INTERNATIONAL JOURNAL FOR NUMERICAL AND ANALYTICAL METHODS IN GEOMECHANICS, VOL.

17, 849-869 (1993)

AN APPROXIMATE ANALYSIS PROCEDURE
FOR PILED RAFT FOUNDATIONS
P. CLANCY AND M. F. RANDOLPH
Department of Civil Engineering, University of Western Australia, Nedlands, Western Australia 6009, Australia
SUMMARY
A piled raft foundation comprises both piles and a pile cap that itself transmits load directly to the ground.
The aim of such a foundation is to reduce the number of piles compared with a more conventional piled
foundation where the bearing effect of the pile cap, or raft, is ignored. This paper describes a 'hybrid'
approach for the analysis of piled raft foundations, based on a load transfer treatment of individual piles,
together with elastic interaction between different piles and with the raft. The numerical analysis is used to
evaluate a simple approximate method of estimating the overall response of the foundation from the
response of the component parts. The method leads to estimates of the overall foundation stiffness, the
proportion of load carried by the pile group and the raft, and an initial assessment of differential settlements.
Parametric studies are presented showing the effect of factors such as raft stiffness and pile spacing, length
and stiffness, and a worked example is included demonstrating the accuracy of the approximate design
approach.
I NTRODUCTI ON
It is common practice to cast the cap of a pile group foundation directly on the ground. Where
competent soil conditions exist at ground level, a significant proportion of the structural load
may then be transmitted directly from the pile cap to the ground. I n effect, the foundation behaves
as a piled raft, although the performance is usually estimated conservatively by ignoring the
bearing effect of the raft. This can lead to an unnecessarily high density of
In order to improve the efficiency of piled raft design, thus minimizing the number of piles, it is
necessary to develop simple analytical approaches that are able to quantify the foundation
performance, in terms of both overall and differential settlements, and the load sharing between
the raft and the piles. However, any simple approach must first be calibrated against a more
rigorous numerical analysis.
Approximate analytical methods for piled rafts have been proposed by Pouios and Davis,3
based on the analysis of individual 'pile-raft' units and appropriate interaction factors, and by
Randolph4 who combined pile group and raft response through a single interaction factor (also
based on an individual pile-raft unit). However, rigorous numerical solutions have been confined
to the work of Hain and Lee,' Weisner and Brown6 and Bilotta et aL7 using the boundary
element method to analyse piled raft foundations with a relatively small number of piles. Both the
boundary element method and the finite element method are limited in the size of problem that
can be addre~sed,~.~ owing to the three-dimensional (3-D) nature of the problem that leads to
very large stiffness matrices.
The present paper describes a 'hybrid' numerical method which has been developed specificaily
to minimize the amount of computation.'' The method combines finite elements and simplified
closed form expressions for the diagonal terms of the stiffness matrices, with off-diagonal terms
0363-9061/93/120849-21$15.50
0 1993 by J ohn Wiley & Sons, Ltd.
Received I7 December I992
Revised 29 March I993
850 P. CLANCY AND M. F. RANDOLPH
calculated using Mindlin's' solution and lumped forces. These simplifications, which have been
checked for accuracy against more refined approaches, allow problems of practical proportions
to be analysed. In addition, a simple design approach for piled rafts is proposed, extending the
work of Randolph4 Lo allow the foundation response to be estimated directly from the separate
responses of the raft and pile group.
METHOD OF ANALYSIS
The numerical method of analysis employed depends on a hybrid model, combining finite
elements to model the structural elements of the foundation with analytical solutions for the soil
response. A brief description of the various elements of the analysis is given here, but a fuller
description has been presented by Griffiths et cri."
The approach is based on a method of analysis for pile groups presented by C h ~w' ~' ' ~ and in
computer code by Smith and Griffiths.14 This employs one-dimensional (I-D) rod finite elements
to represent each axially-loaded pile. coupled with a soil response at each pile node modelled by
discrete laod transfer (or r - z ) springs.' ' An analytical method for calculating the gradient of the
load transfer springs was proposed by Randolph and Wroth.I6 The soil deformation around the
pile shaft was idealized as the shearing of concentric cylinders, and the base response was
analysed separately as a rigid punch on the surface of a semi-infinite half space. The two parts of
the analysis were combined by forcing displacement compatibility at the base of the pile. This
approach has been extended to permit a non-linear (hyperbolic) variation in shear modulus,
leading to the derivation of a formula involving the secant shear modul ~s. ' ~~' ~ ChowI2 de-
veloped this idea further to produce a more efficient formula for computational purposes based
on the tangent shear modulus.
Interaction between piles through the soil is calculated using Mindlin's" elastic continuum
solution. This allows the displacement at a point in the interior of a semi-infinite linear elastic half
space, due to a point load at another point in the half space, to be calculated. Mindlin's solution
has been used in the boundary element method.' 9. 20 where numerical integration of the equation
over the pile surfaces is required. In the present method no integration is involved since the forces
are considered to be lumped at each pile node. This is an acceptable approximation since the use
of load transfer springs eliminates the need to calculate interactions between nodes on the same
pile.
A separate raft analysis has been developed in a consistent manner, using two-dimensional
(2-D) 'thin' plate-bending finite element^'^to model the raft. An equivalent soil 'spring' response
is calculated for each raft node using an analytical solution due to Giroud" for the average
displacement under a uniformly loaded rectangular area. The area of raft contributing to each
node is calculated by summing the area of each raft eiement to which the node is attached, and by
dividing this area by four (since each element has four nodes). The contributing area does not
necessarily centre on the node itself. and so the centre of this area is calculated for each node.
