Acta Geotechnica (2013) 8:3–15

DOI 10.1007/s11440-012-0193-4

RESEARCH PAPER

Re-assessment of foundation settlements for the Burj Khalifa,

Dubai
Gianpiero Russo • Vincenzo Abagnara •

Harry G. Poulos • John C. Small

Received: 14 April 2012 / Accepted: 12 October 2012 / Published online: 30 October 2012
Ó Springer-Verlag Berlin Heidelberg 2012

Abstract This paper deals with the re-assessment of 1 Introduction

foundation settlements for the Burj Khalifa Tower in
Dubai. The foundation system for the tower is a piled raft, The Burj Khalifa in Dubai was officially opened in January
founded on deep deposits of calcareous rocks. Two com- 2010 and, at a height of 828 m, is currently the world’s
puter programs, Geotechnical Analysis of Raft with Piles tallest building. The foundation system is a piled raft, a
(GARP) and Non-linear Analysis of Piled Rafts (NAPRA) form of foundation that is being used increasingly to sup-
have been used for the settlement analyses, and the paper port tall structures where the loads are expected to be
outlines the procedure adopted to re-assess the foundation excessively large for a raft alone and where the raft and the
settlements based on a careful interpretation of load tests piles are able to transfer load to the soil. The foundation
on trial piles in which the interaction effects of the pile test design process for this building has been described by
set-up are allowed for. The paper then examines the Poulos and Bunce [12].
influence of a series of factors on the computed settle- An important component of the design of a piled raft
ments. In order to obtain reasonable estimates of differ- foundation is the detailed assessment of the settlement and
ential settlements within the system, it is found desirable to differential settlement of the foundation system, and their
incorporate the effects of the superstructure stiffness which control by optimising the size, location and arrangement of
act to increase the stiffness of the overall foundation sys- the piles, and the raft thickness. Many different methods of
tem. Values of average and differential settlements for the analysis have been devised in order to predict the behav-
piled raft calculated with GARP and NAPRA were found iour of raft and piled raft foundations [2, 4, 5, 6, 10, 13, 15,
to be in reasonable agreement with measured data on set- 17, 19, 20], and these range from simple hand-based
tlements taken near the end of construction of the tower. methods to complex three-dimensional numerical analyses.
In this paper, attention is focussed on two methods that
Keywords Case history
Footings and foundations
model the raft as an elastic plate and the piles as interacting
Full-scale tests
Piles
Rafts
Settlement non-linear springs. The computer codes implementing
these methods are described very briefly and are then
applied to the Burj Khalifa, which is founded on a piled
raft. The evaluation of the ground modulus values is
described using a combination of field test and laboratory
data and the results of pile load tests. The method of
G. Russo (&)
V. Abagnara
interpreting the pile load test data is discussed, and the
University of Naples, Naples, Italy
e-mail: pierusso@unina.it importance of allowing for interaction between the test pile
and the surrounding reaction piles in emphasised. The two
H. G. Poulos
J. C. Small programs are then used to compare the computed settle-
Coffey Geotechnics, Sydney, Australia
ments with available measurements of foundation settle-
J. C. Small ments, and with the ‘‘Class A’’ predictions made by the
University of Sydney, Sydney, Australia foundation designers and the peer reviewers.

123
4 Acta Geotechnica (2013) 8:3–15

An important objective of the paper is to explore how vertical; accordingly, only the axial stiffness of the springs
pile load test data should be used when predicting the is required.
settlement performance of piled and piled raft foundation The soil is assumed to be a layered elastic continuum.
systems, and to examine some factors that may have an The Boussinesq solution for a point load and the closed
important influence on predicted foundation settlements. form solution for a rectangular uniformly loaded area at the
surface of an elastic half space are used to calculate the soil
displacements produced by the contact pressure developed
2 Computer analyses at the interface between the raft and the soil. The layered
continuum is solved by means of the Steinbrenner
The settlement analyses used in this paper for the Burj approximation [3, 13], and as such, invokes the simple
Khalifa have employed two computer programs, Geo- assumption that the stress distribution within an elastic
technical Analysis of Raft with Piles (GARP) and Non- layer is identical with the Boussinesq distribution for a
linear Analysis of Piled Rafts (NAPRA), which idealise the homogeneous half space [13].
piled raft foundation as a plate supported by non-linear The interaction factor method is used to model pile to
interacting springs. A very brief description of these pro- pile interaction and a preliminary boundary element (BEM)
grams is given below. analysis allows calculation of the interaction factors
between two piles at various spacings. Interaction between
2.1 Program GARP axially loaded piles beneath the raft and the raft elements is
accounted for via pile–soil interaction factors computed
The computer program GARP [18] uses a simplified with a preliminary BEM procedure. The reciprocal theorem
boundary element analysis to compute the behaviour of a is used to maintain that the soil–pile interaction factor is
piled raft when subjected to applied vertical loading, equal to the pile–soil interaction factor.
moment loading, and free-field vertical soil movements. A stepwise incremental procedure is used to simulate the
The raft is represented by a thin elastic plate and is non-linear load-settlement relationship of a single pile, the
discretized via the finite element method using eight-noded total load to be applied is subdivided into a number of
elements. The soil is modelled as a layered elastic con- increments, and the diagonal terms of the pile–soil flexi-
tinuum, and the piles are represented by elastic–plastic or bility matrix are updated at each step. A computation of the
hyperbolic springs, which can interact with each other and nodal reactions vector is made at each step to check for
with the raft. Pile–pile interactions are incorporated via tensile forces between raft and soil and an iterative pro-
interaction factors [9]. Simplifying approximations are cedure is used to make them equal to zero. Basically, this
utilised for the raft–pile and pile–raft interactions. Beneath procedure releases the compatibility of displacements
the raft, limiting values of contact pressure in compression between the raft and the pile–soil system in the node where
and tension can be specified so that some allowance can be tensile forces were detected, although the overall equilib-
made for non-linear raft behaviour. The output of GARP rium is maintained by a re-distribution of forces. An iter-
includes: the settlement at all nodes of the raft; the trans- ative procedure is needed since after the first run some
verse, longitudinal, and torsional bending moments within additional tensile forces may arise in different nodes. The
each element of the raft; the contact pressures below the output of the code is represented by the distribution of the
raft; and the vertical loads on each pile. In its present form, nodal displacements of the raft and the pile–soil system,
GARP can consider vertical and moment loadings, but not the load sharing among the piles and the raft, the bending
lateral loadings or torsion. moments and the shear in the raft, for each load increment.
Abagnara et al. [1] have compared GARP and NAPRA
2.2 Program NAPRA analyses for a simple case, and have concluded that both
programs give comparable results, but that some of the
The computer program NAPRA [13, 15] computes the simplifying assumptions employed in each program give
behaviour of a raft subjected to any combination of vertical rise to differences in detail. For example, the difference in
distributed or concentrated loading and moment loading. raft settlements may be due to the differences in the details
The raft is modelled as a two-dimensional elastic body of calculation of the soil layer stiffness using the Bous-
using the thin plate theory, and utilising the finite element sinesq/Steinbrenner approach. The difference in plate ele-
method, adopting a four- or nine-noded rectangular ment types may also contribute to the differences. For the
element. piled raft, the differences may arise because of differences
The piles and the soil are modelled by means of inter- in the methods used to compute the single pile stiffness
acting linear or non-linear springs. It is assumed that the values, the interaction factors and the pile–raft and raft–
interaction between the raft and the soil (the piles) is purely pile interactions.

