Vous êtes sur la page 1sur 15


Foundation design for the Emirates Twin Towers,

Harry George Poulos and Andrew J. Davids

Abstract: This paper describes the foundation design process adopted for two high-rise buildings in Dubai, the
Emirates Twin Towers. The foundation system for each of the towers was a piled raft, founded on deep deposits of cal-
careous soils and rocks. The paper outlines the geotechnical investigations undertaken, the field and laboratory testing
programs, and the design process and describes how potential issues of low skin friction and cyclic degradation of skin
friction due to wind loading were addressed. An advanced numerical computer analysis was used for the design pro-
cess, which was carried out using a limit state approach. This necessitated analysis of a large number of load cases,
and the paper describes how the information was processed to produce design information. A comprehensive program
of pile load testing was undertaken, and class A predictions of both axial and lateral load–deflection behaviour were in
fair agreement with the load test results. Despite this agreement, the overall settlements of the towers observed during
construction were significantly less than predicted. The possible reasons for the discrepancy are discussed.
Key words: case history, footings and foundations, full-scale tests, piles, rafts, settlement.
Résumé : Cet article décrit le processus de conception des fondations adopté pour deux tours à Dubai, les « Emirates
Twin Towers ». Le système de fondation pour chacune des tours consistait en un radier sur pieux reposant sur des dé-
pôts profonds de sols et roches de carbonate. Cet article donne les grandes lignes des études géotechniques réalisées,
les programmes d’essais en laboratoire et sur le terrain, et le processus de conception, et décrit comment les problèmes
potentiels de faible frottement de surface et de dégradation cyclique du frottement de surface due aux charges de vent
peuvent être traités. Une analyse numérique de pointe à l’ordinateur a été utilisée pour le processus de conception qui
a été réalisé au moyen de l’approche d’état limite. Ceci a nécessité l’analyse d’un grand nombre de cas de charge-
ments, et l’article décrit comment les données ont été traitées pour produire les informations pour la conception. Un
programme élaboré d’essais de chargement sur pieux a été entrepris et des prédictions de classe A du comportement en
déflexion sous chargement tant axial que latéral ont montré une concordance assez bonne avec les résultats des essais
de chargement. En dépit de cette concordance, les tassements globaux des tours observés durant la construction étaient
appréciablement inférieurs à ceux prédits. On discute des raisons possibles de cet inconsistance.
Mots clés : histoire de cas, semelles/fondations, essais à l’échelle naturelle, pieux, radiers, tassement.

[Traduit par la Rédaction] Poulos and Davids 730

Introduction mately 200 000 m2 site, which also incorporates low-level

retail and parking podium areas.
The Emirates project is a twin tower development in Figure 1 shows a photograph of the towers just after the
Dubai, one of the United Arab Emirates (UAE). The towers completion of construction.
are triangular in plan form, with a face dimension of approx- The foundation system for both towers involved the use of
imately 50–54 m. The taller office tower has 52 floors and large-diameter piles, in conjunction with a raft. This paper
rises 355 m above ground level, whereas the shorter hotel describes the geotechnical investigation undertaken for the
tower is 305 m tall. These towers are more than double the project and the process used for the foundation design. It
height of the nearby World Trade Centre, which was for- also presents the results of a major program of pile testing
merly the tallest building in Dubai. The office tower is cur- and compares predicted and observed test pile behaviour.
rently the eighth tallest building in the world, and the hotel Finally, some limited data on settlements during construction
tower is the 17th tallest. The twin towers are on an approxi- of the towers are presented, together with the predicted values.

Received 21 April 2004. Accepted 24 November 2004. Published on the NRC Research Press Web site at http://cgj.nrc.ca on
8 June 2005.
H.G. Poulos.1 Coffey Geosciences Pty Ltd., 8/12 Mars Road Lane Cove West, P.O. Box 125 North Ryde, NSW 1670, Australia
and The University of Sydney, Department of Civil Engineering, NSW 2006, Australia.
A.J. Davids.2 Building Structures, Hyder Consulting Pty Ltd., 116 Miller St., North Sydney, NSW 2065, Australia.
Corresponding author (e-mail: harry_poulos@coffey.com.au).
Present address: Coffey Geosciences Pty Ltd., 8/12 Mars Road Lane Cove West, P.O. Box 125 North Ryde, NSW 1670, Australia.

Can. Geotech. J. 42: 716–730 (2005) doi: 10.1139/T05-004 © 2005 NRC Canada
Poulos and Davids 717

Ground investigation and site Fig. 1. The Emirates Twin Towers soon after completion of con-
characterization struction.

Preliminary geotechnical data for the site were available

from earlier investigations via a series of boreholes drilled to
about 15 m depth. These revealed layers of sand or silty
sand overlying very weak to weak sandstone. For the twin
tower development, it was clear that this preliminary infor-
mation was inadequate; hence, a comprehensive investiga-
tion was carried out. This investigation involved the drilling
of 23 boreholes to a maximum depth of about 80 m. The
deepest boreholes were located below the tower footprints;
boreholes below the low-rise areas tended to be considerably
shallower. Standard penetration tests (SPTs) were carried out
at nominal 1 m depths in the upper 6 m of each borehole and
then at 1.5 m intervals, until an SPT value of 60 was
achieved. The SPT values generally ranged between 5 and
20 in the upper 4 m, increasing to 60 at depths of 8–10 m.
Rotary coring was carried out thereafter. Core recoveries
were typically 60%–100%, and rock quality designation val-
ues were also between about 60% and 100%.
Figure 2 shows the borehole information along a section
that passes through the two towers. It was found that the
stratigraphy was relatively uniform across the whole site, so
it was considered adequate to characterize the site with a
single geotechnical model. The ground surface was typically
at a level of +1 to +3 m Dubai Municipality datum (DMD),
and the groundwater level was relatively close to the surface,
typically between 0 and –0.6 m DMD. The investigation re-
vealed seven main strata, which are summarized in Table 1
by material descriptions commonly adopted in Dubai.

