Computers and Geotechnics 36 (2009) 6–19

Contents lists available at ScienceDirect

Computers and Geotechnics

journal homepage: www.elsevier.com/locate/compgeo

Axisymmetric finite element analysis of pile loading tests

I. Said 1, V. De Gennaro *, R. Frank
Université Paris-Est, Navier, Ecole des Ponts – CERMES 6-8 Avenue Blaise Pascal, Cité Descartes, Champs-sur-Marne, 77455 Marne-la-Vallée, Cedex 2, France

a r t i c l e i n f o a b s t r a c t

Article history: In this paper a typical soil–structure interaction problem is considered, the case of a vertical pile installed
Received 2 December 2007 in sand and submitted to an axial compression loading. Results from two full scale pile tests are analysed
Received in revised form 18 February 2008 and the tests are reproduced by numerical simulations via finite elements method (FEM). The choice of
Accepted 20 February 2008
the mechanical parameters for the soil and the sand–pile interface and the modelling approach are first
Available online 25 April 2008
described. A new numerical strategy is outlined to account for pile installation effects due to jacking and
driving via FEM. The proposed approach is based on the application of existing empirical correlations
available for the quantification of residual radial and shear stresses along the pile shaft as well as residual
Keywords:
Interface behaviour
pressures around the pile base after the installation. This approach is proposed as an alternative to more
Soil–structure interaction complex methods based on the numerical modelling of the pile penetration problem. The role of the con-
Piles stitutive modelling of the interface is also discussed. Finally, comparative analyses of pile loading tests
Finite elements using FEM are provided and the comparisons between numerical and experimental results are presented
Installation effects and discussed.
Elastoplasticity Ó 2008 Elsevier Ltd. All rights reserved.

1. Introduction embedded in various soils, they are not sufficient to assess the asso-
ciated deformation patterns. For this reason, FEM numerical model-
Nowadays many geotechnical structures such as soil reinforce- ling is often adopted to get a deeper understanding of the soil
ments and piles are used to avoid failure or excessive deformations movement, the pile behaviour and especially the mechanical behav-
of constructions. In this work emphasis is given on the behaviour iour of the soil–pile system (et.g. [29,28,4,31,38,14,21,10,39]).
of deep foundations and, more particularly, of the contact between However, numerical simulation of test on piles often leads to er-
a granular soil and a pile. During installation and under service rors and inaccuracies. This is due essentially to the difficulty of taking
loads, piles generate in the surrounding soil stresses and strains, account of installation effects and reproduce soil structure interface
by means of two main mechanisms: the mobilisation of base resis- behaviour. These two problems will be analysed in this paper.
tance and lateral friction. The relative importance shared between The results presented herein are obtained from 2D axisymmet-
these two components of pile behaviour depends, among other fac- ric numerical analyses via finite elements of two full scale tests on
tors, on the type of applied loads (i.e. traction, compression, lateral, piles jacked into sand and subjected to an axial compression load
etc.), the pile type and dimension, the type of soil, the installation [8,7]. The aim of the calculations was to simulate the mechanical
procedure. In this typical soil–structure interaction problem avail- behaviour of the pile considering explicitly: (i) the soils and the
able analyses of the mechanical behaviour of single piles submitted interface geotechnical characterization, and (ii) the installation ef-
to axial loads have shown that the soil–pile interface exerts signif- fects in terms of stress changes within the interface layer and the
icant influence in defining structural stability conditions. surrounding soils mass. The choice of the mechanical parameters
The most fundamental aspects of pile analyses still rely on for the soil and the interface, as well as the modelling strategy
empirical correlations based on experimental observations from are highlighted. The available experimental data are used to repro-
laboratory and full scale in situ testing (e.g. [34]). In both cases, duce the initial state of stress in the soil mass due to jacking. This
the investigations have been often carried out using instrumented simulation, although not representative of the overall installation
piles, allowing for a direct quantification of the base pressure and process, is able to encompass some of the aspects related to instal-
the shaft friction. Although proposed empirical correlations allow lation effects.
to roughly quantifying the expected bearing capacities of piles
2. Interface modelling: MEPI model
* Corresponding author. Tel.: +33 1 64 15 35 52; fax: +33 1 64 15 35 62.
E-mail address: vincenzo.degennaro@enpc.fr (V. De Gennaro). The numerical simulation of pile load tests uses the interface
1
Present address: ENIT-URIG (Ecole Nationale d’Ingénieurs de Tunis, Unité de
Recherche en Ingénierie Géotechnique), Tunisia.
model proposed by De Gennaro and Frank [13]. The model, called

0266-352X/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compgeo.2008.02.011
 I. Said et al. / Computers and Geotechnics 36 (2009) 6–19 7

MEPI-2D, is based on a Mohr–Coulomb oriented failure criterion Parameter B defines the rate of dilatancy stabilization at the inter-
including the principal characteristics of sand–structure interface face. When B increases dilatancy stabilization at the zero value
such as mobilised friction hardening–softening, phase transforma- (i.e. ultimate state) is more rapid. Note that if D = 1 in Eq. (4) the
tion state and ultimate state. single flow rule of the original Cam-clay model is obtained. The fol-
The formulation of the model is two-dimensional; it associates lowing phenomena observed experimentally on sand–structure
normal and tangential relative displacements components un and interfaces can be considered using MEPI-2D model:
ut of the incremental relative displacements vector du = (un,ut)T
to normal and shear stress components rn and s of the incremental (1) the presence of an initial compaction (dun > 0) in a test with
stress vector dr = (drn, ds)T. The elastic behaviour of the interface is constant normal stress or a reduction of the stress rn normal
given by: to the interface layer in a test with imposed constant volume
  or constant normal stiffness;
Kn 0
dr ¼ Ke due ; Ke ¼ ð1Þ (2) the presence of the transformation phase from compaction
0 Kt
to dilatancy ( dun < 0) which corresponds to the increase of
where Kn and Kt are the normal and tangential stiffness of the the normal stress in a test with imposed constant volume
interface. or constant normal stiffness;
The yield surface F (assuming compression positive and neglect- (3) the stabilisation of the normal relative displacement un, or
ing cohesion) depends on the hardening function lðupt Þ. It can be the normal stress rn, on a asymptotic value for large relative
written as: tangential displacements at the interface (dun = 0 or drn = 0),
which corresponds to the ultimate state, where constant vol-
F ¼ s  lðupt Þrn ¼ 0 ð2Þ
ume conditions are assumed at failure.
where
up As formulated MEPI-2D model requires eight parameters: kn, kt,
lðupt Þ ¼ lo þ ðlf  lo Þ   t ð3Þ lo, lf, lco, A, B, and po; being rni, and t given by the initial condi-
A p t þ upt
rni
tions. The validation of the proposed approach for interface model-
o