123
Acta Geotechnica (2013) 8:3–15 5

In this paper, attention will be focussed on analyses The ground conditions at the site comprise a horizon-
carried out with NAPRA, although a comparison will also tally stratified subsurface profile which is complex and
be presented between the GARP and NAPRA analyses. highly variable in terms of the strata thickness due to the
nature of deposition and the prevalent hot arid climatic
conditions. The main strata identified were as follows:
3 Settlement assessment for the Burj Khalifa Tower,
1. Very loose to medium dense silty sand (Marine
Dubai
deposits).
2. Weak to moderately weak calcarenite, generally
3.1 Foundation layout
unweathered with fractures close to medium spaced
interbedded with cemented sand. This material is
The Burj Khalifa project in Dubai, United Arab Emirates
generally underlain by very weak to weak sandstone
(UAE), comprises a 160 storey high rise tower, with a
which is generally unweathered with fractures close to
podium development around the base of the tower,
medium spaced interbedded with cemented sand.
including a 4–6 storey garage. The Burj Khalifa is located
3. Very weak to weak calcarenite, calcareous sandstone,
on a 42,000 m2 site. The tower is founded on a 3.7 m thick
and sandstone; this formation is slightly to highly
raft supported on 194 bored piles, 1.5 m in diameter,
weathered with fractures extremely close to closely
extending 47.45 m below the base of the raft; podium
spaced and interbedded with cemented sand. Bands of
structures are founded on a 0.65 m thick raft (increased to
1–5 m thickness are also present of medium dense to
1 m at column locations) supported on 750 bored piles,
very dense, cemented sand with sandstone bands and
0.9 m in diameter, extending 30–35 m below the base of
locally with bands of silt.
the raft. A plan view of the foundation is shown in Fig. 1.
4. Very weak to weak gypsiferous sandstone, gypsiferous
calcareous sandstone occasionally gypsiferous silt-
3.2 Ground investigation and site characterisation
stone. This material is generally unweathered to
slightly weathered with fractures extremely close to
The investigations involved the drilling of 32 boreholes to
closely spaced and interbedded with cemented sand.
a maximum depth of about 90 m below ground level and 1
The formation is interbedded with dense to very dense,
borehole to a depth of 140 m under the tower footprint.
cemented silty sand and occasionally silt with sand-
Standpipe piezometers were installed to measure the
stone bands.
ground water level which was found to be relatively close
5. Very weak to weak calcisiltite, conglomeritic calcisil-
to the ground surface, typically at a level of 2.5 m DMD.
tite, and calcareous calcisiltite. This material is gen-
The ranges of measured SPT N values are summarised in
erally moderately to highly weathered, occasionally
Table 1. There was a tendency for N values to increase
slightly and completely weathered with fractures
with depth, beyond an elevation of about -8 m DMD.
extremely close to medium spaced. Calcareous silt-
stone was encountered in the majority of the deeper
boreholes comprising very weak to weak occasionally
moderately weak calcareous siltstone in bands with a
thickness of 0.5–14.4 m generally slightly to moder-
ately weathered occasionally highly to extremely
weathered.
6. Very weak to weak and occasionally moderately weak
calcareous siltstone, calcareous conglomerate, con-
glomeritic sandstone, and limestone. This material is

Table 1 Summary of measured SPT values

Elevation (m) Range of SPT values

2.5 to -1 0–40
-1 to -8 50–400
-8 to -14 0–100
-14 to -30 40–200
-30 to -40 100–200
-40 to -80 100–400
Fig. 1 Plan view of the Khalifa Tower foundation system

