Professional Documents
Culture Documents
1 s2.0 0045794986903457 Main PDF
1 s2.0 0045794986903457 Main PDF
00
Pcrpmon Joumda Ltd.
Y. K. CHOW
Department of Civil Engineering, National University of Singapore, Singapore
Abstract-Two discrete clement models are described for the analysis of settlement of vertical pile groups.
In both models, the pilesare represented by discrete elements with an axial model of deformation, and
the soil b&&our for the individual pilesis representedby load-transfer curves. The essential difference
between the models is the manner in which pilbsoil-pile interaction is represented. The first model treats
the soil as independent horizontal layers which permit interaction between piles to take place within the
same layer only. Thus ignoring the continuity of tbe soil medium. The second utiliaca Mindlin’s solution
which determines interaction effects in a homogeneous, isotropic elastic half-space, but an approximate
procedure is used to account for soil inhomogeneity. Comparisons of both approaches with a rigorous
integral equation method for pile groups in a homogeneous. isotropic elastic soil .indicate t&t the second
method generally gives good agreementwhile the fust method tends to underestimateinteraction,
especially for shorter piles. For pile groups in nonhomogeneous soils, where rigorous theoretical solutions
are geaelally not available, parametric studies comparing the two discrete element models solutions are
presented. The studies performed using both methods on the nonlinear khaviour of an instrumented pile
group compare favourably with the field measurements.
is based on Mindlin’s solution for a point load in in which: I = embedded length of pile; v = soil Pois-
a homogeneous, isotropic elastic half-space. The son’s ratio; and p = inhomogeneity factor = G,,?/Gi
essential difference between the two discrete element (i.e. ratio of soil shear modulus at pile mid-depth to
models is the manner in which interaction effects ate that at the pile base). Assuming a constant shear
determined. The first model assumes that interaction stress in a pile segment of length L associated with a
between piles takes place within each soil layer only. particular node, the following expression may be
Thus soil stratification can be dealt with, but the written for the settlement of the shaft node due to a
continuity of the soil medium is ignored. On the other shaft load, P, acting at the node:
hand, the second model [q maintains the continuity
of the soil through the use of Mindlin’s solution, but
handles the soil inhomogeneity in an approximate (4)
manner. In problems where rigorous solutions are
not available (for example, in layered soil profiles), The soil stiffness may be obtained from the above eqn
parametric studies comparing the solutions from (4).
these two models would be of interest. The nonlinear load-transfer curve is constructed
assuming a hyperbolic variation in the mobilized
shear stress (Kraft er al. [13]). The incremental soil
FORMULATION OF DISCRETE ELEMENT MODELS displacement at the pile shaft may be shown to bc
given by the following expression (Chow [q)
Single piles
in which: t ,, = shear stress at the pilosoil interface; Figure la shows a discrete element model based on
r, = pile radius; G = shear modulus of soil; and r, = an extension of the work of Randolph and Wroth
some empirical distance at which the shear stress in [lo]. The soil consists of independent horizontal
the soil becomes negligible. For a pile embedded in layers which can have different properties. The soil
a half-space in which the soil stiffness is uniform or reaction generated at node n due to the loading of the
increases linearly with depth, Randolph and Wroth pile induces additional settlements at the nodes of
[ 141 suggest adjacent piles only in the same soil layer, k. Thus, the
continuity of the soil medium is discounted. Hence,
r, = 2.5 r/I (1 - v) (3) this model shall be designated as a layer model. A
Discrete element analysis of pile group settlement IS9
_____
i-4___---
___--- in which: {wt} = shaft displacement vector for layer
k; [Fs] = flexibility matrix of order NP x NP for
______
____--
k n ______
_____ --__--Gk layer k; and {P,} = shaft load vector for layer k. The
b b flexibility matrix may be inverted to give the stiffness
______ ___--_ me----
L-k
matrix for the soil in the layer k, i.e. [KfJ = [Ff]-‘.
____
(10)
/-I {PI = [WwI (I%
in which f$ = the displacement influence coefficient in which {P} = external applied load vector.
at the pile shaft denoting the settlement of the shaft With the formulation described above, the stiffness
at pile i due lo a unit load a1 pile j, within the layer matrix [K] is symmetrical and banded with a semi-
k, and it may be obtained from eqn (9) with an bandwidth of NP if the nodes are numbered accord-
appropriate value of r for the pile spacing. Equation ing to Fig. la. This feature may be taken advantage
(10) may be written for each of the NP piles in the of to reduce storage requirements in the computer.