The hardening function lðupt Þ gives the evolution of the mobilised ling has been performed using tests at constant normal stress, tests
friction coefficient l during loading; it is assumed of hyperbolic at constant volume and tests at constant stiffness [12,13].
type. The tangential plastic displacement upt is the hardening
parameter, lf is the friction coefficient at failure, lo defines the elas- 3. Numerical modelling of full scale model pile test
tic region, A is a parameter controlling the rate of the deviatoric
hardening, t is the thickness of the interface layer (usually assumed The first pile test results analysed are those reported by Chow
equal to 10–15 times the average grain diameter, D50), rni is the ini- [8] using the Imperial College Pile (ICP). The steel ICP used during
tial normal stress acting on the interface, and po is a reference pres- the test has a diameter D = 0.102 m, and a length Lp = 7.4 m
sure (1 kPa). Note that only hardening is accounted herein, that is: (Fig. 1a). The pile is close ended with a 60° conical shape. Due to
lf the friction coefficient at failure is equal to lr the friction coeffi- the reduced diameter ICP can be considered as a model pile rather
cient at ultimate state; the basic version of the model includes also than a real pile. The instrumentation is concentrated in four clus-
softening behaviour [13]. ters (Leading, Following, Trailing, Lagging; see Fig. 1a) about 1 me-
The plastic flow rule of MEPI-2D model is an extended form of ter spaced along the shaft, the first cluster being 0.2 m distant from
the stress–dilatancy relation d = M  g given by Roscoe et al. [36] in the pile tip. Each cluster contains: (i) an axial load cell measuring
the formulation of the original Cam-clay model. p
The flow rule has the axial load transmitted through the pile and the average shear
u_
been defined using the dilatancy ratio d ¼ u_ np and considering that stress between two clusters (fs), and (ii) a surface stress transducer,
t
the stress ratio lc at phase transformation (transition from com- allowing the measurement of the total radial stress (rr) and the lo-
paction to dilatancy) in sand interfaces does not coincide with cal shear stress (srz). Further details on ICP instrumentation can be
the value of lr at ultimate state, although for both situations con- found in Bond [3] and Chow [8]. The pile was installed in a rather
dition d = 0 is fulfilled. Lings and Dietz [25] and Dietz and Lings homogeneous formation of dense Flandrain sand (Dr ffi 75%). The
[17] suggest that interface apparatuses compliances could be at Flandrian sand horizon is interrupted by an organic layer 0.6 m
the origin of this behaviour, even if the same behaviour has been thick at 7.6 m depth. The effect of this soft layer has been neglected
observed in sands tested using a triaxial cell [24], where appara- in this study. The water table was at 4 m depth (Fig. 1b).
tuses compliances are less severe.
In the circumstance that lc and lr are distinct characteristic for
3.1. Finite element mesh and boundary conditions
the same value of dilatancy (d = 0) the hypothesis of a single flow
rule (i.e. dilatancy d is a unique function of l) is no longer valid
Numerical simulations were performed using the software CE-
and the dependency of lc on a state parameter of the sand has to
SAR-LCPC [23]. The geometry of the problem, the geotechnical data
be considered (e.g. [11,24]). The extended form of the stress–dilat-
and the finite element mesh used for the simulations of the pile
ancy relations proposed in MEPI-2D model accounts for this situa-
test are shown in Fig. 1b. It is defined for axisymmetric conditions
tion admitting that dilatancy depends on a state parameter, which
(Z being the axis of symmetry, coinciding with the pile axis). The
is represented by the relative plastic tangential displacement upt
limits of the domain investigated are fixed at 30 m in the vertical
(e.g. [12,15]). The following stress–dilatancy relationship is
direction (i.e. more than three times the pile length, Lp) and three
considered:
times the pile length in the radial (horizontal) direction. It consists
u_ pn of 1850 axisymmetric isoparametric eight-node solid elements,
d¼ ¼ ðlco  lÞDðupt Þ ð4Þ
u_ pt including 100 pile elements and 50 interface elements.
The interface layer is 3 mm in width (12 times D50 of the sand)
where lco is the stress ratio at phase transformation state, l is the
and its behaviour is simulated by means of eight-node thin-layer
mobilised friction coefficient and D ¼ Dðupt Þ is given by:
    elements [16]. MEPI-2D constitutive law was assigned to the
1 B rni thin-layer elements during pile loading. The aspect ratio L/t of
Dðupt Þ ¼ cosh upt ð5Þ
t po the interface elements (L/t = 48) was chosen in order to reproduce
8 I. Said et al. / Computers and Geotechnics 36 (2009) 6–19

Fig. 1. Model pile test in Dunkirk: (a) instrumented IC pile (after [8]) (b) geometry and geotechnical data and (c) mesh and boundary conditions.

correctly the interface behaviour and avoid numerical ill-condi- rived from the results of the interface tests performed by Kuwano
tioning [33,22,12]. Horizontal displacements are set to zero on [27]. Tests were conducted on sand–steel interface in a
the external vertical boundary, whereas on the bottom limit only 60 mm  60 mm modified direct shear box under constant normal
the vertical displacements are set to zero. stress level ranging between 40 and 300 kPa and with a shear rate
of 0.5 mm/min. A dense sample of Flandrian sand and a steel plate
3.2. Constitutive modelling of sand and sand–steel interface of roughness R = 10 lm (maximum notch depth) were used. Fig. 2
shows typical experimental results of constant normal stress inter-
The Flandrian sand from Dunkirk is supposed elastic perfectly face test (rn = 150 kPa) in terms of interface shear stress vs relative
plastic, obeying the Mohr–Coulomb failure criterion. The mechan- tangential displacement (Fig. 2a) and normal relative displacement
ical parameters were obtained from data by Kuwano [27], quoted vs relative tangential displacement (Fig. 2b). The graphs show a
in Chow [8], who performed triaxial tests and interface direct shear peak shear stress reached after a horizontal displacement of less
tests on this sand. The elastic parameters are Es = 200 MPa and than 1 mm. The evolution of the normal displacement shows a
ms = 0.3. Values of Es were found considering in situ measurements small initial compaction followed by dilatancy and constant vol-
by means of CPT correlations [8]. Besides an initial increase in the ume conditions corresponding to large tangential displacements
first two meters, these values showed an increase with depth of (ultimate state). The set of MEPI-2D model parameters was deter-
about 0.9 MPa/m. For this reason the values of Es were considered mined from the interface test results plotted in Fig. 2. Numerical
almost constant all over the pile length (7 m). The internal friction simulations of the interface test are also plotted in the same figure
angle is /0 = 37° and the dilatancy angle is w0 = 10°. Linear elasticity (bold lines). The parameters are summarized in Table 1. Due to the
was considered for the ICP, with Ep = 195 GPa and mp = 0.28 (steel lack of experimental data the normal stiffness of the interface Kn
pile). was assumed equal to 2Kt. The comparison between experimental
The interface behaviour between Dunkirk sand and ICP is simu- results and simulations is satisfactory with regards to the evolution
lated using MEPI-2D model. The parameters of the model were de- of the shear stress s and the normal displacement un versus the

