shown in Figures 88 and 89 respectively for the case
of liquefaction. In design, maximum pile head displacement is often

Figure 87. Maximum moments along the pile; pile head xed against rotation (Finn,

Figure 88. Deections along pile.

considered the most critical condition. Results are shown for two condi-
tions; the pile head is xed against rotation because of the large grade
beams used in Japan to tie together the pile foundations of the building
and the pile head is essentially free to rotate which is closer to North
American practice which uses much lighter grade beams. The large grade
beams are very effective in controlling displacements because the end x-
ity against rotation mobilizes much higher inherent pile stiffness. The max-

Figure 89. Moments along pile.

imum moment occurs at the pile head, when the pile head is xed against
rotation, but very signicant moments also occur at the boundary between
the softer and stiffer soils. When the pile head is not xed against rota-
tion, the maximum moment occurs at the boundary between the stiffer and
softer soils. This moment is approximately equal to the pile head moment,
when the pile head is xed against rotation. The displacements are more
than twice as large when the pile head is free to rotate.
These results suggest that when designing piles or evaluating pile
foundations in potentially liqueable soils for earthquake loading, it is
important to make a realistic assessment of pile head restraint against rota-
tion and to be aware of the potential for large moments at the interfaces
between soft and hard layers.

15.2.2. Pile head restraint in practice

In North America, grade beams are used to tie the isolated single pile foot-
ings together. These beams are designed to carry axial loads equal to 10%
of the column load. In Japan much larger grade beams are used to tie the
footings together. The grade beam for the 14 storey apartment building in
reclaimed land is shown in Figure 90. Large, well-reinforced beams of this
type provide considerable restraint to the pile head. This restraint should
be included in analysis, especially in liqueed soils, where the relatively high
levels of xity can be important in keeping displacements at an acceptable

Figure 90. Large grade beam for 14 storey building.

The contribution of the grade beams to the rotational stiffness of the

pile head in Figure 85, was estimated to be of the order of 10E + 07 kNm.
In order to bracket this value, the seismic pile displacements were calcu-
lated by Pile-3D for total grade beam rotational stiffnesses of 10E + 08
and 10E + 06 kNm for the case when liquefaction occurs in the upper
loose sands during the earthquake. The distribution of maximum displace-
ments is shown in Figure 91. The maximum displacements, where the pile
head is assumed to be xed against rotation or not, are shown in Fig-
ure 92 for comparison. The pile head displacements in Figures 91 and 92
show that full xity is approximated by a pile head rotational stiffness of
10E + 08 kNm. Clearly the grade beams with an estimated rotational stiff-
ness of 10E+07 provide a high level of restraint and their effects should be
included in any analysis. Rotational restraint is very effective in controlling

Figure 91. Pile displacements for specied pile head rotational restraints.

Figure 92. Pile displacements for limiting pile head rotational restraints.

15.2.3. Pile head stiffnesses in liqueable soils

The time histories of pile stiffnesses for the 1.5 m foundation pile are
shown in Figure 93 for three different conditions; liquefaction, no seismic
PWP and high intermediate pressures. Clearly it is not enough to consider
only liquefaction but to take into account any reasonably high PWP when
evaluating pile head stiffnesses.

Figure 93. Time histories of lateral stiffness for various PWP.

15.3. Effects of non-liqueable surface layer; inertial interaction

At some sites a thick surface layer of non-liqueable soil, a crust, may
lie over the liquefaction zone. Typically this layer will suppress the surface
manifestations of liquefaction. Here the crust is initially assumed to be very
stiff, corresponding to an overconsolidated clay layer or a layer densied to
prevent damages to infrastructure or lighter structures not on piles. Deec-
tions and moments at the instant of maximum pile head displacement are
shown in Figures 94 and 95 respectively. As before, the results are shown
for two pile head conditions; no rotation and free to rotate. In this case,
the pile head displacements are about the same whether the pile head is
xed against rotation or not. Also the deection of the pile head, when the
pile head is xed against rotation, has more than doubled compared to the
previous case of no stiff upper layer. This is due to the restraint of the pile
by the upper layer and the movement of the layer as a rigid body after
liquefaction develops. The stiff upper layer greatly increases the moment
demand on the pile during earthquake shaking. The moments at the pile
head and at the interface between the soft and stiff soils has increased sub-
stantially, compared to the case without the upper layer. When the pile is
xed against rotation the moments at the pile head and the interface are
about the same. The behaviour of the upper layer is claried further in the
next section which presents results from kinematic analyses.