Foundation parameter assessment and the

geotechnical model Test results
From the viewpoint of the foundation design, some of the
In situ and laboratory testing relevant findings from the in situ and laboratory testing were
Because of the relatively good ground conditions near the as follows:
surface, a piled raft system was deemed appropriate for the (i) The site uniformity borehole seismic testing did not re-
foundation of each tower. The design of such a system re- veal any significant variations in seismic velocity, thus
quires information on both the strength and the stiffness of indicating that it was unlikely that major fracturing or
the ground. Consequently, a comprehensive series of in situ voids would be present in the areas tested.
tests was carried out. In addition to standard SPTs and per- (ii) The cemented materials were generally very weak to
meability tests, pressuremeter tests, vertical seismic shear weak, with uniaxial compressive strength (UCS) values
wave testing, and site uniformity borehole seismic testing ranging between about 0.2 and 4 MPa, with most values
were carried out. lying within the range of 0.5–1.5 MPa.
Conventional laboratory tests, including classification (iii) The average angle of internal friction of the near-surface
tests, chemical tests, unconfined compression tests, point load soils, from direct shear box tests, was about 31°.
index tests, drained direct shear tests, and oedometer consol- (iv) The oedometer data for compressibility were considered
idation tests, were carried out. In addition, a considerable unreliable, because of the compressibility of the appara-
amount of more advanced laboratory testing was undertaken. tus being of a similar order to that of some of the sam-
This included stress path triaxial tests for settlement analysis ples.
of the deeper layers, constant normal stiffness (CNS) direct (v) Cyclic triaxial tests were carried out with one-way cy-
shear tests (Lam and Johnston 1982) for pile skin friction clic loading and a maximum stress of up to 930 kPa.
under both static and cyclic loading, resonant column testing These tests indicated that the unit 4 sand deposit had the
for small-strain shear modulus and damping of the founda- potential to generate significant excess pore pressures
tion materials, and undrained static and cyclic triaxial shear under cyclic loading and to accumulate permanent de-
tests to assess the possible influence of cyclic loading on formations under repeated one-way loading. It could
strength and to investigate the variation of soil stiffness and therefore be susceptible to earthquake-induced settle-
damping with axial strain. ments.

© 2005 NRC Canada

718 Can. Geotech. J. Vol. 42, 2005

Fig. 2. Geotechnical conditions. DMD, Dubai Municipality datum.

Table 1. Summary of main strata at site.

Avg. elevation of base
Unit No. Designation Material description of unit (m DMD)
1 Silty sand Uncemented calcareous silty sand, loose to moderately dense –3.3
2 Silty sand Variably and weakly cemented calcareous silty sand –8.1
3 Sandstone Calcareous sandstone, slightly to highly weathered, well cemented –26.8
4 Silty sand Calcareous silty sand, variably cemented, with localized well-cemented bands –33.1
5 Calcisiltite Variably weathered, very weakly to moderately well cemented –53.5
6 Calcisiltite As for unit 5 –68.5
7 Calcisiltite As for unit 5 –79.0
Note: DMD, Dubai Municipality datum.

(vi) The CNS shear tests indicated that cyclic loading had measured values from the CNS tests were within and beyond
the potential to significantly reduce or degrade the skin the range of design values for static skin friction of piles in
friction after initial static failure and that a cyclic stress cemented calcareous soils tentatively suggested by Poulos
of 50% of the initial static resistance could cause failure (1988); that range was between 100 and 500 kPa, depending
during cyclic loading, resulting in a very low postcyclic on the degree of cementation.
residual strength.
Figure 3 summarizes the values of Young’s modulus ob- Geotechnical model
tained from the following tests: (i) seismic data (reduced by The key design parameters for the foundation system were
a factor of 0.2, to account for a strain level appropriate to the the ultimate skin friction of the piles, the ultimate end-
overall behaviour of the pile foundation); (ii) resonant col- bearing resistance of the piles, the ultimate bearing capacity
umn tests (at a strain level of 0.1%); (iii) laboratory stress of the raft, and the Young’s modulus of the soils for both the
path tests, designed to simulate the initial and incremental raft and the pile behaviour under static loading. For the as-
stress states along and below the foundation system; and sessment of dynamic response under wind and seismic load-
(iv) unconfined compression tests (at 50% of ultimate ing conditions, Young’s modulus values for rapid loading
stress). conditions were also required, together with internal damp-
Figure 4 shows the ultimate static shear resistance, de- ing values for the various strata.
rived from the CNS test data, as a function of depth below The geotechnical model for foundation design under static
the surface. With the exception of one sample, all tests loading conditions was based on the relevant available in
showed a maximum shear resistance of at least 500 kPa. The situ and laboratory test data and is shown in Fig. 5. The ulti-

© 2005 NRC Canada

Poulos and Davids 719

Fig. 3. Summary of Young’s modulus values.

mate skin friction values were based largely on the CNS Fig. 4. Ultimate skin friction values from CNS tests.
data, whereas the ultimate end-bearing values for the piles
were assessed on the basis of correlations with the UCS data
(Reese and O’Neill 1988) and also on the basis of previous
experience with similar cemented deposits (Poulos 1988).
The values of Young’s modulus were derived from the data
summarized in Fig. 3. Although inevitable scatter exists
among the different values, there is a reasonably consistent
general pattern of variation of modulus with depth. Consid-
erable emphasis was placed on the laboratory stress path
tests, which should have reflected realistic stress and strain
levels within the various units. The values for the upper two
units were obtained from correlations with the SPT data.
The bearing capacity of the various layers for shallow
foundation loading, pu, was estimated from bearing capacity
theory for the inferred friction angles, the tangents of which
were reduced by a factor of two thirds to allow for the ef-
fects of soil compressibility, as suggested by Poulos and
Chua (1985).

Limit state design approach

Ultimate limit state
The piled raft foundation was designed via a limit state [1] R*s ≥ S*
approach based on Australian Standard AS 2159–1995,
“Piling—Design and installation” (AS 2159 1995). The de-
sign criteria for the ultimate limit state were as follows: [2] R*g ≥ S*

© 2005 NRC Canada

720 Can. Geotech. J. Vol. 42, 2005

Fig. 5. Geotechnical model adopted for design. Eu, undrained Serviceability limit state
Young’s modulus; E′, drained Young’s modulus, v ′, drained The design criteria for the serviceability limit state were
Poisson’s ratio; fs, ultimate shaft friction; fb, ultimate pile end as follows:
bearing; pu, ultimate lateral pile–soil pressure.
[4] ρmax ≤ ρall
[5] θmax ≤ θall
where ρmax is the maximum computed foundation settle-
ment; ρall is the allowable foundation settlement, taken to be
150 mm; θmax is the maximum computed local angular rota-
tion; and θall is the allowable angular rotation, taken to be
1/350 here.