Fig. 2. Comparison of model prediction and experimental results: interface test at constant normal stress (rn = 150 kPa) carried out on a modified direct shear box (Dense
Dunkirk sand, experimental data after [27]).
 I. Said et al. / Computers and Geotechnics 36 (2009) 6–19 9

Table 1 et al. [26], that gives the final value of the radial stress along the
Constitutive parameters for steel-Dunkirk sand interface (rn = 150 kPa) pile shaft after the equalisation period, it reads:
Kn (kPa m1) Kt (kPa m1) lo lf lco A B t (mm)
 0 0:13  0:38
501612 250806 0.017 0.51 0.25 0.00008 0.065 3 rvo h
r0rc ¼ 0:016qc ð6Þ
pa R

where h is the vertical distance from the pile tip, R is the pile radius,
tangential displacement ut at the interface (Fig. 2). Note that the r0vo the effective vertical stress, r0rc the radial effective stress at the
softening phase observed experimentally is not captured by the end of installation and qc is the net cone resistance from CPT and
numerical simulation since the implemented version of the model pa is a reference pressure equal to 1 kPa. The applicability of Eq.
does not include strain-softening. (6) for jacked and driven piles has been verified on the basis of a
As already mentioned MEPI-2D model was associated to the quite large number of high quality pile tests (e.g. [34,26]) and could
eight-node thin-layer elements used to simulate the interface be considered well representative of the distribution that one might
layer. This was done following the approach proposed by Frank get around a pile installed in sand.
et al. [19] and further developed by Sharma and Desai [37]. In this The radial effective stress r0rc was imposed numerically as an
approach the two stress components s and rn acting at the inter- external load on the interface layer considering a linearization of
face are calculated in each integration point of the thin-layer ele- stresses in various horizontal soil horizons (Fig. 3b). At the end of
ment as a projection on the directions perpendicular and parallel this calculation the radial effective stress values after pile installa-
to the interface layer of the general stress tensor acting on that tion matched the measured ones, allowing the reproduction of the
point. In other words stresses are ‘‘oriented” in the direction of final effect of jacking on the radial confinement of the pile. The dis-
the interface and the interface constitutive law links the two com- tribution of radial stress against depth imposed during the numer-
ponents of stress rn and s, perpendicular and parallel to the inter- ical simulations is compared satisfactorily to the experimental
face layer. During the initialisation of the state of stress in the soil values in Fig. 4a.
(geostatic stresses and stresses induced by installation effects), the The distribution of the negative residual friction imposed along
interface layer was simulated assuming Mohr–Coulomb oriented the pile shaft is shown in Fig. 4b. The shape of the curve can be
failure criterion and non-associated plasticity. MEPI-2D model approximated by a bilinear fitting as suggested by Alawneh and
was used to simulate the interface behaviour during pile loading Malkawi [1], with the point with zero residual friction close to
test. the pile tip. Note that this fitting is only valid for long or flexible
piles. The values of residual friction sres were imposed numerically
3.3. Calculation phases along the pile shaft (Fig. 3b), simultaneously with the application
of the radial stress after installation given in (6). This initialisation
In order to simulate the overall behaviour of the pile prior and led to the final state of stress r2o;i (Fig. 3b).
during the loading test three phases were identified and repro- Finally, the residual pressure developed below the pile tip
duced numerically using FEM. These phases are summarised in during installation was considered (r2o;p , Fig. 3b). The value of
Fig. 3, they correspond to the geostatic initialisation of the state qp-res results from the equilibrium of the pile at the end of jack-
of stress within the soils mass (Fig. 3a), the simulation of the instal- ing considering that the action of the residual base pressures
lation effects (Fig. 3b) and, finally, the simulation of pile loading (upward) balances out the overall negative residual shaft friction
test (Fig. 3c). (downward) and the pile weight. This change in base load after
installation was considered during the numerical analyses
3.3.1. Geostatic initialisation imposing at the pile base the pressure qp-res ffi 6300 kPa mea-
To reproduce the initial conditions in terms of effective stresses sured at the end of the equalisation period [8]. Alternatively,
in the soil mass before loading (step 1, Fig. 3a), sand weights c in the absence of direct measurements, the value qp-res can be
above and below the water table (4 m depth) were activated (geo- determined adopting available empirical formulations (e.g.
static initialisation) using the geotechnical data of Fig. 1. The initial [5,1]). It is to be noted that during step 2, the pile elements
horizontal stress was taken into account numerically considering are not yet activated.
0
the K0 condition given by K 0 ¼ ð1  sin u0 ÞOCRsin u [30]. As indi-
cated by Chow [8], in Dunkirk site Flandrian sand was normally 3.3.3. Loading test
consolidated (OCR = 1). The state of stress within the soil mass at Step 3 consisted in the simulation of the static loading test.
the end of this phase is designated as r1o;s , while r1o;i and r1o;p repre- Compression loads were applied following a variant of the LCPC
sent the state of stress within the interface and below the pile tip, procedure given by Bustamante [6]. On the site loads were applied
respectively. by successive increments of 5–10% of the expected maximum load,
and maintained for 10–20 min. Failure was reached when the rate
3.3.2. Installation effects of pile movement increased significantly with time. The failure ax-
On the site, pile installation was performed ‘‘step by step”, using ial load was 309 kN, which corresponds to a pile head displace-
a hydraulic jack, imposing a constant pile head displacement rate ment of 4.2 mm [8]. Numerically, loading is applied by activating
of 600 mm/min. At the end of the installation, an equalisation per- the pile elements (weight and stiffness) and applying displacement
iod of about 15 h was considered, based on the time necessary to increments at the pile head (Step 3, Fig. 3c). During this phase the
stabilise the values of the measured radial effective stress. Three state of stress is modified in the soil mass (r3o;s ), the interface (r3o;i )
important phenomena were observed during jacking and were and the pile tip (r3o;p ).
considered numerically in this numerical step: (i) the increase of
the radial effective stress along the pile shaft, (ii) the mobilisation 3.4. Discussion of results
of a negative residual friction along the pile shaft and (iii) the resid-
ual pressure around the pile base. The total head displacement during compression loading was
The overall increase of the radial effective stress r0rc along the achieved in 80 increments, (during each increment 1.25% of the fi-
pile shaft at the end of the equalisation period was imposed con- nal vertical displacement was applied). With regard to the mobili-
sidering the empirical formula proposed by Chow [8] and Jardine sation of the shear stress along the pile shaft, the results of
10 I. Said et al. / Computers and Geotechnics 36 (2009) 6–19