15.4. Effects of non-liqueable layer: kinematic interaction

Kinematic analyses were conducted on the 1.5 m diameter pile with and
without the stiff surface layer to assess the importance of kinematic
interaction. In each case, the pile head was considered either xed against

Figure 94. Pile displacements in layered soils.

Figure 95. Pile moments in layered soils.

rotation or not. The kinematic analyses were conducted by removing the

super-structural mass in Figure 85.
The free eld displacements of the foundation soils and the pile dis-
placements, both at the instants of maximum pile head displacement, are
shown in Figures 96 and 97. It is evident from Figure 97 that the stiff
upper layer is moving at this time as a rigid body.
The displacement patterns of the soil in the free eld and pile are
different. The compatibility of displacements must be preserved along the

Figure 96. Displacements of pile and free eld.

Figure 97. Displacements of pile and free eld.

interfaces, wherever the soil and pile maintain contact during earthquake
shaking. Signicant moments and shears may develop in the pile in order
to maintain this compatibility requirement as shown in Figures 98 and 99.
In the case of the stiff upper layer, the moments are about twice as large as
for the case with no such layer. In both cases, the moments are of the same
order as when the inertial mass is incorporated. This indicates that, for
these conditions, the kinematic moments dominate the moment response

Figure 98. Kinematic moments along the pile with no stiff upper layer.

Figure 99. Kinematic moments along the pile with stiff upper layer.

of the foundation. Therefore analyses that neglect kinematic effects may

underestimate signicantly design moments and shearing forces in founda-
tion piles.
These analyses were repeated with a much softer surface layer. The 4 m
surface layer was assigned stiffness compatible with the gravitational effec-

tive stresses in the layer, Moments from kinematic analyses only are shown
in Figures 100 and 101 for the cases where the pile head is free to rotate
and is xed against rotation. The results for no non-liqueable surface
layer are also shown for comparison in each case.
When the pile is free to rotate, the kinematic moment at the bound-
ary between the soft soils and the stiff soils below is increased 50% by

Figure 100. Distribution of kinematic moments: pile head free to rotate.

Figure 101. Distribution of kinematic moments: pile head xed against rotation.

the presence of the surface crust. When the pile head is xed against
rotation, the pile head moment is increased by 75% and the moment at
the boundary between soft and stiff soils is increased by 50%. Thus the
effects of a nonliqueable surface layer can have a major effect on the
dynamic moments and shears in the piles. The pile needs to be analysed
by a method that can pick up these effects. The analyses also make clear
that kinematic interaction should not be routinely neglected as in the basic
pseudo-static analysis procedure. Therefore the preferred type of analysis, if
a py model is to be used, is the dynamic model, which includes the effect
of ground motions.

16. Summing up
16.1. Single pile analysis: non-liqueable soils
In engineering practice, the moments, shears and deections in a pile foun-
dation caused by earthquake shaking are calculated using a pseudo-static
analysis in which the pilesoil interaction is modeled by non-linear springs.
The most commonly used springs are the non-linear py curves recom-
mended by the American Petroleum Institute. The py curves were devel-
oped for monotonic and slow cyclic loading conditions by correlation with
full-scale eld tests and achieved their present form by the mid-seventies.
Two major studies in the mid-eighties, involving data from 35 monotonic
tests and 19 cyclic loading tests of piles in sands and clays, concluded that
the py constitutive model gave poor predictions and was fairly unreliable.
The model has not been calibrated for seismic loading conditions and can-
not be expected to perform better in the seismic environment.
In pseudo-static analysis, the foundation is loaded by the base moment
and shear from the super-structure. This top down analysis neglects two
important features of seismic response: kinematic interaction between pile
and soil and the effects of the ground motions on the stiffness of the foun-
dation soils. Kinematic moments are important whenever there is a sharp
difference in stiffness between adjacent layers. It is particularly important,
when a site which experiences liquefaction, has a non-liqueable layer at
the surface. In these conditions, there can be a big increase in moment
demand. The potential importance of kinematic moments is recognized by
Eurocode 8: Part 5 which species the conditions under which kinematic
interaction should be taken into account.
The shear modulus of soil is strain dependent. Therefore shear strains
generated by strong shaking can cause signicant reduction in soil stiffness.
The top down pseudo-static analysis takes into account only the stiffness
reduction due to the inertial loads, neglecting the effects of ground motions
on soil stiffness. These omissions further degrade the predictive capability

