Modern Geotechnical Design Codes of Practice Implementation, Application

MODERN GEOTECHNICAL DESIGN CODES
OF PRACTICE
Copyright © 2012. IOS Press, Incorporated. All rights reserved.
@seismicisolation
@seismicisolation
Modern Geotechnical Design Codes of Practice : Implementation, Application and Development, IOS Press, Incorporated, 2012.
Advances in Soil Mechanics and
Geotechnical Engineering
Advances in Soil Mechanics and Geotechnical Engineering (ASMGE) is a peer-reviewed book series
covering the developments in the key application areas of geotechnical engineering. ASMGE will focus on
theoretical, experimental and case history-based research, and its application in engineering practice. The
series will include proceedings and edited volumes of interest to researchers in academia, as well as industry.
The series is published by IOS Press under the imprint Millpress.
Volume 1
ISSN 2212-781X (print)

ISSN 2212-7828 (online)
@seismicisolation
@seismicisolation
Modern Geotechnical Design Codes
of Practice
Implementation, Application and Development
Edited by
Patrick Arnold
Delft University of Technology, Geo-Engineering Section, Delft, Netherlands
University of Manchester, Geo-Engineering Expert Group, Manchester,
United Kingdom
Gordon A. Fenton
Visiting Professor, Delft University of Technology, Geo-Engineering Section,
Delft, Netherlands
Dalhousie University, Department of Civil and Resource Engineering
and Department of Engineering Mathematics, Halifax, Nova Scotia, Canada
Michael A. Hicks
Delft University of Technology, Geo-Engineering Section, Delft, Netherlands
Timo Schweckendiek
Deltares, Unit Geo-engineering, Delft, Netherlands
Delft University of Technology, Section of Hydraulic Engineering, Delft,
Netherlands
and
Brian Simpson
Arup Geotechnics, London, United Kingdom
Amsterdam • Berlin • Tokyo • Washington, DC
@seismicisolation
@seismicisolation
© 2013 The authors and IOS Press.
All rights reserved. No part of this book may be reproduced, stored in a retrieval system,
or transmitted, in any form or by any means, without prior written permission from the publisher.
ISBN 978-1-61499-162-5 (print)

ISBN 978-1-61499-163-2 (online)
Library of Congress Control Number: 2012952418
Publisher
IOS Press BV
Nieuwe Hemweg 6B
1013 BG Amsterdam
Netherlands
fax: +31 20 687 0019
e-mail: order@iospress.nl
Distributor in the USA and Canada

IOS Press, Inc.
4502 Rachael Manor Drive
Fairfax, VA 22032
USA
fax: +1 703 323 3668
e-mail: iosbooks@iospress.com
LEGAL NOTICE
The publisher is not responsible for the use which might be made of the following information.
PRINTED IN THE NETHERLANDS
@seismicisolation
@seismicisolation
Modern Geotechnical Design Codes of Practice v
P. Arnold et al. (Eds.)
IOS Press, 2013
© 2013 The authors and IOS Press. All rights reserved.
Preface
Geotechnical design is a product of local history, engineering practice, availability of
construction materials and, of course, the geology of each site. All of these factors vary
from region to region, which is why a standard recipe for developing geotechnical codes
of practice does not exist. It is probably fair to say that the differences in geotechnical
design codes worldwide are much larger than exist between steel or concrete design
codes. Steel and concrete are quality controlled materials and the uncertainty in their
engineering behaviour is very similar from region to region. Thus, concrete and steel
design codes have been able to take advantage of worldwide research efforts in their
calibration over the decades.
Modern geotechnical design codes are generally striving towards a similar harmoni-
sation, both with their counterpart structural design codes and between regional geotech-
nical codes. However, harmonising geotechnical design codes is not an easy task. An
excellent example of the challenges faced in harmonisation is presented by Eurocode 7.
Although aiming to create common terms of reference, Eurocode 7 still required sev-
eral design approaches to accommodate the needs of all member states, along with na-
tional annexes enabling each member state to define their own set of safety factors. Why
was this diversity in design approaches necessary? And what do code developers have
in mind when they make their choices in adopting a design approach or set of safety
factors? Some answers to these questions will be given in this book.
The impetus for this publication started with an international workshop on Safety
Concepts and Calibration of Partial Factors in European and North American Codes of
Practice, which was held on November 30 to December 1, 2011 at Delft University of
Technology, the Netherlands. The aim of the workshop was to exchange experience and
transfer knowledge between code developers, practitioners, and researchers on code de-
velopment, safety concepts and the calibration of partial factors in modern geotechnical
codes of practice. The attendees, who were leading authorities from Europe and North
America, provided interesting and valuable insights into the development of their own
national codes. This workshop led to the idea of collecting contributions from geotech-
nical code developers worldwide into a single book, providing a resource that can be
referred to as a guide in the years to come.
The papers collected in this book are organised into three sections: Code Imple-
mentation describes choices relating to safety concepts, target reliabilities, and design
approaches; Code Application addresses their application to specific geotechnical prob-
lems; and Code Development includes papers discussing directions for future develop-
ments.
The editors would like to acknowledge the support of the following committees who
have substantially contributed to and supported this publication: the International Soci-
ety for Soil Mechanics and Geotechnical Engineering (ISSMGE) Technical Committee
for Safety and Serviceability in Geotechnical Design (TC 205), Technical Committee for
Engineering Practice of Risk Assessment and Management (TC 304), and the Comité Eu-
ropéen de Normalisation (CEN) Technical Committee for Structural Eurocodes (TC250).
@seismicisolation
@seismicisolation
This page intentionally left blank
@seismicisolation
@seismicisolation
vii
Contents
Preface v
Code Implementation
Implementation and Evolution of Eurocode 7 3

Andrew Bond
Harmonisation of Anchor Design Within Eurocode 15
Eric R. Farrell
Eurocode 7 and Polish Practice: Implementation of Eurocode 7 in Poland 25
Beata Gajewska
An Explanation of Characteristic Values of Soil Properties in Eurocode 7 36
Michael A. Hicks
Implementation of Eurocode 7 in German Geotechnical Design Practice 46
Kerstin Lesny
Implementation of Eurocode 7 in French Practice by Means of National
Additional Standards 60
Jean-Pierre Magnan and Sébastien Burlon
Implementing Eurocode 7 to Achieve Reliable Geotechnical Designs 72
Trevor Orr
Dealing with Uncertainties in EC7 with Emphasis on Determination of
Characteristic Soil Properties 87
Hans R. Schneider and Mark A. Schneider
The Safety Concept in German Geotechnical Design Codes 102

Bernd Schuppener
British Choices of Geotechnical Design Approach and Partial Factors for EC7 116
Brian Simpson
Dutch Approach to Geotechnical Design by Eurocode 7, Based on Probabilistic
Analyses 128
Ton Vrouwenvelder, Adriaan van Seters and Geerhard Hannink
Code Application
Limit State Design of the Foundations of Concrete Gravity Dams –

A Case Study 143
Laura Caldeira, Maria Luísa B. Farinha, Emanuel Maranha das Neves
and José V. Lemos
Using Numerical Analysis with Geotechnical Design Codes 157
Andrew Lees
@seismicisolation
@seismicisolation
viii
Influence of Ground Water Level on Shallow Foundation Design. Application

of EC7 Probabilistic and Deterministic Methods 171
Carlos Pereira and Laura Caldeira
Reliability Based Design of Drilled Shafts: LRFD and Performance Based
Design 183
Lance A. Roberts and Anil Misra
Application of Computational Limit Analysis in Ultimate Limit State Design 195
Colin Smith
Probabilistic Assignment of Design Strength for Sands from In-Situ Testing
Data 214
Marco Uzielli, Paul W. Mayne and Mark J. Cassidy
Experiences with Limit State Approach for Design of Spread Foundations
in the Czech Republic 228
Martin Vaníček and Ivan Vaníček
Code Development
AASHTO Geotechnical Design Specification Development in the USA 243

Tony M. Allen
Lessons Learned from LRFD Calibration of Reinforced Soil Wall Structures 261
Richard J. Bathurst, Tony M. Allen, Yoshihisa Miyata
and Bingquan Huang
Geotechnical Design Code Development in Canada 277
Gordon A. Fenton
Can We Do Better than the Constant Partial Factor Design Format? 295
Kok-Kwang Phoon and Jianye Ching
Target Reliabilities and Partial Factors for Flood Defenses in the Netherlands 311
Timo Schweckendiek, Ton Vrouwenvelder, Ed Calle, Wim Kanning
and Ruben Jongejan
Subject Index 329

Author Index 331
@seismicisolation
@seismicisolation
Code Implementation
@seismicisolation
@seismicisolation
@seismicisolation
@seismicisolation
Modern Geotechnical Design Codes of Practice 3
IOS Press, 2013
doi:10.3233/978-1-61499-163-2-3
Implementation and evolution of

Eurocode 7
Andrew BOND
Director, Geocentrix Ltd, Banstead, UK and Chairman of TC250/SC7
Abstract. The paper describes the ways in which Eurocode 7 has been
implemented by National Standards Bodies in Europe. It identifies key differences
between the 33 countries involved in their choice of Design Approach for different
foundation types and the introduction of different factors based on the design
situation being considered and the consequences of failure. Finally, the paper gives
details about the plans that are in place to develop the Eurocodes as a whole and
Eurocode 7 in particular.
Keywords. Partial factors, material strength design, load and resistance factor
design, consequences of failure
Introduction
In 1975, the Commission of the European Community (known, at the time, as the
European Economic Community, EEC), decided to create an action programme in the
field of construction, with the objective of promoting free trade between the member
states by the elimination of technical obstacles and harmonization of technical
specifications. The fruits of that programme are the suite of European Standards for
structural (and geotechnical) design known as the ‘Eurocodes’.
The story of the development of Eurocode 7 has been reported in detail by Orr
(2008) and will not be repeated here, other than to note the immense contribution of the
late Prof. Niels Krebs-Ovesen – whose ‘clear and deep understanding of geotechnical
design principles, … excellent social and diplomatic skills, and … enthusiasm and
ability to motivate people’ (Orr, loc cit.) were key to the successful development of
Eurocode 7.
1. Implementation of Eurocode 7
The European Standard for geotechnical design, EN 1997 (better known as ‘Eurocode
7’), was published by CEN, the European Standards Organization, as a full Euronorm
(EN) between 2004 and 2007, following more than 30 years of development. The
standard is divided into two parts: Part 1 (designated EN 1997-1: 2004) giving General
Rules for geotechnical design, and Part 2 (EN 1997-2: 2007) covering Ground
Investigation and Testing.
CEN’s rules regarding publication of European Standards normally require such
standards to be implemented in Member States within six months of them being made
available. Because of the complexity and interconnection of the 58 standards in the
@seismicisolation
@seismicisolation
4 A. Bond / Implementation and Evolution of Eurocode 7
Eurocode family, these rules were relaxed, thereby allowing a phased introduction of
the codes throughout Europe. A deadline of April 2010 was set for full introduction of
the Eurocodes and withdrawal of any national standards that conflicted with them. This
deadline has not been universally met and so the take up of Eurocode 7 into
geotechnical design practice in Europe has been slower than anticipated.
1.1. Choice of Design Approach for GEO/STR ultimate limit state verifications
Eurocode 7 requires ultimate limit states that involve the strength of structural
materials or the ground (so-called limit states STR and GEO) to be verified using the
inequality:
Ed ≤ Rd (1)
where Ed is the design effect of actions (e.g. bending moment, shear force, bearing
pressure, etc.) and Rd is the corresponding resistance to that effect (bending capacity,
shear capacity, bearing capacity, etc.). Reliability is introduced in a deterministic
manner, by the introduction of partial factors into this expression:
R { Fd , X d , ad }
γ E E { Fd , X d , ad } ≤ (2)
γR
where Fd, Xd, and ad represent design values of actions, material properties, and
geometry, respectively; γE and γR are partial factors on the effects of actions and
resistance; and E{…} and R{…} denote appropriate functions of the enclosed variables.
The enclosed variables are themselves modified by partial factors:
Xk
Fd = γ F Frep , X d = , ad = anom ± Δa (3)
γM
where Frep, Xk, and anom represent representative values of actions, characteristic
material properties, and nominal dimensions, respectively; γF and γM are partial factors
on the actions and material properties; and Δa is a tolerance or safety margin on the
geometry.
The particular partial factors (and their values) that must be used in a country are
specified in National Annexes to Eurocode 7 published by the various National
Standards Bodies. Eurocode 7 provides a choice between three Design Approaches as
follows:
• Design Approach 1 (DA1) requires two separate verifications to be performed

using different combinations of partial factors: in Combination 1, factors > 1.0
are applied to actions only; in Combination 2, factors > 1.0 are applied
primarily to ground strengths (except in the case of piles and anchors, for
which factors > 1.0 are applied to resistances instead)
• Design Approach 2 (DA2) requires a single verification using partial factors >
1.0 being applied to actions and resistances simultaneously
@seismicisolation
@seismicisolation
A. Bond / Implementation and Evolution of Eurocode 7 5
• Design Approach 3 (DA3) requires a single verification using partial factors >
1.0 applied to structural actions and ground strengths simultaneously
Design Approaches 1 and 3 adopt what is known as the ‘material strength design’
(MSD) method, which was introduced into European practice by Brinch Hansen (1956).
In this method, partial factors are applied (either separately or simultaneously) to loads
– called actions in the Eurocode system – and material strengths.
By contrast, Design Approach 2 is similar to the ‘load and resistance factor design’
(LRFD) method that will be familiar to users of modern American codes, such as the
AASHTO LRFD Bridge Design Specifications. In this method, partial factors are
applied simultaneously to loads and resistance.
Figure 1. Design Approaches adopted by different European countries for design of shallow foundations
Figure 1 shows the choice of Design Approach that has been made for the design
of shallow foundations by the 33 countries who are members of CEN:
• 6 countries have chosen DA1 (Belgium, Iceland, Lithuania, Portugal,

Romania, and the UK)
• 11 countries have chosen DA2 (Austria, Cyprus, Estonia, Finland, Germany,
Greece, Hungary, Poland, Slovakia, Slovenia, and Spain)
• 4 countries have chosen DA3 (Denmark, Netherland, Norway, and Sweden)
• 1 country allows a choice of DAs 1 and 2 (Italy)
@seismicisolation
@seismicisolation
• 1 country allows a choice of DAs 2 and 3 (France)

• 2 countries allow a choice of all three DAs (Czech Republic and Ireland)
• 8 countries have yet to decide – or their decision is unknown to the author
(Bulgaria, Croatia, Latvia, Luxembourg, Macedonia, Switzerland, and Turkey
– plus Malta, which is not shown the map)
The division of Design Approaches indicates an almost equal number of countries

favouring the MSD method as there are favouring the LRFD method. This balance
changes, however, when considering the design of slopes and embankments, as Figure
2 illustrates. For these foundations, there is almost universal adoption of the MSD
method (i.e. Design Approaches 1 and 3).
Figure 2. Design Approaches adopted by different European countries for design of slopes and embankments
Finally, Figure 3 shows the choice of Design Approach that has been made for the
design of pile foundations by the members of CEN. A strong majority has specified the
use of Design Approach 2 and – since in Design Approach 1 partial factors are applied
to pile resistances and not ground properties – this means there is almost universal
adoption of the LRFD method for pile foundations.
@seismicisolation
@seismicisolation
Figure 3. Design Approaches adopted by European different countries for design of pile foundations
1.2. Reliability discrimination based on design situation
The head Eurocode, EN 1990, uses the term design situation to describe ‘physical
conditions … occurring during a certain time interval for which the design will
demonstrate that relevant limit states are not exceeded’. Four design situations are
specified in EN 1990, as listed in Table 1.
Table 1. Design situations defined in EN 1990