Fig. 3. Calculation phases for FEM modelling of jacked/driven pile tests: (a) geostatic initialisation, (b) installation effects and (c) loading test (all stresses are effective
stresses).

calculations are compared in Fig. 5 with the experimental data. The predicted initial stiffness of the interface, given by the ini-
Measurements refer to the shear stress s measured locally, in each tial slope of the shear curves in Fig. 5a, is also in good agreement
of the cluster located on the ICP shaft (Fig. 5a) and to the average with the measurements. The initial stiffness depends on the con-
shear stress at the pile shaft sav (Fig. 5b). tractancy and/or dilatancy of the interface layer and, furthermore,
We may notice that the experimental curves start from a on the surrounding soil stiffness. It seems then that the constitu-
residual negative friction due to the relaxation of the soil–pile tive parameters of the interface and the surrounding soil are real-
system after jacking. As mentioned earlier, shear stress changes istic. It is observed that the interface tests revealed strong dilatant
on the pile shaft due to jacking were considered during step 2 behaviour [8], in agreement with the high density of the sand de-
in FEM calculations (Fig. 3). The shape of the mobilised shear posit (Dr = 75%). These data were further corroborated during load-
stress curves appear in good agreement with the measured ones. ing test, where an average increase of the radial stress Dr0r ¼ 60 kPa
The final plateau of the predicted shear stress curves tend to be was measured. As shown in Fig. 6a, the dilatant behaviour of the
overestimated for the Following cluster measurement, and interface is well captured by the numerical simulations. Clearly,
slightly underestimated for the Lagging cluster measurements the increase of the radial stress, normal to the interface layer, illus-
(Fig. 1a). Nonetheless, the average shear stress along the pile shaft trated in Fig. 6a, has a major role in the mobilised shear stress
compares well with the experimental value. The average shear curves presented in Fig. 5a.
stress found at failure is sav = 89 kPa against an experimental va- The prediction of the axial capacity of the pile during the test is
lue of about 92 kPa (Fig. 5b). shown in Fig. 6b. Both the pile head load (Qt) vs the pile head dis-
 I. Said et al. / Computers and Geotechnics 36 (2009) 6–19 11

Fig. 4. (a) Numerical conditions imposed along the pile shaft at the end of the installation: (a) radial effective stress (b) negative residual friction (data after [8]).

Fig. 5. Mobilisation of the shear stress during loading: (a) local values and (b) average value along the pile shaft.

Fig. 6. Numerical predictions vs. experimental results ICP test: (a) radial stress in the four clusters along the pile shaft during loading (experimental data not available) and
(b) load displacement responses during loading (without densified zone below pile base).

placement and the end-bearing load (Qp) are compared with the tween the measured base resistance at failure (92 kN) and the pre-
results from numerical simulation using FEM. Base reaction devel- dicted one (about 71 kN), corresponding to a pile head
ops starting from the value of Qp imposed at the end of installation displacement of about 4 mm. It’s quite likely that the discrepancy
(about 51 kN). However, there appears to be a discrepancy be- between predicted and measured base resistance at failure is the
12 I. Said et al. / Computers and Geotechnics 36 (2009) 6–19

consequence of neglecting a basic phenomenon associated to the

installation by driving/jacking: the densification of the sand
around the pile base. Indeed, this induces an increase in the soil
stiffness which is not taken into account by the numerical simula-
tions of the pile test. The same phenomenon takes place also along
the pile shaft, but it is partially encompassed numerically by
means of the application of the radial stress given in (6). In an at-
tempt to account for the increase of the base stiffness in a simpli-
fied way, the Young’s modulus of the zone below the pile base has
been modified. The extension of this zone, almost entirely com-
pressed, has been early identified by Robinsky and Morrison [35]
and more recently by Allersma and Broere [2], White and Bolton
[40], Dijkstra et al. [18]. It appears that during installation the
movement of the grains is set on a zone that extend below the pile
tip up to a distance of 2 to 3.5 pile diameter D and in the side direc-
tion from 3D to 4D from the pile axis (Fig. 7a). It was then decided
to consider a rectangular zone 3.5D height and 4D large and in-
crease the Young’s modulus of this zone before starting the simu-
lation of the loading test (step 3) in order to account for soil
densification due to installation. To quantify the amount of in-
crease of Young’s modulus, the variation of E with density and con-
solidation stress (rc) established by De Gennaro [12] on Fig. 8. Numerical predictions vs. experimental results ICP test: load displacement
Fontainebleau sand during unloading–reloading phases (Eur) in tri- responses during pile loading with densified zone below the pile base.