These centroidal points are then used to determine the interaction between raft nodes through the
soil, which again makes use of Mindlin's equation. The adequacy of this approach has been
demonstrated by Griffiths et al."
The two analyses are combined by attaching piles to the raft via common nodes at the
connecting points. The wrtical freedoms are linked, resulting in only axial load being transmitted
to the piles. It is assumed that there is no raft-soil contact at the common nodes. Interaction
between pile nodes and raft nodes is calculated using Mindlin's equation.
This has resulted in a relatively rigorous and yet considerably more efficient method of analysis
for piled rafts than has previously been available, and allows for variable geometry, pile stiffness,
ANALYSIS PROCEDURE FOR PILED RAFT FOUNDATIONS 85 1
0 One-dimensional pile element
@ Ground resistance at each pile node represented by non-linear 'T-Z' springs
@ Two-dimensional plaie-bending tinite element ratt mesh
@ Ground resistance at each raft node represented by an eqiiiavslerii !qiring
@ Pile-soil-pile interaction effects calculated betu-een p:tirs of odes tiring
Mindlin's equation
@ Pile-soil-raft interaction
@ Raft-soil-raft iriteraction
Figure I . Numerical representation of piled raft
soil stiffness and raft stiffness. Figure 1 illustrates the various components of the analysis. In the
present paper only vertical loading is considered, but i t would be a relatively simple matter to
allow for horizontal or inclined loading in a consistent manner (at the expense of increasing the
size of the numerical problem). Although only linear elastic soil conditions have been considered
to date, non-linear load transfer springs may be used at the pile-soil interface to model local
yielding.
PARA MGTERS
In the development stage of the analysis, attention has been restricted to homogeneous soil
conditions, and square groups of piles at uniform spacing. The performance of the piled raft is
then determined by the material and geometrical properties of the soil, piles and raft, which have
been summarized in Table I. These parameters may be grouped in the appropriate non-
dimensional ratios summarized in Table 11.The form of the dimensionless group for the raft-soil
stiffness ratio (I&) is based on the work of Brown.22
852 P. CLANCY AND M. F. RANDOLPH
Table I. Parameters for piled raft foundation
Soil Pile Raft
~
Youngs modulus E, Youngs modulus E, Youngs modulus E,
Poissons ratio y Length L Poissons ratio v,
Diameter d Thickness t
Spacing s Length Lr
Breadth
Table 11. Dimensionless groups for piled raft foundation
- ~ ~~
Dimensionless group Definition Practical range
Pile spacing ratio sl d 2.5-8
Pile slenderness ratio Ll d
10-100
Pile-soil stiffness ratio K, , =E , P S 100-10000
4E,B,t3(1 - v : )
37cE,LP
Raft-soil stiffness ratio K, , = 0.01-10
ACCURACY OF PRESENT ANALYSIS
A sensitivity study was undertaken in order to explore the accuracy of the numerical analysis, and
to assess the level of discretization necessary to yield acceptable results. This was done in
a number of stages, the first involving a single pile analysis. A medium length relatively flexible
pile was investigated, with the materail and geometrical properties given in Table 111.
The number of rod finite elements was varied from 5 to 25 and the load induced due to an
applied unit displacement of the pile head (the overall pile stiffness) was calcualted in each case.
The results have been tabulated (Table IV) along with solutions due to Randolph and WrothI6
and Poulos and Davis.3 The present method shows good comparison, even when only five pile
elements were used. The results converge on the solution due to Randolph and Wroth because the
load transfer approach is used in both cases, whereas the solution due to Poulos and Davis is
calculated from a full boundary element analysis.
The effect of mesh refinement on pile-soil-pile interaction was also investigated using the same
pile properties. Calculations were performed for two adjacent piles at a spacing of 2.5d, and for
a 2 x 2 square pile group with a similar spacing. Again, the results for individual pile stiffnesses
were compared with those due to Fleming et aLZ3 and Poulos and Davis3 (Table V).
Table 111. Properties for single pile and pile group mesh
refinement analyses
-
Soil Pile Dimensionless
groups
E, 280MPa E, 35000MPa LJd 25
vS 0.4 L 20m K,, 125
d 0.8 m
ANALYSIS PROCEDURE FOR PILED RAFT FOUNDATIONS
853
Table IV. Results of mesh refinement analysis for singlepile
Number of rod
finite elements
Overall pile
stiffness (MN/mm)
5
10
15
20
25
Randolph and Wroth
Poulos and Davis
1.569
1.548
1.544
1.543
1.542
1.542
1.697
Table V. Results of mesh refinement analysis for pilegroup
Number of rod Two piles 2 x2 pile group
finite elements single pilestiffness singlepilestiffness
(MNImm) (MN/mm)
5
10
15
20
25
Fleming et al.
Poulos and Davis
1.166
1.151
1.148
1.147
1.147
1.077
1.237
0.797
0.789
0787
0.787
0.787
0,752
0817
Finally, three different meshes were used to analyse the effect of mesh refinement on the raft in
a piled raft situation (Figure 2). The piles had similar material and geometrical properties to those
used previously, and were each discretized into 15 rod finite elements. A flexible raft was used in
the analysis with the properties given in Table, VI.