of the pseudo-static approach. Data from centrifuge tests show that the
py curves may need considerable adjustment to match recorded response
parameters with some adjustment factors reaching a value of 2.
The py based analysis can be improved to cope more effectively with
kinematic interaction and ground motion effects. If an appropriate distri-
bution of ground displacements is applied as input to the ends of the near
eld py springs, a useful approximation to kinematic interaction can be
achieved in a pseudo-static analysis. An appropriate displacement distribu-
tion might be, for example, the distribution when the surface displacement
is a maximum which can be obtained easily by a SHAKE analysis. If a
dynamic Winkler model is used, the time histories of ground motions at
the locations of the py curves can be used as input and kinematic inter-
action and ground motion effects are included. The residual problems with
reliability are now conned to the constitutive model itself.
Analyses of centrifuge tests using the dynamic py model suggest that
the API curves are too stiff for the analysis of strong shaking. The pre-
diction of maximum moment is improved by using the Terzaghi subgrade
moduli in the equations for the py curves, rather than the API moduli
but the general distribution of moments along the pile is still not well pre-
dicted. The MCEER/ACT recommendations for the design of bridge foun-
dations support the view that the use of the Terzaghi subgrade moduli for
analysis is the more appropriate of the two options currently in use. Lim-
ited centrifuge test data suggests that by assuming that the modulus of sub-
grade reaction at any depth is proportional to the soil modulus at the same
depth signicantly improves prediction of response.

16.2. Pile group analysis

A pile group is not analysed as a group. Single pile analysis is modied by
empirical adjustment factors to give the response of the piles in the group.
There are two different approaches based on the concepts of efciency or
p-multipliers. Efciency is most useful in estimating group displacement. It
may be dened as the ratio of the average load per pile in the group to
the load on a single pile to cause the same displacement. Efciencies have
been established on the basis of static tests for a limited range of condi-
tions and at specied pile displacements to take pile soilpile interaction
into account. However it is well-known that dynamic and static interaction
can be very different. The efciencies used in practice are typically in the
range 0.70.8 but efciencies as low as 0.40.5 have been measured in tests.
The p-multiplier concept also relies on data from static tests. In this
approach, the p-ordinate of the py curve is multiplied by a factor less
than one depending on the position of the pile in the group. Thereafter, the
analysis proceeds as for the single pile.

There is little published data on how well these methods predict seismic
responses of pile groups. API, which does offer recommendations for sin-
gle pile analysis, does not recommend a method for group analysis. It sug-
gests that the designer use two or more appropriate methods with upper
and lower bounds on soil properties. Then, armed with an appreciation of
the uncertainties involved, the designer is advised to use his judgment for
the structural design of the foundation.