axial compression tests were considered (Fig. 7b). The following

relation was assumed between Young’s modulus and consolidation
Figs. 9 and 10 show the distribution of the vertical displace-
stress:
ments and the evolution of the plastic deformation around the pile
Ed ¼ mr0:55
c ð7Þ for various loading steps. A detailed observation of the zone around
the pile shows that numerical calculations predict a bulb formed
In the absence of similar data for Dunkirk sand, Fontainebleau
around the inclusion characterized by vertical displacements di-
curves were considered valid. The maximum densification is sup-
rected downwards. The lateral extension of this bulb decreases
posed to be reached for Dr = 100%. In this case, starting from the re-
gradually around the inclusion during pile loading. At the end of
sults of Fig. 7b, a linear interpolation between the state parameter
the loading phase, the bulb is almost completely localised around
m associated to the current density and the corresponding relative
the pile, in the interface layer, and slightly below the pile base,
density Dr = 100% gave a value of m ffi 20. The modified Young’s
and vertical displacements become negligible elsewhere. Note that
modulus was determined using (7) and considering the value of
for a better estimation of the shape and the extension of this zone
the radial stress measured in situ close to the pile base after instal-
below the pile base a large deformation analysis would have been
lation ðrc ¼ r0rc ¼ 215 kPaÞ. The modified value was Eur = 420 MPa;
more suitable.
compared to the initial value Eo ffi 200 MPa an increase of about
The evolution of the plastic zones (Fig. 10) indicates that the
110% was then considered.
yielding of the interface and the surrounding sand increases for
As shown in Fig. 8 and Table 3 the change in Eur value has a clear
increasing values of the pile settlement (the percentage of the
effect on the numerical prediction of the pile base load Qp and the
maximum plastic deformations reached during the loading is indi-
application of the new elastic parameters clearly improve the
cated for each loading step).
numerical predictions at failure.

Fig. 7. (a) Volumetric strain (compression is negative) during pile penetration in a medium-dense granular material (after [18]) and (b) evolution of Young’s modulus of
Fontainebleau sand according to consolidation stress and density [12].
 I. Said et al. / Computers and Geotechnics 36 (2009) 6–19 13

Fig. 9. Numerical simulation of the evolution of vertical displacements around the ICP during loading.

Fig. 10. Evolution of the plastic deformations around the ICP during loading.

3.5. Influence of the interface model of this model, known as ‘‘oriented criterion” (OC) are given in Table
2. The comparison of the calculations is presented in Fig. 11. As ex-
To highlight the influence of the interface model on the results pected, neglecting the relative displacement between the pile and
of simulations of Dunkirk pile test, two series of calculations were the sand (perfect adherence) the pile settlement at failure is under-
performed. In the first series the interface behaviour is neglected estimated (about 1 mm assuming perfect adherence, against a
(i.e. perfect adherence between the pile and the sand is assumed). measured settlement of 4.3 mm). If the interface behaviour is con-
In the second series perfect plasticity was considered, admitting sidered using the OC model, the experimental load–settlement
Mohr–Coulomb failure criterion and non-associated plastic flow curve is better reproduced, although the values of settlement re-
‘‘oriented” in the direction of the interface [19]. The parameters main on the whole underestimated.
14 I. Said et al. / Computers and Geotechnics 36 (2009) 6–19

Table 2 tained with the OC model are the consequence of the absence of
Constitutive parameters used for the numerical study of ICP test stabilization of the shear stress (Fig. 12). This increase in the shear
Parameter stress is caused by the non stabilization of dilatancy in the inter-
Sand (Mohr–Coulomb) face layer (w = cste). Indeed, the constant dilatancy predicted by
Young‘s modulus Es = 200 Mpa the non-associated Mohr–Coulomb model, constrained by the stiff-
Poisson’s ratio ms = 0.3 ness of the adjacent sand generates a continuous increase in the ra-
Cohesion cs = 0.1 kPa dial stress (Fig. 13b) and consequently in the shear stress. This
Friction angle /0 = 37°
Dilatancy angle w = 10°
situation is not met in the case of the numerical simulations using
MEPI-2D model for the interface because the model considers a
Pile (linear elastic)
Young’s modulus Ep = 195 Gpa
stabilization of dilatancy at large tangential displacements (i.e.
Poisson’s ratio mp = 0.28 D ? 0 when ut ? 1, Eq. (5)).
Interface (oriented criterion) (step 1 and 2: stress initialisation)
The evolution of the average shear stress at the pile shaft
Ei 200 MPa (Fig. 13a) confirms the strong dilatant behaviour given by the ori-
mi 0.3 ented criterion (CO) and shows that simulations with MEPI-2D
ci 0.1 kPa model reproduce more satisfactorily the interface behaviour.
d0i 27°
A synthesis of the various load–settlement curves obtained
wi 10°
varying the behaviour of the interface is shown in Fig. 13b. It is
Interface (MEPI-2D) (step 3: pile loading)
noted that the base resistance is the same for all the simulations.
E* 1500 kPa
m* 0 The load–settlement curve is well reproduced using MEPI-2D mod-
lf = lr 0.51 el for the interface and generally over-estimated adopting the ori-
lco 0.25 ented criterion ‘‘OC”. Also, the shape of the load–settlement curve
lo 0.017 shows that the failure condition represented by a final plateau is
A 0.00008
B 0.065
not fulfilled using OC model. Once again, the effect of a constant
t 3 mm constrained dilatancy is at the origin of this behaviour.

4. Numerical modelling of full scale loading pile test

Table 3 The second set of numerical analyses concern a driven box pile.
Comparison of the numerical and experimental loads at failure
Loading test is described by Bustamante and Gianeselli [7]. A less
Friction load Qs Pile base load Qp Total load Qt detailed geotechnical characterization of the site is available for
(kN) (kN) (kN) this example. This will allow to checking and better appreciate
Experiment 217 92 309 the extent of previous assumptions, in particular concerning the
Numerical 211 89 300 simulation of installation effects.
modelling

4.1. Pile geometry and site characterisation

The pile loading test analysed herein was conducted on a driven

pile installed in Dunkirk [7]. The pile length is Lp = 7.76 m and its
geometry is hexagonal (Table 4, Fig. 14). In order to perform axi-
symmetric modelling in CESAR-LCPC, the hexagonal pile has been
replaced by a cylindrical one, using the equivalence of the lateral
surface to determine the equivalent pile diameter. The lateral sur-
face is then expressed as follows:
Slat ¼ P  Lp ¼ 2pReq Lp ð8Þ

where Slat is the lateral surface of the box pile, P is the perimeter of
the box pile, Lp is the pile length and Req is the equivalent radius
used for the circular pile. Req is found equal to 26 cm. Furthermore,
the equivalent Young’s modulus of the pile is determined in order to
have the same stiffness as the real pile:
ðEAÞpile
Eeq ¼ ð9Þ
pR2eq