123
6 Acta Geotechnica (2013) 8:3–15

generally slightly weathered and occasionally shear, and constant normal stiffness (CNS) direct shear
unweathered and moderately weathered to highly tests.
weathered. Occasionally encountered as calcisiltite Some of the relevant findings from the in situ and lab-
interbedded with bands of siltstone and conglomerate. oratory testing are as follows:
7. Very weak to moderately weak claystone interbedded
1. The cemented materials were generally very weak to
with siltstone. This material is generally slightly
weak; unconfined compressive strength (UCS) values
weathered with close to medium-spaced fractures.
ranged mostly between about 0.1 and 6 MPa, the
Between -112.2 and -128.2 m occasional bands of
average values for each layer being the ones reported
up to 500 mm thick gypsum were encountered. Below
in Table 2.
-128.2 m the stratum was encountered as weak to
2. Values of the Young’s modulus from pressuremeter
moderately weak siltstone with medium to widely
tests (first and second reload cycle) were found to be in
spaced fractures.
good agreement with values from correlation with
Table 2 summarises the stratigraphy adopted for the shear waves velocities. From calcarenite (0 to -7.5 m)
foundation settlement analyses. to sandstone (-7.5 to -24 m), Young’s modulus is
approximately constant with depth; at greater depths,
3.3 In situ and laboratory test results the average values decrease in the gypsiferous sand-
stone (-24 to -28.5 m) then they slightly increase in
A comprehensive series of in situ tests was carried out, the calcisiltite (from -28.5 to -68.5 m) and finally
including pressuremeter tests, down-hole seismic, cross- decrease in the siltstone (from -68.5 to -91 m).
hole seismic, and cross-hole tomography to determine 3. Triaxial stress path testing (at strain levels of 0.01 and
compression (P) and shear (S) wave velocities through the 0.1 %) was found to give results for Young’s modulus
ground profile. The vertical profile of P-wave velocity with that were in good agreement with pressuremeter and
depth gave a useful indication of variations in the nature of geophysics testing results.
the strata between the borelogs. 4. Resonant column testing was found to give lower
Conventional laboratory classification tests (moisture values for the Young’s modulus when compared with
content of soil and rock, Atterberg limits, particle size values from pressuremeter tests, geophysics tests, and
distribution, and hydrometer) and laboratory tests for triaxial stress tests.
determining physical (porosity tests, intact dry density, 5. Constant normal stiffness (CNS) tests were carried out
specific gravity, particle density) and chemical properties on three samples taken from stratum 5a to assess the
were carried out. In addition, unconfined compression tests, ultimate skin friction values and the potential for
point load index tests, and drained direct shear tests were cyclic degradation at the pile–soil interface. These
carried out. A considerable amount of more advanced tests indicated values of peak monotonic shear stress
laboratory testing was undertaken, including stress path ranging from 360 to 558 kPa, with only a little
triaxial tests, resonant column testing for small-strain shear difference between the peak monotonic and the
modulus, undrained cyclic triaxial tests, cyclic simple residual cyclic shear stress values.

Table 2 Stratigraphic model adopted for settlement assessment

Stratum Description Level at the top of Thickness (m) Adopted level at top UCS qu (MPa)
the stratum (m DMD) of layer (m DMD)

1 Marine deposits 1.15 to 2.96 1.85 to 4.3 2.5

2 Calcarenite/calcareous sandstone -0.27 to -1.95 2.87 to 10.75 -1.2 2
3a Calcareous sandstone/sandstone -4.13 to -12.06 10.5 to 21.43 -7.3 –
3b -13.5 1
4 Gypsiferous sandstone -21.54 to -26.69 1.7 to 7.75 -24 2
5a Calcisiltite/conglomeritic calcisiltite -27.64 to -31.15 39.2 to 46.75 -28.5 1.3
5b Calcareous siltstone -50 1.7
6 Calcareous/conglomeritic strata -67.19 to -76.04 31 (from 140 m deep -68.5 2.5
BH only)
7 Claystone/siltstone interbedded -98.19 Proved to 39.6 m -90 –
with gypsum layers thickness