16.3. Piles in liqueable soils

If the site liquees, large moments may occur at the boundary between
liqueable and non-liqueable soils due to kinematic interaction. The
moments may be increased further if there is a non-liqueable surface
layer. These moments can occur at considerable depth in contrast to pile
moments in non-liqueable soils. In the latter case, the maximum moments
occur at the pile head or within the top 34 diameters. Evidence of kine-
matic moments at depths of 7 m has been found in eld investigations.
Piles liqueable sites undergo may undergo large displacements. The
level of displacement depends very strongly on the xity of the pile head.
In liqueed sites, the xity is usually controlled by the grade beams that tie
the pile caps together and any frame action from the structure. The effects
of rocking of the structure or the foundation itself should be taken into
account in any analysis of the foundation because lateral displacement and
rocking are coupled. The PILE-3D analysis described in the paper auto-
matically takes rocking into account.
The lateral spreading of soils, when there is a non-liqueable surface
layer, can generate large kinematic moments. There are two approaches to
the calculation of the kinematic moments: displacement analysis or force
analysis. In displacement analyses, the ground displacements are estimated
by empirical methods or by site analyses. There are great uncertainties
associated with these methods and there is as yet no consensus on appro-
priate soil properties for the analyses or how the analyses should be con-
ducted. A determinate force method, which has been calibrated by case
histories in the eld, has been introduced into the latest Japanese codes.
For large displacements, the non-liqueable surface layer is assumed to
apply passive pressure to the pile foundation and the liqueable material is
assumed to apply 30% of overburden pressure. The forces can be ignored
for quite small displacements. The dependence of the force magnitude on
displacement amplitude has been conrmed in centrifuge tests.
The pile cap stiffnesses are also affected by the development of high
PWP or liquefaction. Effective stress analyses to determine time histories
of pile stiffnesses are described and examples of the effects of high PWP
and liquefaction on stiffness are given.

16.4. Pile cap stiffnesses

The determination of pile cap stiffnesses was investigated in considerable
detail in the context of pile foundations for bridges. Time histories of
pile stiffnesses were determined by a simplied 3-D non-linear response.
The response of a 3-span bridge to seismic excitation, including the
time-dependent stiffnesses, was calculated to provide benchmark data for
checking approximate methods for representing the stiffnesses by single
value springs for use in mainstream structural analysis programs. It was
found that, if inertial interaction effects on pile foundation response are
not important, single value estimates of stiffnesses equivalent to the time
histories can be calculated by using as moduli in the stiffness analyses the
distribution of effective moduli from a free eld SHAKE analysis. The
inuence of inertial interaction on pile foundation response depends on
the relative stiffness of the pile foundation and the supports of the super-
structure. The period shift from the xed based period may be used as a
criterion for the inuence of inertial interaction. A chart is provided for
estimating the period shift.
The effects of foundation exibility on structural response are shown
by acceleration and displacement spectra of the example bridge for differ-
ent assumptions of site conditions when estimating pile cap stiffnesses:
xed base, stiffnesses using initial small strain moduli of the site, time his-
tories of stiffnesses from non-linear analyses and stiffnesses from moduli
obtained from free eld SHAKE analyses. Almost identical acceleration
and displacement spectra were obtained using time histories of stiffnesses
from non-linear PILE 3-D analyses and effective moduli from a free eld
SHAKE analysis, when inertial interaction was not important for pile foun-
dation response. Even using stiffnesses based on initial site moduli with no
recognition of non-linear effects was a big improvement over the xed base

16.5. Moving forward

There is no simple alternative to the py constitutive model and, despite its
limitations and uncertainties, it will continue to be used. The model could
be improved. It seems that by assuming that the modulus of subgrade
reaction at any depth is proportional to the soil modulus at the same depth
improves prediction of response considerably. The additional deciencies
introduced into pseudo-static analysis by neglecting kinematic interaction
and the effects of ground shaking can be avoided by using a dynamic ver-
sion of the Winkler model.
It is difcult to make any quantitative judgment about the methods of
analysis for a pile group based on the response of a single pile from a

py analysis. The correlation factors used to develop the group response

were all determined by static tests and for a prescribed displacement of
the pile cap. In the use of pseudo-static methods for group response, the
advice given to designers in the API recommendations reects the prob-
lems: use two or more methods of analysis, get a measure of the uncertain-
ties involved and the use judgment.
If the engineer is to use judgment, he needs a good general understand-
ing of the characteristics of pile foundation response to earthquake load-
ing. Very good information is now available to inform his judgment from
centrifuge tests, shake table tests, eld tests and theoretical parametric stud-
ies. In addition for important projects, consideration should be given to
nal conrmatory analyses using more sophisticated methods.