where A is the average section of the pile, E is Young’s modulus of

the real steel pile (E = 195 GPa). We obtain Eeq = 16 GPa. Thus the
Fig. 11. Comparison between experimental data and numerical simulations of the circular pile analysed by finite elements is circular, of diameter
load–settlement curve of ICP at Dunkirk, effect of the interface modelling with D = 0.52 m, length Lp = 7.76 m and Young’s modulus E = 16 GPa.
oriented criterion (OC). The pile was instrumented using the LCPC extensometers and
delimiting five measuring sections. Note that, for this test, residual
stresses after pile driving along the shaft and below the pile tip are
A comparison of local shear and normal stresses at the interface not measured. However, their values have been determined using
using OC model and MEPI-2D model allows a better understanding available correlations and applied numerically. This will be shown
of this underestimation (Fig. 12). The mobilization of the local in the section devoted to the numerical analysis and discussion of
shear stress in the 4 ICP clusters shows that the higher values ob- results.
 I. Said et al. / Computers and Geotechnics 36 (2009) 6–19 15

500
200
450
MEPI 2D
MEPI 2D

Normal stress σn (kPa)

160 OC 400
OC
350
Shear stress τ (kPa)

120 300

250
80 200

150
40
100

50
0
0
0 2 4 6 8 0 2 4 6 8
-40 Pile head displacement (mm) Pile head displacement (mm)

Fig. 12. Comparison between simulations using the oriented criterion (OC) and MEPI-2D for interface modelling: (a) local shear stress at clusters and (b) normal stress at
clusters.

500
180
450 QQ (OC)
OC t (OC)
t (CO)
150 Experiment 400
MEPI 2D
350
120
300
Qt (MEPI 2D)
LOAD (kN)

90
τav (kPa)

250

60
200 Qt (Exp)
150
30 Qp(Exp)(rupture)
100

0 50 Qp(Num)
0 2 4 6 8 0
-30 0 2 4 6 8
Pile head displacement, wto (mm) Pile head displacement, wto (mm)

Fig. 13. Comparisons between experimental results and numerical simulations using oriented criterion (OC) and MEPI 2D for the interface: (a) average shear stress against
pile settlement and (b) load–settlement curve.

Table 4
Geometrical characteristics of the LCPC box pile

B H E S Perimeter Weight (kg/ Section steel

(mm) (mm) (mm) (mm) (cm) m) (cm2)
533 385 12.3 342 164 139 177

4.2. Finite element mesh and boundary conditions

The finite element mesh used for the simulations of the pile test
is shown in Fig. 15. The mesh is similar to that used for the ICP
analysis. It consists of 2729 axisymmetric isoparametric eight-
node solid elements, including 128 pile elements and 64 interface
elements. The interface layer is 3 mm in width and its behaviour is
simulated by means of eight-node thin-layer elements. The aspect
ratio L/t of the interface elements is fixed at 40.

4.3. Constitutive modelling of sand and sand–steel interface

Sand from Dunkirk was supposed elastic perfectly plastic,

obeying the Mohr–Coulomb failure criterion. The mechanical Fig. 14. Geometry of the LCPC box pile [7].
16 I. Said et al. / Computers and Geotechnics 36 (2009) 6–19

Layer 1 w to
couche1

Interface
L =7, 76 m
Layer 2
couche2
Pile
p

Layer 3
couche3
Layer 4
Layer 5

u=0 u=0

v=0 R

Fig. 15. LCPC pile test in Dunkirk: (a) geometry and (b) mesh and boundary conditions.

parameters were obtained from pressuremeter tests results. where q0 is the initial vertical stress. An average friction angle value
Young’s modulus is determined from the pressuremeter modulus obtained from (12) and (13) was then considered in each layer.
EM. The analyses of Frank [20], which compares the Ménard Young’s modulus and the friction angle values are given in Table 5.
pressuremeter and the self-boring pressuremeter (PAF) for the Linear elasticity was considered for the LCPC pile (Table 5). The
estimation of the pressuremeter modulus EM were considered. methodology outlined in Fig. 3 was adopted to reproduce the initial
It is suggested that: conditions prior to pile loading. During the initialisation of the state
 of stress in the soil (geostatic stresses and installation effect), the
Gpo 4:1 for sands
¼ ð10Þ interface layer was simulated assuming Mohr–Coulomb oriented
GM 11:3 for clays
failure criterion and non-associated plasticity, like in the case of
where Gpo is the secant shear modulus with PAF corresponding to a the ICP test. During the simulation of the loading test, non-associ-
strain of 0.02% (almost the initial tangent modulus); GM is the shear
modulus with Ménard pressuremeter. The self-boring pressureme- Table 5
ter (PAF) is used to avoid the drawbacks of the Ménard pressureme- Constitutive parameters used for the numerical study of LCPC pile test
ter installation which remoulds the soil. Since with PAF, the
Parameter Layer 1 Layer 2 Layer 3 Layer 4 Layer 5
mechanical characteristics of the soil remain almost intact, it is
Sand (Mohr–Coulomb)
admitted that the initial modulus of PAF is almost equal to Young’s
Young’s modulus Es (MPa) 29 37 53 62 94
modulus of the soil (small deformations). Consequently, the follow- Poisson’s ratio m 0.3 0.3 0.3 0.3 0.3
ing relations between Young’s modulus and the pressuremeter Cohesion: c (kPa) 0.1 0.1 0.1 0.1 0.1
modulus can be written: Friction angle: /0 (°) 35.8 38.7 41.2 40.5 41
Dilatancy angle w (°) 5.8 8.7 11.2 10.5 11

E 4:1 for sands Value
¼ ð11Þ Pile (linear elasic)
EM 11:3 for clays
Young’s modulus Ep (GPa) 16
The friction angle of sand is related to the limit pressure of the Poisson’s ratio mp 0.28

Ménard pressuremeter as follows [32]: Interface (Mohr–Coulomb orienté)

Ei(MPa) 200
p  p0
u0 ¼ 25 þ 5:77 log LM ð12Þ mi 0.3
b ci (kPa) 0.1
d0i ð Þ 27
where pLM is the limit pressure measured with the pressuremeter,
wi (°) 10
p0 is the horizontal total stress before the test (b = 1.8 for wet
Interface (MEPI 2D)
homogenous soils, 3.5 for dry heterogeneous soils and 2.5 for other
E* (kPa) 1500
soils). m* 0
The friction angle was also calculated using the empirical for- lf 0.51
mula proposed by Combarieu [9] for sands: lco 0.25
2 3 lo 0.017
9 41 ln qpl A 0.00008