123
Acta Geotechnica (2013) 8:3–15 7

3.4 Geotechnical model constructed under polymer fluid. A permanent casing, 6 m

long, was installed from the top of each pile to just above
The key parameters for the assessment of the settlement the highest strain gauge level for all the trial piles tested in
behaviour of the Khalifa Tower piled raft foundation sys- compression and tension. Five piles, designated as P1, P2,
tem are the values of the Young’s modulus of the strata for P3, P4, and P5, were tested in compression; two, P3 and
both raft and pile behaviour under static loading. In a non- P5, were shaft grouted. Test pile P6 was tested in tension
linear analysis, the values of ultimate skin friction of piles, and test pile P7 was laterally loaded.
the ultimate end-bearing resistance of the piles, and the Only the compression load tests on trial piles P1, P2, and
ultimate bearing capacity of the raft would also be P4 have been considered for the present paper. Table 4
required, but in this paper, only linear elastic analyses have summarises the main features of these piles. Figure 4a shows
been undertaken using NAPRA and GARP analyses, hav- the load test arrangements for piles P1 and P2, which con-
ing explored the little influence of non-linearity up to the sisted of the test pile and six reaction piles, while Fig. 4b
maximum observed load level. Attention has thus been shows the set-up for pile P4, which consisted of the test pile
focussed on evaluating relevant values of Young’s modulus and four reaction piles. All the reaction piles were 1.5 m in
for each stratum. diameter. Steel load distribution plates were grouted to the
As a first step in obtaining these values, the relative top of the test piles and hydraulic jacks were placed between
stiffness of the various soil layers was assessed considering the steel plates and the reaction beams. Steel reaction beams
values of the Young’s modulus from the following data: were used to transfer the load from the hydraulic jacks to the
installed reaction piles. Macalloy bars were used as reaction
1. pressuremeter tests (initial loading, first reload, second
anchors to transfer the load from the beams to the reaction
reload cycles);
piles. Six cycles of loading were applied to trial piles P1 and
2. geophysics tests (correlation with shear wave
P2 while nine cycles of loading were applied to trial pile P4,
velocities);
which was the pile designated to be tested cyclically.
3. resonant column tests (Initial, 0.0001, 0.001, 0.01 %
Four main types of instrumentation were used in the
strain levels);
compression test piles:
4. triaxial stress path tests (0.01 and 0.1 % strain levels);
1. concrete embedment vibrating wire strain gauges, to
Values of the various Young’s modulus values are
allow measurement of axial strains at six levels along
plotted in Fig. 2, and although inevitable scatter exists
the pile shafts and hence estimation of the axial load
among the different values, there is a reasonably consistent
distribution;
general pattern of variation with depth.
2. extensometers, to measure change in length of the
Layer 3b (see Table 2) has arbitrarily been chosen as the
piles, and installed at the same levels as the vibrating
reference layer, and for each type of test, values of the
wire strain gauges to provide back-up information on
Young’s modulus for a layer i, Ei, have been related to the
axial load distribution with depth;
value for layer 3b, E3b. The values of Ei/E3b have then been
3. displacement transducers at the top of piles, to measure
averaged using the following data: reload cycles from
the vertical movement at the pile heads;
pressuremeter testing; seismic data; resonant column data
4. load cells, to monitor the load applied to the pile via
at a strain level of 0.01 %, and the triaxial stress path tests.
the jacks.
Figure 3 shows the different assessed relative stiffness
profiles so obtained, and Table 3 summarises the average
values of relative Young’s modulus that were adopted for 3.6 Back-analysis interpretation of load tests to obtain
the analyses and the interpretation of the pile load test data. Young’s modulus values
The absolute values of Young’s modulus for each of the
different layers have been then obtained by fitting the load- The computer program NAPRA was used to carry out the
settlement curves of the single piles obtained from the load back analyses of compression tests on the three test piles
tests, and the process of fitting the load-settlement curves considered. Since a detailed soil profile at each trial pile
to obtain the Young’s modulus values is described below. location was not available, the same geotechnical model
was adopted for all three piles.
3.5 Pile load tests For comparison purposes, the three load tests were back-
analysed both taking and not taking into account interac-
A program of pile load testing was undertaken which tion between test piles and reaction piles. It is now well
involved the installation of seven test piles in the podium recognised that ignoring interaction between the test pile
area near the location of the Khalifa Tower. All the test and the reaction piles can lead to an over-estimation of the
piles and reaction piles were bored cast in situ and pile head stiffness [8, 9, 11, 14].

123
8 Acta Geotechnica (2013) 8:3–15

0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
0

-10

-20

-30
Depth [m]

Press Init AVE

-40 Press First AVE

Press Second AVE

-50 Seismic (x 0.2)

TRXL 0.01%
-60 TRXL 0.1%

RC Initial
-70
RC 0.0001%

RC 0.001%
-80
RC 0.01%

adopted profile for drained modulus E'

-90
Young's Modulus [MPa]

Fig. 2 Summary of Young’s modulus values

0 1 2 3 4
0

-10

-20

-30
Depth [m]

Press Init AVE

-40 Press First AVE

Press Second AVE

Seismic (x 0.2)
-50
TRXL 0.01%

TRXL 0.1%
-60
RC Initial

RC 0.0001%
-70 RC 0.001%

RC 0.01%

-80 ALL

NO INITIAL PRESS

NO INITIAL PRESS & NO TRXL0.01%

-90
Relative Stiffness E/E3b [-]

Fig. 3 Assessed soil relative stiffness

Both linear elastic (LE) and non-linear analyses (NL) 1. Qlim was estimated as the asymptote to a hyperbola
were carried out. In all analyses, Young’s modulus for the fitted to the whole measured load-settlement curve
piles, Ep, was assumed to be 31.8 GPa. For the linear (HYP);
analyses, the theoretical behaviour was fitted to the 2. Qlim was based on the load transfer deduced by strain
observed load-settlement behaviour at pile head displace- gauges readings (SG);
ments of about 0.08 % of diameter and 0.2 % of diameter. 3. Qlim was based on the load transfer deduced by
In the non-linear analyses, in order to assess the sensi- extensometer readings (EX).
tivity of the back-calculated values of soil stiffness to the
Ultimate skin friction values inferred from the axial load
value of ultimate capacity, Qlim, three different values were
distribution and from the extensometer readings were
adopted in the analyses:
employed to assess pile shaft capacity up to depths above