The nancial and material support of Anabuki Komutem is gratefully
acknowledged. The major cited contributions of former PhD. Students
W.B. Gohl, G. Wu and T. Thavaraj are deeply appreciated. The help of
Noboru Fujita, Research Assistant, was invaluable in the analyses of the
CIDH piles and in putting this paper together. The support and patience
of my wife, Tomris, made it all possible.

AASHTO (1983) Guide Specications for Seismic Design of Highway Bridges, American
Association of State Highway and Transportation Ofcials, Washington, DC, USA.
Abghari, A. and Chai, J. (1995) Modeling of soilpile superstructure interaction for bridge
foundations, Performance of Deep Foundations under Seismic Loading. ASCE Geotech-
nical Special Publication No. 51, pp. 4559.
API (1993) Recommended Practice for Planning, Designing, and Constructing Fixed Offshore
Platforms, American Petroleum Institute.
ATC-3 (1978) Tentative Provisions for the Development of Seismic Regulations for Buildings
(ATC-306), The Applied Technology Council, Redwood, California.
ATC (1996) Improved Seismic Design Criteria for California Bridges: Provisional Recommen-
dations, ATC 32, Applied Technology Council, Redwood City, California.
Bardet, J.P., Tobita, T., Mace, N. and Hu, J. (2002) Regional modeling of liquefaction
induced ground deformation. Earthquake Spectra 18(1), 1946.
Barton, Y.O. (1982) Laterally loaded model piles in sand: Centrifuge tests and nite element
analyses. PhD. Thesis, University of Cambridge, UK.
Brandenburg S.J., Singh, P., Boulanger, R.W, Kutter. B.L. (2001) Behavior of piles in laterally
spreading ground during earthquakes. CDROM Proc. 6th CALTRANS Seismic Research
Workshop, Sacramento, CA.
Brandenburg, S.J., Boulanger, R.W., Chang, D. and Kutter, B.L. (2005) Mechanisms of load
transfer between pile groups and laterally spreading non-liqueed crust layers, Proceed-
ings, International Symposium on Earthquake Engineering Commemorating 10th Anniver-
sary of the 1995 Kobe Earthquake (ISEE Kobe 2005), Abstract Volume, p 59.