5
0
sin u ¼ þ 0
ð13Þ B 0.065
8 8 ln pl þ 3 3 t (mm) 3
q0 2
 I. Said et al. / Computers and Geotechnics 36 (2009) 6–19 17

ated elastoplastic behaviour was considered at the interface, based qpres ¼ 533:4Lp b
sffiffiffiffiffiffiffiffi
on the interface model MEPI-2D. Since no interface test results were ð14Þ
KsP
available, the same interface constitutive parameter as for the ICP b¼ ; K s ¼ 188:9ðN side Þ0:27
AEp
load test were considered, assuming the same density for the Dun-
kirk sand and the same rugosity for the LCPC pile. where Ks is the initial slope of the friction curve during the loading
(kPa/cm), P is the pile perimeter (cm), Ep is the pile elastic modulus
4.4. Numerical analysis and discussion of results (kPa), A is the pile section (m2), Lp is the pile length (m), Nside is the
average value of SPT sand blow count around the pile. The second
4.4.1. Initial conditions and installation effects formula [1] considers the pile flexibility:
Initial conditions in terms of geostatic initialisation and pile
installation effects are based on the approach proposed in Fig. 3, qp-res ¼ 13158g0:724 ð15Þ
   
as for the previous simulations of the ICP test. The sand weight L Ap G
and horizontal initial stress (Ko condition) are first accounted g¼ ð16Þ
D A Ep
numerically (Fig. 3a). Pile installation effects are then imposed
numerically following Fig. 3b. where g is the flexibility factor, D is the pile diameter, G is the sand
The residual radial stress was determined along the pile shaft shear modulus, Ap is the cross-section of the pile. The two methods
using Eq. (6). In the absence of direct measurements the residual give an average residual base resistance qp-res = 660 kPa. This value
base pressure qp-res was calculated as an average value between was applied at the pile base in the second step (Fig. 3b). Residual
the two empirical formulations proposed by Briaud and Tucker shear stress sres after driving is calculated considering the equilib-
[5] and Alawneh and Malkawi [1], Briaud and Tucker [5] formula rium between the residual base force, the residual friction and the
writes: pile weight W.

Fig. 16. LCPC pile test in Dunkirk, mobilisation of the shear stress during loading: (a) local values and (b) average shaft friction.

Fig. 17. LCPC pile test in Dunkirk: (a) radial stress in the five sections along the pile shaft during loading and (b) load–settlement curves: numerical predictions vs.
experimental results (after [7]).
18 I. Said et al. / Computers and Geotechnics 36 (2009) 6–19