123
Acta Geotechnica (2013) 8:3–15 9

Table 3 Relative values of Young’s modulus used in pile load test

interpretation
Stratum Young’s modulus,
relative to value for layer 3b

2 2.3
3a 0.6
3b 1.0
3c 1.0
4 0.8
5a 0.7
5b 0.8 Fig. 4 Set-up for pile load tests. a Piles P1 and P2. b Pile P4
6 0.7
In the cases of piles P2 and P4, values of back-calculated
soil stiffness are in close agreement with values back cal-
-30, -38, and -30 m for piles P1, P2, and P4, respec- culated in the non-linear analysis (values of w/D are
tively. From pile–soil interface load–strain curves at vari- 0.0008–0.0009), while in the case of pile P1, the first point
ous depths along the shafts, these values were found to be is at a higher displacement (0.21 %), and so the back-cal-
representative of the ultimate values in the upper (cased) culated value is lower. It should be noted that had the
part of the shaft. Below these depths, ultimate values of interaction between the test pile and the reaction piles not
shaft friction were estimated from correlations with the been taken into account; the back-calculated values of pile-
unconfined compressive strength (UCS) of the rock. soil relative stiffness would have been considerably larger.
Table 5 summarises the values of Qlim obtained from these From Table 6, the following points can be noted:
three procedures. As might be expected, the hyperbolic
1. The consideration of interaction between the test pile
extrapolation procedure gives the largest values, and probably
and the reaction piles results in back-figured modulus
over-estimates the capacity. There is some difference between
values which are considerably less than those for
the values assessed on the basis of the strain gauge and
which interaction has been ignored. Thus, there would
extensometer readings, but from the point of view of settle-
be a tendency to under-estimate foundation settlements
ment prediction, such differences are not very significant.
if interaction effects are ignored.
Figures 5 and 6 show typical fits (for pile P2) of the
2. The back-calculated values from the three tests are
computed non-linear load-settlement behaviour and the
scattered.
observed load-settlement behaviour. Figure 5 is for the
interpretation taking account of interaction, while Fig. 6 In order to partially overcome the described limitation in
shows the corresponding fit with interaction between the the back analysis of load tests, and to show its effects on
test pile and reaction piles being ignored. In both cases, the average settlement assessment, sensitivity analyses
very reasonable fits are obtained with the measured data. have been carried out with NAPRA by adopting two dif-
Back-calculated values of the Young’s modulus for ferent values of soil stiffness assessed to be representative
stratum 3b, E3b, are reported in Table 6. In the linear of lower- and upper-bound values.
elastic analyses, the first point on the measured load-set- In the GARP and NAPRA analyses described below, for
tlement curve has been considered. In this way, back-cal- the assessment of the average and differential settlements,
culated values of soil stiffness in linear analyses are the values of E3b shown in Table 7 were adopted on the
affected by the loading procedure adopted in the load tests. basis of the non-linear analysis of the load test results.

Table 4 Summary of pile load tests

Trial Diam. Cut-off level Toe level Length Load test layout DWL* DML** No. of cycles
pile (m) (m DMD) (m DMD) (m) (t) (t)

1 1.5 -4.85 -50 45.15 6 RP circle with a 4.5-m radius 3,000 6,000 6 (50–150 % DWL)
2 1.5 -4.85 -60 55.15 6 RP circle with a 4.5-m radius 3,000 6,000 6 (50–150 % DWL)
4 0.9 -2.90 -50 47.1 4 RP square with a 9-m side 1,000 3,500 9 (100–150 % DWL)
* Designated working load, ** designated maximum test load

123
10 Acta Geotechnica (2013) 8:3–15

Table 5 Assessed pile capacity with different methods

Pile Qlim (kN)
Hyperbolic extrapolation (HYP) Strain gauge readings (SG) Extensometer readings (EX)

TP1 108,800 93,800 73,200

TP2 115,900 97,300 100,200
TP4 82,600 50,500 59,900

Q [kN] Q [kN]
0 10000 20000 30000 40000 50000 60000 70000 0 10000 20000 30000 40000 50000 60000 70000
0 0

5 5

w [mm]
w [mm]

10 10

15 15

20 20

measured E3b=1000 Mpa (NL) E3b=1200 Mpa (NL) measured E3b=1500 Mpa (NL) E3b=1700 Mpa (NL)
HYP SG HYP SG
E3b=1200 Mpa (NL) E3b=1700 Mpa (NL)
EX EX

Fig. 5 Predicted and measured load settlement for pile P2 (interac- Fig. 6 Predicted and measured load settlement for pile P2 (interac-
tion considered) tion ignored)

4 Procedure for foundation settlement re-assessment

The load sharing between the raft and the piles was almost
The majority of the foundation settlement re-assessment negligible, the load carried by the piles ranging in all the
was carried out using linear elastic analyses with the cases between 94 and 100 % of the total applied load.
computer program NAPRA. The mesh adopted for the A series of sensitivity analyses was undertaken to
NAPRA analyses is shown in Fig. 7; it is based on examine the following issues:
square elements whose side is 1.7 m long. Preliminary 1. the influence of non-linear pile response;
analyses indicated that using a finer mesh than this 2. the influence of the range of back-figured values of
produced no change in the results. The actual shape Young’s modulus;
of the raft was modelled by adopting a piecewise 3. the influence of using ‘‘correct’’ and ‘‘incorrect’’ back-
approximation. figured values of Young’s modulus;
Only long-term conditions have been considered, and 4. the effect of not considering the raft in the foundation
for the majority of the early analyses, an average load per settlement analysis;
pile of 23.21 MN has been used (this is representative of 5. the differences between analyses using NAPRA and
the design dead plus live loading) and has been applied as a GARP;
point load on each of the 194 piles. This load corresponds 6. the influence of the assumed loading pattern;
to a uniformly distributed load on the tower raft of about 7. the effect of incorporating the stiffness of the
1,250 kPa. superstructure;
The majority of analyses were undertaken using the 8. the effect of including the podium loading.
best-estimate modulus value of 900 MPa derived from the
proper interpretation of the load test data (see Table 7). The results of these sensitivity studies are described below.