Brown, D.A. and Bollman, H.T. (1996) Lateral behavior of a pile group in sand. Journal of
Geotechnical Engineering, ASCE 114(11), 12611276.
Byrne, P.M. (1991) A cyclic shear-volume coupling and pore pressure model for sand. In:
Proc. 2nd Int. Conf. on Recent Advances in Geotechnical Earthquake Eng. and Soil Dynam-
ics, St. Louis, Report 1.24, Vol. 1, pp. 4756.
Chang, Y.L. (1937) Discussion of lateral pile loading tests. Transactions of ASCE 102, 272
Dobry, R. and Abdoun, T. (2001) Recent studies of centrifuge modeling of liquefaction and
its effect on deep foundations. In: S. Prakash (ed.), 4th Int. Conf. on Recent Advances in
Geotechnical Earthquake Engineering and Soil Dynamics, pp. 2631, San Diego CA.
Dobry, R., Taboada, V. and Liu, L. (1995) Centrifuge modeling of liquefaction effects dur-
ing earthquakes. In: K. Ishihara (ed.), Proc. 1st Int. Conf. on Earthquake Engineering, Vol.
3, pp. 12911324, Tokyo, Japan.
ECN (2003) prEN 19985:2003, Eurocode 8: Design of Structures for Earthquake Resistance,
Part 5: Foundations, Retaining Structures and Geotechnical Aspects, European Commission
for Standardisation.
Feagin, L.W. (1937) Lateral pile loading tests. Transactions of ACSE 102.
Finn, W.D. Liam (1999) Lessons from recent earthquakes on foundation performance, design
and analysis, In: Proc., Inaugural Anabuki Chair Symposium, Faculty of Engineering, Kag-
awa University, FEKU TN009.
Finn, W.D. Liam, Lee, K.W. and Martin, G.R. (1977) An effective stress model for lique-
faction. Journal of the Geotechnical Engineering Division, ASCE, Proc. Paper, 13008, 103,
Finn, W.D. Liam and Gohl, W.H. (1987) Centrifuge model studies of piles under simulated
earthquake loading. In: Proceedings of Symposium on Dynamic Response of Pile Founda-
tions: Experiment, Observation and Analysis, ASCE Special Technical Publication.
Finn, W.D. Liam and Yogendrakumar, M. (1988) Tara-3FL: computer program for the anal-
ysis of large strain ow deformations, Report, Soil Mechanics Group, Department of Civil
Engineering, University of British Columbia, Vancouver, BC, Canada
Finn, W.D. Liam, Thavaraj, T., Wilson, D.W., Boulanger, R.W. and Kutter, B. (1999) Seismic
analysis of piles and pile groups in liqueable sand. In Proceedings, 7th International Sym-
posium on Numerical Models in Geomechanics, NUMOG VI, Graz, pp. 287292, Austria,
Finn, W.D. Liam and Thavaraj, T. (2001) Deep foundations in liqueable soils: case histo-
ries, centrifuge tests and methods of analysis. In CD-ROM Proceedings, 4th Int. Conf. on
Recent Advances in Geotechnical Earthquake Engineering and Soil Dynamics, San Diego,
CA, March 2631.
Fleming, W.G.K., Weltman, A.J. and Randolph, M.F. (1992) Piling Engineering, second edi-
tion. Blackie and Son, Glasgow, UK.
Gazetas, G. and Dobry, R. (1984) Horizontal response of piles in layered soils. Journal of
Geotechnology Engineering, ASCE, 110(1), pp. 2040.
Gazetas, G., Fan, K. and Kaynia, A. (1993) Dynamic response of pile groups with different
congurations. Soil Dynamics and Earthquake Engineering 12: 239257.
Gazioglu, S.M. and ONeill, M.W. (1984) An evaluation of py relationships in
cohesive soils. In: J.R. Meyer (ed.), Proceedings of the ASCE Symposium on Analysis and
Design of Pile Foundations, ASCE National Convention, San Francisco, California, Oct.
15, 1984.
Gohl, W.B. (1991) Response of pile foundations to simulated earthquake loading: experi-
mental and analytical results. PhD. Thesis, Department of Civil Engineering, University
of British Columbia, Vancouver, Canada.