Load combinations
For each tower, 18 load combinations were analyzed: 1
loading set for the ultimate dead and live loading only; four
groups of 4 loading sets for various combinations of dead,
live, and wind loading for the ultimate limit state; and 1
loading set for the long-term serviceability limit state (dead
plus live loading).

Conventional pile capacity analyses were used to assess
the ultimate geotechnical capacity of the piles and raft. For
the piles, this capacity was taken as the sum of the shaft and
base capacities. For the raft, account was taken of the layer-
ing of the geotechnical profile and the large size of the foun-
dation, and a value of 2.0 MPa was adopted for the ultimate
bearing capacity. In these conventional analyses, it was as-
sumed that the portion of the raft providing additional bear-
ing capacity had a diameter of 3.6 m (three pile diameters)
where R*s is design structural strength; R*g is design around each pile.
geotechnical strength; and S* is design action effect (fac- In additional to the conventional analyses, more complete
tored load combination). The above two criteria were ap- analyses of the foundation system were undertaken with the
plied to the entire foundation system, and the structural computer program geotechnical analysis of raft with piles
strength criterion (eq. [1]) was also applied to each individ- (GARP) (Poulos 1994). This program uses a simplified
ual pile. The values for R*s and R*g were obtained from the boundary element analysis to compute the behaviour of a
estimated ultimate structural and geotechnical capacities, rectangular piled raft when subjected to applied vertical
multiplied by appropriate reduction factors. In this case, the loading, moment loading, and free-field vertical soil move-
structural reduction factor was taken as 0.6. For the ments. The raft is represented by an elastic plate, the soil is
geotechnical ultimate limit state, the worst response arising modelled as a layered elastic continuum, and the piles are
from the pile–soil–raft interaction may not occur when the represented by elastic–plastic or hyperbolic springs, which
pile and raft capacities are factored downwards. As a conse- can interact with each other and with the raft. Pile–pile inter-
quence, additional calculations were carried out for actions are incorporated via interaction factors. Beneath the
geotechnical reduction factors of both less than 1 (0.6) and raft, limiting values of contact pressure in compression and
equal to 1. tension can be specified so that some allowance can be made
In addition to the normal design criteria, as expressed by for nonlinear raft behaviour. The output of GARP includes
eqs. [1] and [2], a criterion was imposed for the whole foun- the settlement at all nodes of the raft; the transverse, longitu-
dation to cope with the effects of repetitive loading from dinal, and torsional bending moments at each node in the
wind action, as follows: raft; the contact pressures below the raft; and the vertical
loads on each pile. In addition to GARP, the simplified
boundary element program deformation analysis of pile
[3] ηR*gs ≥ S *c groups (DEFPIG) (Poulos and Davis 1980) was used to ob-
tain the required input values of the pile stiffness and pile–
where R*gs is design geotechnical shaft capacity; S *c is maxi- pile interaction factors for GARP and for computing the
mum amplitude of wind loading; and η is a factor assessed overall lateral response of the foundation system (ignoring
from geotechnical laboratory testing. The value selected for the effect of the raft in this case).
η, on the basis of laboratory data from CNS tests, was 0.5. Both GARP and DEFPIG were used for the ultimate limit
The value for S *c was obtained from computer analyses, state, with undrained soil parameters for the wind loading
which gave the cyclic component of load on each pile for cases and with drained soil parameters for the dead and live
various wind loading cases. loading only cases. The pile and raft capacities were fac-

© 2005 NRC Canada

Poulos and Davids 721

Fig. 6. Computed final settlement contours for the hotel tower. Table 2. Computed maximum settlement and angu-
Contour interval: 5 mm. lar rotation ultimate limit state.
Max. settlement Max. angular
Tower (mm) rotation
Office 185 1/273
Hotel 181 1/256

Table 3. Computed maximum settlement and angu-

lar rotation serviceability limit state.
Max. settlement Max. angular
Tower (mm) rotation
Office 134 1/384
Hotel 138 1/378

collapse (or very large settlements and angular rotations) in-

dicates that the main geotechnical design criterion in eq. [2]
is satisfied: that is, the reduced foundation resistances
clearly exceed the worst ultimate limit state loadings. For
both foundation systems, the average ratio of cyclic compo-
nent of load to design shaft resistance was found to be less
than 0.5, thus satisfying the requirements of the criterion in
eq. [3] for cyclic loading.

Serviceability limit state: overall foundation

Table 3 summarizes the computed maximum settlement
tored, as discussed in the “Ultimate limit state” section and angular rotation under serviceability loading conditions
above. They were also used for the serviceability limit state, as determined by the GARP analyses. Although the com-
but the pile and raft resistances were unfactored in this case. puted values are relatively large, they nevertheless satisfy the
design criteria set out in the “Serviceability limit state” sec-
Foundation design tion, above. Thus, the overall foundation systems were as-
sessed to be satisfactory from the viewpoint of
Pile layout serviceability.
The number, depth, diameter, and locations of the founda- Figure 6 shows the contours of settlement for the hotel
tion piles were altered several times during the design pro- tower after 24 months as computed by the GARP analyses.
cess. The geotechnical and structural designers collaborated Similar settlement contours were developed for the office
closely in an iterative process of computing structural loads tower, which had a somewhat different pile layout. It can be
and foundation response. In the final design, the piles were observed that for both towers, the predicted settlements
primarily 1.2 m in diameter and extended 40 or 45 m below showed a “dishing” pattern, with the settlements near the
the base of the raft. In general, the piles were located di- centre being significantly greater than those near the edge of
rectly below 4.5 m deep walls, which spanned the raft and the foundation.
the first-level floor slab. These walls acted as “webs”, which
forced the raft and the slab to act as the flanges of a deep Raft design
box structure. This deep box structure created a relatively During the design process, the effects of raft thickness
stiff base for the tower superstructure, although the raft itself were studied, but it was found that the performance of the
was only 1.5 m thick. Figure 6 shows the foundation layout foundation was not greatly affected by raft thickness within
for the hotel tower; the piles are generally located beneath the range of feasible thicknesses considered. It was therefore
the load-bearing walls. Also shown in this figure are the decided to use a raft 1.5 m thick for the final design.
contours of predicted final settlement, which will be dis- Initially, GARP was used to obtain estimates of the largest
cussed later. bending moments and shears in the raft for any of the com-
binations of ultimate limit state loadings. Subsequently, it
Ultimate limit state: overall foundation was realized that the moments thus computed were likely to
Table 2 summarizes the maximum computed settlement be greater than the actual moments, because no account was
and angular rotation for each tower under the worst ultimate taken of the effects of the stiffness of the structure itself in
limit state loadings as determined by the GARP analyses. these calculations. Therefore, for the final assessment of raft
For the cases here, the pile and raft capacities have been re- moments and shears, the computed pile stiffnesses and raft
duced by a geotechnical reduction factor of 0.6. The calcu- contact pressures were provided to the structural engineer,
lated values of settlement and angular rotation are not who put them into a program for the complete analysis of
meaningful, but the fact that the analysis does not predict the structure and foundation. Although the settlements were