The results of the piled raft analyses (Table VII) show that even the coarsest raft mesh is
sufficiently fine for the calculation of displacements and load distribution. In later analyses, 15
pile elements were used together with a raft mesh in which there were two plate bending finite
elements between each pile.
DEVELOPMENT OF A SI MPLI FI ED APPROACH
The present method provides a rigorous approach to the piled raft problem, but is limited because
of the amount of computing power required to analyse larger pile groups. This arises from the fact
that the problem is attacked from a flexibilility approach, requiring inversion of a fully-populated
flexibility matrix before the stiffness matrix can be formed. For practical purposes, the analysis is
limited to groups of around 50 piles or less.
I n order to allow routine design of piled rafts with a larger number of piles, it is necessary to
develop approximate methods which allow extrapolation of the rigorous analyses. To this end, an
extensive series of analyses have been carried out for smaller piled-raft systems of square
geometry and up to 36 piles. The influence of varying a number of parameters (as defined above)
was investigated: pile group size, pile spacing ratio, pile slenderness ratio, pile-soil stiffness ratio,
and raft-soil stiffness ratio.
H d l O a N V U ' M I a N V A 3 N V 7 3 . d
854 P. CLANCY AND M. F. RANDOLPH
- - - _ - - _ _ _ _ _ _ . . .
One wch currently awilahle approximate method is the approach described by Randolph'
employing a 'flexibility' matrix method to combine the individual st i f hesses (i.e. load- displace-
ment response) of pile group and raft. Two lactors are introduced, apr and a,,, which descrrhc t he
interaction of the two foundation elements when combined.
ANALYSIS PROCEDURE FOR PILED RAFT FOUNDATIONS 855
Table VII. Results of mesh refinement analysis for raft
-
Refinement Mesh 1 Mesh 2 Mesh 3
wp (mm)
7.978 797 1 7.967
w, (mm)
8 W Y 8.088 8.083
p, (MN)
29.66 29.21 29.05
p, (MN)
6.34 6.79 6.95
Run time (min) 5.3 415 404.3
Thus,
where
w, =average displacement of pile grcbup in combined foundation
w, =average displacement of raft in combined foundation
P, =total load carried by pile group in combined foundation
P, =total load carried by raft in combined foundation
k , =overall stiffness (P/w) of pile group in isolation
k, =overall stiffness ( P/ w) of raft in isolation
clrP =interaction factor of pile group on raft
a,, =interaction factor of raft on pile group
For a rigid raft w, =w, , and the overall pileeraft displacement will subsequently be referred to
as wpr. This is not generally true for systems with more flexible rafts, since the average displace-
ment of the pile group will usually not be equal to the average displacement of the raft. I n order to
simplify matters, the difference has been ignored at this stage.
If appropriate work-compatible average displacements of the pile group and the raft are used in
equation (I ), then the off-diagonal terms of the flexibility matrix must be equal (i.e.
npr/kr =zr,/k,). This can easily be demonstrated by considering the trivial example of two
unequal piles (Figure 3). The two piles have stiffnesses of k , and k 2 , respectively, and are loaded
with P, and P,. According to the liexibility matrix ofequation ( I ) , this causes a total displacement
oC( P, / kl +PzgIz;L2) i n Pile 1, and ( P , zL1 . XI i- t? , A 2 } in Pile 2. I f Pilc 1 is 1o;tded before Pile 2.
the total work done by the loads is equal to C 1 . l . where
It Pile 2 is loaded before Pile 1, then the total work done is equal to M >
HI the. prin\iple of superpovticm. M, mu5t be equd to I + - since the cl.istrc encrp) \toted in each
case 1s the same, and therefore X , ~ / L , and z 2 , A , must also be equal
856 P. CLANCY AND M. F. RANDOLPH
I
I
Kl Ki I
Pile 1 Pile 2
-E I
I
I
Pile 1 load PI
Pile 2 load 0
disp. PdKi
I
I
I I I
- I
-
I
I
1
i
11
I Load pile 1: I
Pile 1 load PI I
I Pile 2 load 0 I
disp. PdKI
Load pile 2:
Pile 1 load 0
Pile 2 load Pz
J disp. PdKz
disp. PzcLIdKz
Load pile 1, pile 2:
Pile 1 load PI
Pile 2 load PZ
disp. PI/KI+P~CCI~KI
disp. Qlazi/KI+PyKz
I I
Figure 3. Reciprocal displacements between two adjacent piles
This reciprocal theorem is only valid if work-compatible displacements of the various compo-
nents are used. For general piled raft systems, the overall settlements of the pile group and raft are
calculated by an averaging method which does not satisfy this requirement. In addition, each of
the components deforms differently in isolation from their behaviour, as part of a piled raft
system. When combined, the influence of the pile group on the raft is more likely to result in an
average displacement of the raft similar to the work-compatible value simply because the average
displacement of a raft foundation is relatively independent of the raft-soil stiffness ratio. The net
result of this is that arp/kp is a more reliable parameter for determining piled-raft behaviour than
is upr/kr.