Design Description Examples Partial factors for
situation ultimate limit state STR
γG γQ
Persistent Normal use 1.35 1.5
Transient Temporary conditions Execution or repair 1.35 1.5
Accidental Exceptional conditions Fire, explosion, impact, or 1.0 1.0
the consequences of
localized failure
Seismic When subject to seismic events 1.0 1.0
Values of the partial factors γG and γQ – which are applied to permanent and
variable actions, respectively – are summarized in Table 1 for limit state STR. EN
1990 recommends one set of values for persistent and transient design situations; and a
different set (all equal to 1.0) for accidental and seismic situations; i.e.:
@seismicisolation
@seismicisolation
(γ F , persistent = γ F ,transient ) > ( γ F , accidental = γ F , seismic )
Two countries – Austria and Germany – have decided to specify alternative

values for γG and γQ depending on the design situation being considered, with these
values obeying the general relationship:
γ F , persistent > γ F ,transient > ( γ F , accidental = γ F , seismic )
For example, in Germany, the partial factor applied to variable actions is

specified as γQ = 1.5 for persistent, 1.2 for transient, and 1.0 for accidental design
situations. Likewise, these same countries specify alternative values for the resistance
factors γR to those given in Eurocode 7 (EN 1997-1) with:
γ R , persistent > γ R ,transient > ( γ R , accidental = γ R , seismic )
For example, in Germany, the partial factor applied to vertical bearing resistance
is specified as γRv = 1.4 for persistent, 1.3 for transient, and 1.2 for accidental design
situations.
1.3. Reliability discrimination based on consequences
The head Eurocode also allows the level of reliability used for the verification of
ultimate limit states to be varied according to ‘classes of consequence’, as summarized
in Table 2.
Table 2. Consequence classes and their associated reliability indices, as defined in EN 1990
Consequence Description Examples Associated minimum KFI

Class reliability index, β,
for given reference
period
1 year 50 years
CC3 High consequence for Grandstands, public 5.2 4.3 1.1
loss of human life, or buildings where
economic, social or consequences of failure
environmental are high (e.g. a
consequences very great concert hall)
CC2 Medium consequence for Residential and office 4.7 3.8 1.0
loss of human buildings, public
life, economic, social or buildings where
environmental consequences of failure
consequences are medium (e.g. an
considerable office building)
CC1 Low consequence for Agricultural buildings 4.2 3.3 0.9
loss of human life, where people do
and economic, social or not normally enter (e.g.
environmental storage
consequences small or buildings), greenhouses
negligible
@seismicisolation
@seismicisolation
One way of introducing reliability discrimination is to introduce an extra partial

factor KFI into the calculation of design actions, such that:
Fd = K FI × γ F × Frep (4)
with the values of KFI given in Table 2. Five countries – Austria, Denmark, Finland,
the Netherlands, and Sweden – have decided to adopt this approach. So, for example,
in these countries the partial factor applied to variable actions is γQ = 1.35 for CC1, 1.5
for CC2, and 1.65 for CC3.
This method of reliability discrimination does not work for the design of slopes
and embankments, where the governing parameter is the strength of the ground, rather
than imposed loads. Three of the five countries listed above – Austria, Denmark, and
the Netherlands – have therefore extended this concept to the calculation of design
strengths, by increasing or decreasing the value of γM by a factor akin to KFI:
Xk
Xd ≈ (5)
K FI × γ M
For example, in the Netherlands, the partial factor applied to the coefficient of
shearing resistance (i.e. tan ϕ) is specified as γϕ = 1.2 for CC1, 1.25 for CC2, and 1.3
for CC3.
1.4. Summarizing reliability discrimination
Figure 4 indicates which countries have chosen different values for partial factors on
actions, material properties, and resistance to account for an increase or decrease in risk
brought about by the either a) the design situation and/or b) the consequence of failure
(KFI). In total, 7 of the 33 countries within CEN have chosen to do this (just over 20%).
1.5. Other changes to partial factors for ground properties

Several countries have chosen to adopt slightly different values for the partial factors to
be applied to ground properties in their jurisdiction. The motivation and justification
for this is beyond the scope of this paper, but, to give an idea of the changes that have
been made, Table 3 summarizes the values specified for slope design in selected
countries’ National Annexes. There is a slight tendency for γϕ to be reduced from the
recommended value given in Eurocode 7 and for γcu to be increased.
1.6. Some lessons learnt from implementation of Eurocode 7
The preceding review of the implementation of Eurocode 7 in various countries

throughout Europe reveals considerable variation from the ‘template’ provided by the
code for the level of reliability required in geotechnical design. It is clear that countries
have chosen to preserve their existing national practice as far as they can and have
adapted the Eurocode rules in order to do so. The challenge going forward will be to
explain the reasons for the apparent differences in national practice and to reduce these
differences where they are based on outdated experience.
@seismicisolation
@seismicisolation
Figure 4. Countries where partial factors vary according to the consequence of failure and/or design situation
Table 3. Values of partial factors that differ from Eurocode 7’s recommended values
Standard/Country Design Approach adopted for Partial material factors for design of slopes
slope design in persistent design situation, CC2
γϕ γc γcu
Eurocode 7 Values given for DA3 1.25 1.25 1.4

Austria DA3 1.15 1.1 1.25
Denmark DA3 1.2 1.2 1.8
Finland DA3 1.25 1.25 1.5
Germany DA3 1.25 1.25 1.25
Netherlands DA3 1.25 1.45 1.75
Portugal DA3 1.1 1.1 1.15
Sweden DA3 1.3 1.3 1.5
Underlined values differ from the recommended values given in Eurocode 7
2. Evolution of Eurocode 7
In May 2010, the European Commission issued a ‘Programming Mandate’ M/466 to

initiate further development of the Structural Eurocodes. This Mandate invited CEN,
the European Standards Organization, to submit proposals for creating new Eurocodes
and evolving existing Eurocodes. Responsibility for preparing CEN’s response to the
Mandate was delegated to CEN Technical Committee TC250.
@seismicisolation
@seismicisolation
2.1. Response to the Mandate
CEN’s initial response to the Commission’s Programming Mandate proposed new

Eurocodes for:
1. design of structural glass
2. structural use of fibre-reinforced polymers
3. design of membrane structures1
CEN’s response also proposed various improvements to existing Eurocodes, including:

1. reducing the number of Nationally Determined Parameters (NDPs)
2. incorporating recent research on innovation (e.g. performance-based and
sustainability concepts)
3. incorporating recent research on sustainability
4. simplifying rules for limited and well identified fields of application
Nationally Determined Parameters allow the Member Countries of CEN to decide

on their own safety levels and give national geographic and climatic data via National
Annexes to the appropriate Eurocodes. The use of NDPs and the publication of
supporting documents and standards (so-called ‘non-contradictory complementary
information’ or ‘NCCI’) have been far more extensive than was originally envisaged. A
priority for the evolution of the Eurocodes is to reduce national alternatives and thus
aid simplification.
2.2. How the evolution of Eurocode 7 is being organized
At its meeting in Cambridge, UK, in March 2011, TC250/SC7 – the subcommittee that
is responsible for the maintenance and development of Eurocode 7 – decided to
establish a number of ‘Evolution Groups’ to prepare for the ‘second generation’ of
Eurocode 7. The brief for each Evolution Group (EG) was to prepare the best possible
advice to allow SC7 to make the necessary changes, additions, and/or deletions to
Eurocode 7 to meet the aims set out in CEN’s response to the Mandate. The Evolution
Groups that were established – with one or two amendments – are shown in Figure 4.2
As Figure 5 shows, the Evolution Groups can be divided into four broad groups.
EGs 2 and 8 are concerned mainly with editorial matters, such as simplification and
harmonization of rules throughout both parts of the code; EGs 4, 6, 9, 10, and 11 are
concerned with generic technical subjects that are independent of foundation type;
while EGs 1, 5, 7, 13, and 14 deal with specific technical subjects which generally have
their own sub-section with EN 1997-1 (reinforced soil and rock mechanics are the
exceptions to this – but plans are in place to change this during evolution of the code).
EG3 is unusual amongst this company, since its purpose is to develop worked
examples based on the current Eurocode text, to establish current best practice. This
group is continuing the work of ETC10, the European Technical Committee
established by the International Society of Soil Mechanics and Geotechnical
Engineering to research the implementation of Eurocode 7. Finally, EG0 is an over-
1
At the time of writing, it seems likely that European Commission will only fund the creation of a new
Eurocode for the first of these subjects and not for the other two
2
Creation of EG12 Tunnelling is pending; EG14 was created after the March 2011 meeting of SC7
@seismicisolation
@seismicisolation
arching group whose task is to coordinate the work of all the other groups and to act as
a single link to the main technical committee TC250/SC7.
Figure 5. SC7’s Evolution Groups

A micro-site has been setup to support the work of these Evolution Groups at
www.eurocode7.com/sc7/evolutionsgroups.html, where you will find the names of the
delegates who are working in each group.
2.3. Maintenance and simplification
The Structural Eurocodes have generally been regarding as complicated documents that
are not easy for designers to use. In its response to Mandate M/466, TC250 stated:
The Eurocodes are the result of a consensus between many European experts after
enormous exposure to examination. The Eurocodes are technically advanced standards
which are intended to allow the design of most structures that are likely to be needed
now and in the future. Accordingly they are very comprehensive, which can lead to
their appearing to be more complicated than is necessary when a limited range of types
of structure is to be designed.
@seismicisolation
@seismicisolation
In view of the above, TC250 agreed to ensure simplification in the future

development of the Eurocodes by:
1. improving clarity
2. simplifying routes through the Eurocodes
3. limiting, where possible, the inclusion of alternative application rules
4. avoiding or removing rules of little practical use in design
To date, EG2 has concentrated its efforts on improving the contents of Eurocode 7
Part 2, which covers ground investigation and testing and has been accused of
including too much ‘text book’ material that would be better left out of the code. The
EG is also attempting to remove as much duplication as possible between Parts 1 and 2,
in order to simplify both documents.
2.4. Harmonization
In many ways, EG8 has the most important and most difficult tasks of all the Evolution
Groups – the reduction in national variation introduced via the National Annexes to
Eurocode 7. This task, however, is the European Commission’s number one priority in
commissioning evolution of the Eurocodes, since national variation is perceived as a
barrier to trade.
EG8 is investigating ways in which the need for different Design Approaches
introduced in EN 1997-1 might be avoided. The almost universal adoption of material
strength (MSD) design for slopes and embankments and load and resistance factor
design (LRFD) for piles suggests that progress towards this goal could be achieved on
a case-by-case basis, considering each foundation type in turn.
There appears to be a strong case for introducing reliability discrimination into
Eurocode 7, taking the best ideas from (mainly Scandinavian) practice. One proposal
under consideration is to ‘build up’ the value of the partial factors applied to ground
properties γM from various elements, like this:
γ M = γ M ,basic × kCC × k DS
where γM,basic would be the base value for γM for consequence class CC2 in persistent
design situations. The factor kCC would then modify γM for different consequence
classes; and kDS would do likewise for different design situations. Table 4 gives some
proposed values for these factors, based on the review of national practice described in
the first part of this paper. (The values given in Table 4 reproduce to a large extent the
specific values of γM already adopted in those countries which have already introduced
reliability discrimination, as indicated on Figure 4.)
Table 4. Some proposed values for the modifiers to be applied to basic partial material factors
Modifier for… Symbol Specific values
Case Symbol Value
Low consequence (CC1) kCC1 0.95 = 1/ 1.05
Consequence Class kCC Medium consequence (CC2) kCC2 1.0
High consequence (CC3) kCC3 1.1
Persistent kpers 1.0
Transient ktran 0.95 = 1/ 1.05
Design Situation kDS
Accidental kacc 0.91 = 1/ 1.1
Seismic kseis 0.91 = 1/ 1.1
Underlined = ‘default’ consequence class and design situation
@seismicisolation
@seismicisolation
3. Conclusion
If the European Commission’s goal of free trade between member states in the field of
construction is to be achieved, then technical obstacles – such as a choice of Design
Approaches, a multitude of Nationally Determined Parameters, and further national
rules presented as non-contradictory complementary information (NCCI) – must be
reduced or eliminated. The best way to achieve this is to understand the purpose of
these national variations – and to provide alternative means of satisfying that purpose.
References
Bond, A.J. and Harris, A.J. (2008). Decoding Eurocode 7, Taylor and Francis, London.
Brinch Hansen, J. (1956). Limit design and safety factors in soil mechanics, Bulletin No. 1, Danish
Geotechnical Institute.
Orr, T.L.L. (2008). The story of Eurocode 7, 14th European Conference on Soil Mechanics and Geotechnical
Engineering.
@seismicisolation
@seismicisolation
IOS Press, 2013
doi:10.3233/978-1-61499-163-2-15
Harmonisation of Anchor Design within