C.&S %il--K
160 Y. K. CHOW
This approach for the layer model is formulated in a NN x NN consisting of displacement influence co-
manner which can cater for generally layered soil efficients; and {is} = soil reaction load vector. The
profiles, through the use of a suitable value of r, in main diagonal elements of the flexibility matrix corre-
eqn (9). In soil where the shear modulus is uniform spond to the displacement influence coefficients for
or increases linearly with depth, r,,, may be obtained the single piles, and under linear elastic conditions
from qn (3). Randolph and Wroth [16] have sug- they may be obtained from qns (4) and (6) for the
gested a suitable expression for r, for piles bearing on pile shaft and the pile base respectively, but for the
a stiffer stratum. However, for a generally layered nonlinear case, qns (5) and (7) may be used and the
soil, r, would need to be obtained from a more problem needs to be solved incrementally. The off-
rigorous numerical method of analysis such as the diagonal or interaction terms are determined based
finite element method. on Mindlin’s solution [I I]. It should be noted that
The extension of the above linear elastic approach the off-diagonal terms associated with nodal points
to include nonlinear soil response needs some within the same pile are zero because of the inherent
elaboration. The general concept will follow closely assumption in the load-transfer method for modelling
the ideas presented in an earlier paper (Chow [A), and soil behaviour in the single pile. Following the earlier
the problem needs to be solved incrementally. Non- work, the continuously distributed stresses at the pile
linearity at the pile+soil interface is generally confined shaft are replaced by quivalent nodal point loads.
to a narrow zone of soil adjacent to the pile shaft. An advantage of using the point load solutions is that
Hence, interaction effects will remain essentially the flexibility matrix is symmetrical for general pile
elastic. Thus, nonlinear behaviour of the group is groups, thus reducing storage requirements in the
dominated by the nonlinear response of the individ- computer. The soil stiffness matrix in the group is
ual piles, and the relevant influence coefficients in the obtained by inverting the flexibility matrix in qn (16)
main diagonals (i.e. fu) of the soil flexibility matrices i.e. [x,] = [F,]-i. This matrix is assembled with the
in qns (11) and (14) may be obtained from qns (5) pile stiffness matrices to give the load-displacement
and (7) respectively at the appropriate load levels. relationship for the pile group system. Thus, in this
The off-diagonal terms remain unchanged until the continuum model the computer storage requirement
ultimate capacities at the nodes are reached, after is much greater than that of the layer model in
which the appropriate terms are set to zero. The which the stiffness matrix of the pile group system
detailed procedure is similar to that described by is banded. The corresponding solution time of the
Chow [7’J continuum model is also longer.
Conrinuwn model. Pile-soil-pile interaction in the The essential features of these two discrete element
continuum model is based on Mindlin’s solution [1 1] models are summarized in Table 1.
for a vertical point load in a homogeneous, isotropic
elastic half-space. Soil inhomogeneity is handled COMPAIUSONS OF DISCRETE ELEMENT MODELS
using an approximate averaging procedure. A de-
tailed description of the approach has been presented Theoretically rigorous solutions are only available
by Chow [I. Hence, a brief account is summarized for pile groups in a homogeneous, isotropic elastic
here. The relationship between the soil settlement at half-space (Poulos [ 11; Butterfield and Banerjee [2);
the pile nodes and the loads acting at the nodes may Poulos and Davis [17]). In soil where the shear
be represented by the following matrix equation: modulus is proportional to depth, only approximate
solutions are available (Banerjee and Davies [3];
{WI = El{P> (16) Banerjee [18]; Poulos [19]). However, in a generally
layered soil, such published solutions are not readily
in which: {w} = soil displacement vector for the NN available. In this section, the discrete element model
nodes in the group; [F,] = flexibility matrix of order solutions are compared with available published
0.6
(I
0.5
s
- 04
! ._ t b
0.1 t *--_
O.?
0.6 - k.200
s
8 0.5 -
5
e 04 -
.s
2
P ::.
0.1 -
I 1
0 lo 20 Jo 90 500 0 20 30 40
solutions. In problems where rigorous or approxi- The dimensionless parameters of interests here
mate solutions are not available, a comparison of the are 1, l/r,,, s/r,, and v, in which: A = EJG = stiffness
computed solutions using these two discrete element ratio; l/r0 = slenderness ratio; and s/r,, = normalized
models is presented. spacing of piles in the group, centre to centre.