Randolphs4 analytical work was based on a rigid raft, so that wp =w, =wpr. This also meant
that the assumption of work-compatible displacements was only a small approximation, and he
was able to show that arp/kr z upr/kp. Assuming these two conditions to hold true, the load
ANALYSIS PROCEDURE FOR PILED RAFT FOUNDATIONS 857
distribution between the pile group and the raft can be found in terms of arp by solving equations
( 1 ) simultaneously for Pp and Pr:
The load distribution and overall stiffness can be derived in terms of apr in a similar manner.
Randolph4 suggested a method of calculating the two factors arp and apr by considering a single
pile-raft unit. The method involves superposing the displacement fields induced by a single pile
and by a circular raft. This results in the following formulae for the calculation of arp and apr:
k r
apr =arp --
fi P
L
t
c
#-
0 Pile spacing, s
0 Pile diameter, d
@ Pile length, L
@ Raft length, Lr
0 Raft breadth, Br
@ Raft thickness, t
@ Radius to give
equivalent raft
area per piIe
Figure 4. Piled raft divided into single pile-raft units
858 P. CLANCY AND M. F. RANDOLPH
where n is the ratio of the circular raft diameter to the pile diameter, rm is a measure of the radius
of influence of the pile [rm =2.5pL(1 - us)] and p is the degree of homogeneity of the For
very slender piles, the pile length L is replaced by a limiting effective pile length which may be
calculated according to Fleming et dZ3:
L, =1.5dJ2(1 +v,)E,/E,.
In order to calculate the overall stiffness of a piled raft foundation having more than one pile,
the values of arP and up, for a single pile-raft unit are assumed to be directly applicable. The single
pile-raft unit is taken to have a raft area equal to the mean raft area per pile in the complete
system (Figure 4). The pile group stiffness may be calculated in a manner similar to that for
a single pile, allowing for interaction effects between the piles, and the raft stiffness for a uniformly
loaded rectangular or circular raft may be found directly using closed form analytical solutions.
The present work tackles the problem from the opposite direction, i.e. to calculate real values
of arP and apr from knowledge of the results of full piled raft analyses (i.e. Pp, P, and wpr). The
results are compared with the approximate values calculated from Randolphs analytical ap-
proach. Values of and apr can be back-calculated, if wpr, Pp and P, are known, by rearranging
equations (1).
RESULTS
A comprehensive range of values for arp and apr were calculated for a single pile and circular raft
using the analytical method of Randolph: and are plotted in Figure 5. The pile spacing
indicated on the abscissae is an indicator of the relative raft size, and is similar to the rectangular
pile-pile spacing used in larger pile groups. Thus, for a single pile-raft unit with a raft radius r, the
equivalent rectangular pile spacing in a larger piled raft system is given by s/d =& r/ d.
As has been mentioned, the analytical method is based on a single pile-raft unit and assumes
a rigid raft. Figure 6 shows the numerical results which are directly comparable, i.e. a single pile
under a rigid raft (Krs =10). Since rectangular plate-bending finite elements were used to model
the raft, these results were produced for a set of square rafts, each having an area equal to that of
the corresponding circular raft. To derive each point on the curves it was necessary to perform
three separate analyses, isolating the pile group and the raft in turn to obtain k, and k, , and then
performing a full piled raft analysis to obtain kpr, Pp and P,.
It can be seen that there is good agreement between the two sets of results, both in general
trends and in numerical values, particularly for arP. This indicates the validity of the approximate
method when applied to single pile-raft units, but it is necessary to confirm that the method can
be extended to larger pile groups. Observing the trends, arP decreases as the pile spacing is
increased, or as the pile stiffness is decreased; apr shows the opposite behaviour. Since
arp represents the interaction of the pile group on the raft, these trends are intuitively correct.
Increasing the slenderness ratio of the pile merely increases the spread of values at a given pile
spacing, resulting from pile stiffness being more important in longer piles.
Figure 7 is a similar set of plots obtained numerically for a 3 x 3 pile group. In all cases the
values of arP and apr are higher than for the corresponding single pile-raft unit, and the spread of
values is tighter. Thus, the interaction between the separate elements increases as the size of the
pile group is increased, leading to a reduction in the overall stiffness of the system per pile-raft
ANALYSIS PROCEDURE FOR PILED RAFT FOUNDATIONS 859
1.0.
0.8
0.6
0.4
- 0.2
yi -0.2
pile 'spacing'. r/d
-0.8
-1.0 Lld =10
1.0,
08
0 6
0 4
^p 02
Zpopzka s 6 7 b b
pile 'spacing'. sl d
-0.4
-0.6 1
-0.8
-1.0 L/d =25
1.0.
0.8
0.6
0.4
^a 0.2
pile 'spacing'. sl d
-0.6
-0.8
-1.0 Lld =100
1.0
4 5 6 7 8 9 7 -02
pile 'npcmg', sld
-0 8
- I 0 Lld =10
1.0
pile 'spang', sld
-0.8
- 1 0 L/d =25
1.0,
pile 'spacing'. $Id
-0.4
-0.6 1
-0.8
-1.0 Lld =100
~ apr I
Figure 5. Randolph analytical values for all pile group sizes and raft stiffnesses
unit in addition to the reduction in stiffness of each element (per pile-raft unit). As the size of the
foundation is increased, the effects of pile stiffness and pile slenderness on the interaction are
reduced. It is worth noting that the values of a,, become negative when the pile group consists of
long, stiff piles. The assumption of reciprocity breaks down in these cases because the raft is no
longer rigid in comparison to the pile group, allowing the piles to displace by different amounts,
and the average pile displacement is no longer a good approximation to the work-compatible
value. Numerical errors are likely to occur if the stiffness of the raft is increased, but applying
a uniform displacement instead of a uniform load would overcome the problem. However, it was
noted earlier that apr was likely to be a less reliable parameter than arp, and for this reason all
calculations will use arp.