Hamada, M., Yasuda, S., Isoyama, R. and Emoto, K. (1986) Study on Liquefaction Induced
Permanent Ground Displacements. Association for Development of Earthquake.
Liu, L. and Dobry, R. (1995) Effect of liquefaction on lateral response of piles by centri-
fuge tests. National Centre for Earthquake Engineering Research (NCEER) Bulletin, 9(1),
pp. 711.
JRA (1996) Specications for Highway Bridges, Part V, Seismic Design. Japan Road Associ-
JWWA (1997) Seismic Design and Construction Guidelines for Water Supply Facilities. Japan
Water Works Association, Tokyo.
Imbsen, R.A. and Penzien, J. (1986) Evaluation of Energy Absorption Characteristics of High-
way Bridges under Seismic Conditions, Vol. I. Report No. EERC-1984/17, Earthquake
Engineering Research Center, University of California, Berkeley, CA.
Kagawa, T. and Kraft, L. (1980) Seismic py response of exible piles. Journal of Geotech-
nical Engineering, ASCE, 106, GT8, 899919.
Kimura, M., Kosa, K. and Morita, Y. (1994) Full-scale failure tests on laterally loaded
group piles. In: Proc., 3rd Int. Conf. on Deep Foundation Practice incorporating PILE-
TALK International 94, Singapore, pp. 147154.
Lam, I and Martin, G.R. (1986) Seismic design of highway bridge foundations, Vol. II, Design
procedures and guidelines. Report No. FHWA/RD-86/102, Federal Highway Administra-
tion, Virginia.
Liu, l. and Dobry, R. (1995) Effect of liquefaction on lateral response of piles by
centrifuge tests. National Centre for Earthquake Engineering Research. (NCEER) Bulletin,
9(1), 711.
Martin, G.R. (2004) The seismic design of bridges geotechnical and foundation design
issues. In: Proc., ASCE Specialty Conference on Geotechnical Engineering for Transporta-
tion Projects, Vol. 1 pp. 137166, Los Angeles, July, ASCE Geotechnical Special Publica-
tion, No. 126.
Martin, G.R., W.D. Liam Finn and H.B. Seed (1975) Fundamentals of liquefaction under
cyclic loading. Journal of Geotechnical Engineering Division, ASCE 101 (GT5) 423438.
Masing, G. (1926) Eigenspannungen und verfestigung beim messing. In: Proc., 2nd Int. Con-
gress of Applied Mechanics, Zurich, Switzerland.
Matsui, T. and Oda, K. (1996) Foundation damage of structures. Special Issue of Soils and
Foundations, pp. 189200, Japanese Geotechnical Society, January.
Matlock, H. (1970) Correlations for design of laterally loaded piles in soft clay. Proc., 2nd
Annual Offshore Technology Conf, Houston, Tex. pp. 577594.
Matlock, H., Foo, S.H.C. and Bryant, L.M. (1978) Simulation of lateral pile behavior under
earthquake loading. In Proc., Earthquake Engineering and Soil Dynamics, ASCE Specialty
Conference, pp. 601619, Pasadena, California.
MCEER/ATC (2003) Recommended LFRD Guidelines for the Seismic Design of High-
way Bridges, Part I: Specications, Part I: Commentary and Appendices. MCEER/ATC49.
MCEER Report No. MCEER-03SP03, University of Buffalo, Buffalo, NY.
Murchison, J.M. and ONeill, M.W. (1984) An evaluation of py relationships in cohe-
sionless soils, In J.M. Meyer (ed.), Proceedings of the ASCE Symposium on Analysis and
Design of Pile Foundations, pp. 174191, ASCE National Convention, San Francisco, Cal-
ifornia, Oct 15.
ONeill M.W. and Murchison, J.M. (1983) An Evaluation of py relationships in Sands. A
report to the American Petroleum Institute, PRAC 82-41-1.
Pinto, P., McVay, M. and Lai, P. (1997) Centrifuge testing of plumb and battered pile groups
in sand. In Proc., Transportation Research Board, 17th Annual Meeting, pp. 117, Wash-
ington, Disrict of Columbia, Paper No. 551.