Eurocode
Eric R FARRELL a,1
a
AGL Consulting and Department of Civil Structural and Environmental Engineering,
Trinity College, Dublin, Ireland
Abstract. The harmonisation of anchor design within the limit state framework
of the Eurocode presented a challenge due to the different design and testing
practices in use in the various countries. An Evolution Group (EG1) was set up by
TC250/SC7 tasked with developing a standard that conformed with the limit state
philosophy and which could be used in the member countries. This paper presents
the basis of a draft standard that has been prepared by EG1, which considers the
ULS and SLS of the restrained structure and of the anchor itself and introduces the
force FServ.k which considers the effect of prestress on the anchor force. The
application of the recommendations is illustrated by a design example
Keywords. Anchor, Eurocode.
Introduction
The development of a European standard for anchor design presented two major
challenges, namely developing a standard for the design of anchors that conformed
with the philosophy of the limit state culture of the Eurocode and secondly
accommodating the different approaches to anchor design which are used within the
various countries. It had been recognized that Section 8:Anchorages of the current
Eurocode (EN1997-1:2004) did not adequately cover the design of anchors and an
Evolution Group (EG1) was set up by TC250/SC7 tasked with revising this section.
The members of this group are listed in the acknowledgements and whilst the views
expressed in the paper are solely those of the author, the experience and knowledge of
the members was a major source of information.
The Evolution Group has been very active over the last year or so and has
produced a draft document for circulation within the membership of TC250/SC7. This
document has been taken as the basis of this paper and will be referred to as Sec 8:July
2012 in the text. Also, the document EN1997-1 ‘Geotechnical Design-Part 1’ will be
abbreviated to EC7.
The design of anchors is inextricably linked to their construction (execution) and
to the methods used to test the anchors. EN1537:1999 ‘Execution of special
geotechnical work-Ground Anchors’, which is currently being revised by CEN/TC288
WG14, standardizes the execution procedures and the material properties required for
grouted anchors. EN1537 is at the final stages of a review process and comments on
prEN 1537:2011 have been received by TC288/WG14 and this execution standard will
1
Corresponding Author.
@seismicisolation
@seismicisolation
16 E.R. Farrell / Harmonisation of Anchor Design Within Eurocode
be updated in the near future. No standard is currently available on the testing of

anchors, however prEN ISO 22477-5 ‘Geotechnical Investigation and Testing –Testing
of geotechnical structures – Part 5:Testing of anchorages’ has been circulated for
comment and hopefully will be developed into an EN in the near future.
This paper summarises the limit state design process set out in Sec 8:July 2012 and
the adaptation of that standard to include the variety of design methods that are used in
the various countries. The application of the recommendations is illustrated by a design
example using the partial and correlation factors given in that draft standard.
1 An Anchor
The term anchor is generally used to cover such installations as anchor walls (also
called deadman anchors), grouted prestressed anchors, non-prestressed anchors
(sometimes called passive anchors), duck-bill anchors, and rock anchors, among others.
The feature that distinguishes an anchor from other tension members, such as a
micropile, is that an anchor transfers a load from the anchor head to a remote stratum
using a free length. In order to cover a broad range of anchors the definition adopted in
Sec 8:July 2012 is an ‘Installation capable of transmitting an applied tensile load
through a free length to a load bearing stratum’. Ideally all of the different types of
anchors should be covered in this section of EC7, however, whilst Sec 8:July 2012
covers the basic principles of the design of all anchors it is mainly directed at ‘grouted
anchors’. The designer is referred to Section 9 ’Retaining structures’ for the design of
anchor walls.
A grouted anchor is defined as ‘An anchor that uses a bonded length formed of
cement grout, resin or similar material to transmit the tensile force to the ground.’ It
should be noted that EN1537 is limited to grouted anchors only and the use of the term
‘ground anchor’ in that standard refers solely to that type of anchor.
The presence of a free tendon length means that an anchor can a) be prestressed to
minimize deformations in a structure or supporting structure and b) that generally it can
readily be load tested using a jack, thus avoiding the need for expensive reaction beams
that are required to test tension piles in order to take the reaction away from the tested
element. These tests can assess, not only the load carrying capacity of the anchor tested,
but also that the bonded length is correctly located by comparing the measured
extension under load with that calculated using the designed free length. The maximum
test load to which each anchor is subjected (Proof load) has implications on the value
of the partial factors that can be adopted in design.
One aspect of the anchor standardization which the author considers is poorly
addressed in the current draft of Sec 8:July 2012 is the grey area between the definition
of an anchor and a micropile. The requirement in Sec 8:July 2012 that all anchors be
subjected to acceptance tests, including non-prestressed anchors, does not give a
smooth transition with the testing requirements of micropiles and could cause
confusion should the latter be given free length for a special reason, such as to prevent
downdrag or to allow for ground expansion.
@seismicisolation
@seismicisolation
E.R. Farrell / Harmonisation of Anchor Design Within Eurocode 17
2 Limit states in anchor design
This section presents a summary of the limit state design of anchors to Sec8:2012. The
standardization of anchor design within the limit state framework must consider:-
Ultimate Limit States (ULS) and Serviceability Limit States (SLS) of the
anchor, together with ULS and SLS of the supported structure.
Prestress forces and the effect of prestress forces, where relevant.
2.1 Ultimate limit state (ULS) design force to be resisted by an anchor
Anchors are required to have the capacity to resist not only the force required to
prevent an ULS in the structure and supporting structures, but also, they must have the
capacity to resist the maximum force that could be transferred to the anchor during its
service life (FServ,k) with an adequate margin of safety. The ULS resistance of an anchor
(RULS;d) must be capable of resisting the ultimate limit state (ULS) design force (E ULS,d).
EULS,d is the greater of the force required to prevent any ULS in the supported structure
(FULS,d) and FServ,d which is the design value of the maximum anchor force expected
within the design life of the anchor, including effect of lock off load, and sufficient to
prevent a SLS in the supported structure. These requirements are expressed in Eqs. (1)
to (3) where γServ is a partial factor:-
RULS;d ≥ EULS,d (1)

where E ULS,d = Max(FULS,d; FServ,d) (2)
and FServ,d = γServFServ,k (3)
All the load conditions that would apply during the life of the anchor must be
considered when deriving the design value FULS,d, and where relevant, the limit states
considered must include those that could arise from any prestressing force which may
be applied.
The introduction of FServ,k is a novel concept within Eurocode, but is required to
ensure that an anchor has the required SLS and ULS resistance under the maximum
working load likely to be experienced over its working life. It is similar to the concept
of a ‘service load’ used in the French practice and to the ‘working load’ used in the UK.
It is needed because anchors may well sustain in service forces that are bigger than the
force required to prevent an ULS in the structure and supporting structures. Where a
prestress force is required in an anchor to prevent a SLS in the structure, F Serv,k would
include the effect of that prestress and any increase that may subsequently occur if
staged construction is involved. Depending on the value of γServ adopted, the value of
FServ,d in some design situations may exceed FULS,d. Consequently, where applicable, the
anchor must be designed to resist this higher value with a sufficient margin of safety.
Some countries, for example Germany, do not require FServ,d as the prestress force
is considered when determining FULS,d.
@seismicisolation
@seismicisolation
2.2 Serviceability limit state (SLS) design force to be resisted by an anchor
A separate validation of the SLS of an anchor is not required in some countries as it is

considered to be satisfied by the margins in the ULS calculation, i.e by Eq.1. When
stated in the National Annex of a country, as well as satisfying Eq.1, an anchor shall be
designed to ensure that it has the required design SLS resistance (R SLS;d) such that the
appropriate SLS limiting creep or load loss criteria are not exceeded at FServ,k, with an
appropriate margin of safety, i.e Eq.4 must be satisfied, where γF,SLS is a partial factor
(normally equally to unity). France and the UK have such requirements but this
calculation is not required in Germany.
RSLS;d ≥ γF,SLS FServ,k (4)
2.3 Geotechnical ultimate limit state anchor resistance
The design value of the geotechnical ultimate limit state anchor resistance (R ULS;d) can
only be obtained from values measured (RULS;m) in investigation and/or suitability tests
on anchors. The assessment of measured values depends on the particular criteria used
in each country and is discussed in further detail in Section 4. The design value is
determined from the measured values using Eqs.5 and 6.
RULS;k = RULS;m,min / ξULS (5)
where RULS,m,min is the lowest value of RULS;m measured in investigation and/or

suitability tests, for each distinct condition of the ground and structure, RULS;k is the
characteristic value of the ULS anchor resistance and ξULS is a correlation factor. Also,
RULS;d = RULS,k / γa,ULS (6)
where γa,ULS is a partial factor.

2.4 Geotechnical serviceability limit state anchor resistance
In those countries that consider the SLS of an anchor in their design, the design value
of the geotechnical serviceability limit state anchor resistance (R SLS;d) can only be
obtained from values measured (RSLS;m ) in investigation and/or suitability tests on
anchors in a similar manner to that adopted for ULS design, which is:-
RSLS;k = RSLS;m,min / ξSLS (7)
RSLS,d = RSLS,k / γa,SLS (8)
Where RSLS;k is the characteristic value of the SLS anchor resistance, RSLS,m,min is the
lowest value of RSLS;m measured in investigation and/or suitability tests, ξSLS a
correlation factor and γa,SLS is a partial factor which would commonly have a value of
unity.
@seismicisolation
@seismicisolation
3 Determining RULS,m and RSLS,m
Sec 8:July 2012 requires that the geotechnical resistance of an anchor be determined
from anchor tests, however such tests must be taken to a Proof load (PP) which
confirms that the anchor has sufficient resistance under a specified load sequence and
time hold periods. In order to satisfy Eq.1 using Eqs.5 and 6, the Proof load in
investigation and/or suitability tests must satisfy Eq.9.
PP ≥ ξULS γa,ULS EULS,d (9)
The difference between investigation and suitability tests is mainly that

investigation tests are carried out on a sacrificial anchor (sometimes called preliminary
anchors) and are designed to be taken to a failure criterion, whereas suitability tests are
carried out on working anchors and not necessarily to failure. The definition of an
investigation test in the standard is ‘Load test to establish the geotechnical ultimate
resistance of a ground anchor at the grout/ground interface and to determine the
characteristics of the anchor in the working load range’. Such tests are typically taken
to failure of the ultimate grout/ground interface, possibly using a stronger tendon than
is required for a working anchor. The definition of a suitability test is ‘Load test to
confirm that a particular anchor design will be adequate in particular ground
conditions’. Suitability tests are not generally expected to be taken to a failure criterion
of an anchor but must show that the anchor has sufficient resistance to satisfy the
requirements of Eq.1.
Sec 8:July 2012 requires a minimum number of investigations/suitability tests to
be carried out, but the number required can be set in the National Annex. Typically an
anchor design should be based on a minimum of two or three such tests.
Different anchor test methods are in use in Europe and different limiting criteria
are interpreted from these tests. Not all ULS criteria adopted could be said to be a true
‘pull-out resistance’ of an anchor. This has significance when it comes to the level to
which an acceptance test can be taken. An acceptance test is used to verify the
resistance of each individual anchor and is discussed in Section 4.
The interpretation of anchor capacity from load tests is based on the relationship
between load and creep rate or load loss rather than the load displacement relationship
which is used for piles. The creep rate is determined from the displacement/time
measurements where the load is maintained or from the loss of load/time measurements
where the displacement of the anchor head is held constant.
Three test methods are recognised in pr EN ISO 22477-5 and are referred to as
Test Methods 1 to 3 (TM1, TM2 and TM3). An indication of the measurements to be
made in these anchor tests and the type of loading regime applied can be got from those
required for Investigation tests, which can be taken to failure, and which are
summarised in Table 1.
The definition of RULS,m and RSLS,m adopted in Sec 8:July 2012 states that the limit
state values of these resistances are those complying with the relevant ULS or SLS
criteria. This wording has been carefully chosen to avoid the difficult task of defining
the ultimate failure or serviceability condition and allows each country to continue with
current practice. The provisional limiting criteria for the three test methods in Annex 8
of Sec8:July 2012 are given in Table 1.
@seismicisolation
@seismicisolation
The measured ultimate resistance is taken as that anchor load that induces the
limiting condition, or the Proof load applied (PP) if that criterion has not been reached
i.e.
RULS;m = Min { Rm (αULS or kl ULS) and PP} (10)

Note: α and kl defined in footnote to Table 1
A similar approach is taken in the determination of RSLS;d, for those countries

which consider a SLS resistance of an anchor.
Table 1. Summary of anchor test methods 1 to 3 for investigation tests.
Test method TM1 TM2 TM3

Type of loading Cycle loading Cycle loading In steps
Rest periods Maintained loads Maintained Maintained load

deflection
Tendon head Load-loss vs time at Anchor head
displacement the highest load of displacement vs
vs applied load at end each cycle anchor load at the
of each cycle beginning and end of
each load step.,
Measurements
Tendon head k1 versus anchor load Anchor head
displacement vs time displacement vs time
for each load step.
α versus anchor load α versus anchor load

or bond load, if
possible.
Displacement vs load
for all cycles
SLS
Limiting values NA k1,SLS =2% per log Pc

in Sec8:July cycle of time
2012
ULS
Limiting values α1,ULS = 2mm k1,ULS =2% per log α3,ULS =5mm
in Sec8:July cycle of time.
2012
Note:-
α = slope of ‘creep displacement versus logarithm of time.
k1= load loss expressed as percentage of applied load;
NA= Not Applicable
@seismicisolation
@seismicisolation
4 Acceptance testing of anchors
The values of the partial and correlation factors used in the design of anchors is
inextricably linked to the degree to which the capacity of each working anchor is
verified in an acceptance test. Lower partial factors can be used in the design of
anchors when the Proof load is such that RULS,d ≥ EULS,d is verified than if the test were
taken to a lower proof load. Some countries relate PP of acceptance tests to the ‘service
load’ or ‘working load’ and the degree to which RULS,d is verified in such cases has to
be considered.
Sec 8:July 2012 requires that an acceptance test is carried out on every anchor and
permits countries to select whether PP is based on EULS;d or FServ,k from Eqs.11 & 12:-
PP ≥ ξa,acc,ULS γa,acc,ULSEULS;d (11)
or
PP ≥ ξa,acc,SLS γa,acc,SLS, γF,SLS FServ,k (12)
where ξa,acc,ULS ; ξa,acc,SLS are correlation factors
γa,acc,ULS; γa,acc,SLS are partial factors
Germany, for example, uses Eq.11 and requires acceptance tests to be loaded to
the same PP as investigation/suitability tests and to satisfy the same criterion (α1=2mm),
albeit which shorter holding periods. France on the other hand, load the anchors to a
lower PP for acceptance tests which is 1.25 FServ,k for permanent anchors but requires an
anchor to satisfy a creep criterion that is less than for a ULS condition. The results of
tests taken to failure, which are mandatory in France, are used together with their
experience of anchor performance, to be satisfied that anchors that comply with this
criterion satisfy the ULS requirements.
The current draft of Sec 8:July 2012 does not address the design of anchors
should one of the anchors fail an acceptance test, even if only marginally. Should this
anchor test be incorporated into the design process of all the anchors or should it be
essentially ignored and treated as an outlier.
5 Discussion
The application of the principles of Sec 8:July 2012 can best be illustrated using an
example of the design of an anchored sheetpile wall as illustrated on Figure 1. Taking
an anchor spacing of 3m and using a simple limit equilibrium analysis (LEM) and the
free earth anchor support method, a design value of the ULS force (FULS,d) of
270kN/anchor is determined using Design Approach 1. Interestingly, Germany uses a
different calculation approach that determines a characteristic permanent (F G,k) and
(FQ,k) variable anchor components and FULS,d using Eq.13. Any prestress force is
included in the determination of FULS,d in Germany.
@seismicisolation
@seismicisolation
FULS,d = 1.35 FG,k +1.5 FQ,k .. (13)
As there is no particular requirement on serviceability in this example, there is no need

to apply a particular prestress into the anchor at lock-off. Typically the prestress force
may be related to FServ,k in those countries which determine that force.
Figure 1. Design example of quay wall.
FServ,k must be determined in those countries which elect to consider this force in
the design of anchors. In practice for the simple design situation of the Quay Wall
given in Figure 1, the value of FServ,k is frequently estimated by carrying out a simple
LEM analysis using characteristic values of actions and soil parameters/resistances and
using the length of the sheetpile wall required for equilibrium with these forces, i.e.
with a sheet pile shorter than required for ULS. For the Quay Wall example, this would
give FServ,k = 158 kN/anchor, provided it is not prestressed to a higher force. The design
value of this force for ULS design is obtained from Eq.3. Using γserv=1.35, then FServ,d =
213.3 kN/anchor for use in Eq.2. The SLS resistance of an anchor (RSLS,d) would be
determined from Eq.4 using γF,SLS of 1.0. Note that a different value of FServ,k would be
obtained from a finite element method, the actual value being dependent on the soil
model and the sequence of loading adopted.
@seismicisolation
@seismicisolation
In summary, RULS,d ≥ EULS = Max(270 , 213.3) = 270kN/anchor
and , if required in the NA, RSLS,d = 1.0x158kN/anchor = 158kN/anchor
As stated previously, Sec 8:July 2012 requires that anchors be designed from
investigation/suitability tests, and with the provisional values given in Annex 8 of a
minimum of 3 if test methods 1 & 2 are adopted or a minimum of 2 if test method 3 is
used. Irrespective of the test method to be employed, and using the provisional values
given in Annex 8 (γa,ULS = 1.1, ξULS = 1.0 for acceptance tests on all anchors) in order to
verify the ULS resistance, the Proof load in investigation/suitability tests (I/S tests) in
all test methods must be taken to be:-
PP (I/S tests) ≥ ξULS γa,ULS EULS,d = 1.0x1.1x270 = 297 kN
The value of PP applied in acceptance tests depends on the test methods selected
for use in each country. For instance, in Germany, PP is based on EULS,d (Eq.11) and
TM 1 (γa,acc,ULS = 1.1, ξa,acc,ULS =1.0), which gives:-
PP (acceptance TM1) ≥ 1.0x 1.1x270 = 297 kN and must meet a ULS criterion
The Proof load for acceptance tests in France is based on FServ (Eq.12) and uses TM3
(γa,acc,SLS = 1.25, ξa,acc,SLS =1.0, γF,SLS =1.0 ), which for permanent anchors gives:-
PP (acceptance TM3) = PP ≥1.0x1.25x1.0x158 = 198kN but must meet a creep