The trend from the single pile to a 6 x 6 pile group has been generated for piles of slenderness
ratio L/d of 25 and is presented in Figure 8. It is clear that convergence of the interaction factors is
approached as the size of the pile group is increased. Of particular interest is that the value of
arp approaches a value of approximately 0.8 for all pile spacings and across the full range of pile
860 P. CLANCY AND M. F. RANDOLPH
I 1.07 I
0.8
0.6
~ 0.4
^p 0.2
0.8
0.6
~ a 0 . 4 * 0.2
2 f l
pile 'rpcing', r/d
-1.0 Lld =25
0.8
06
I a 0 . 4 k 0.2 i3
pile 'spacing'. dd
-0.6
-0.8
-1 0 Lld =100
L.
arp-
i
1.0
08
06
0.4
c 0.2
% -0.2 4 5 6 7 8 9
2 0.0 , - - -
' ,
pile 'rpcing'. r/d
-0.8
- I 0 Lld =10
I .o
0.8 T
pile 'spang'. dd
-0.4
-0.6
-0.8
-1.0 Wd =25
1 .o
0.8 T
id
, __
v
- 0.0
4.0.2i 3 4 3 6 7 8 9
pile 'spcmg'. r/d
-0 8
-I 0 Lld =100
Figure 6. Computed values for single pileeraft unit (Kr3 =10)
stiffnesses. Figure 9 is a replot of the data shown in Figure 7, this time highlighting that variation
of pile slenderness ratio has little effect on the value of arp.
It is known that the average displacement of a raft foundation is approximately constant over
the range of raft ~tiffness.'~ The small effect of raft-soil stiffness on the parameters arp and apr is
demonstrated in Figure 10, which shows the results for a 3 x 3 pile group of L/d =25, with raft
stiffnesses ranging from very flexible (Krs =0.01) to very stiff ('rigid' K, , =10).
The physical significance of clip remaining at a constant value while the geometry and material
properties are varied is difficult to extract due to the complex interplay of a number of factors.
However, it is possible to determine whether it would be reasonable to accept the validity of such
results. Normalizing equation (9) for aIp with respect to the total load ( P ) carried by the piled raft
system gives
ANALYSIS PROCEDURE FOR PILED RAFT FOUNDATIONS 861
~
1.0
0.8
0.6
0.4
^e 0.2
g 0 0
2 -0.2 6 7 8 9
-0.4 pile 'spacing', s/d
-0.6 1
-0.8
-1.0 Lld =10
1.0
0.8
0.6
0.4
0.0
-0 4
-0.6
^e 0.2
-
-0.8 f
- I 0 L/d =25
pile 'spacing', dd
Lld =100
1.0,
0.8
0.6
0.4
pile 'spacing'. r/d
-0.8
-1.0 Lld =10
10,
: --+.--
3
4 0.0
=.o.2 3 4 5 6 1 8 9
pile 'apacmg'. r/d
-0.8
-1.0 Lld =25
1.0
0.8 I
06
04
- 00
g 0 2
%O*
pile 'spacing'. s/d
-0.8
- 1 0 Lld =100
Figure 7. Computed values for 3 x 3 pile group ( Kr T =10)
As an example, Figure 11 analyses in detail the variation with pile spacing of all the factors
1. As s/d is increased, both k , and k, increase, as might be anticipated.
2. k, increases at a smaller rate than k, , so that the proportion of load taken by the pile group
(P, / P) should decrease while the proportion of load taken by the raft ( P, / P) should increase,
and this is also observed.
involved in the calculation of a,, for a 6 x 6 pile group.
3. Since k , is increasing and Pp/P is decreasing, kp/ ( P, / P) must increase.
4. k, and P, are both increasing, so any variation in ( P, / P) / k, is dependant on the relative
magnitudes of increase. In fact ( P, / P) / k, can be seen to decrease slightly with increasing s / d.
5. Since both k , and k, are increasing, it is obvious that wpr/P must decrease, and it does this at
a greater rate than ( P, / P) / k, .
6. Thus, [(wp,/P) - ( P, / P) / k, ] decreases while kp/Pp increases, and the combined effect results
in a constant value of a,,.