123
Acta Geotechnica (2013) 8:3–15 11

Table 6 Young’s modulus values derived from load tests

Test pile E3b (MPa) with interaction accounted for E3b (MPa) with interaction not accounted for
Linear analysis Linear analysis Non-linear analysis Linear analysis Non-linear analysis
(w/D = 0.0008) (w/D = 0.002) (w/D = 0.0008)

P1 – 350 650 (HYP) to 850 (SG) – 900 (HYP) to 1,100 (SG)

P2 700 650 1,000 (HYP) to 1,200 (SG–EX) 1,200 1,500 (HYP) to 1,700 (SG–EX)
P4 850 550 650(EX) to 850(SG) 1,100 850(EX) to 1,100(SG)
HYP denotes values derived from hyperbolic extrapolation
EX denotes values derived from extensometer readings
SG denotes values derived via strain gauges

Table 7 Values of Young’s modulus (E3b) of layer 3b adopted for foundation analyses
Case Young’s modulus of layer 3b (E3b) MPa
Best estimate Upper-bound value Lower-bound value

Reaction pile interaction considered 900 1,000 650

Reaction pile interaction ignored 1,200 1,500 900

5.2 The influence of the range of back-figured values

of Young’s modulus
Section
adopted for Table 9 summarises the computed settlements from the
settlement
profiles NAPRA analysis, using the range of values of Young’s
81.6 m modulus backfigured from the correct interpretation of the
pile load tests (see Table 7). As would be expected, the
computed settlement for the lower-bound modulus value is
considerably greater than that for the upper-bound modulus
value, although the ratio of the computed settlement values is
less than the ratio of the modulus values. This may be
91.8 m explained by the non-linearity source provided by the itera-
tive check on the tensile forces at the raft–soil interface.
Fig. 7 Model adopted in NAPRA analyses
5.3 The influence of using ‘‘correct’’ and ‘‘incorrect’’
back-figured values of Young’s modulus
5 Results of sensitivity studies
Table 10 shows the influence on the computed settlements
5.1 Influence of non-linear pile response of using the best-estimate modulus values for layer 3b
obtained from the ‘‘correct’’ interpretation (considering test
For the non-linear NAPRA analyses, the ultimate axial pile–reaction pile interaction) and the ‘‘incorrect’’ interpre-
capacity of each pile has been assumed to be 112.5 MN. tation (ignoring this interaction). The settlements computed
Table 8 compares the computed maximum (S max) and using the ‘‘incorrect’’ modulus interpretation are about 21 %
central settlements (S centre), and the maximum differen- less than those using the ‘‘correct’’ interpretation, and it is
tial settlement (DS max), from the linear and non-linear therefore important to properly interpret the test pile load-
analyses. There is very little difference between the two
Table 8 Computed settlements (mm) from linear and non-linear
analyses in this case, as it could be expected being the analyses
global safety factor on each pile in the range 3–5. Thus, the
Linear analysis Non-linear analysis
comparison indicates that the foundation response is
essentially elastic under the dead plus live loadings. S max S centre DS max S max S centre DS max
Accordingly, only linear analyses have been employed for
52 51 27 53 53 27
the remainder of the sensitivity studies.

123
12 Acta Geotechnica (2013) 8:3–15

Table 9 Influence of using upper and lower bounds of back-figured Table 11 Influence of ignoring pile cap on computed settlements
modulus values
Case analysed Computed settlements (mm)
Modulus value for layer 3b S max S centre DS max
S max S centre DS max
Lower bound (E3b = 650 MPa) 67 67 34
Pile group (no raft) 61 51 38
Upper bound (E3b = 1,000 MPa) 48 47 25
Piled raft 52 51 27

wall and column locations, and consequently, NAPRA has

Table 10 Influence of modulus value on computed settlements
been used to examine the influence of the loading pattern
Modulus value used Computed settlements (mm) on the computed settlement profile for two cases:
S max S centre DS max
(a) equal loads on all the piles;
‘‘Correct’’ interpretation 52 51 27 (b) the actual design loadings are applied at the wall and
E3b = 900 MPa column locations.
‘‘Incorrect’’ interpretation 41 40 22
The computed settlement profiles along Wing C in Fig. 9
E3b = 1,200 MPa
show a difference in the computed settlement patterns, with
the equal load assumption giving smaller maximum settle-
settlement data to avoid the tendency to under-estimate the ment than the other case. Thus, it would appear desirable to
foundation settlements and differential settlements. employ the actual load pattern in design calculations.

5.4 Effect of not considering the raft in the analysis 5.7 The influence of the podium structure on the tower
settlement
Non-linear analysis of piled rafts has been used to analyse
the foundation system, both as a piled raft and pile group in The podium structure was assumed to be founded at the
which there is no raft joining the piles. The ‘‘correct’’ best- same depth of the tower raft, but the two rafts were
estimate modulus of layer 3b of 900 MPa has been used. assumed to be unconnected. The length of the podium piles
Table 11 shows the computed settlements for both these was taken to be 30 m, and the columns and rows in the
cases. The difference between the computed central set- NAPRA mesh were spaced at 3 m up to a distance of 30
tlements is negligible, but there is a considerable difference diameters and 4 m at larger distances. An average load per
between the computed maximum settlements and differ- pile of 23.21 MN was applied as a point load on each of the
ential settlements. In this case, the conservatism introduced 194 tower piles while the loads acting on the low-rise area
by ignoring the raft would lead to a 17 % increase in the were modelled as point loads of between 2 and 8 MN
computed maximum settlement but a 40 % increase in the acting on the podium piles, depending on their location.
maximum differential settlement. Therefore, it is desirable The maximum computed settlement was 54 mm, which
to incorporate the effect of the raft when computing the was only 2 mm larger than the value computed for the
settlement distribution within the foundation system. tower only. The effect of the 750 piles of the podium was

5.5 Analyses using NAPRA and GARP

Distance along wing cross section [m]
0 10 20 30 40 50 60 70
Similar meshes have been used for both the NAPRA and 0
GARP analyses and identical analysis assumptions have
10
been made in both cases. Figure 8 shows the computed
profile of settlement along Wing C from both analyses and 20
w [mm]

reveals that they are almost identical. Thus, pleasingly, for 30

the same input, each program is capable of giving very
40
similar results.
50
5.6 The influence of the loading pattern 60
GARP AV 23210 kN NAPRA AV 23210 kN
The preceding results have all been obtained assuming that
the average design load (23.21 MN) has been applied to Fig. 8 Calculated settlements along Wing C from NAPRA and
each pile location. In reality, the loads will be applied via GARP

123
Acta Geotechnica (2013) 8:3–15 13

d [m]
0 10 20 30 40 50 60 70
0

10

20

30
w [mm]

40

50

60

70

80
Average pile loads applied Loads applied at wall &
column locations Fig. 10 Raft Model 2

Fig. 9 Influence of assumed loading pattern on computed settlement

profile

thus very small in this case, primarily because of the sig-

nificant distance of many of the podium piles from the
tower and relatively small loads that they carried.