© 2005 NRC Canada

722 Can. Geotech. J. Vol. 42, 2005

generally similar to those computed by GARP, the resulting quency range of interest (up to about 0.2 Hz), dynamic ef-
values of moment and shear from the structural analysis fects on stiffness were minor, and in general the static stiff-
were significantly smaller than those from the oversimplistic ness values provided an adequate approximation of the
modelling of the raft as a uniform flat plate in the GARP dynamic foundation stiffness.
analysis. The range of frequencies was assessed to be lower than
the natural frequency of the soil profile, which was of the or-
Pile design der of 0.7–0.8 Hz. As a consequence, little or no radiation
To enable assessment of the piles from the standpoint of damping could be relied on from the piles, so all the damp-
structural design, the maximum axial force, lateral force, and ing would be derived from internal damping of the soil.
bending moment in each pile were computed by the follow- From the resonant column laboratory test data, the average
ing process: (i) the maximum axial force was computed value of the internal damping ratio was found to be about
from the GARP analyses for the various loading combina- 0.05. Following the recommendations of Gazetas (1991), the
tions; and (ii) the maximum lateral shear force and bending foundation damping ratio was taken to be 0.05 for vertical
moment were computed with the program DEFPIG, allow- and rocking motions and 0.04 for lateral and torsional mo-
ing for interaction effects among the piles but ignoring any tions.
contribution of the raft to the lateral resistance. The overall
group was analyzed under the action of the various wind Seismic hazard assessment
loadings. A seismic hazard assessment was carried out by a special-
It was found that the largest axial forces were developed ist consultant, and for a 500 year return period, the peak
in the piles near the corners and in two of the core piles. A ground acceleration was assessed to be 0.075g. Assessments
number of the piles reached their full geotechnical design re- were then made of the potential for ground motion amplifi-
sistance, but the foundation as a whole could still support cation and for liquefaction at the site. Because of the lack of
the imposed ultimate design loads and therefore satisfied the detailed information on likely earthquake time histories, the
design criterion in eq. [2]. potential for site amplification was estimated simply on the
Combined with the moments developed by the lateral basis of the site geology, related to the shear wave velocity
loading, the load on some of the piles fell outside the origi- within the upper 30 m of the geotechnical profile (Joyner
nal design envelope for a 1.2 m diameter pile with 4% rein- and Fumal 1984). On this basis, the site was assessed to
forcement, as supplied by the structural engineer. To address have a relatively low potential for amplification.
the problem of overstressing of the piles, a number of op- The presence of uncemented sands near the ground sur-
tions were considered, including increasing the reinforce- face and below the water table suggested that the possibility
ment in the 1.2 m diameter piles, increasing the number of of liquefaction during a strong seismic event. The grading
1.2 m diameter piles in the problem areas, and increasing the curves for these soils indicated that they might fall into the
diameter of the “problem” piles to 1.5 m. The second option range commonly considered to be easily liquefied. The pro-
was adopted, and for the office tower, the total number of cedure described by Seed and de Alba (1986) was used as a
piles was increased from the original 91 to 102, and for the basis for assessing liquefaction resistance with SPT data.
hotel tower, the number of piles was increased from 68 to 92. Because of the greater propensity of the calcareous sand to
generate excess pore pressures under cyclic loading, a con-
Dynamic response servative approach was adopted, in which only a small
The structural design required information on the vertical amount of fines was considered, and the design earthquake
and lateral stiffness of the individual piles in the two tower magnitude was assumed to be 7.5. The overall risk of lique-
blocks for a dynamic response analysis of the entire struc- faction was assessed on the basis of the liquefaction poten-
ture–foundation system. For the pile stiffness calculations, tial index defined by Iwasaki et al. (1984). This index
the program DEFPIG was used, with the following simplify- considers the factor of safety against liquefaction within the
ing assumptions: upper 20 m of the soil profile. On this basis, the risk of liq-
uefaction was judged to be low to very low, depending on
(i) Each pile carried an equal share of the vertical and lat-
the borehole considered. Consequently, there appeared to be
eral loads (although this might not have been an entirely
no need to consider special measures to mitigate possible ef-
realistic assumption, it did enable the stiffness of each
fects of liquefaction within the upper uncemented soil lay-
pile within the group, allowing for interaction effects, to
be evaluated straightforwardly).
(ii) The loadings from the most severe case of wind loading
were considered. Site settlement study
(iii) The loading was very rapid, so undrained conditions
prevailed in the soil profile. An assessment was made of the settlement over the entire
(iv) The pile heads were fixed against rotation to simulate site at various times after the commencement of construc-
the effect of the restraint provided by the raft. tion. This study was undertaken to facilitate the design of
(v) The dynamic stiffness of the piles in the group environ- structural interfaces between various parts of the project and
ment was equal to the static stiffness. in particular the interface between the towers and the po-
To check the latter assumption, approximate dynamic dium structure.
analyses were also undertaken, using the approach outlined The methodology involved the integration of the settle-
by Gazetas (1991), incorporating dynamic interaction factors ment due to each of the towers (considered as distributed
and dynamic pile stiffnesses. It was found that in the fre- loadings) and due to the low-rise structures (considered as a
© 2005 NRC Canada
Poulos and Davids 723

Fig. 7. Computed settlement contours around the site after Table 4. Summary of pile load tests.
24 months. Contour interval: 10 mm.
Test Diameter Length Max. test
Tower pile No. (m) (m) Test type load (MN)
Hotel P3(H) 0.9 40 Compression 30.00
P1(H) 0.6 25 Static tension 6.50a
P2(H) 0.6 25 Cyclic tension 3.25b
P2(H) 0.6 25 Lateral 0.20
Office P3(O) 0.9 40 Compression 30.00
P1(O) 0.7 25 Static tension 6.50a
P2(O) 0.7 25 Cyclic tension 3.25b
P2(O) 0.7 25 Lateral 0.20
Initial estimated value; actual value was different.
Maximum load in cyclic test = 0.5 × maximum load in static tension