criterion less stringent than that for a ULS condition.
When using TM3 for acceptance tests, the confidence that each anchor has the
required RULS,d relies on a combination of the database/experience from results at this
level of Proof load, and from the results of the investigation tests which are mandatory
in France.
It is also a requirement that the apparent free length of a grouted anchor comply
with the requirements in EN 1537.
The question of differing value of PP in acceptance tests for temporary and
permanent anchors has not been addressed in Sec8:July2012 but could be covered in
the National Annex of each country. While a definition of a temporary anchor is given
(anchor with a design life of 2 years or less), it is considered by some that this relates to
corrosion protection and that the partial and correlation factors for acceptance tests
would more correctly be related to the consequences of a failure and amount of
monitoring employed, rather than to a time interval.
6 Conclusion
The current draft of a standard for anchor design (Sec8:July2012 ) that has been
prepared for circulation with SC7 of TC250 presents a rational and comprehensive
framework for the limit state design of an anchor that considers the ULS and SLS limit
states of the structure and of the anchor.
@seismicisolation
@seismicisolation
The preparation of the standard on anchors that harmonizes the European

experience has been difficult owing to the varying practices that have developed over
the years in the different countries. These differences relate to the methods used to
determine the force for which the anchor design is based and on the criteria used to
assess the anchor performance from testing. It was considered necessary to introduce a
force, FServ,k , which considers the effects of prestress on the anchor force, as well as the
anchor force required to ensure that SLS requirements of the structure and supported
structure.
Many of the anchor testing methods in use at present have been developed for a
different design philosophy and there is scope for reviewing the practices to make them
more consistent with that of a limit state design method.
Acknowledgements
The author would like to acknowledge the information exchanged in EG1 with
members Klaus Dietz, Yves Legendre, Caesar Merrifield, Ole Møller, Pierre Schmitt,
Bernd Schuppener, Arne Schram Simonsen and Brian Simpson. The author would also
like to thank his colleagues Conor O’Donnell and David Gill for their assistance.
References
EN 1537:1999, ‘Execution of special geotechnical work – Ground anchors’

prEN 1537:2011 Execution of special geotechnical work – Ground anchors’
EN1997-1:2004 Eurocode 7:Geotechnical Design-Part 1:General Rules.
prEN ISO 22477-5 ‘Geotechnical investigation and testing – Testing of geotechnical structures – Part
5:Testing of pre-stressed ground anchors.
@seismicisolation
@seismicisolation
IOS Press, 2013
doi:10.3233/978-1-61499-163-2-25
Eurocode 7 and Polish Practice:

Implementation of Eurocode 7 in Poland
Beata GAJEWSKA1
Road and Bridge Research Institute, Poland
Abstract. In Poland, designing with the use of limit states and partial factors has
been a common practice for over 30 years. However, there are many differences
between Eurocode 7 and Polish practice and previous codes. The paper presents
some of these differences. The state of Eurocode 7 implementation in Poland is
presented in general. Eurocode gives opportunity to choose one of the three Design
Approaches. The reasons of the choice of the Design Approaches in Poland are
presented. The Polish National Annex to EC7, as well as the Polish EC7 guide, are
discussed. Reliability issues and the National TC Geotechnics’s structure
reliability approach are also raised.
Keywords. Eurocode 7, Polish Standards PN-B, Polish practice, EC7

implementation, limit state design, partial factors
Introduction
In Poland, the designing with limit states and partial factors was introduced in 1974.
Foundations, retaining walls and similar structures were designed on the basis of a set
of Polish Standards (PN-B) of design. The main of these standards include:
• PN-B 03020:1981 Shallow foundations. Geotechnical design,
• PN-B 02482:1983 Bearing capacity of piles and pile foundations,

• PN-B 03010:1983 Retaining structures. Geotechnical design.
These standards are presently not harmonized with Eurocode 7.
The PN-B standards generally use the concept of limit states and partial factors.
However, in practice there are exceptions for some problems (e.g. for slope stability),
in which the use of global safety factor is considered more reasonable.
There are also several standards which are harmonized with the Eurocodes (with
the ENV EC generation). They include:
• PN-B-02481:1998 Geotechnics. Basic terms and definitions,
• PN-B- 06050:1999 Geotechnics. Earthworks.
Previous Polish Standards of design include calculation models and detailed
procedures.
1
Corresponding Author: Beata Gajewska, Road and Bridge Research Institute, Instytutowa 1, 03-302
Warsaw, Poland; E-mail: bgajewska@ibdim.edu.pl
@seismicisolation
@seismicisolation
26 B. Gajewska / Eurocode 7 and Polish Practice: Implementation of Eurocode 7 in Poland
1. Structure of EC7
Eurocode 7 should be used for all structure-ground interaction issues, foundations and
retaining structures. The Standard concerns buildings, bridges and other engineering
structures.
In Eurocode 7-1 (PN-EN 1997-1) the method of limit states is used. Partial factors
in EN 1997-1 are analogous to the load factors (usually greater than 1) and reduction
factors (usually less than 1) in the previous PN-B standards. However, the system of
factors (partial and correlation) in the Eurocode 7-1 is more developed and diversified.
For each of the three design approaches EC-7 gives different sets of partial factors. The
recommended values of the factors are in Annex A. They can be determined nationally.
Detailed rules for the design or calculation models (i.e. specific formulas or charts) are
included in informative Annexes only. Eurocode 7 is a quite general document, which
gives only the geotechnical design principles. It allows to calculate the geotechnical
actions on the structures, as well as ground resistances. It also contains the rules and
principles of good practice. Standard EN 1997-1 does not include many commonly
used structures such as reinforced soil with nails or geosynthetics, or ground
strengthened by columns or other inclusions.
Eurocode introduces a term “geotechnical design”, which is new to the Polish
practice.
2. The main differences between PN-B and EC7
Eurocode is quite close in philosophy to the previous Polish standards. In both cases,
namely EC7 and PN-B, the limit states are checked and partial safety factors are used.
However, there are many significant differences. They have been reported in many
papers (Kłosiński 2007, Kłosiński and Rychlewski 2009, Konderla 2008, Kotlicki 2005,
Wysokiński 2002, Wysokiński 2006). This paper discusses only a few examples.
Comparative calculations (Wysokiński et al. 2011, Krzyczkowska et al. 2004,
Wysokiński 2008) indicate the designing according to the previous PN-B to be usually
more economical than those based on Eurocode 7. Nonetheless the structures are
generally safe. A few cases of failure were caused by reasons not related to used values
of safety factors (Wysokiński 2007).
2.1. Design of spread foundations
The provisions of Section 6 of EC7 apply to spread foundations including pads,

strips and rafts. This Section contains a list of the limit states. The actions listed in
2.4.2(4) of EC7 should be considered.
Informative annexes provide detailed examples of calculation models for bearing
resistance calculation (Appendix D), semi-empirical method for bearing resistance
estimation (Appendix E), settlement evaluation (Appendix F), deriving presumed
bearing resistance for spread foundations on rock (Appendix G).
The provisions of Chapter 6 concern the design of spread foundations in all
possible cases. Given calculation models concern verification of basic limit states.
The scope of PN-B 03020:1981 is restricted to the commonly occurring cases.
Analyses were reduced to demonstrate that exceeding the elementary limit states are
sufficiently improbable, i.e. loss of bearing resistance of soil under the foundation
@seismicisolation
@seismicisolation
B. Gajewska / Eurocode 7 and Polish Practice: Implementation of Eurocode 7 in Poland 27
(including slip of foundation) and the settlement limit state. For these cases, the
standard sets out detailed procedures to be followed. The Standard indicated that for
specific cases it is needed to incorporate additional recommendations that have not
been determined.
Formulas for the dimensioning of the spread foundations in drained conditions are
similar to those previously used. But in fact, they differ significantly from PN-B
03020:1981 due to different foundation shape factors and inclination of the load factors
(according to PN-B the load inclination factors was determined from nomographs). The
verification of bearing resistance limit state in undrained conditions is novel for the
Polish standard.
2.2. Design of pile foundations
The provisions of Section 7 of EC7 apply to end-bearing piles, friction piles, tension
piles and transversely loaded piles installed by driving, by jacking, and by screwing or
boring with or without grouting. Piles should be constructed in accordance to
standards: EN 1536 – Bored piles, EN 12063 - Sheet pile walls and EN 12699 -
Displacement piles.
Eurocode 7-1 contains the general patterns and recommended values of partial
safety factors and the rules for determining the resistances (correlation factors).
The pile design should be generally based on load tests (static and dynamic), or the
observed performance of a comparable pile foundation. Calculations based on the
results of the ground tests have a supporting role, and there are no indications how to
calculate resistances. EC 7 contains general requirements for the interpretation of pile
tests and determination of characteristic and design resistances. In difference to PN-B,
there are no numerical values to determine the resistances or displacements of piles.
Extensive information about the design of steel piles and sheet piling is provided in EN
1993-5.
EN 1997-1 provisions are quite significantly different from the previous Polish
standards and unsatisfactory for practical design of piles. It will be necessary to use
withdrawn national standards, guidelines and technical literature. However, pile
standard PN-B-02482:1983 requires a substantial change, updates and complements, as
well as adjustments to the requirements of Eurocode.

The equation 7.9 (characteristic values of base resistance and shaft friction
obtained by calculation) according to EN 1997-1 is similar to the equation in Polish
standard PN-B-02482:1983. The base resistance qb;k and shaft friction qs;i;k are not
entirely synonymous with resistance qi and ti in the PN-B standard. For the practical
application of Eurocode 7-1 it will be necessary to supplement the National Annex and
amend the previous pile standard PN-B-02482:1983.
Standard PN-B-02482 is already regarded as too conservative. The application of
unit base resistance and shaft friction of piles, taken directly from the PN-B-02482:
1983 to EN 1997-1, results in an additional factor of safety. Ultimate resistances in the
PN-B-02482: 1983 are not true ultimate values in the meaning of EN 1997-1, but
“design limit values” (i.e. reduced).
@seismicisolation
@seismicisolation
2.3. Design of retaining structures and anchorages.
The standard distinguishes between retaining structures: gravity walls, embedded walls
and composite retaining structures. Section 8 contains list of limit states (ULS and
SLS), actions (inter alia weight of backfill material, surcharges, weight of water, wave
and ice forces, thermal effects) and design situations, which should be considered.
It describes in detail the rules for determining the earth pressure on the walls (at
rest, active, passive, intermediate) The informative Annex C contains example
calculation formulas and charts for analytical determination of earth pressures. The
rules are generally similar to those given in the PN-B 03010:1983; however, they are
substantially different in many ways. Different values of displacement needed for
mobilization of earth pressure may be an example. Displacement values necessary for
raising passive earth pressure are at least twice as large in EC7 (Wysokiński et al.
2011).
2.4. Characteristic values of soil parameters
In geotechnical design proper determination of geotechnical parameters, i.e. the

numerical values of the ground properties, is a key issue. The rules for determining
characteristic values according to EC 7-1 are significantly different from previous
national standards PN-B. Till now they were equivalent to the most probable i.e.
average values. The characteristic value of the parameter did not depend on the type
and number of tests. The design value was determined by reducing (or increasing) the
characteristic value by applying the relevant factors or their combinations.
EC 7-1 states that the characteristic value of a geotechnical parameter shall be
selected as a cautious estimate of the value affecting the occurrence of the limit state.
Further, it is said that if statistical methods are used, the characteristic value should be
derived such that the calculated probability of a worse value governing the occurrence
of the limit state is not greater than 5%. A cautious estimate of the mean value is a
selection of the mean value of the limited set of geotechnical parameter values, with a
confidence level of 95%. So these are significantly reduced (or increased) values. It is
an important difference compared to the previous national practice.
Design values of geotechnical parameters (according to 2.4.6.2 of EC7) can also be

assessed directly (expert’s values). In this case, the values of the partial factors
recommended in Annex A should be used as a guide for required level of safety.
2.5. Other design issues
Eurocode 7-1 contains rules for several important design problems that have not been
included in previous national standards, or were considered marginal. These are:
hydraulic failure (Section 10), overall stability (Section 11) and the design of
embankments (Section 12). These issues are presented in some detail, but not
exhaustively. EC-7 provide lists of limit states, actions and design situations. Detailed
rules for checking the resistance and overall stability may be new for Polish designers.
There is no comprehensive national literature on these subjects, especially including
provisions of EC 7. Instruction ITB 424/2011 concern the evaluation of overall stability
of slopes.
@seismicisolation
@seismicisolation
The design of anchorages is a new issue, which was not available in Polish
standards before. Often the DIN Standard was used. Eurocode provides requirements
concerning the verification of the limit states: ultimate (pull-out resistance, structural
resistance, overall stability) and serviceability, calculation models, and requirements
concerning tests of anchorages.
3. PL National Annex to Eurocode 7-1
The role of the National Annex, among others things, is to indicate Design Approach
and to include the decision of the nationally determined values. The National Annex
may also give the normative status to one or more of the informative Annexes. In this
case, they become mandatory within the country. Each country may supplement
Eurocode 7 general rules by national standards to clarify the calculation models and
design rules used in the country. However, these standards must be consistent with
Eurocode.
One of the tasks of the National Annex is to provide partial coefficients and
models to be used in Poland. The proper selection of safety factors values is important
to ensure the durability and safety of structures, without excessive costs. In order to
establish numerical values of partial factors, the research and comparison analysis with
existing national standards are needed. The choices are limited because the values of
partial factors for actions have already been included in the standard PN EN
1990:2004, which Poland accepted without changes. Those values are more cautious
than in previous Polish Standards.
In the Polish NA the following choices were made:
• National choice of Design Approach for slopes - DA3;
• Other than slopes - DA2* (GEO - Design bearing resistance equation 2.7b,
characteristic values and γF = 1,0);
• HYD Limit state – Hydraulic failure by heave - the use of equation 2.9b was
recommended;
• Posted the scope of Ground Investigation for Geotechnical Categories.
Reasons for DA2* selection:

• is the most similar to our previous (current) practice;
• simples calculation due to less loading combinations (checks with γG <1 and
>1 are not needed);
• avoiding multiplying twice the eccentricity of the resultant action by partial
factor;
• was chosen by many countries, including Germany, whose practice is most
similar to ours.
Reasons for selection of DA3 for slopes:
• it is more rational then DA2 for slopes
• is similar to previous (current) rules of stability verification
@seismicisolation
@seismicisolation
Table 1. Minimum requirements for the scope of the ground investigation (according to PL National Annex).
Geotechnical category Scope of the subsoil investigation
Qualitative determination of the ground properties

on the basis of:
Structures classified to the 1st geotechnical • archival data analysis

category in simple ground conditions • comparable experience considered
• field investigation
Quantitative determination of numerical values of

geotechnical parameters on the basis of:
Structures classified to the 2nd geotechnical • archival data analysis and comparable
category in simple and complex ground experience considered
conditions • field investigation results
• laboratory investigation results
Quantitative determination of numerical values of

geotechnical parameters on the basis of:
• archival data analysis and comparable
experience considered
Structures classified to the
3rd geotechnical category • field investigation results
in simple, complex and
complicated ground conditions • laboratory investigation results
• special investigation results with allowance
made for direct correlations from
investigation
There is a general practice that a geotechnical investigation report contains a table

with geotechnical parameters values for each layer. These values were often taken from
table in standard PN-B 03020:1981 – method B. The method consists of determining
the value of the parameter based on the correlation between strength or physical
parameters and so cold “leading” parameters (e.g. liquidity index or density index). The
National Annex to EC7 states that the scope and methods of the ground investigations
as well as geotechnical parameters for the structural design are determined by the
author of geotechnical design report (geotechnical conditions of foundation) with co-
operation with the structural designer. Minimum requirements for the scope of the
ground investigation are presented in Table 1.
Annex H was supplemented by the following recommendations:
• It is recommended to verify the limit states for construction settlements on the
basis of displacement and deformation values: smax, θmax, Δmax, ω.
• Limit displacement and deformation values for buildings:
• for commonly used structures
• with no special requirements referring to the settlements
@seismicisolation
@seismicisolation
The limit values of measurements of settlements for buildings recommended in the

Polish National Annex are presented in Table 2.
Table 2. The limit values of measurements of settlements for buildings according to the PL National Annex.
smax [mm] θmax [rad] Δmax [mm] ω [rad]
50 0.002 10 0.003
Recommended limit values smax are close to the limit values according to PN-B
03020:1981. So far practice does not indicate these values to be inappropriate. The θmax
value is in the range of values recommended in Annex H of EC7 for relative rotation
(βmax).
The Polish NA in general accepts the partial factor values recommended in EC 7-1
Annex A. In two cases the use of different values of partial factors than those given in
Annex A may be justified:
• more unfavorable values in the case of specific risk, complicated soil
conditions and/or unusual loads (structures classified to the 3 rd geotechnical
category);
• less severe values for temporary structures or transient design situations.
The use of more unfavorable or less severe values of partial factors should be
justified in the design.
4. Polish guide to EC7
In 2011 “Geotechnical design according to Eurocode 7 – Guidance” was published

(Wysokiński et al. 2011). The guide explains requirements, recommendations and
geotechnical design procedures provided in EC7. It describes predominantly cases of
design foundations of Geotechnical Category 2. The Guide also compares Eurocode
with the previous PN-B Standards, noting Polish practice and experience, especially in
the case of shallow foundations and retaining structures. A great part of the publication
concerns methods of ground testing and the interpretation of results. The Guide
contains three comprehensive sections on foundation designing. These are: design of
shallow foundations, pile foundation and design of retaining structures. Attention was
also paid to the overall stability of slopes. The guide presents examples of "step by
step" calculations of typical, commonly used types of foundations. The guide does not
include issues related to hydraulic failure and embankments.
5. Status of Eurocodes in Poland
Until 31 December 1993 the use of Polish Standards (PN-B) was mandatory.
Disobedience to provisions of standard was a violation of the law. From 1 January
1994 the use of Polish standards is voluntary. However, up to 31 December 2002 it was
possible, by the relevant ministers and in some cases, to impose the use of PN. From 1
@seismicisolation
@seismicisolation
January 2003 the use of PN-B is completely voluntary (Journal of Laws of 2002 No
169 item 1386, the Law on Standardization), beyond a few exceptions.
The following rules are applied in national (Polish) standardization (Law on
Standardization):
• transparency and public accessibility;
• taking into account the public interest;
• voluntary participation in the development and application of standards.
According to the Law on Standardization, Polish standards may be established in
the law after their publication in Polish (art. 5, paragraph 4). Thus, standards may be
referred to in the mandatory documents. Polish Building Law and Ministries’
regulations in many instances state that design shall be made according to standards.
The Minister of Transport, Construction and Maritime Economy Regulation (2012)
gives the requirements for the content of geotechnical design according to Polish
standards PN-EN 1997-1: Eurocode 7: Geotechnical design - Part 1: General principles
and PN-EN 1997-2: Eurocode 7: Geotechnical design - Part 2: Ground investigation
and testing.
6. Implementation of Eurocode 7
Both parts of Eurocode 7 are published in Polish. PN-EN 1997-1 was published on 6th
May 2008 under the Polish title: “Eurokod 7 - Projektowanie geotechniczne - Część 1:
Zasady ogólne”. PN-EN 1997-2 was published on 10th April 2009 under the Polish
title: “Eurokod 7 - Projektowanie geotechniczne - Część 2: Rozpoznanie i badanie
podłoża gruntowego”. The National Annex to Eurocode 7-1 PN-EN 1997-
1:2008/NA:2011 was published on 27th October 2011. The English version of the
National Annex is under preparation by the Polish Standardization Committee on
Geotechnics (KT 254).
The implementation of the standard EN 1997 'Geotechnical design' has long been
discussed. For several years, the specific work has been undertaken in the Building
Research Institute ITB (e.g. Kotlicki 2005, Wysokiński 2006) and in the Road and
Bridge Research Institute IBDiM (e.g. Assumptions to NA 2002, Kłosiński 2006), as

well as in the others.
National practice shows that previous standards provide a sufficient (if not
excessive) level of safety (Wysokiński 2002, Kotlicki 2005). Therefore, there is no
reason for radical changes in the value of safety factors. In foundation design according
to PN-B Standards, the total safety factor was a combination of partial factors applied
to the loadings (different in the case of buildings and bridges) and to the soil
parameters (material parameters), corrections factors m (additional partial factor,
having no equivalent in EC7) and others. This results in large number of variables and
complicates the comparison of the effects of designing according to Eurocode-7.
Additionally, the values of partial factors to be applied in Poland to the actions have
already been established in the National Annex to the standard PN-EN 1990.
PKN Technical Committee No. 254 on Geotechnics consider that it is reasonable
to complement Eurocode 7 by national standards harmonized with the EC7,
complementary its provision. There is a need to clarify the rules for determining
geotechnical parameters, designing spread foundations, designing pile foundations,
@seismicisolation
@seismicisolation
designing retaining structures (including deep excavations) and checking the overall
stability.
Some national geotechnical standards, e.g. PN-B-03020, PN-B-02482, PN-B-
03010 will still be needed (as in other EU countries). However, those standards should
be completely rewritten: harmonized with Eurocodes, updated and supplemented.
In order to promote Eurocode numerous conferences, seminars and workshops are
organized. To familiarize designers with design according to Eurocode, postgraduate
studies „Geotechnical design” and courses are organized.
EC7 Implementation Commission was established in the Polish Geotechnical
Committee (PKG). Tasks of the Commission are as follows:
• Clarification and analysis of Eurocode 7.
• Opinion and advisory role in the preparation of standards.
• The collection of opinions.
• Popularization of European Standard solutions.
• Collection and dissemination of information on Eurocode 7.
7. Structure reliability aproach
Limit state methods, in previous Polish Standards, were methods in which an

appropriate level of reliability of a structure was achieved by using a set of partial
factors modifying representative values of variables deciding upon the condition of a
structure. The values of safety factors have been established on the basis of previous
experiences.
Now Poland has adopted values of partial factors recommended in Annex A. As
previously mentioned, Polish Standards were generally more economical. Calibration
of current reliability measures, i.e. calibration of partial factor values recommended in
the Polish NA, is needed, especially in the case of some types of foundations, e.g. piles.
Simplified probabilistic methods with the use of the reliability ratio β (provided in
Eurocode 1990) and probabilistic methods in Polish geotechnical practice are rather not
used.
After 30 years from the introduction of Polish Standards generation based on semi
probabilistic method of limit states there exists a general view, that acceptance of a
slight risk of failure of every structure is unavoidable, and that fundamental variables
taken into account in the design process are usually uncertain (Woliński 2011).
The Polish TC 254 for Geotechnics initiated action aiming at a common approach
to the reliability of the structure from the loadings point of view (EC0 and EC1) and
geotechnical resistances (EC7), and, consequently, a common calibration of reliability
measures.
In Table 3 a comparison of the safety factors in PN-B-02482 and EC7 is presented.
For this comparison it was assumed (for a building) that 70% of loading is dead loads
and 30 % is live loads. In this case the total safety factor obtained from applied partial
factors according to EC7 is about 40 to 50 % higher than according to PN-B. The shaft
and base resistances given in PN-B-02482 are design limit values, i.e. reduced
(equivalent settlement of about 5% of the diameter). The EC7 limit resistance
corresponds to a settlement equal to 10% of the diameter. The number of tests or
profiles has a significant impact on the result. Total factor according to EC7 is even
greater because of the way of determining the characteristic resistances. In the Polish
@seismicisolation
@seismicisolation
Standard PN-B the characteristic value is a mean value, not a cautious estimate. And
pile foundations are in general safe and rather oversized.
Table 3. Comparison of the safety factors for piles in PN-B-02482 and EC7.
Piles - partial factors for: PN-B-02482 Eurocode 7-1
*
loads (0,7 x1,1+0,3x1,2)=1,13 (0,7x1,35 + 0,3x1,5)=1,395
parameters 1/0,9 = 1,11 1,4 - 1,25
resistance 1/0,9** (0,8; 0,7) = 1,11 1,1
total factor 1,13:0,9:0,9=1,4 1,395x(1,25-1,4)x1,1=1,92-2,15
Ratio (1,92 – 2,15) : 1,4 = 1,37 – 1,54
*
70% of loadings is dead weight, 30 % is live loads
**
m=0,9 for 3 and more piles, 0,8 for 2 piles, 0,7 for 1 pile
8. Conclusions
Eurocode 7 is not easy to use, despite similarity to previous Polish Standards. Previous
Polish Standards include calculation models and detailed procedures to be followed.
They were also less extensive than EC7. The statements in EC7 are often more general
than a designer (used to more detailed standards) would expect. More details on the
determination of characteristic values of geotechnical parameters are needed. Parameter
values should be adjusted for the type and features of a specific structure.
Calibration of current reliability measures (partial factor values recommended in
the Polish NA) is still needed, especially in the case of some types of foundations, e.g.
piles, anchors. A common approach to the reliability of a structure, from the loadings
point of view (EC0 and EC1) and geotechnical resistances (EC7), can contribute to
improving the design’s economy, while providing safety of a structure.
References
EN 1997-1 Eurokod 7 – National Annex NA – Assumptions. The implementation of European standards in

road and bridge construction in Poland (bridges). IBDiM, Item SN-1, Stage 2002.
Instruction 424/2011(author: Wysokiński, L.). Evaluation of slope stability. Rules for selection of protection.
Building Research Institute ITB, Warsaw.
Minister of Infrastructure Regulation, 10th of December 2010, changing the regulation on the technical
requirements to be met by buildings and their location. Journal of Laws of 2010 No. 239 item 1597
Minister of Transport, Construction and Maritime Economy Regulation, 25th of April 2012, on the
establishing of geotechnical conditions for the foundation of building structures. Journal of Laws of
2012 No. 0 item 463.
Kłosiński, B. (2006): Perspectives for the implementation of the geotechnical Eurocodes, Inżynieria i
Budownictwo No 6, p. 318-322.
Kłosiński, B. (2007). Problems of implementation of EN 1997 'Geotechnical design', Inżynieria i
Budownictwo No 7-8, p. 361-364.
Kłosiński, B., Rychlewski, P. (2009). Characterization of new European geotechnical standards. XXIV
National Workshop of Structural Designers. Wisła, 17-20 of March.
Konderla, H. (2008). Slope stability in terms of Eurocode 7, Geoinżynieria drogi mosty tunele No 2, s. 26-28.
Kotlicki, W. (2005). Design of spread foundations in terms of Eurocode 7. XX National Workshop of
Structural Designers, Wisła-Ustroń, 1-4 of March.
@seismicisolation
@seismicisolation
Krzyczkowska, A., Gajewska, B., Kłosiński, B. (2004). Examples of calculations of foundations according to
ENV 1997-1. Seminar on "Design of bridges in European standards", IBDiM, Warsaw, 15th of
November.
PN-B-02481:1998 Geotechnics. Basic terms and definitions.
PN-B 02482:1983 Bearing capacity of piles and pile foundations.
PN-B 03010:1983 Retaining structures. Geotechnical design.
PN-B 03020:1981 Shallow foundations. Geotechnical design.
PN-B- 06050:1999 Geotechnics. Earthworks.
PN-EN 1536 Bored piles.
EN 1990 Eurocode. Basis of structural design.
PN-EN 1993-5 Eurocode 3. Design of steel structures. Piling.
PN-EN 1997-1 Eurocode 7. Geotechnical design. General rules.
PN-EN 1997-1:2008/NA:2011 Eurocode 7. Geotechnical design. General rules. (PL National Annex to EC7-
1).
PN-EN 1997-2 Eurocode 7. Ground investigation and testing.
PN-EN 12063 Sheet pile walls.
PN-EN 12699 Displacement piles.
Rymsza, J. (2011). Procedure of fast implementation of Eurocodes in bridge structures in Poland, Rzeszów
University of Technology Scientific Papers, Civil and Environmental Engineering z 58 (3/11/I), s. 235-
248.
The Law on Standardization, Journal of Laws of 2002 No 169 item 1386.
Woliński, Sz. (2011). Probabilistic basis of contemporary design codes, Rzeszów University of Technology
Scientific Papers, Civil and Environmental Engineering z 58 (3/11/I), s. 269-288.
Wysokiński, L. (2002). Issues relating to the implementation of European geotechnical standards in Poland,
Inżynieria i Budownictwo, No 11, p. 625-630.
Wysokiński L. (2006). The implementation of European standards in the Polish geotechnical practice.
International Geological Fair „Geologia-2006”, Warsaw, 6-7 of June.
Wysokiński, L. (2007). Systematic errors in the ground investigation and their impact on building design.
XXIII Conference of Building Failures, Szczecin-Międzyzdroje, p. 527-541.
Wysokiński, L. (2008). Information of Technical Committee of Polish Committee for Standardization, TC
254 (Geotechnics) about the Eurocode 7 implementation in Poland. ITB Warsaw, 11 p.
Wysokiński, L., Kotlicki, W., Godlewski, T. (2011) Geotechnical design according to Eurocode 7 –
Guidance. Building Research Institute ITB, Warsaw.
@seismicisolation
@seismicisolation
36 Modern Geotechnical Design Codes of Practice
IOS Press, 2013
doi:10.3233/978-1-61499-163-2-36
An Explanation of Characteristic Values of