862 P. CLANCY AND M. F. RANDOLPH
1.0
0.8
0.6
0 4
^a 0.2
pile 'spacing'. r/d
-0 8
EplEs =100
10 ~~
0.8
0.6
0 4
~0. 4
-0.6 I
-0.8
EplEs =100000
pile 'rpeing', s/d
I L O T I
0.8
I :::
2 0.2
pile 'ipemg'. s/d
-0.8
-1.0 Ep/Es =100
0.8
0.6
0.4
6 1 8 9
pile'spang'. sl d
EplEs =100000
-
'1x1' pile group 0 2x2 pile group
0 4x4 pile group A 5x5 pile group
3x3 pile group
6x6 pile group
Figure 8. Computed values: L/d =25; K,, =10
1.00
0.8
0.6 0 ,-, 7
0.4
@
% -0.2 7 8 9
pllc 'spacing', sld
-0.8
Ep/Es =100
pile 'spacing', rld
I
1.0
pile 'spacing', r/d
1.0
pile 'spacing'. dd
-0.8 Ep/Es =1OOOOO
aPr
Figure 9. Computed values for 3 x 3 pile group (Krs =10)
ANALYSIS PROCEDURE FOR PILED RAFT FOUNDATIONS
g o z - ~
-
8 0.0-
z.o.2:L
-0.4
-0.6
-0 8
.I.o:
863
+ - - - I ' ~ : :
3 4 5 6 7 8 9
pile 'spacing'. sld
~~
~~
EpiEs =100
1.0 I
08
0.6
0.4
pik 'spacing'. d d
08
Ep/Es =100000
0.8
0.6
0.4
f.ip . 6 ._ ---. 7 8 . -+ 9
pile 'spacing', d d
-0.6
-0.8
.,,,, Ep/Es =100
_ _ _ _ _ _ _ ~ _ _ _ ~~ _.
1.0
nR I _ _
06
0 4
-z 0 2
4 00
pile 'spcmg'. r/d
-
0 Krs=O.l Krs=1
~ ~ _ _
Figure 10. Computed values for 3 x 3 pile group (L/d =25)
l . k+06
8.0e+05
6.Oe+05
4.0e+05
2.0e+05
o.oe+oo
2 3 4 5 6 7 8 9
Pile spacing, s/d
~ _ _ _ _ _
5.k+05
o.oe+oo I-
2 3 4 5 6 1 8 9
Pile spacing, s/d
2.5e-06 T
2 3 4 5 6 7 8 9
Pile spacing, s/d
1 .k+Oo
8.k-01
6.Oe-01
4.0e-01
2 3 4 5 6 7 8 9
Pile spacing, s/d
- WPrP
3.oe-06
2.5e-06
2.k-06
1.5e-06
1.k-06
5. k-07
2 3 4 5 6 7 8 9
Pile spacing, s/d
l.oe+Oo
8.k-01
6.Oe-01
4.k-01
2.k-01
O.k+Oo
alpha(rp)
r 1
2 3 4 5 6 1 8 9
Pile spacing, s/d
Figure 11. Investigation of factors contributing to urp, 6 x 6 pile group: L/d =25; Ep / E, =1ooO; Kro =10
864 P. CLANCY AND M. F. RANDOLPH
DIFFERENTIAL SETTLEMENTS
The approximate method outlined above focuses entirely on average settlement and does not
address the key problem of differential settlement. A method of predicting the differential
settlements of a piled raft system is currently being investigated, and will be dealt with in
a separate publication. However, a very simple method that appears to work well for groups of up
to 80 piles is to factor the raft displacements by k, / k, , . For convenience of plotting the variation of
settlement across foundations of differing size, a normalized co-ordinate is introduced (Figure 12).
This has a value of 0 at the corner of the raft, 0.5 at the mid-edge, and 1 at the centre. Figure 13
shows a series of results for a square 3 x 3 pile group across the range of raft stiffnesses as pile
stiffness is varied. The displacements have been normalized by a factor of L, E, / P( l - uf ). The
differential settlements are predicted satisfactorily, with a tendency to overpredict the central
displacement and underpredict that at the corner, for all but the most flexible raft. This is to be
expected, since the local stiffening due to the piles will be more pronounced under a raft of low
stiffness.
A study of the effect of pile spacing, pile slenderness and group size was also made (Figure 14),
showing that this method is generally applicable except in the case of very small pile groups. Here
the local stiffening effect of the piles again becomes considerable.
WORKED EXAMPLE
A worked example is presented, considering a 9 x 9 square pile group (Figure 15). If five rod finite
elements are used to model each pile and the coarsest raft mesh is used, the piled raft problem
requires approximately 10 MB of computer memory to run in single precision (4 bytes per real
number). This compares with 2 MB for a similar 6 x 6 pile group analysis and 0.2 MB for a 3 x 3
pile group. The analysis takes 15 h to solve on a Sun SparcStation IPC (similar speed to IBM
486), compared with less than 5 min for the 3 x 3 pile group. Obviously, it is advantageous to have
a reliable approximate method for the analysis of larger systems.
The material and geometrical properties used in the worked example are given in Table VIII.
0.0 - 0.5
r - r - i
I t I
I I I
+--+-I
I I I
I I I I
c -
I
c -
L - L - L - L - I - I - I
- +- +- 4 - A - -I
1 1 I I I I I
Figure 12. Normalized co-ordinate applied to 3 x 3 pile group. Raft nodes to be plotted marked by x
ANALYSIS PROCEDURE FOR PILED RAFT FOUNDATIONS 865
normalimd coordinate
0 0 0 5 10
_i
, ~ _ _
E 0.2
m
B
3 0.4
P
2 0.6
.u
0
0.8 Krs =0.01
normalised coordinate
0.0 0.5 1 .o
0.0 +- . . . . , - - -
$ 1 B 0.2
M
5
0.4
P
a 0.6
U
.-
0.8 Krs =0.1
normalised coordinate
0.0 0.5 1 .o
0.0 1 - ~ i
$ 1 B 0.2
M
3 0.4
Fi
P
0.6
0
0.8 Krs =1.0
norrnalised coordinate
0.0 0.5 1 .o
0.0 1 - -
$ . I 0.2
piled raft, Kps =1 0 0
0 estimate, Kps =100
piled raft, Kps =loo0
0 estimate, Kps =loo0
A piled raft, Kps =loo00
A estimate, Kps =loo00
piled raft, Kps =1OOOOO
0 estimate, Kps =IOOOOO
Figure 13. Comparison between actual piled raft displacements and estimates from raft alone. Variation of pile stiffness
and raft stiffness. 3 x 3 pile group: L/d =25; s/ d =2
The first step is to analyse the pile group alone and calculate its overall stiffness, k,. Next the
raft is analysed in isolation to find k, . Finally, a full piled raft analysis is performed to calculate
P,, P, and wpr. A uniform load of 1 MPa was applied to the raft, giving a total load of 1296 MN.