For the PIGS and FLEA analyses, linear soil behaviour

was assumed. A large Excel spreadsheet was developed to
allow the summation of the effects of all 188 circular loads
and two towers assumed in the model. The settlement at 289
points over the site was computed for 6-month intervals after
the commencement of construction. A typical contour plot
series of equivalent uniform loadings) at defined points for 24 months is shown in Fig. 7.
across the site. For each of these loadings, the relationship
between settlement and distance was obtained. The follow- Pile load test program
ing procedure was developed:
(i) For the towers themselves, the settlements were avail- As part of the foundation design process, a program of
able from the GARP analyses for the serviceability pile load testing was undertaken, the main purpose being to
loadings. assess the validity of the design assumptions and parameters.
(ii) The settlements due to tower loadings that occurred at The test program involved the installation of three test piles
points outside the towers were computed with the pile at or near the location of each of the two towers. Table 4
group settlement (PIGS) computer program (developed summarizes the tests carried out. All piles were drilled under
in-house by the first author). This program uses a sim- bentonite slurry support, with steel casing being provided in
plified approach to compute the settlements both within the upper 3–4 m of each shaft. Because of the very large de-
and outside pile groups subjected to vertical loading. sign loads on the piles, it was not considered feasible to test
The PIGS program uses the equations of Randolph and full-size piles in compression, and as a consequence, the
Wroth (1978) to compute the single-pile stiffness val- maximum pile diameter for the pile load tests was 0.9 m.
ues, and the approximate approach described in Fleming Nevertheless, it will be observed from Table 4 that the two
et al. (1992) is used to compute pile interaction factors. compression tests on the 0.9 m diameter piles involved a
The Mindlin equations are then used to compute ground very high maximum test load of 30 MN.
settlements outside the loaded area.
(iii) The loads acting on the low-rise areas were modelled as Test details
a series of uniformly loaded circular areas. The com- Figure 8 shows the test setup for the 0.9 m diameter test
puter program finite layer elastic analysis (FLEA) piles. For the compression tests, the loading was supplied by
(Small 1984) was used to compute the variation of sur- a series of jacks, and the reaction was provided by 22 an-
face settlement with distance from each loaded area. chors drilled into the underlying unit 5 calcisiltite. Each an-
(iv) The rate of settlement over time for both the tower foun- chor had a diameter of 100 mm and a total length of 40–
dations and the distributed loads was calculated on the 45 m. The anchors were connected to the test pile via two
basis of two- and three-dimensional consolidation the- crowns (a larger one above a smaller unit) located above the
ory, allowing for the gradual increase of load with time jacks and load cells. For the tension tests, the reaction was
(Taylor 1948). For these calculations, a coefficient of supplied by a pair of reaction piles 12 m long, with a cross-
consolidation of 800 m2/year was assumed on the basis beam connecting the heads of the test and reaction piles. In
of field permeability tests and the assessed Young’s the lateral load tests, the test pile was jacked against the ad-
modulus values for the various layers. The rate of settle- jacent 0.9 m diameter compression test pile, the centre-to-
ment of the towers was based on the solution for an centre spacing between the piles being 4.5 m.
equivalent isolated surface circular load, 40 m in diame- For piles P2(O) and P(2)H, the cyclic tension tests were
ter, located above a compressible material 60 m deep, carried out prior to the lateral loading test. In each of the cy-
with an impermeable base layer and a free-draining sur- clic tension tests, four parcels of uniform one-way cyclic
face layer (Davis and Poulos 1972). For the low-rise ar- load were applied.
eas, the solutions for the rate of settlement of a strip Four main types of instrumentation were used in the test
foundation were used, to allow for the continuity of piles: (i) strain gauges (concrete embedment vibrating wire
loading. type), to allow measurement of strains along the pile shafts
© 2005 NRC Canada
724 Can. Geotech. J. Vol. 42, 2005

Fig. 8. Setup for axial pile load tests.

and hence estimation of the axial load distribution: (ii) rod installation of the test piles. The following programs were
extensometers, to provide additional information on axial used to make the predictions: (i) PIES, for static compres-
load distribution with depth; (iii) inclinometers (a pair, at sion and tension tests (Poulos 1989); (ii) Static and Cyclic
180°, for each pile for the lateral load tests), to enable mea- Axial Response of Piles (SCARP), for cyclic tension tests
surement of rotation with depth and hence assessment of lat- (Poulos 1990); and (iii) ERCAP, for lateral load tests (CPI
eral displacement with depth; and (iv) displacement 1992). All three programs were based on simplified bound-
transducers, to measure vertical and lateral displacements. ary element analyses that represented the soil as a layered
For the two P3 piles, 44 strain gauges were used, 4 at continuum and were capable of incorporating nonlinear
each of 11 levels, and extensometers were installed at 8 lev- pile–soil responses and considering the effects of the reac-
els; for the P1 and P2 piles, there were 32 strain gauges, 4 at tion piles. Young’s modulus for the piles was assumed to be
each of 8 levels, and extensometers at 5 levels. In general, 30 000 MPa. The input geotechnical parameters for the pre-
the strain gauges performed reasonably reliably. For the of- dictions were those used for the design, as shown in Fig. 5.
fice piles, only 3 of the strain gauges, all on P3(O), did not The program SCARP, however, required additional data on
function properly, and for the hotel piles, 13 strain gauges cyclic degradation characteristics for skin friction and end
(out of a total of 108) did not function properly: 1 on P3(H), bearing. Some indication of skin friction degradation was
4 on P2(H), and 8 on P1(H). The strain gauge readings were available from the CNS test data, but some of the parameters
generally consistent with the extensometer readings. relating to displacement accumulation had to be assessed on
the basis of judgement and previous experience with similar
deposits (Poulos 1988). It was therefore expected that the
Class A predictions predictions for the cyclic tension test would be less accurate
To provide some guidance on the expected behaviour of than those for the static tests.
the piles during the test pile program, class A predictions of
the load–deflection response of the test piles were calculated Predicted and measured test pile behaviour
and communicated to the main consultant, prior to the com-
mencement of testing. The geotechnical model was similar Compression tests
to that used for design (see Fig. 5), with some minor modifi- Comparisons between predicted and measured test pile
cations to allow for the specific conditions revealed during behaviour were made after the results of the tests were made