Soil Properties in Eurocode 7
Michael A. HICKS
Faculty of Civil Engineering and Geosciences, Delft University of Technology, Delft,
Netherlands
Abstract. The paper investigates the philosophy and use of characteristic soil
property values in Eurocode 7. Due to the spatial nature of soil variability,
characteristic property values are shown to be problem-dependent and a function
of two competing factors: the spatial averaging of properties along potential failure
planes, which reduces the coefficient of variation of property values; and the
tendency for failure to follow the path of least resistance, which causes an apparent
reduction in the property mean. The Random Finite Element Method provides a
self-consistent framework for quantifying and understanding this behaviour, and
for deriving characteristic values satisfying the requirements of Eurocode 7. It is
widely accepted that characteristic values may be over-conservative if they do not
account for the spatial averaging of property values. Conversely, this paper argues
that characteristic values based only on variance reduction techniques may be un-
conservative, if no account is taken of the apparent reduction of the mean along
potential failure planes. Simpler probabilistic methods can be effective in guiding
design through quantifying uncertainty. However, further research is needed to
assess when such methods are applicable, and when they are significantly in error
and require further attention.
Keywords. Characteristic value, Eurocode 7, geotechnical design, heterogeneity,

reliability, soil properties
Introduction
Engineers have traditionally followed a conservative approach to geotechnical design,

involving mostly subjective estimates of design soil properties and global factors of
safety (Hicks, 2007; Schneider, 2010; Schneider & Fitze, 2011). Eurocode 7 (CEN,
2004) is innovative, in that it requires uncertainties in the design process to be
considered more explicitly and in a consistent manner. However, it is also controversial
in that it provides little guidance as to how this should be achieved, especially with
respect to the determination of so-called characteristic soil property values used in
design.
This paper focuses on uncertainty due to soil variability and, in particular, on
Section 2.4.5.2 of Eurocode 7, “Characteristic values of geotechnical parameters”
(CEN, 2004). It highlights and explains selected paragraphs, clarifies the relationship
between paragraphs and addresses areas of potential confusion. In particular, it clearly
demonstrates the influence of soil heterogeneity and scale of fluctuation on the
problem-dependency of characteristic values, and demonstrates how the Random Finite
Element Method may be used for improving understanding and for deriving
characteristic values satisfying the requirements of Eurocode 7 (Hicks & Samy, 2002b).
@seismicisolation
@seismicisolation
M.A. Hicks / An Explanation of Characteristic Values of Soil Properties in Eurocode 7 37
1. Heterogeneity and Eurocode 7
Eurocode 7 states that the design value, Xd, of the soil property, X, is given by
X
Xd = k (1)
m
where Xk is the characteristic value of X and γm is the partial safety factor. The value of
the partial safety factor is defined by the Code. Hence, all engineering judgement is
focused on the chosen characteristic value. In terms of the mean property value, Xm, it
may be represented by
Xk = αX Xm (2)
where αX is a factor which, for strength properties, generally lies in the range 0 to 1.
Table 1 lists a selection of paragraphs on characteristic values of geotechnical
parameters, taken from Section 2.4.5.2 of EC7 (CEN, 2004) which comprises twelve
paragraphs in total. Firstly, Paragraph 1 asserts the principle that characteristic values
be “based on results and derived values from laboratory and field tests, complemented
by well established experience”; whereas Paragraph 2 asserts the principle that the
characteristic value be “selected as a cautious estimate of the value affecting the
occurrence of the limit state”. The apparent vagueness of the term “cautious estimate”
has led to some debate; while it seems to support the continued application of previous
good practice, implying “no change”, there is little specific guidance within EC7 as to
how characteristic values should be derived.
However, Paragraphs 3-12 present some points of reference, including Paragraph 4
which lists items to account for. In particular, “the variability of the measured property
values” highlights the variable nature of soils; whereas, “the extent of the field and
laboratory investigation” and “the type and number of samples” highlights the
uncertainties that exist when determining property values, partly due to the variability
itself and partly due to the constraints of field and laboratory testing. Meanwhile, “the
extent of the zone of ground governing the behaviour of the geotechnical structure”
indicates that the spatial characteristic of the variability is important. And finally, “the
ability of the geotechnical structure to transfer loads from weak to strong zones in the
ground” implies that the characteristic value is problem-dependent; that is, for the same
ground conditions, the characteristic value will be different for different structures and
different loading conditions.
Paragraphs 7-8 are an attempt to simplify matters for two limiting scenarios.
Firstly, Paragraph 7 considers the case when the zone of ground governing the structure
performance is large, relative to the size of soil specimens tested in the laboratory and
the zone of ground affected in an in situ test. It states that, in this instance, “the value of
the governing parameter is often the mean of a range of values covering a large surface
or volume of the ground”. However, there are two things that should be noted: (a) it is
the domain size relative to the spatial scale of fluctuation of the property value that is
important in assessing the relevance of the mean property value, not the domain size
itself, nor its size relative to test samples and field tests although this may have an
influence; (b) the mean value of a property over a potential failure surface may indeed
be a reasonable choice for the characteristic value (e.g. Tang, 1993), but this value may
differ significantly from the mean for the domain of influence as a whole, due to the
tendency of failure mechanisms to follow the weakest path. Hence, the challenge is to
identify the potential failure mechanisms and whether, and to what extent, the
@seismicisolation
@seismicisolation
38 M.A. Hicks / An Explanation of Characteristic Values of Soil Properties in Eurocode 7
mechanisms and characteristic values are influenced by the spatial structure of the
heterogeneity.
Table 1. Extracts from Section 2.4.5.2 of Eurocode 7 (CEN, 2004).
No. Paragraph Author Comments
(1)P The selection of characteristic values for geotechnical No change to previous practice.
parameters shall be based on results and derived values from
laboratory and field tests, complemented by well-established
experience.
(2)P The characteristic value of a geotechnical parameter shall be No change to previous practice.
selected as a cautious estimate of the value affecting the But what is meant by a cautious
occurrence of the limit state. estimate?
(4)P The selection of characteristic values for geotechnical

parameters shall take account of the following:
geological and other background information, such as data Reduces uncertainty.
from previous projects;
the variability of measured property values and other Should account for soil
relevant information, e.g. from existing knowledge; variability.
the extent of the field and laboratory investigation; Affects uncertainty.
the type and number of samples; Affects uncertainty.
the extent of the zone of ground governing the behaviour Spatial aspect of soil variability
of the geotechnical structure at the limit state being is important.
considered;
the ability of the geotechnical structure to transfer loads Characteristic values are
from weak to strong zones in the ground. problem-dependent.
(7) The zone of ground governing the behaviour of a Important to consider the mean
geotechnical structure at a limit state is usually much larger over potential failure surfaces;
than a test sample or the zone of ground affected in an in situ this could be very different to
test. Consequently the value of the governing parameter is the mean over the domain of
often the mean of the range of values covering a large surface influence.
or volume of the ground. The characteristic value should be a
cautious estimate of this mean value.
(8) If the behaviour of the geotechnical structure at the limit state Extreme scenario implying
considered is governed by the lowest or highest value of the local failure.
ground property, the characteristic value should be a cautious
estimate of the lowest or highest value occurring in the zone
governing the behaviour.
(11) If statistical methods are used, the characteristic value should 5% refers to probability of
be derived such that the calculated probability of a worse failure of the structure; not to
value governing the occurrence of the limit state under parameter values.
consideration is not greater than 5%.
NOTE: In this respect, a cautious estimate of the mean value Percentages refer to parameter
is a selection of the mean value of the limited set of values; not to structure
geotechnical parameter values, with a confidence level of performance.
95%; where local failure is concerned, a cautious estimate of
the low value is a 5% fractile.
@seismicisolation
@seismicisolation
For the special case of when the prevailing mechanism dictates a failure surface
that is long relative to the scale of fluctuation, a conservative estimate of the mean may
be warranted. However, the scale of fluctuation will generally be much larger in the
horizontal plane than in the vertical direction. Hence for applications in which the
prevailing mechanism has a significant horizontal component, the weaker material
zones can have a considerable influence on geo-structural performance as has been
demonstrated by Hicks & Samy (2002a, 2002b, 2004), Hicks & Onisiphorou (2005),
Spencer & Hicks (2007), Hicks et al. (2008) and Hicks & Spencer (2010).
Paragraph 8 considers the other limiting scenario, when a cautious estimate of the
extreme value of the ground property may be appropriate. This may be applicable in
the following situations: (a) when the domain of influence is small relative to the scale
of fluctuation, as is the case in local failure; (b) when deposition-induced anisotropy of
the heterogeneity results in semi-continuous weaker zones through which unstable
failure mechanisms such as liquefaction can propagate (Hicks & Onisiphorou, 2005).
Paragraphs 10-11 concern the possible use of statistics. In particular, Paragraph 11
states that “the characteristic value should be derived such that the calculated
probability of a worse value governing the occurrence of the limit state under
consideration is not greater than 5%”. This implies a minimum reliability of 95% for
the geotechnical structure (before application of partial safety factors). It is important to
realise that Paragraph 11 is not referring to a 95% confidence level for the property
value itself, even though the footnote itself is.
The footnote to Paragraph 11 is an attempt to explain the implication of the
paragraph for the two limiting cases discussed previously in Paragraphs 7-8. Firstly, it
states that “a cautious estimate of the mean value is a selection of the mean value of the
limited set of geotechnical parameter values, with a confidence level of 95%”; this
situation applies when the domain of influence is large compared with the spatial scale
of fluctuation characterising the variability. Secondly, it states that “when local failure
is concerned, a cautious estimate of the low value is a 5% fractile”; this refers to the
underlying property distribution (rather than the mean) and applies when the domain of
influence is small compared to the scale of fluctuation. For these special cases, it is
reasonable to use the respective mean and property value distributions to satisfy the
requirements of Paragraph 11. However, most practical situations require a more
general approach.
2. Reliability-Based Characteristic Values
Figure 1 illustrates two approaches to deriving reliability-based characteristic property

values. Firstly, Fig. 1(a) shows the probability density function of a material property X,
which, to simplify the illustration, is assumed to be normal. The simplest way to derive
a reliability-based characteristic value of X is to proportion the areas under the
distribution, as indicated in the figure. For example, for a reliability of R=95%, the
characteristic value Xk is that value of X that subdivides the area under the distribution
into the ratio 1:19, as illustrated in the figure. However, this simple approach is only of
limited use, as, for most practical situations, Xk will merely be the value of X for which
there is a 95% probability of a higher value occurring; it will generally not represent
the reliability of the structure itself and will lead to an over-conservative value of Xk.
@seismicisolation
@seismicisolation
(a) Basic definition of Xk

(b) General definition of Xk
Figure 1. Derivation of characteristic property values satisfying Eurocode 7
@seismicisolation
@seismicisolation
In order to satisfy the Eurocode 7 requirement for a reliability of 95% with respect
to the performance of the structure, a modified distribution of X is required that takes
account of the problem being analysed. Figure 1(b) shows that this distribution is
different to the underlying property distribution in two respects: it is narrower, to
account for the averaging of property values over potential failure surfaces within the
soil mass; and its centre of gravity is shifted, to account for failure being attracted to
weaker zones. In other words, with respect to the underlying distribution of X, the new
“effective” distribution has a reduced variance and a lower mean; this is in contrast to
traditional variance reduction techniques, in which no change in the mean is assumed,
thereby raising the possibility of an un-conservative value of Xk. The reliability-based
value of Xk satisfying EC7 is found by proportioning the area under the effective
distribution of X, as indicated in the figure.
Note that the distribution of effective X has two limits. When the scale of
fluctuation is very small relative to the size of the domain of influence D, failure
mechanisms pass through weak and strong zones alike and there is much averaging of
soil properties. This leads to a very narrow distribution and, in the limit when /D tends
to zero, to a distribution with a variance of zero. Moreover, due to the assumption of a
normal distribution of X, it leads to a mean that is equal to the mean of the underlying
distribution. In this case, the characteristic value is indeed a cautious estimate of the
mean, as advocated by EC7. Conversely, when the scale of fluctuation is large relative
to the domain of influence, there is a wide range of possible solutions and, in the limit
when /D tends to infinity, the distribution of effective X tends to the underlying
distribution of X. In this case, the characteristic value can be taken directly from the
underlying distribution, also in line with EC7.
However, for intermediate values of /D the distribution is as shown in Fig. 1(b),
with an intermediate variance and lower mean. In this case, the distribution is a
function of the point statistics of the soil property, the horizontal and vertical scales of
fluctuation of the soil property and the problem being analysed, as demonstrated by
Hicks & Samy (2002a, 2002b, 2004) and Hicks & Spencer (2010). It is also a function
of the degree of knowledge about a site. Whereas the underlying property distribution
quantifies the actual variation in soil property values, the distribution of effective X is
the result of uncertainty in the structure response, which, in turn, results from
uncertainty in the property values across the domain. This implies that, regardless of
spatial variability, if the real value of X is known at every point in the domain the
distribution of effective X would become very narrow and, in the limit when all
information is known, tend to a single representative value. So, the choice of
characteristic value is as much a function of the site investigation as of the spatial
variability itself.
3. Stochastic Approach to Eurocode 7
Hicks & Samy (2002b) described a stochastic approach for deriving reliability-based
characteristic values satisfying EC7 and demonstrated its use for a simple 2D slope
stability problem. The approach uses the Random Finite Element Method (Fenton &
Griffiths, 2008), which links random field theory for generating spatially varying
property distributions with finite elements for analysing structural response. This
section summarises the main features of the approach and, through reference to Hicks
@seismicisolation
@seismicisolation
& Samy (2002b), summarises how the modified property distribution illustrated in Fig.
1(b) may be derived.
3.1. Statistical Definitions
The geotechnical problem domain is subdivided into distinct material layers (or zones).
In traditional deterministic analysis, each layer is represented by a single representative
soil property value, thereby leading to geotechnical performance being defined in terms
of a single factor of safety for which there is no information regarding probability of
failure. In contrast, stochastic analysis may be used to quantify the uncertainty that
arises in geotechnical performance, partly (but not exclusively) due to the variability of
property values within the layer. In this case, the soil property variability is represented
most simply by the property mean Xm and standard deviation SD, and by a probability
density function such as the normal distribution used for illustrative purposes in Fig. 1.
However, many researchers have demonstrated the importance of including the spatial
nature of soil variability in geotechnical analysis. This is often represented by the scale
of fluctuation , which is a measure of the distance over which property values are
significantly correlated (Vanmarcke, 1983). In the vertical direction the scale of
fluctuation v is generally small and often less than one metre (Hicks & Samy, 2002a;
Hicks & Onisiphorou, 2005). In the horizontal direction, the scale of fluctuation h is
generally much larger, due to the natural or engineered process of deposition (Hicks &
Onisiphorou, 2005; van den Eijnden & Hicks, 2011; Lloret-Cabot et al., 2012).
3.2. Summary of Stochastic Process
Hicks & Samy (2002b) outlined the following three-stage strategy for the stochastic
analysis of geotechnical problems:
• The pre-analysis stage involves the determination of statistical properties for
the site; for example, the point statistics, Xm and SD, the probability density
function, and the vertical and horizontal scales of fluctuation, v and h, taking
account of any depth-dependent trends in the statistics.
• The analysis stage involves a Monte Carlo simulation. It uses the material
property statistics to generate multiple predictions (i.e. random fields) of the
spatial variability of soil properties over the whole problem domain (Fenton &
Vanmarcke, 1990) and, for each prediction, the problem is analysed, for
example using finite elements as in the Random Finite Element Method
(Fenton & Griffiths, 2008). Hence the result is multiple predictions for the
response of the geotechnical problem, in which each prediction of
geotechnical performance is based on a different prediction of the spatial
variability at the site.
• The post-analysis stage involves expressing the results of the Monte Carlo
simulation probabilistically. This generally involves defining geotechnical
performance in terms of probability of failure; or in terms of reliability, the
probability of failure not occurring.
Note that, while it may seem counter intuitive to generate numerical predictions of
spatial variability for a site based on site-specific statistics determined from field data
@seismicisolation
@seismicisolation
obtained for the same site, it should be remembered that the statistics will have been
derived from data obtained at discrete (e.g. CPT) locations. In contrast, the numerical
simulations predict soil property values at every location. Hence the purpose of the
stochastic analysis is to quantify the uncertainty in the response of a geotechnical
structure, not due to the heterogeneity itself, but due to the uncertainty that arises due to
having incomplete knowledge about the nature of the spatial variability of material
properties (as mentioned earlier). This, in turn, leads to two obvious questions (van den
Eijnden & Hicks, 2011; Lloret-Cabot et al., 2012): (a) what is the required intensity of
in situ testing to give reasonable estimates of the soil property statistics; and (b), how
may the uncertainty in geotechnical performance be reduced through the optimal use of
available data?
3.3. Derivation of Characteristic Values
Hicks & Samy (2002b) conducted detailed analyses of a simple slope stability problem,
characterised by a spatially varying undrained shear strength, to illustrate how the
above stochastic framework can be utilised to derive characteristic soil property values.
This involved tackling the problem from two different (albeit equivalent) directions.
In the first case, the statistical properties of undrained shear strength for the slope
were used to analyse the problem using the Random Finite Element Method.
Specifically, for each realisation of the spatial variability of undrained shear strength,
the factor of safety of the slope was found by using finite elements and the strength
reduction method, so leading to a distribution of factors of safety for the slope. The
results demonstrated that, for most realisations, the factor of safety was lower than the
factor of safety of the slope based on the mean undrained shear strength, due to the
influence of the weaker zones. Using this type of analysis, the distribution of effective
property values may easily be backfigured from the distribution of factors of safety (e.g.
by using a stability chart). That is, for any realisation, the effective value of X is that
single value giving the same factor of safety as the analysis based on the random
property field. The result of such a simulation is the effective property distribution
illustrated in Fig. 1(b) and, as stated previously, the characteristic value satisfying EC7
is then the value corresponding to the 5% fractile.
In the second case considered by Hicks & Samy (2002b), the value of Xk is found
by first deriving a relationship between reliability and factor of safety based on the
mean property value. This involves the following calculation steps:
• The undrained shear strength corresponding to the slope at the point of failure,
for the special case of no spatial variability, is determined; either directly, for
example using a stability chart, or numerically, for example using finite
elements and the strength reduction method.
• The influence of heterogeneity for F = 1.0 based on the mean property value is
quantified by generating multiple realisations of the spatial variation of X,
based on the mean , and on the site statistics V, v and h, in which V is the
coefficient of variation and is equal to the value of undrained shear strength
that would just cause the slope to fail in the absence of heterogeneity (i.e. as
computed in the previous step). For each random field, the slope is analysed
by finite elements to see if it remains standing under its own self weight. The
reliability is then the percentage of realisations in which the slope remains
stable.
@seismicisolation
@seismicisolation
• The process is repeated for a range of values of F, to enable the complete