5.8 The influence of superstructure stiffness

In order to investigate the effect on the computed settlement

and differential settlement and to try and obtain a more
accurate estimate of the pattern of settlement, the stiffening
effect exerted by the superstructure on the raft was taken into
account, in various ways, by increasing the bending stiffness Fig. 11 Raft Model 3
of the raft in each wing (estimated by the structural designers
to be equivalent to an increase in 25,200 kNm2 per wing). Six Figure 12 compares the various computed profiles of
alternative methods of incorporating this increased bending settlement across the tower, together with those in which
stiffness were adopted: no account is taken of superstructure stiffness. Also shown
is the design profile developed by Poulos and Bunce [12],
(a) Increasing the thickness of the whole raft to reflect the
which was for combined dead plus live load and therefore
bending stiffness of the entire tower (Model 1).
not directly comparable. Clearly, there is a considerable
(b) Increasing the raft thickness over the central part of
difference between the extreme profiles, those taking all
the wings and on the core, as shown in Fig. 10, to
the superstructure stiffness into account, and that in which
reflect the bending stiffness of the entire tower. This
no account is taken of the superstructure stiffness. It would
is denoted as Model 2.
appear reasonable to assume that the profiles from Models
(c) Increasing the raft thickness only below the shear
1M, 2M, and 3M may be more reasonable approximations
walls (see Fig. 11) to reflect the bending stiffness of
to reality, and this appears to be borne out by the com-
the entire tower. This case is denoted as Model 3.
parisons with the measured settlements as described below.
(d) Model 1, with only 10 % of the stiffness of the tower
considered (Model 1M).
(e) Model 2, with only 10 % of the stiffness of the tower
6 Comparisons between calculated and measured
considered (Model 2M).
settlements
(f) Model 3, with only 10 % of the stiffness of the tower
considered (Model 3M).
Detailed settlement measurements were only available up
In each case, the actual pattern of loading via the col- to February 2008, before the end of construction and well
umns and walls was applied, with only the dead load before the commissioning and occupation of the building in
component considered. January 2010. Nevertheless, anecdotal evidence indicated

123
14 Acta Geotechnica (2013) 8:3–15

Distance along wing [m] Distance along wing [m]

0 20 40 60 0 20 40 60
0 0

10
10
20
Settlement [mm]

30

Settlement [mm]
20
40

50 30

60
40
70

80
50
90

NAPRA Model 1 NAPRA Model 1M 60

NAPRA Model 2 NAPRA Model 2M NAPRA Model 1M NAPRA Model 3M

NAPRA Model 3 NAPRA Model 3M Measured (February 2008)
NAPRA Model 2M
NAPRA-No Structure Stiffness
Allowance
Design Values (Poulos & Bunce Fig. 13 Measured and computed settlement profiles along Wing C
(2008)

Fig. 12 Comparison between various calculated settlement profiles

there is better agreement between the computed and
measured profiles, with a computed maximum differ-
that the additional settlements between February 2008 and
ential settlement ranging between 15 and 21 mm for
January 2010 were relatively small of the order of 1–2 mm.
the three models, similar to the measured value. In
Figure 13 shows comparisons between the latest avail-
this case, the stiffness of the raft is approximately
able measured profile of settlement in February 2008 and
53 times its original value, and this latter value is
the calculated settlement profiles from Models 1M, 2M and
much larger than the value of 10 times adopted by
3M. The following observations are made from an exam-
Hooper [7] for the Hyde Park Barracks in London and
ination of Figs. 12 and 13:
by Sales et al. [16] for the Skyper Building in
1. Without allowance for superstructure stiffness, the Frankfurt. Interestingly, and almost certainly coinci-
calculated maximum final differential settlement is dentally, the profile for this case is rather similar to
about 35 mm, which is considerably larger than the that obtained for the case when the average load is
measured value of about 14 mm. The computed imposed on each pile.
maximum settlement is also larger than the measured 4. There remain some differences between the measured
value, although some additional settlement would be and computed settlement profiles in the vicinity of the
expected after the building has been in operation for edge of the wing. There may well be scope to refine the
some years. method by which the superstructure is incorporated
2. When allowance is made for the superstructure into the geotechnical foundation analysis.
stiffness, the computed maximum settlement is similar 5. The calculated settlements from the design phase are
in magnitude to the measured value. However, for considerably greater than those obtained from the
Models 1, 2 and 3 in which the full superstructure analyses in this paper. The main reason for these larger
stiffness is incorporated (albeit approximately), the settlements is that the settlements were for both dead
computed settlement pattern differs somewhat from and live load acting, and in addition, conservative
the measured pattern, and the computed differential values of Young’s modulus were used in these
settlements are significantly smaller than those mea- analyses, with a somewhat different distribution of
sured. It seems clear that it is not appropriate to allow ground stiffness with depth being adopted in those
for the bending stiffness of the entire structure when calculations. Once again, this comparison emphasises
trying to modify the foundation stiffness. the importance of appropriate selection of the ground
3. When the allowance for superstructure stiffness is stiffness values if accurate foundation settlement
reduced by a factor of 10 (Models 1M, 2M and 3M), predictions are to be made.