© 2005 NRC Canada

Poulos and Davids 725

Fig. 9. Predicted and measured load–settlement behaviour for Fig. 10. Predicted and measured axial load distribution for pile
pile P3(H). P3(H). DMD, Dubai Municipality datum.

available. Figure 9 compares the measured and predicted

load–settlement curves for test P3(H) and reveals a fair mea-
sure of agreement in the early stages. The predicted settle-
ments, however, exceed the measured values, and the Fig. 11. Predicted and measured load–uplift behaviour for ten-
maximum applied load of 30 MN exceeded the estimated ul- sion test on pile P1(H).
timate load capacity of about 23 MN. The corresponding
comparison for office tower test pile P3(O) also revealed
good agreement in the early stages, but again, the predicted
ultimate load capacity of 23 MN was exceeded. Indeed, it is
clear from Fig. 9 that the actual ultimate load capacity is
likely to be well in excess of the maximum applied load of
30 MN.
The fact that the actual capacity exceeded the predicted
value was significant, because the values of ultimate skin
friction used for the predictions were well in excess of val-
ues commonly used for bored pile design at that time in
Figure 10 shows the measured and predicted distributions
of axial load with depth for two applied load levels. The
agreement at 15 MN load is reasonable, but at 23 MN the
measured loads at depth are less than those predicted, indi-
cating that the actual load transfer to the soil (i.e., the ulti-
mate shaft friction) was greater than predicted.

Static tension tests

Figure 11 compares the measured and predicted load–dis-
placement curves for the static tension test on pile P1(H) and
indicates good agreement up to about 2 MN. At higher
loads, the actual displacement exceeded the predicted value, Figure 12 shows the values of ultimate skin friction in-
but the maximum applied load of 5.5 MN exceeded the pre- ferred from the axial load distribution measurements for
dicted ultimate value of about 4.7 MN. For the office tower both the compression and the tension tests. These values are
test pile, a similar measure of agreement was obtained, al- derived for the maximum applied test loads and are likely to
though the maximum load in that case was about 7.5 MN, be less than the actual ultimate values. Also shown are the
because the test pile had a larger diameter (700 mm) than ultimate values adopted for the design process, and these are
the originally planned 600 mm on which the predictions in reasonable agreement with the measured values; indeed,
were based. the values used for design appear to be comfortably conser-

© 2005 NRC Canada

726 Can. Geotech. J. Vol. 42, 2005

Fig. 12. Ultimate skin friction values: design values and values Fig. 13. Measured and predicted load–uplift behaviour for cyclic
derived from load tests. uplift test on pile P2(H).

Fig. 14. Measured and predicted lateral load versus displacement

for pile P2(H).

vative. It is interesting to note that the design values were

substantially larger (by about a factor of two) than the de-
sign values commonly used in the UAE prior to the project.
It appears that the CNS tests, which were used as the pri-
mary basis for selecting the design values of skin friction,
hold significant promise as a means of measuring relevant
pile skin friction characteristics in the laboratory.

Cyclic tension tests

Figure 13 shows the results of the cyclic tension test for
hotel tower pile P2(H). Four parcels of one-way cyclic load
were applied, and for each parcel there was an accumulation
of displacement with increasing number of cycles; this accu-
mulation was more pronounced at higher load levels. The
predictions from the SCARP analysis are also shown in
Fig. 13. Although the predictions at loads of <1 MN are rea-
sonable, the theory significantly underestimates the accumu-
lation of displacement at higher load levels. A similar (and
limited) level of agreement was obtained for the test on of-
fice tower test pile P2(O). It had been expected that predic-
tions of cyclic response might not be accurate, and this
expectation was borne out by the comparisons. Clearly, the the distribution of load along the pile. The approach is very
extent of displacement accumulation under cyclic loading approximate and clearly has been unable to reproduce the
was underestimated in the SCARP analysis. This analysis, real behaviour of the pile under cyclic tension loading.
based on an empirical approach originally developed by Nevertheless, from a practical viewpoint, the important
Diyaljee and Raymond (1982), assumes that the accumu- feature of the cyclic tension tests was that a load of about
lated settlement of the soil at each element of the pile is re- 50% of the static ultimate load could be applied without the
lated to the number of cycles of loading and the cyclic stress pile failing (i.e., reaching an upward displacement of the or-
level at that element. The computed settlements are imposed der of 1%–2% of diameter). However, the tests indicated
on the pile as external soil movements, and the analysis that the foundation rotations under repeated wind loading
computes the resulting accumulated pile head settlement and could be larger than predicted if the piles were subjected to a

© 2005 NRC Canada

Poulos and Davids 727

Fig. 15. Measured and predicted deflection distributions for pile Fig. 16. Measured and predicted time–settlement behaviour for
P2(H). DMD, Dubai Municipality datum. the hotel tower.

cyclic tension in excess of about 25% of the ultimate static

uplift load capacity.

Lateral load tests dicted performance of the test piles gave rise to expectations
Figure 14 shows the predicted and measured load– of similar levels of agreement for the entire tower structure
displacement curves for the hotel tower test pile. Both the foundations. Unfortunately, this was not the case. Measure-
test pile and the reaction pile responses are plotted. The ments were available only for a limited period during the
agreement in both cases is reasonably good, although there construction process, and these are compared with the pre-
is a tendency for the predicted deflections to be smaller than dicted time–settlement relationships in Fig. 16 for two typi-
the measured values as the load level increases. A similar cal points within the hotel tower. The time–settlement
measure of agreement was found for the office tower pile, predictions were based on the predicted distribution of final
although the initial prediction had to be modified to allow settlement (Fig. 6), an assumed rate of construction, and a
for the larger as-constructed diameter of the test pile. It rate of settlement computed from three-dimensional consoli-
should be noted that the predictions took account of the in- dation theory. At the time of the last available measure-
teraction between the test pile and the reaction pile. Had this ments, the tower had reached about 70% of its final height
interaction not been taken into account, the predicted deflec- (i.e., a height of about 215 m). Figure 16 shows that the ac-
tions would have been considerably larger than those mea- tual measured settlements were significantly smaller than
sured. those predicted, being only about 25% of the predicted val-
Figure 15 shows the predicted and measured deflection ues after 10–12 months. A similar level of disagreement was
profiles along the hotel tower test pile at an applied load of found for the office tower.
150 kN. The agreement is generally good, although the mea- Figure 17 shows the contours of measured settlement at a
surements indicate a reversal of direction of deflection at particular time during construction for the hotel tower. Al-
about 3.5 m depth, a characteristic that was not predicted. though the magnitude of the measured settlements is far
This characteristic has been observed from other analyses smaller than predicted, the distribution bears some similarity
(for example, Matlock and Reese 1960) and may also reflect to that predicted. The predicted ratio of final settlement at
the fact that the stiffness of the ground beyond about re- T4 to that at T15 is about 0.7, which is a similar order to
duced level (RL) –4 m was greater than assumed in the anal- that measured. Thus, despite the considerable thickness of
ysis. The sharp kink in the measured deflection profile may the raft and the apparent stiffness of the structure, the foun-
also be due to the change in stiffness caused by the transi- dation experienced a dishing distribution of settlement,
tion from a cased to an uncased pile. which is similar to that measured on some other high-rise
structures on piled raft foundations, particularly the
Measured and predicted building Messeturm in Frankfurt, Germany (Sommer 1993; Franke et
settlements al. 1994).