k, =14.31 MN/mm
k, =13.08 MN/mm
Pp =867.9 MN
P, =428.1 MN
wpr =84.03 mm (w, =84.11 mm)
866 P. CLANCY AND M. F. RANDOLPH
normalisedcoordinate
0.0 0.5 1.0
0.0 I - '
-
normalised coordinate
0.0 0.5 1 .o
-i
0.0 1
"
0 .*+ A A
0.8 '
normalisedcoordinate
0.0 0.5 1 .o
0.0 t - - - i
0.8 1
I piled raft, s/d =2
7 estimate, s/d =2
piled raft, s/d =4
> estimate, s/d =4
piled raft, s/d =6
1 estimate, s/d =6
3x3 pile group
Krs =0.1
Lld =25
Kps =10000
piled raft, L/d =10
0 estimate, L/d =10
piled raft, Ud =25
0 estimate, L/d =25
A piled raft, L/d =100
_ _ ~~
piled raft, 2x2 group
0 estimate, 2x2 group
piled raft, 4x4 group
0 estimate, 4x4 group
A piled raft, 6x6 group
A estimate, 6x6 group
~
3x3 pile group
d d = 2
Kps =10000
Krs =0.1
L/d =25
d d = 2
Kps =10000
Krs =0.1
Figure 14. Comparison between actual piled raft displacements and estimates from raft alone. Variation of pile spacing,
pile length and size of pile group
These results may then be used to calculate a,, and a,,, using equations (9):
a,, =0.85
apr =0.71
In this case a,, is greater than 0.8, as might be expected for a relatively flexible system. The effect
this has on the prediction of settlement and load distribution can be investigated by substituting
the calculated values of k , and k, into equation (6). Assuming a,, =0.8 gives k,, =15.42 MN/mm,
allowing wpr under the applied total load of 1296 MN to be calculated. This value of wpr may then
be used in equations (4) and (5) to calculate Pp and P,. A comparison of the accurate analysis and
approximate results is given in Table IX, showing that there is a very small error in estimating
ANALYSIS PROCEDURE FOR PILED RAFT FOUNDATIONS 867
36 m
1
o Pile l l Raft element
Figure 15. 9 x 9 piled raft mesh
Table VIII. Properties for 9 x 9 piled raft worked example
Soil Pile Raft Dimensionless
groups
E, 280MPa Ep 35000MPa E, 35000MPa sJd 5
v, 0.4 L 20m 0, 0.3 Lld 25
d 0.8 m t 5m Kps 125
S 4m L, 36m K S , 0.1 19
B, 36m
Table IX. Comparison of actual and predicted results 9 x 9 piled
raft worked example
Full piled raft Predicted Prediction
analysis values error ( YO)
% 0.85 0.8 - 5.9
Wpr 84 mm 84 mm 0
PP 868MN 778MN - 10
pr 428 MN 518 MN 21
wpr. The estimation of load distribution errs by 90 MN, or 6.9% of the total load, resulting in an
underestimation of the load taken by the pile group.
To estimate the differential settlements of the piled raft, the results of the isolated raft analysis
were factored by k, / k, , (using the predicted value of k,, gives k, / k, , =0.889). The accurate and
868 P. CLANCY AND M. F. RANDOLPH
normalised coordinate
-1
-0.8
8 -0.6
3 -0.4
3 0
4 0.4
0.6
1
-g -0.2
-g 0.2
s 0.8
. full piled raft 0 esti mate from
factored raft anal ysi s
Figure 16. Comparison between actual piled raft displacements and estimates from raft alone. 9 x 9 square pile group
approximate displacements have been plotted in Figure 15, using the normalized co-ordinate as
per Figure 12, and normalizing the displacements with respect to the range of differential
displacements as follows:
where
w =displacement (actual piled raft or factored raft)
wav(pr), wrnax(pr), w ~ ~ ~ ( ~ ~ ) =average, maximum and minimum displacement of piled raft
The approximate displacements underestimate the range of piled raft displacements, with
a maximum difference of 0.87 mm, or 1 . 1 % . The corner-centre differential displacement is
underestimated by 1.68 mm, or 13%.
In summary, the approximate method requires the following procedure:
(1) calculate the pile group stiffness and raft stiffness in isolation,
( 2 ) combine these stiffnesses using equation (6) and an clrp value of 0.8 to calculate the full piled
(3) the load sharing between the pile group and raft in the combined foundation may now be
raft stiffness kpr,
found from equations (4) and (5).
CONCLUSIONS
A rigorous numerical method of piled raft analysis has been developed, based on a hybrid
approach of load transfer analysis of individual piles, together with elastic interaction between
piles and between the various raft elements and the piles. A comprehensive range of foundation
geometries has been analysed, within the practical limitations of the method, where computa-
tional times become excessive for groups of more than 100 piles.