Aqp-res  W Further developments would be necessary at more fundamental

sres ¼ ¼ 10 kPa ð17Þ
Slat level to be able to reproduce the overall installation process. This
was out of the scope of this work.
The average residual shear stress obtained was applied numerically
as a uniform stress on the interface limits (step 2, Fig. 3b).
References
4.4.2. Loading test
Moving from the initial conditions obtained during the previous [1] Alawneh AS, Malkawi AIH. Estimation of post-driving residual stresses along
driven piles in sand. Geotech Test J 2000;23(3):313–26.
numerical modelling (step 2), compression loading was simulated [2] Allersma HGB, Broere W. Optical analysis of stress around a penetrating probe
in the third step applying a total displacement wto = 50 mm to the in granular material. In: Philips R, Guo P, Popescu R, editors. Physical modelling
pile head (Fig. 3c). The total head displacement was reached in 100 in geotechnics: proceedings of the international conference on physical
modelling in geotechnics. Swets and Zeitlinger; 2002. p. 149–54.
increments, (during each increment 0.5 mm was applied). Note
[3] Bond AJ. Behaviour of displacement piles in overconsolidated clays. PhD
that, as for ICP test, a modified Young’s modulus E ffi 400 MPa in Thesis, Imperial College, London; 1989.
the rectangular sand zone around the pile base has been consid- [4] Bransby MF, Springman SM. 3D finite element modelling of piles groups
ered (Fig. 7a), assuming that the sand below the LCPC pile base adjacent to surcharge loads. Comput Geotech 1996;19:301–24.
[5] Briaud JL, Tucker AM. Piles in sand: a method including residual stresses. J
reaches the same stiffness at maximum density like for ICP test. Geotech Eng 1984;110(11):1666–80.
The results of the calculations of the mobilisation of the shear [6] Bustamante M. The pile loading test. In: Pilot G, editor. Foundation
stress along the pile shaft are compared with the experimental engineering, vol. 1. Presse de l’école Nationale des Ponts et Chaussées; 1982.
p. 263–73.
data in Fig. 16. Simulations refer to the shear stress measured lo- [7] Bustamante M, Gianeselli L. Predicting the bearing capacity of sheet piles
cally, in each of the sections located on the pile shaft (Fig. 16a) under vertical load. In: Proceedings of the 4th international conference on
and to the average shaft friction (Fig. 16b). Comparisons between piling and deep found, Stresa, Italy, 7–12 April; 1991.
[8] Chow FC. Investigations in the behaviour of displacement piles for offshore
experimental data and calculations show that numerical curves foundations, PhD thesis, Imperial College, London; 1997.
starts from non-zero residual values. This is not the case for the [9] Combarieu F. Estimation rapide de l’angle de frottement d’un sol granulaire au
test, as measurements were set to zero at the beginning of loading pressiomètre. Bull des Laborat des Ponts et Chaussées 1996;203:121–3.
[10] Comodromos EM, Anagnostopoulos CT, Georgiadis MK. Numerical assessment
test, withdrawing installation effects in terms of residual values. of axial group response based on load test. Comput Geotech 2003;30:505–15.
The evolution of experimental shaft friction is well reproduced [11] Cubrinovski M, Ishihara K. Modelling of sand behaviour based on state
by the numerical calculations. Like for the ICP test, the increase concept. Soils Found 1998;38(3):115–27.
[12] De Gennaro V. Etude du comportement mécanique des interfaces sol-
of the radial stress normal to the interface layer, illustrated in
structure. Application à l’analyse du comportement des pieux. PhD thesis,
Fig. 17a, has a major role in the mobilised shear stress presented Ecole Nationale des Ponts et Chaussées, Paris, France; 1999. p. 427.
in Fig. 18a. [13] De Gennaro V, Frank F. Elasto-plastic analysis of the interface behaviour
The evolution of the total base resistance Qp and of the total between granular media and structure. Comput Geotech 2002;29:547–72.
[14] De Gennaro V, Frank, F. Insight into the simulation of calibration chamber
load Qt are compared to the experimental results in Fig. 17b. tests. In: Mestat, editor. Proceedings of the European conference on numerical
Although values at failure (large pile head displacements) are well methods in geotechnical engineering (NUMGE 2002), Presses de l’ENPC/LCPC,
reproduced, the initial slopes of both numerical curves are slightly Paris; 2002b. p. 169–77.
[15] De Gennaro V, Frank R. Elastoplastic modelling of the behaviour of the
underestimated. For the base load numerical curve it’s quite likely granular soil–structure interface. In: Frémond M, Maceri F, editors. Novel
that the increase in the overall stiffness during loading, related to approaches in civil engineering. Springer Verlag; 2003. p. 133–58.
the sand densification around the pile base after driving is at the [16] Desai CS, Zaman MM, Lightner JG, Sirirwardane HJ. Thin-layer element for
interfaces and joints. Int J Numer Anal Meth Geomech 1984;8:19–43.
origin of this scatter. Concerning the initial slope of the total load [17] Dietz MS, Lings ML. Post peak strength of interfaces in a stress–dilatancy
numerical curve Qt(Num), the underestimation seems to be related framework. J Geotech Geoenviron Eng, ASCE 2006;132(11).
to the scatter observed on the base load numerical prediction. In [18] Dijkstra J, Broere W, Frits van Tol A. Experimental investigation into the stress
and strain development around a displacement pile; 2006.
fact the local behaviour in terms of mobilized shaft friction www.geo.citg.tudelft.nl/broere/pdf/dijkstra_experimental_pi_dfi2006.pdf.
(Fig. 16) agrees well with in situ measurements. [19] Frank R, Guenot A, Humbert P. Numerical analysis of contacts in geomechnics.
In: Proceedings of the fourth international conference on numerical methods
in geomechanics, Edmonton, May 31–June 4; 1982. p. 37–45.
5. Conclusion [20] Frank R. Recent developments in the prediction of pile behaviour from
pressiometer results. Rapport interne du LCPC; 1985.
[21] Georgiadis K, Potts DM, Zdravkovic L. The influence of partial soil saturation on
In this paper the results of the numerical analysis of load tests pile behaviour. Géotechnique 2003;53(1):11–25.
on instrumented jacked/driven in situ piles installed in sand are [22] Hohberg J-M, Schweiger H. On the penalty behaviour of thin-layer elements.
In: Proceedings of the fourth international symposium on numerical models in
presented.
geometry (NUMOG IV), Swansea, UK; 1992. p. 241–48.
A first attempt to account for installation effects during numer- [23] Humbert P. CESAR-LCPC, un code général de calcul par éléments finis. Bull.
ical simulation via finite elements has been shown. Empirical cor- liaison Labo. Ponts et Chaussées 1989;160:112–6.
relations based on field data have been used to reproduce the [24] Li XS, Dafalias YF. Dilatancy for cohesionless soils. Géoetchnique
2000;50(4):449–60.
change in radial stress around the pile, consecutive to jacking/driv- [25] Lings ML, Dietz MS. An improved direct shear apparatus for sand.
ing. Similar correlations are employed to reproduce the installation Géotechnique 2004;54(4):245–56.
effects in terms of residual shaft friction and base resistance. [26] Jardine R, Chow FC, Overy R, Standing J. ICP design methods for driven piles in
sands and clays, 112. Thomas Telford Ltd; 2005.
The importance of the constitutive modelling of the interface [27] Kuwano R. Laboratory tests on Dunkirk sand. Imperial College Internal Report;
layer has been clearly illustrated, mostly in relation with the mod- 1996.
elling of the volumetric behaviour at the interface, in terms of [28] Mabsout ME, Reese LC, Tassoulas L. Study of pile driving by finite element
method. J Geotech Eng 1995;121(7):535–43.
dilatancy and/or contractancy, using a deviatoric hardening model. [29] Maqtadir A, Desai CS. Three-dimensional analysis of a pile-group foundation.
Friction at the interface layer is well reproduced numerically. Int J Numer Anal Meth Geomech 1986;10:41–58.
Base resistance results are improved considering the change of [30] Mayne PM, Kulhawy FH. K0-OCR relationships in soils. J Geotech Eng Div, ASCE
1982;108(GT6):851–72.
material mechanical properties due to pile installation (i.e. soil [31] Mohamedzein YEA, Mohamed MG, El Sharief AM. Finite element analysis of
densification). Although this approach remains quite crude and short piles in expansive soils. Comput Geotech 1999;24:231–43.
simple, its efficiency looks satisfactorily from a practical point of [32] Muller H. Baugrunduntersuchung mit dem Pressiometerverfahren nach
Menard, die Bautechnik, Heft 9, Berlin 1970.
view. Load–settlement curves are well reproduced by the numeri-
[33] Pande GN, Sharma KG. On joint/interface elements and associated problems of
cal modelling and the results seem promising in view of the numerical ill-conditioning. Int J Numer Anal Meth Geomech 1979;3:
numerical modelling of other tests on full scale in situ piles. 293–300.
 I. Said et al. / Computers and Geotechnics 36 (2009) 6–19 19

[34] Randolph MF. Science and empiricism in pile foundation design. Géotechnique [38] Wakai A, Gose S, Ugai K. 3D elasto-plastic finite element analyses of
2003;53(10):847–75. pile foundations subjected to lateral loading. Soils Found 1999;39(1):
[35] Robinsky EI, Morrison CF. Sand displacement and compaction around model 97–111.
friction piles. Can Geotech J 1964;1(2):81–93. [39] Wehnert M, Vermeer PA. Numerical analyses of load tests on bored piles. Num
[36] Roscoe KH, Schofield AN, Thurairajh A. Yielding of clays in states wetter than Models Geomech (NUMOG IX) 2004:505–11.
critical. Géotechnique 1963;13(3):211–40. [40] White DJ, Bolton MD. Observing friction fatigue on a jacked pile. In centrifuge
[37] Sharma KG, Desai CS. Analysis and implementation of thin-layer element for and constitutive modelling, Rotterdam; 2002. p. 347–54.
interfaces and joints. J Eng Mech, ASCE 1992;118(12):2442–62.