The degree of interaction between two equally
Homogeneous isotropic elastic hal/zrpoce loaded, identical piles may be expressed as an inter-
The accuracy of the layer model and the con- action factor, a, defined as the ratio of the additional
tinuum model is compared with the more rigorous settlement induced in the single pile due to load on
integral equation method (Poulos and Davis [ 17))for an adjacent pile, to the settlement of the single pile
pile groups in a homogeneous, isotropic elastic half- under its own load. Figures 2, 3 and 4 compare the
space. It is noted that verification of the continuum interaction factors determined using the three meth-
model was previously presented by this author ods for rigid and compressible piles with different
(Chow [7]). slenderness ratio, I/ro. For the shorter piles (l/r0 = 20
@nthuum modal
0.7
L -20
$0 06-
05. +o
lo
o 0.6 -
0 0 xl 30 40 500 IO m 30 40 50
Fig. 4. Comparison of interaction factors for compressible piles in a homogeneous, isotropic elastic
half-space (A = 300; v = 0.5).
and SO),the layeimodel generally underestimates the Mindlin’s solution is strictly valid for a half-space,
interaction factors. When the piles are rigid, the layer but it may be modified to account for the existence
model solutions improve with an increase in the pile of a rigid stratum at some depth by employing
slenderness ratio. It may be noted that the layer Steinbrenner’s approximation (see for example
model solutions for the rigid piles are identical to the Poulos and Davis (171). However, this was not done
analytical solutions given by Randolph and Wroth since this procedure would destroy the symmetry of
[lo]. However, for compressible piles where the the soil flexibility matrix in eqn (16). While it can be
analytical solutions are approximate similar solutions argued that a hard layer can be expected at some
are not available for comparison. The continuum depth beneath the soil surface, the choice of H/l = 2
model generally agrees well with the integral equation is quite arbitrary. In practice, maximum bore hole
method except for long, very compressible piles penetrations in a site investigation for pile founda-
(e.g. 1 = 300; I/r, = 100 and 200) when both discrete tions seldom reach such depth for economic reasons.
element models do not give good accuracy. For these
cases, the problem lies with the inadequacy of the
expression for rm given by eqn (3) for long, very
compressible, single piles (Chow [20]). This point was
also noted by Randolph and Wroth [14] who attrib-
uted this to the instability of their method. For these
cases, a suitable modification of the influence distance
r,,, is necessary, and this may be determined from an
integral equation analysis (Mattes and Poulos [21]).
With the adjusted value of r,,,, the solutions from the
continuum model are in good agreement with the 0071
integral equation analysis (Fig. 5). However, for these 06 1
0 6 16 24 0 6 I6 24
Rle spacing s/r, Ale Spacing s/r,
0 6 16 240
Pile spociq s/r, Pita spacing s/r,
Fig. 8. Comparison of interaction factors in a half-space
with soil stiffness proportional to depth (E,/G,=260, Fig. 10. Comparison of interaction factors for rigid piles in
v = 0.3; p = 0.5). a two-layer soil medium (I/r, = 50; G, /G2 = 0.2; v = 0.5).
164 Y. K. CHOW
6
. I
16
, L
casr2
groups in a three-layer soil
- fiti maarvmnt[2ll
------- continuumm&l
f
s
9 l2 ol A hlrr-1
both the discrete element model results compare strictly applicable, but the actual trend is well
favourably with the measured values. The layer reflected in the nonlinear solutions. Of course, in this
model shows a stiffer response! than the continuum case, the degree of accuracy would to a large extent
model which is more evident in the nine-pile group.
This is in accordance with the lower degree of W IkN) LwdtkN) l&od IkN)
0 Kx)2cO303 0 lmmxxJ 0 100200300
interaction estimated by the layer model observed
in the previous section of this paper. Tables 2 and 3
show the load distribution in the nine-pile group at
a working load of 2.58 MN and at a load of 5.66 MN
which is close to failure. Under linear elastic soil
conditions, the load distribution predicted by the
layer mode1 shows a more even distribution com-
pared to the continuum model. Even at a working
load level of 2.58 MN, the mild nonlinearity in the
computed solutions has an important influence on the
load distribution, and improves on the agreement
_____ canthum -
between the computed and measured values. Closer -FM musuremmt w
load distribution amongst the piles is fairly even. At Fig. IS. Axial load distribution along piles in nine-pile
this load level, the linear elastic solutions are not group at a working group load of 2.58 MN.
Table 2. Comparison of load distribution to individual piles Table 3. Comparison of load distribution to individual piles
in nine-pile group at a workingload of 2.58MN in nine-pile group near failure at a load of 5.66 MN
Averagepile loads (kN) Average pile loads (kN)
Comer age Center Comer Edge Center
p&s piles pile piles pik pile
(Piles It) (Piles 27) (Pile 3t) (pile, It) (piles 3) cpile 3)
Measured [24] 294 285 261 MasuraJ [24] 635 608 696
Layer model 291 282 261 Layer model 630 627 636
(nonlinear) (nonlinear)
Continuum model 295 284 269 Continuum model 631 629 626
(nonlinear) (nonlinear)
Layer model 301 283 241 Layer made1 660 621 542
(linear) oincar)
Continuum model 315 274 229 Continuum model 690 600 502
(linear) (linaar)
t See Fig. 14. t See Fig. 14.
166 Y. K. CHOW