Comparisons during construction Possible reasons for discrepancies

The generally good agreement between measured and pre- The disappointing lack of agreement between measured

© 2005 NRC Canada

728 Can. Geotech. J. Vol. 42, 2005

Table 5. Summary of revised calculations for the hotel tower.

Ratio of max. Max. Min.
Modulus below modulus to near- settlement settlement % load
Case 53 m (MPa) pile modulus (mm) (mm) on piles
Original calculations 80 1 138 91 93
Case 2 80 5 122 85 93
Case 3 200 5 74 50 92
Case 4 600 5 40 23 92
Case 5 600 1 58 32 92

Fig. 17. Measured settlement contours for the hotel tower. Con- Fig. 18. Sensitivity of computed interaction factors to analysis
tour interval: 1 mm. assumptions. s, pile spacing; d, pile diameter.

and predicted settlement of the towers prompted a “postmor-

tem” investigation of possible reasons for the poor predic-
tions. At least five reasons were suggested:
(i) Some settlements may have occurred prior to the com-
mencement of measurements.
(ii) The time–load pattern may have differed from that as- reality, the ground between the piles is likely to be stiffer
sumed. than that near the piles, because of the lower levels of strain
(iii) The rate of consolidation may have been much slower within the rock below the pile tips is also likely to increase
than predicted. significantly with depth, because of both the increasing level
(iv) The effects of pile interaction within the piled raft foun- of overburden stress and the decreasing level of strain. The
dation may have been overestimated. DEFPIG program was therefore used to compute the interac-
(v) The stiffness of the ground below RL –53 m may have tion factors for a series of alternative (but credible) assump-
been underestimated. tions regarding the distribution of stiffness both radially and
On the basis of the information available during construc- with depth. The ratio of the modulus of the soil between the
tion, the first two of these suggested reasons did not seem to piles to that near the piles was increased to 5, and the modu-
be likely, and the last two were considered to be the most lus of the material below the pile tips was increased from
likely. Calculations were therefore carried out to assess the the original 80 MPa to 600 MPa (the value assessed for the
sensitivity of the predicted settlements to the assumptions rock at depth). The various cases are summarized in Table 5.
made in deriving interaction factors for the piled raft analy- Figure 18 shows the computed relationships between in-
sis with GARP. When the program DEFPIG was used to de- teraction factor and spacing for a variety of parameter as-
rive the interaction factors originally used, it had been sumptions. It can be seen that the original interaction curve
assumed that the soil or rock between the piles had the same (No. 1) used for the predictions lies considerably above those
stiffness as that around the pile and that the rock below the for what are considered (in retrospect) more realistic as-
pile tips had a constant stiffness for a considerable depth. In sumptions. Since the foundations analyzed contained many