123
Acta Geotechnica (2013) 8:3–15 15

7 Conclusions References

1. The case history of the Burj Khalifa Tower in Dubai has 1. Abagnara V, Poulos HG, Small JC (2012) Comparison of two
piled raft analysis programs. Submitted for 12th Australia-New
been re-assessed for the prediction of average and dif- Zealand conference geomechanics, Melbourne
ferential settlement of the piled raft foundation system. 2. Clancy P, Randolph MF (1993) An approximate analysis proce-
The comprehensive ground investigation and pile testing dure for piled raft foundations. Int J Numer Anal Methods
program carried out for this project has enabled the site to Geomech 17(12):849–869
3. De Sanctis L, Russo G, Viggiani C (2002) Piled raft on layered
be characterised in some detail. The pile load tests in soils. In: Proceedings of the ninth international conference on
particular have been an important factor in enabling piling and deep foundations, Nice, pp 279–286
reasonable settlement prediction to be made. 4. De Sanctis L, Russo G (2008) Analysis and performance of piled
2. The ground stiffness or modulus is a key factor in the rafts designed using innovative criteria. J Geotech Geoenviron
Eng 134(8):1118–1128. doi:10.1061/(ASCE)1090-0241(2008)
prediction of foundation settlements. If this is to be 134:8(1118)
derived from pile load test data, then the interpretation 5. Hemsley JA (1998) Elastic analysis of raft foundations. Thomas
of the load settlement should take into account Telford, London
interaction effects with the reaction system, otherwise, 6. Hemsley JA (ed) (2000) Design applications of raft foundations.
Thomas Telford, London
the ground stiffness is likely to be over-estimated and 7. Hooper JA (1973) Observations on the behaviour of a pile raft
the foundation settlements subsequently under- foundation on London clay. Proc Inst Civ Eng, part 2
predicted. 55(4):855–877
3. Sensitivity studies have been carried out to explore the 8. Kitiyodom P, Matsumoto T, Kanefusa N (2004) Influence of
reaction piles on the behaviour of a test pile in static load testing.
effects of a number of factors on predicted settlement Can Geotech J 41:408–420
behaviour of the Burj Khalifa Tower. In addition to the 9. Poulos HG, Davis EH (1980) Pile foundation analysis and design.
ground stiffness or modulus, the consideration of the John Wiley, New York
effects of the raft and superstructure stiffness may be 10. Poulos HG (1994) An approximate numerical analysis of pile-raft
interaction. Int J Numer Anal Methods Geomech 18:73–92
important factors in influencing both the maximum 11. Poulos HG (2000) Pile testing—from the designer’s viewpoint.
settlement and the maximum differential settlement. In: Statnamic loading test’98, Balkema, Rotterdam, pp 3–21
4. The assessment of the differential settlement of the 12. Poulos HG, Bunce G (2008) Foundation design for the Burj
Khalifa Tower foundation has been explored by Dubai—the world’s tallest building. In: Proceedings of the 6th
international conference case histories in geotechnical engineer-
adopting three different models to account for the ing, Arlington, Virginia, Paper 1.47 CD volume
stiffening effect of the superstructure, and they have 13. Russo G (1998) Numerical analysis of piled rafts. Int J Numer
been found to give reasonably similar results. How- Anal Methods Geomech 22:477–493
ever, if the foundation, or parts of it, is stiffened to 14. Russo G (2012) Experimental investigations and analysis on
different pile load testing procedures. Acta Geotechnica, ISSN
represent the bending stiffness of the entire structure, 1861-1125, accepted for publication May 2012. doi:10.1007/
the consequent foundation response is too rigid, and s11440-012-0177-4
the differential settlements tend to be under-predicted 15. Russo G, Viggiani C (1998) Factors controlling soil-structure
considerably. For the Burj Khalifa Tower, an addi- interaction for piled rafts. In: Katzenbach R, Arslan U (eds)
Proceedings of the international conference on soil-structure
tional stiffness equivalent to about 10 % of the entire interaction in urban civil engineering, Darmstadt, pp 79–102
bending stiffness has been found to give improved, but 16. Sales MM, Small JC, Poulos HG (2010) Compensated piled rafts
by no means perfect, results when compared with in clayey soils: behaviour, measurements, and predictions. Can
measured settlement profiles. This result suggests that Geotech J 47:327–345
17. Selvadurai APS (1979) Elastic analysis of soil-foundation inter-
there is a limit to the stiffness that the structure can action. Elsevier Publishing Co., New York
provide to the foundation system, and so the full 18. Small JC, Poulos HG (2007) Nonlinear analysis of piled raft
structure should not be taken into account when foundations. Geotechnical Special Publication GSP158, ASCE,
calculating the effective increase in stiffness. CD Volume, GeoDenver 2007
19. Ta LD, Small JC (1996) Analysis of piled raft systems in layered
5. In this case at least, consideration of the low-rise soils. Int J Numer Anal Methods Geomech 20:57–72
podium structure leads to only a small increase in the 20. Viggiani C (1998) Pile groups and piled rafts behaviour. In:
settlement under the tower footprint. Proceedings of the 3rd international geotechnical seminar on
6. The method of analysis may be a less significant factor deep foundations on bored and auger piles, Ghent, pp 77–94
in the prediction of piled rafts settlements, provided
the method is sound. For the same input data, the
computer programs GARP and NAPRA produced
similar settlements of the tower.

123