reliability versus F relationship to be determined. For a given value of F, the
mean undrained shear strength is the value of the mean derived in the first
step, multiplied by F, whereas V, v and h are the same (i.e. site) statistics as
used for all other values of F.
• For the site in question, the characteristic value for a given level of reliability
may be obtained by dividing the mean property value Xm by the appropriate
reliability-based value of F (i.e. the value of F corresponding to the required
level of reliability).
Note that, for both of the above cases, the derived value of Xk does not correspond
to a reliability of the structure of 95%. Rather, it corresponds to the characteristic value
that should be used in the geotechnical design. If the structure is calculated to be safe
using this value, then the reliability (before additional factoring) will be at least 95%.
Conversely, if the structure is calculated to be unsafe the reliability will be less than
95%.
4. Conclusions
Some researchers have reasonably postulated that failure is approximately governed by

the mean property value along the failure plane and that the uncertainty regarding
property values along the plane is reduced due to local averaging. This leads to the use
of variance reduction techniques that reduce the variance of property values as a
function of the ratio between the failure plane length and scale of fluctuation, thereby
leading to an apparent reduction in uncertainty.
However, other researchers have noted that failure mechanisms follow the path of
least resistance. Hence the mean property value along a failure plane will generally be
lower than the mean property value in the immediate vicinity of the failure zone, with
the scale of this reduction depending on the nature of the soil heterogeneity as has been
demonstrated in numerous studies. The author supports the view that the mean strength
along a failure plane is important, as is the local averaging of properties to reduce

uncertainty. However, due to heterogeneity there may be considerable uncertainty in
the mean along potential failure planes; so, the use of variance reduction techniques
could lead to unconservative estimates of the characteristic value if no account is taken
of the apparent reduction in the mean.
In summary, due to the heterogeneous nature of soil variability, characteristic
property values are shown to be problem-dependent and a function of two competing
factors: the spatial averaging of properties along potential failure planes, which reduces
the coefficient of variation of property values; and the tendency for failure to follow the
path of least resistance, which causes an apparent reduction in the property mean. The
Random Finite Element Method provides a self-consistent framework for quantifying
and understanding this behaviour, and for deriving characteristic values satisfying the
requirements of Eurocode 7. It is widely accepted that characteristic values may be
over-conservative if they do not account for the spatial averaging of property values.
Conversely, this paper argues that characteristic values based only on variance
reduction techniques may be un-conservative if no account is taken of the apparent
reduction of the mean along potential failure planes.
@seismicisolation
@seismicisolation
References
CEN (2004). Eurocode 7: Geotechnical design. Part 1: General rules, EN 1997-1, European Committee for
Standardisation (CEN).
Eijnden, A.P. van den & Hicks, M.A. (2011). Conditional simulation for characterizing the spatial variability
of sand state, Proceedings of the 2nd International Symposium on Computational Geomechanics,
Dubrovnik, Croatia, 288–296.
Fenton, G.A. & Griffiths, D.V. (2008). Risk assessment in geotechnical engineering, John Wiley and Sons.
Fenton, G.A. & Vanmarcke, E.H. (1990). Simulation of random fields via Local Average Subdivision,
Journal of Engineering Mechanics, ASCE 116 (8), 1733–1749.
Hicks, M.A. (2007) (editor). Risk and variability in geotechnical engineering, Thomas Telford, London.
Hicks, M.A., Chen, J. & Spencer, W.A. (2008). Influence of spatial variability on 3D slope failures,
Proceedings of the 6th International Conference on Computer Simulation Risk Analysis and Hazard
Mitigation, Kefalonia, Greece, 335–342.
Hicks, M.A. & Onisiphorou, C. (2005). Stochastic evaluation of static liquefaction in a predominantly
dilative sand fill, Géotechnique 55 (2), 123–133.
Hicks, M.A. & Samy, K. (2002a). Influence of heterogeneity on undrained clay slope stability, Quarterly
Journal of Engineering Geology and Hydrogeology 35 (1), 41–49.
Hicks, M.A. & Samy, K. (2002b).Reliability-based characteristic values: a stochastic approach to Eurocode 7,
Ground Engineering 35 (12), 30–34.
Hicks, M.A. & Samy, K. (2004). Stochastic evaluation of heterogeneous slope stability, Italian Geotechnical
Journal 38 (1), 54–66.
Hicks, M.A. & Spencer, W.A. (2010). Influence of heterogeneity on the reliability and failure of a long 3D
slope, Computers and Geotechnics 37, 948–955.
Lloret-Cabot, M., Hicks, M.A. & Eijnden, A.P. van den (2012). Investigation of the reduction in uncertainty
due to soil variability when conditioning a random field using Kriging, Géotechnique Letters 2 (in
press).
Schneider, H.R. (2010). Characteristic soil properties for EC7: Influence of quality of test results and soil
volume involved, Proceedings of the 14th Danube-European Conference on Geotechnical Engineering,
Bratislava, Slovakia.
Schneider, H.R. & Fitze, P. (2011). Characteristic shear strength values for EC7: Guidelines based on a
statistical framework, Proceedings of the 15th European Conference on Soil Mechanics and
Geotechnical Engineering, Athens, Greece.
Spencer, W.A. & Hicks, M.A. (2007). A 3D finite element study of slope reliability, Proceedings of the 10th
International Symposium on Numerical Models in Geomechanics, Rhodes, Greece, 539–543.
Tang, W.H. (1993). Recent developments in geotechnical reliability, Balkema, Amsterdam.
Vanmarcke, E.H. (1983). Random fields: Analysis and synthesis, The MIT Press, Cambridge, Massachusetts.
@seismicisolation
@seismicisolation
IOS Press, 2013
doi:10.3233/978-1-61499-163-2-46
Implementation of Eurocode 7 in German

Geotechnical Design Practice
Kerstin LESNY,1
Institute of Geotechnical Engineering, University of Duisburg-Essen, Germany
Abstract. For a long time the framework of geotechnical design in Germany was
provided by the basic code DIN 1054 and various specific design and construction
codes together with code-like recommendations for specific geotechnical
structures. All these regulations were based on a global safety concept. This design
philosophy was commonly supported by long-term experience and was understood
to be not only safe but also economical. Under these circumstances the transition
to the new design philosophy of Eurocode 7 and the Limit State Design was
accompanied by a lot of discussions and adjustments until finally a broad
acceptance was achieved. This contribution highlights the long tradition of
geotechnical design in Germany based on the global safety concept. It shows the
difficulties involved in the transition to the Eurocode 7 design principles and
illustrates the current status of German Limit State Design in geotechnical
engineering.
Keywords. Limit State Design, Eurocode 7, global safety factor, partial factor,
design approach, shallow foundation, pile foundation
Introduction
Germany has a comparably long history of codified design in geotechnical engineering.

The basic German geotechnical design code used for a long period of time was
DIN 1054. In addition to this code, a set of specific design and construction codes
completed the standardization in this field. Furthermore, various recommendations, e.g.
those for excavation pits or waterfront structures, were used for design, which were just
as binding as standards. This system of regulations formed the framework for
geotechnical design in Germany. It was originally based on a global safety concept
with an overall factor of safety which included possible sources of uncertainties related
to actions or their effects, to material parameters as well as to the calculation model in
use.
Minimum values of the overall safety factors for the respective design problems
were defined as experience values in the relevant design codes depending on three so
called load cases. These load cases were a typical feature of German geotechnical
design. The intention of this concept was to introduce different levels of safety
depending on the character of actions and the design situation.
Because of their long-term experience with this design concept and its related
design philosophy, German geotechnical engineers were convinced that geotechnical
design was not only safe, but also economical. Hence, the implementation of the
1
@seismicisolation
@seismicisolation
K. Lesny / Implementation of Eurocode 7 in German Geotechnical Design Practice 47
probability based Limit State Design (LSD) was accompanied by a lot of doubts and
criticism from the very beginning. In the end it was concluded that the use of
probabilistic methods, e.g. for deriving partial factors, was not acceptable for code
development, but it was agreed that the Limit State Design and the partial factor
concept should be adopted.
In the following, the traditional concept of German geotechnical design is
highlighted briefly. Afterwards, the procedure of implementing LSD in Germany
parallel to the European developments and the final transition to Eurocode 7 is outlined.
Some resulting principles of geotechnical design are introduced for shallow and pile
foundations and the consequences are discussed on the basis of two design examples.
1. Traditional Structure of German Geotechnical Design Codes
In Germany standards and codes are developed in various committees and published by
the “Deutsche Institut fuer Normung” (DIN), which is the German institution for
standardization. In geotechnical engineering the standardization goes back more than
70 years. Since then it has been continuously optimized, reaching a very high quality
till now (Vogt et. al, 2008). Correspondingly, DIN 1054 was the basic design code for a
very long period of time. A first version of it was published already in 1934. Its edition
of 1976 was used till the end of 2007 with a transition period of about two years after
the LSD-based DIN 1054:2005 came into effect (Weißenbach, 2012).
DIN 1054 made reference to various specific design codes such as those for
bearing resistance (DIN 4017) or slope stability (DIN 4084) or to construction codes
such as the DIN-code for bored piles (DIN 4014). In addition to the DIN-codes, various
recommendations e.g. for excavation pits (EAB, 2006), piles (EAP, 2012) or waterfront
structures (EAU, 2004) were used in the design, which were just as binding as the
codes published by the DIN. These codes and recommendations formed the framework
for geotechnical design in Germany.
The design in geotechnical engineering was traditionally based on the global safety
concept, i.e. on an overall factor of safety K defined as the ratio between the resultant
resistance R and the effect of actions E for the system in question. In the design it had
to be verified that the existing safety did not fall below a certain minimum value Kmin:
R R
tEK t Kmin (1)
K E
R and E were so-called deterministic values. Consequently, the inequality in Eq.

(1) often led to the assumption that failure could not occur, i.e. that the design is safe in
all cases, but the actual safety level was unknown and could only be estimated.
Within a design based on the global safety concept actions usually were considered
with their highest value or on the safe side. The safety level to be achieved was applied
to the resistance only and was supposed to cover all kinds of uncertainties concerning
e.g. load assumptions, the calculation model, the design procedure etc.
Different safety levels were introduced in a normative way by the definition of
load cases depending on the duration of an action and the frequency of its occurrence
as defined in Table 1. The global safety factors for various structures according to DIN
1054:1976 and other standards are given in Table 2.
@seismicisolation
@seismicisolation
48 K. Lesny / Implementation of Eurocode 7 in German Geotechnical Design Practice
Table 1. Definition of load cases according to DIN 1054:1976

Load Case Definition
LC1 Permanent actions and regularly occurring variable actions
LC2 Actions of LC1 plus irregular variable actions possibly occurring
at the same time or actions during construction or repair
LC3 Actions of LC2 plus extraordinary actions possibly occurring at
the same time
There were no standardized regulations for the definition of the deterministic

values of E and R. In a former version the recommendations for waterfront structures
(EAU) included a procedure for the definition of soil parameters:
x Deterministic values had to be derived directly from the results of soil

mechanics tests;
x Basic values had to be defined as reduced arithmetic averages from n tests;
x Appropriate additions or deductions had to be applied, which considered the
inhomogeneity of the ground and uncertainties during soil sampling and
testing in the laboratory.
The following so-called cal-values were recommended for the shear strength
parameters:
cu cc tan Mc
cal c u cal cc cal tan Mc (2)
1,3 1,3 1,1
Table 2. Global safety factors for various geotechnical structures