© 2005 NRC Canada

Poulos and Davids 729

piles, the potential for overprediction of settlements was were made about the modulus values for soil between and
considerable, as small inaccuracies in the interaction factors below the piles, a much closer match to the measured settle-
can translate into large errors in the predicted group settle- ments was possible. The importance of taking proper ac-
ment (for example, Poulos 1993). In addition, Al-Douri and count of interaction effects in pile group analyses and of
Poulos (1994) indicated that the interaction between piles in allowing for a more realistic distribution of ground stiffness
calcareous deposits may be much lower than that between at depth was therefore reemphasized.
piles in a laterally and vertically homogeneous soil. Unfortu- The Emirates project involved close interaction between
nately, this experience was not incorporated into the class A the structural and geotechnical designers in designing piled
pile group settlement predictions for the towers. raft foundations for two complex and significant high-rise
Revised settlement calculations, on the basis of these in- structures. Such interaction has some major benefits in
teraction factors, gave the results shown in Table 5. The in- avoiding oversimplification of geotechnical matters by the
teraction factors used clearly have a great influence on the structural engineer and oversimplification of structural mat-
predicted foundation settlements, although they have almost ters by the geotechnical engineer. Such interaction therefore
no effect on load sharing between the raft and the piles. The promotes the development of effective and economical foun-
maximum settlement for case 4 is reduced to 29% of the dation and structural designs.
value originally predicted, and the minimum settlement is
about 25% of the original value. If this case had been used
for the calculation of the settlements during construction, the Acknowledgements
settlement at point T15 would have been about 12 mm after The permission of His Highness General Shaikh Moham-
11 months, which is in much closer agreement with the mea- med bin Rashid al-Maktoum to publish this paper is grate-
sured value of about 10 mm than the original predictions. fully acknowledged. Dr. James Apted provided expert advice
The importance of proper assessment of the geotechnical in relation to pile testing. Patrick Wong, Robert Lumsdaine,
model in computing the effects of group interaction has Leanne Petersen, Paul Gildea, Jeff Forse, and Strath Clarke
again been emphasized by this case history. were involved in various aspects of the field and design
work and the subsequent supervision of pile construction
Conclusions and testing. Dr. Julian Seidel carried out the CNS testing at
Monash University, Australia.
The comprehensive investigation and testing program for
the Emirates project enabled the site to be characterized
more completely than is usually possible with many pro- References
jects. Modern methods of in situ and laboratory testing were Al-Douri, R., and Poulos, H.G. 1994. Interaction between jacked
used in conjunction with advanced methods of foundation piles in calcareous sediments. In Proceedings of the 13th Inter-
analysis to design the piled raft foundations. The limit state national Conference on Soil Mechanics and Foundation Engi-
approach used for the foundation design involved a great neering, New Delhi, January 1994. Oxford Press, UK. Vol. 3,
deal of analysis (particularly because of the large number of pp. 1669–1672.
load combinations to be considered). One of the major chal- AS 2159. 1995. Piling—Design and installation (AS 2159–1995).
lenges was to process and portray the results from the analy- Standards Australia, Sydney, Australia.
ses as useful design information. CPI. 1992. ERCAP users’ manual. Coffey Partners International,
The substantial design effort was complemented by a Sydney, Australia.
comprehensive program of pile load testing, including static Davis, E.H., and Poulos, H.G. 1972. Rate of settlement under three
compression tests, static and cyclic tension tests, and lateral dimensional conditions. Géotechnique, 22(1): 95–114.
load tests. The compression tests were among the largest Diyaljee, V.A., and Raymond, G.P. 1982. Repetitive load deforma-
carried out in the Middle East and involved loads up to 30 tion of cohesionless soil. Journal of the Geotechnical Engi-
MN. Class A predictions of the performance of the test piles neering Division, ASCE, 108(GT10): 1215–1229.
were found to be in fair agreement with the measurements, Fleming, W.G.K., Weltman, A.J., Randolph, M.F., and Elson, W.K.
although generally more conservative. In particular, the val- 1992. Piling engineering. 2nd ed. Halsted Press, New York.
ues of ultimate skin friction along the pile inferred from the Franke, E., Lutz, B., and El-Mossallamy, Y. 1994. Measurements
load tests were in good agreement with the values used for and numerical modeling of high-rise buildings on Frankfurt
clay. In Vertical and Horizontal Deformations of Foundations
design, which were derived from CNS laboratory tests.
and Embankments: Proceedings of Settlement ’94, College Sta-
There appears to be potential for this type of test to provide
tion, Tex., 16–18 June 1994. Edited by A.T. Yeung and G.Y.
a rational means of measuring pile skin friction characteris- Felio. American Society of Civil Engineers, Geotechnical Spe-
tics in the laboratory. cial Publication 40, pp. 1325–1336.
The expectation that the tower settlements would be as Gazetas, G. 1991. Foundation vibrations. In Foundation engineer-
well predicted as the settlements in the load tests was not re- ing handbook. 2nd ed. Edited by H.V. Fang. Van Nostrand
alized. The measured values during construction were only Reinhold, New York. pp. 553–593.
about 25% of the predicted values. Possible reasons for the Iwasaki, T., Arakawa, T., and Tokida, K.-I. 1984. Simplified proce-
significant discrepancy were investigated, and it was found dure for assessing liquefaction during earthquakes. Soil Dynam-
that at least some of the differences could be attributed to ics and Earthquake Engineering, 3(1): 49–58.
the conservative assumptions made in deriving the pile set- Joyner, W.R., and Fumal, T. 1984. Use of measured shear-wave ve-
tlement interaction factors that were used in the piled raft locity for predicting geological site effects on strong motion. In
analysis. When more realistic (in retrospect) assumptions Proceedings of the 8th World Conference on Earthquake Engi-

© 2005 NRC Canada

730 Can. Geotech. J. Vol. 42, 2005

neering, San Francisco, Calif. Prentice-Hall, Englewood Cliffs, Poulos, H.G., and Davis, E.H. 1980. Pile foundation analysis and
N.J. Vol. 2, pp. 777–783. design. John Wiley, New York.
Lam, T.S.K., and Johnston, I.W. 1982. A constant normal stiffness Randolph, M.F., and Wroth, C.P. 1978. Analysis of deformation of
direct shear machine. In Proceedings of the 7th South East vertically loaded piles. Journal of Geotechnical Engineering,
Asian Conference on Soil Engineering, Hong Kong. pp. 805– ASCE, 104(GT12): 1465–1488.
820. Reese, L.C., and O’Neill, M.W. 1988. Drilled shafts: construction
Matlock, H., and Reese, L.C. 1960. Generalized solution for later- procedures and design methods. US Department of Transporta-
ally loaded piles. Journal of the Soil Mechanics and Foundations tion, Washington, D.C. Publication No. FHWA-H1-88-042.
Division, ASCE, 86(SM5): 63–91. Seed, H.B., and de Alba, P. 1986. Use of SPT and CPT tests for
Poulos, H.G. 1988. The mechanics of calcareous sediments. John evaluating the liquefaction resistance of sands. In Use of in Situ
Jaeger Memorial Lecture, Australian Geomechanics, Special Tests in Geotechnical Engineering: Proceedings of In Situ ’86, a
Edition. pp. 8–41. Specialty Conference, Blacksburg, Va. Edited by S.P. Clemence.
Poulos, H.G. 1989. PIES user’s manual. Centre for Geotechnical American Society of Civil Engineers, New York. Geotechnical
Research, University of Sydney, Sydney, NSW, Australia. Publication No. 6, pp. 281–302.
Poulos, H.G. 1990. SCARP user’s manual. Centre for Geotechnical Small, J.C. 1984. User’s manual: program FLEA—Finite Layer
Research, University of Sydney, Sydney, NSW, Australia. Elastic Analysis. School of Civil and Mining Engineering, Uni-
Poulos, H.G. 1993. Settlement of bored pile groups. In Deep foun- versity of Sydney, Sydney, NSW, Australia.
dations on bored and auger piles. Edited by W.F. van Impe. A.A. Sommer, H. 1993. Development of locked stresses and negative
Balkema, Rotterdam, The Netherlands. pp.103–117. shaft resistance at the piled raft foundation—Messeturm Frank-
Poulos, H.G. 1994. An approximate analysis of pile–raft interac- furt Main. In Deep foundations on bored and auger piles. Edited
tion. International Journal for Numerical and Analytical by W.F. van Impe. A.A. Balkema, Rotterdam, The Netherlands.
Methods in Geomechanics, 18: 73–92. pp. 347–349.
Poulos, H.G., and Chua, E.W. 1985. Bearing capacity of founda- Taylor, D.W. 1948. Fundamentals of soil mechanics. John Wiley,
tions on calcareous sand. In Proceedings, 11th International New York.
Conference on Soil Mechanics and Foundation Engineering, San
Francisco, Calif. Vol. 3, pp. 1619–1622.

© 2005 NRC Canada