PMB Engineering (1988) Par Pile Analysis Program, PMB Engineering, San Francisco, Cal-
Poulos, H.G. (1971) Behaviour of laterally loaded piles: I-single piles and II- pile groups.
Journal of Soil Mechanics and Foundation Divsion, ASCE, 97(SM5), 711731, 733751.
Ramos, R., Abdoun, T. and Dobry, R. (1999) Centrifuge modeling effects of super-structure
stiffness on pile bending moments due to lateral spreading. In Proc., ASCE Symposium
on Analysis and Design of Lifeline Facilities and Countermeasures Against Liquefaction, pp.
Randolph, M.F. (1981) The response of piles to lateral loading. Geotechnique 31(2): 247259.
Reese, L.C., Cox, W.R. and Koop, F.D. (1974) Analysis of laterally loaded piles in sand.
Proc., 6th Annual Offshore Technology Conf, Houston, Tex., Paper No. 80.
Rollins, Kyle M., Peterson, Kris T. and Weaver, Thomas J. (1998) Lateral load behavior of
full scale pile group in clay. Journal of Geotechnology and Geoenvironmental Engineering,
ASCE 124(6), 468478.
Shapiro, D., Rojahn, C., Reavely, L.D., Smith, J.R. and Morelli, U. (2000) NEHRP Guide-
lines and Commentary for Seismic Rehabilitation of Buildings. Earthquake Spectra 16(1),
pp. 227240.
Schnabel, P.B., Lysmer, J. and Seed, H.B. (1972) SHAKE: A Computer Program for Earth-
quake Response Analysis of Horizontally Layered Sites. Report No. EERC72-12, Earth-
quake Engineering Research Center, University of California, Berkeley, CA.
Seed, H.B. and Idriss, I.M. (1970) Soil Moduli and Damping Factors for Dynamic Response
Analysis. Report #EERC70-10, Earthquake Engineering Research Center, Berkeley, CA.
Terzaghi, K. (1955) Evaluation of coefcient of subgrade reaction, Geotechnique 5(4), 297326.
Thavaraj, T. (2001) Seismic analysis of pile foundations for bridges. PhD. Thesis, University
of British Columbia, Vancouver, Canada.
Veletsos, A.S., Prasad, A.M. and Tang, Y. (1988) Design approaches for soilstructure inter-
action. In Proc. 9th World Conference on Earthquake Engineering, Tokyo.
Veletsos, A.S. and Wei, Y.T. (1971) Lateral and rocking vibration of footings. Journal of Soil
Mechanics and Foundation Division, ASCE 97(SM9), 12271248.
Ventura, C.E., Felber A.J. and Prion H.G.L. (1994a) Dynamic Evaluation of a Medium
Span Bridge by Modal Testing. In Proceedings of the 4th International Conference on
Short and Medium Span Bridges, pp. 545556, Halifax, Nova Scotia, August 811.
Ventura C.E., Felber A.J. and Stiemer S.F. (1994b) Dynamic Characteristics of Bridges by
Experimental Investigations of Ambient Vibrations Queensborough Bridge. In Proceed-
ings, 5th US National Conference on Earthquake Engineering, Vol. II, pp. 733742, Chi-
cago, Illinois, July.
Wallace, J.W., Moehle, J.P. and Martinez-Cruzado, J. (1990) Implications for the design of
shear wall buildings using data from recent earthquakes. In Proc. 4th U.S. National Con-
ference on Earthquake Engineering, Palm Springs, CA.
Weaver, T.J., Ashford, S.A. and Rollins, K.M. (2001) Development of py curves for a 0.6 m
diameter CISS pile in liqueed sand, CDROM Proc. 6th CALTRANS Seismic Research
Workshop, Sacramento, CA.
Wilson, D.W. (1998) Soil-pile superstructure interaction in liquefying sand and soft clay.
PhD. Thesis, Department of Civil and Environmental Engineering, University of Califor-
nia, Berkeley. CA.
Wilson, D.W., Boulanger, R.W., Kutter, B.L. and Abghari, A. (1995) Dynamic centrifuge
tests of pile supported structures in liqueable sand. In Proc National Seismic Conf. on
Bridges and Highways. Sponsored by Federal Highways Admin. and CALTRANS, San
Diego, CA.

Wilson, D.W., Boulanger, R.W. and Kutter, B.L. (2000) Observed lateral resistance of lique-
fying sand. Journal of Geotechnology and Geoenvironmental Engineering, ASCE, 126(10).
pp. 898906.
Wu, G. and Finn, W.D. Liam (1997a) Dynamic elastic analysis of pile foundations using
the nite element method in the frequency domain. Canadian Geotechnical Journal 34(1),
Wu, G. and Finn, W.D. Liam (1997b) Dynamic nonlinear analysis of pile foundations using
the nite element method in the time domain. Canadian Geotechnical Journal 34(1), 4452.
Yoshida, N, and Yoshida, H. (1986) TARA-3: A computer program to compute the response
of 2-D embankments and soil-structure interaction systems to seismic loading. Report,
Soil Mechanics Group, Department of Civil Engineering, University of British Columbia,
Vancouver, BC Canada.
Youd, T.L., Hansen, C.M. and Bartlett, S.F. (1999) Revised MLR equations for predicting
lateral spread displacement. In Proc., 7th U.S.-Japan Workshop on Earthquake Resistant
Design of Lifeline Facilities and Countermeasures against Liquefaction, Technical Report
MCEER-990019, MCEER, University of Buffalo, Buffalo, NY.