The numerical analysis has been used to evaluate an existing approximate method for
estimating the overall stiffness of the foundation and the proportion of load carried by the pile
ANALYSIS PROCEDURE FOR PILED RAFT FOUNDATIONS 869
group and by the raft. The approximate method was found to be satisfactory for single pile units
but became progressively less accurate as the size of the pile group increased. However, a simple
modification to the method has allowed it to be extended to pile groups of practical proportions.
The basis of the method allows the overall foundation response to be estimated from the
response of the component parts, through the use of an appropriate interaction factor. While this
allows the average settlement of the foundation to be estimated, it gives no information
concerning the differential settlements. At present, a simple factoring of the differential settle-
ments that would occur for a raft foundation alone is proposed. This gives satisfactory results for
groups of up to 9 x 9 piles, except in the extreme cases of a very flexible raft, or very small pile
groups (2 x 2 or less) where the pattern of settlement is rather different from a raft alone. Further
refinement of this approach is in hand, and will be addressed in a separate publication.
REFERENCES
1. R. W. Cooke, D. W. Bryden-Smith, M. N. Gooch and D. F. Sillett, Some observations of the foundation loading and
settlement of a multi-storey building on a piled raft foundation in London clay, Proc. l nst . of Ci v. Engnrs. Part 1,
2. C. J . Padfield and M. J . Sharrock, Settlement of structures on clay soils, Special Publication 27, Construction
3. H. G. Poulos and E. H. Davis, Pile Foundation Analysis and Design, Wiley, New York, 1980.
4. M. F. Randolph, Design of piled raft foundations, Proc. Int. Symp. on Recent Developments in Laboratory and Field
Tests and Analysis of Geotechnical Problems, Bangkok, 6-9 December 1983, pp. 525-537.
5. S. J. Hain and I. K. Lee, The analysis of flexible pile-raft systems, Geotechnique, 28 (I), 65-83 (1978).
6. T. J . Weisner and P. T. Brown, Behaviour of piled strip footings subject to concentrated loads, Australian Geomech. J .
7. E. Bilotta, V. Caputo and C. Viggiani, Analysis of soil-structure interaction for piled rafts, Proc. 10th European Conf:
8. M. Ottaviani, Three-dimensional finite element analysis of vertically loaded pile groups, Geotechnique, 25 (2),
9. P. Clancy, Computer analysis of pile groups, M.Sc. Thesis, University of Manchester, 1990.
10. D. V. Griffiths, P. Clancy and M. F. Randolph, Piled raft foundation analysis by finite elements, Proc. 7t h Int. Con$
11. R. D. Mindlin, Force at a point in the interior of a semi-infinite solid, Phys. 7, 195-202 (1936).
12. Y. K. Chow, Analysis of vertically-loaded pile groups, Int. J . Numer. Analytic. Met h. Geomech., 10, 59-72 (1986).
13. Y. K. Chow, Three-dimensional analysis of pilegroups, ASCE J . Geotech. Eny. Diu., 113(6), 637-651 (1987).
14. 1. M. Smith and D. V. Griffiths, Programming the Finite Element Method, 2nd edn, Wiley, New York, 1988.
15. H. B. Seed and L. C. Reese, The actions of soft clay along friction piles, Trans. ASCE, 122, 731-754 (1957).
16. M. F. Randolph and C. P. Wroth, Analysis of deformation of vertically loaded piles, ASCE J . Geotech. Eny. Diu, 104,
17. M. F. Randolph, A theoretical study of the performance of piles, Ph.D. Thesis, University of Cambridge, 1977.
18. L. M. Kraft, R. P. Ray and T. Kagawa, Theoretical t - z curves, ASCE J . Geotech. Eng. Div., 107, GT11, 1543-1561
19. H. G. Poulos and E. H. Davis, The settlement behaviour of single axially loaded incompressible piles and piers,
20. R. Butterfield and P. K. Banerjee, The elastic analysis of compressible piles and pilegroups, Geotechnique, 21 (I ),
21. J .-P. Giroud, Settlement of a linearly loaded rectangular area, ASCE J . Soil Mech. Found. Eng Diu., 94, SM4,
22. P. T. Brown, Strip footing with concentrated loads on deep elastic foundations, Geotech. Eng., 6, 1-13 (1975).
23. W. G. K. Fleming, A. J . Weltman, M. F. Randolph and W. K. Elson, Piling Engineering, Wiley, New York, 1992.
24. M. F. Randolph and C. P. Wroth, An analysis of the vertical deformation of pile groups, Geotechnique, 29(4),
25. D. D. Barkan, Dynamics of Bases and Foundations, McGraw-Hill, New York, 1962.
1981, Vol. 70, pp. 443-460.
Industry Research and Information Association, London, 1983.
G6, NO. 1, 1-5 (1976).
on Soil Mech. Foundn Eng., Florence, 26-30 May 1991, Vol. 1, pp. 315-318.
159-174 (1975).
on Computer Met hods and Advances in Geomechanics, Cairns, 6-10 May 1991, Vol. 2, 1153-1157.
GT, 1465-1488 (1978).
(1981).
Geotechnique, 18(3), 351-371 (1968).
43-60 (1971).
813-832 (1968).
423-439 (1979).