Safety factor
Type of Structure
LC1 LC2 LC3
Shallow foundations:
Bearing resistance 2 1,5 1,3
Sliding 1,5 1,35 1,2
Compression piles
(total resistance, one load test) 2 1,75 1,5
Slopes
(global stability analysis) 1,4 1,3 1,2
2. Implementation of Limit State Design
2.1. Development
In Germany, first attempts at implementing the use of probabilistic methods and later
the LSD in civil engineering already started in the early 1980s and accompanied the
progress on European level from the very beginning. These developments stimulated
critical and controversial discussions among the geotechnical engineering community,
which felt quite comfortable with the traditional design concept.
In the early years, several attempts were made to derive partial factors from
probabilistic methods. In various research projects probability-based partial factors
were derived e.g. for the bearing resistance of shallow foundations (Schultze &
Pottharst, 1981) or for retaining structures (Gaessler & Gudehus, 1983). On the other
@seismicisolation
@seismicisolation
hand, further analyses showed that the reliability level achieved was not consistent in
all relevant design situations. Ongoing discussions in various publications and on
national conferences (a summary is given in Schuppener & Heibaum, 2011) finally led
to the conclusion that probabilistic methods should not be explicitly used for code
development, but it was agreed that LSD and the partial factor concept should be
adopted (Weißenbach, 2012).
As a consequence, a LSD concept was developed in Germany parallel to the
development of Eurocode 7 as a new version of the basic design code DIN 1054,
supplementing the work on Eurocode 7 and conserving traditional German experience.
After several pre-standards had been issued and discussed within the geotechnical
engineering community, DIN 1054 was officially introduced and approved by the
building authorities in 2003 and shortly afterwards revised in 2005. Schuppener &
Heibaum (2011) as well as Weißenbach (2012) outline the development of LSD in
Germany, which was directly connected to the redevelopment of the basic design code
DIN 1054.
This process was called the “German Way” as several aspects of the old German
design concept, e.g. the concept of load cases or the design approach later named as
DA2* (see section 3.1) were retained. The most important reason for following a
separate way, however, was to maintain the safety level of the global safety concept,
which was commonly considered by geotechnical engineers to be not only safe but also
economical. This means that the dimensions of a geotechnical structure designed with
the new DIN 1054:2005 should be roughly the same as if designed using the old DIN
1054:1976. National standardization committees were worried that a different safety
level (which may have been the result if partial factors would have been derived from
probabilistic analyses) would impact the acceptance of Eurocode 7, leading to a
significant delay in the transition process (Vogt et al., 2008). Based on this prerequisite,
the partial factors on the resistance were derived from the former global safety factors
(see section 3.1).
After some years of experience German geotechnical engineers were familiar with
the LSD. The design according to DIN 1054:2005 together with its set of design and
construction codes was broadly accepted, because DIN 1054:2005 still included
regulations representing traditional German design experience, e.g. allowable values
for the base pressure of shallow foundations or experience values for pile resistances.
On the other hand, DIN 1054:2005 included various regulations which were also
specified in the final version of Eurocode 7 published in 2009 (see Figure 1).
2.2. Current Situation
After end of the implementation process geotechnical design in Germany is now based
on three codes, i.e. Eurocode 7 (DIN EN 1997-1:2009), its National Annex (DIN EN
1997-1/NA:2010) and DIN 1054:2010 as a supplement to Eurocode 7. The National
Annex (NA) and DIN 1054:2010 are not independent and can therefore only be used
together with Eurocode 7, which is inconvenient. Hence, an edited summary of these
codes was published in 2011 as a standard-handbook which includes Eurocode 7 as the
basic text supplemented by the regulations of the NA and of DIN 1054:2010, visible by
a shade of grey (Normenhandbuch Eurocode 7, 2011).
@seismicisolation
@seismicisolation
Eurocode 7 DIN EN 1054:2005
not adopted
design approaches particular German
and informative joint regulations:
e.g. limit states, experiences: e.g.
annexes allowable base pressures,
partial factors,
geotechnical pile resistances
categories
Figure 1. Extent of regulations of Eurocode 7 and DIN 1054:2005 (after Schuppener & Ruppert, 2007)
In addition to this, the framework of codes for geotechnical engineering still

includes further design and construction codes as well as recommendations which have
been or still need to be revised as necessary. Schuppener (2010a) illustrates the
resulting system of geotechnical design codes relevant for waterway engineering (see
Figure 2).
Figure 2. New system of geotechnical design codes for waterway structures (Schuppener, 2010a)
German authorities decided on a fixed deadline for the final implementation of the
new standards. In most states this was July 1st, 2012. By this date DIN 1054: 2005 has
been withdrawn and most of the Eurocodes and their National Annexes have officially
been approved and introduced by the state building authorities.
For geotechnical engineers this means that they again have to adjust to a new code
system including not only new and sometimes unclear regulations, but also new
definitions and terminologies, which prevents the acceptance of such codes. The most
important changes are discussed in the following section 3. In section 4 the new design
procedure will be compared to the old design philosophy based on some of the design
examples circulated by the European Technical Committee 10 (ETC10).
@seismicisolation
@seismicisolation
3. New Regulations for Geotechnical Design According to Eurocode 7
3.1. General
In the version of DIN 1054:2005 the concept of load cases was still maintained, but the
load cases were attributed to combinations of actions depending on their occurrence
frequency and safety classes depending on the particular condition of the structure
relevant for specific time periods.
With the new DIN 1054:2010 the concept of load cases has finally been abandoned
and partial factors are now defined for several design situations to adjust to DIN EN
1990:2010. However, the design situations mostly coincide with the traditional load
case definitions as shown in Table 3.
Table 3. Design situations defined in DIN 1054:2010 compared to the load cases of DIN 1054:2005
Load Case acc. to
Design Situation Denotation
DIN 1054:2005
Permanent design situation BS-P LC1
Transient design situation BS-T LC2
Accidental design situation BS-A LC3
Design situation for earthquake BS-E LC3
In DIN 1054:2005 three limit states 1A, 1B and 1C were differentiated in the Ultimate
Limit State (ULS). With the final implementation of Eurocode 7 a new terminology has
been introduced and the classification of the limit states is to some extent different than
in DIN 1054:2005. The main differences refer to limit state 1A (problems dealing with
a loss of equilibrium), which was split up into the limit states EQU (general
equilibrium problems such as overturning), UPL (failure due to vertical forces acting in
upward direction) and HYD (failure caused by flow gradients, e.g. hydraulic heave).
Limit state 1B which was formerly regarded as the limit state that usually determines
the dimensions of a geotechnical structure was split up into the limit states STR
(internal failure of the structure) and GEO-2 (failure of the ground). A comparison of
the former and the new definition of the limit states in the ULS is given in Table 4.
The National Annex DIN EN 1997-1/NA:2010 states that the use of design
approach 1 (DA1 – partial factors on actions or on material parameters, two
combinations have to be checked) according to Eurocode 7 is not permitted in
Germany. For usual design situations within limit state GEO, which have formerly
been assigned to limit state 1B, design approach 2 (DA2, i.e. GEO-2) is now used
whereas design approach 3 (DA3 - partial factors on actions and material parameters)
has to be applied to overall stability analyses (GEO-3).
DA2 is used with the “German” modification (commonly known as DA2*), in
which the partial factors are applied not until the very end of the calculation. This
means the whole calculation is performed using characteristic values of the effects of
actions and the resistance. The partial factors are applied finally when the limit state
equation Ed d Rd has to be checked. If the resistance is load-dependent, the results of a
calculation using DA2* are not the same as the results of a calculation using DA2. This
applies especially to the bearing resistance of shallow foundations in case of combined
loading caused by permanent as well as variable actions.
@seismicisolation
@seismicisolation
Table 4. Comparison of Limit States according to DIN 1054:2005 and Eurocode 7/DIN 1054:2010
Limit State
Eurocode 7/ Definition
DIN 1054:2005
DIN 1054:2010
Loss of equilibrium of the structure or of the
foundation ground, where the strength of the
EQU
structural material or the ground is not decisive for
the resistance
1A Loss of equilibrium of the structure or of the
UPL foundation ground due to uplift or by the effect of
other vertical forces
Hydraulic heave, inner erosion and piping in the
HYD
ground caused by flow gradients
Internal failure of the structure where the strength of
STR
1B the structural material is decisive for the resistance
GEO-2 failure or very large deformations of the structure,
where the strength of the foundation ground is
GEO-3 1C
decisive for the resistance
As mentioned in section 2.1 the partial factors in the ULS are based on the
condition that the safety level of the former global safety concept is maintained. With
Eq. (1) the resistance factors JR have been derived in the following way:
R
Ed d R d E J E d
JR (3)
with E d R/K following from Eq. (1) it is:
R R R K
E JE d Ed Ed JR
JR JE JR K JE (4)
In Eq. (3) and Eq. (4) JE is a weighted mean of the partial factors for the effects of
permanent and variable actions which have been adopted from structural engineering.
The aim was to use a common factor in all disciplines of civil engineering instead of
specific values for geotechnical engineering (Vogt et al., 2008). Factors of JG = 1,35
and JQ = 1,5, for example, lead to a mean value of JE of 1,40. Tables 5 and 6 show
some partial factors for actions and resistances defined in DIN 1054:2010 for the
different design situations. It can be seen that for BS-P they coincide with the
recommended values of Eurocode 7, Annex A. In the Serviceability Limit State (SLS)
no partial factors are applied.
With DIN 1054:2010 combination factors are introduced in geotechnical design
for the first time to adjust to DIN EN 1990:2010. The consequences of the occurrence
of more than one variable action were formerly incorporated in the concept of load
cases (see section 1). With DIN 1054:2010 the combination factors defined in DIN EN
1990:2010 (\0 = 0,8, \1 = 0,7, \2 = 0,5) are adopted to maintain the same level as in
structural engineering.
Consequently, in case of more than one independent variable action the relevant
combination in the respective limit state has to be determined by checking all possible
combinations of actions with interchanging leading action and corresponding
accompanying actions.
@seismicisolation
@seismicisolation
In the following specifically German geotechnical design requirements for shallow

and pile foundations are highlighted.
Table 5. Some partial factors for the effects of actions in the limit states GEO-2 and STR according to DIN
1054:2010 (recommended values of Eurocode 7, Annex A in parentheses)
Actions or Effect of Actions Symbol Design Situation
BS-P BS-T BS-A
Effects of actions from JG 1,35 1,20 1,10
permanent actions, general (1,35) (1,35) (1,35)
Effects of actions from JG,inf 1,00 1,00 1,00
favourable permanent actions (1,00) (1,00) (1,00)
Effects of actions from JQ 1,50 1,30 1,10
unfavourable variable actions (1,50) (1,50) (1,50)
Table 6. Some partial factors for resistances in the limit states GEO-2 and STR according to DIN 1054:2010
(recommended values of Eurocode 7, Annex A in parentheses)
Resistance Symbol Design Situation
BS-P BS-T BS-A
Soil resistances
Passive earth pressure and 1,40 1,30 1,20
JR,e, JR,v
bearing resistance (1,40) (1,40) (1,40)
1,10 1,10 1,10
Sliding resistance JR,h
(1,10) (1,10) (1,10)
Pile resistance from static and dynamic pile load tests
1,10 1,10 1,10
Base resistance Jb
(1,10) (1,10) (1,10)
1,10 1,10 1,10
Shaft resistance (compression) Js
(1,10) (1,10) (1,10)
1,10 1,10 1,10
Total resistance (compression) Jt
(1,10) (1,10) (1,10)
3.2. Shallow Foundations
A specific feature of German geotechnical design of shallow foundations is the

verification of the allowable eccentricity of the resultant vertical loading. In DIN

1054:2005 this proof was attributed to the ULS, but not to a specific limit state and all
partial factors were set to 1,0. The allowable eccentricity was defined as 1/3 of the
foundation width, i.e. under eccentric loading a maximum gap may occur in the footing
base up to its center of gravity. The aim of this proof was to guarantee that a
significantly large portion of the foundation base participates in the load transfer to the
ground. In DIN 1054:2010 this requirement is maintained but is now part of the proof
of the SLS to limit the rotation of eccentrically loaded foundations.
As a new requirement according to DIN 1054:2010 overturning of a foundation
around the foundation edge additionally has to be checked for limit state EQU in the
ULS. This verification may become relevant in the design situation BS-P compared to
the proof of the allowable eccentricity in the SLS (for details see Schuppener, 2010b).
The foundation edge is the true rotation axis only in case of foundations on rock. In
case of foundations on or in soils it is assumed to be a fictitious rotation axis.
DIN 1054 traditionally included a specific German design procedure that
substitutes the proof of ULS and SLS for regular design cases by providing tables with
allowable base pressures for shallow foundations for certain geometric and soil
@seismicisolation
@seismicisolation
conditions. This procedure, which is very familiar in Germany, has been maintained as
well, but in DIN 1054:2010 design values of base pressures are given instead of
allowable values. The design base pressure has to be compared with the design value
resulting from the current foundation loading. The tabulated values are valid for BS-P
and are assumed to be on the safe side for BS-T and BS-A.
3.3. Pile Foundations
Following Eurocode 7 the pile resistance according to DIN 1054:2010 and the
recommendations for piles (EAP, 2012) can be determined by static or dynamic pile
load tests. Another method is the determination of the pile resistance based on ground
test results.
In comparison to DIN 1054:2005 the partial factors for pile resistances from pile
load tests have been reduced, because the use of the scattering factors is more
differentiated than before. The aim is to better reward the performance of pile load
tests. In DIN 1054:2010 all scattering factors are increased, but they decrease with the
number of load tests. A maximum of five load tests is rewarded, whereas in DIN
1054:2005 only two load tests have been considered. Despite these changes in the
design methodology the former safety level is still maintained (see Kempfert, 2012).
In the traditional German design it was also permitted to calculate the resultant pile
resistance using fixed characteristic values for the base resistance and the shaft
resistance depending on the type of pile and the soil characteristics. These values were
considered as experience values which were based on an evaluation of numerous pile
load tests. In the methodology of Eurocode 7 this procedure represents a method of
determining the pile resistance based on ground test results.
As these values represent long-term national experience this traditional concept
has been maintained in DIN 1054:2010 as well. The experience values are tabulated in
EAP (2012) together with detailed regulations for their application. For this method
increased partial factors are to be applied according to DIN 1054:2010 which already
include a model factor of about 1,3 as it is required according to Eurocode 7. This
model factor roughly coincides with the scattering factors [1 and [2 for static pile load
tests (Kempfert, 2012).
4. Consequences for Practical Design
The consequences of LSD according to the new set of standards for geotechnical
engineering problems shall be illustrated on the basis of two out of six design examples
which were distributed by ETC10 together with a questionnaire among the European
Member States to gain information on the progress of harmonization as well as to
analyse the national differences in the implementation of Eurocode 7. This was the
second survey after a first survey with ten design examples had been conducted in 2005
(Orr, 2005). The results of the second survey were presented in a workshop in Pavia in
2010 (ETC10, 2010) and have been discussed frequently since (e.g. Orr, 2011 and Orr
et al., 2011).
The two examples presented in the following are a square pad foundation
(Example 2.2 in the survey) and a pile group of bored piles (Example 2.6 in the
survey). In both cases the characteristic soil parameters for the design have to be
derived from site investigations results, which are provided with the examples.
@seismicisolation
@seismicisolation
The design examples are calculated here by applying all three design approaches
as far as possible using the specific German design approach. Based on the results of
the design overall factors of safety are given as well which are compared with the
global safety factors according to DIN 1054:1976.
4.1. Example of a Shallow Foundation
The square pad foundation of the ETC10 Example 2.2 is illustrated in Figure 3.
characteristic loads:
Gv,k = 1000 kN Gh,k = 0
Qv,k = 750 kN Qh,k = 500 kN
soil:
boulder clay
site investigations:
5 SPTs, water contents and index tests
bulk weight of soil:
J = 21,4 kN/m³
ground water level:
1,0 m below subsurface
Figure 3. Example 2.2 – square pad foundation (www.eurocode7.com/etc10)
For the design approaches DA1 and DA3 the partial factors according to Eurocode
7, Annex A are used. The bearing resistance of the pad foundation is calculated
according to the German DIN 4017:2006.
The evaluation of the SPT test results to determine the shear strength of the
foundation in drained and undrained conditions reveals some problems as the relevant
German DIN EN ISO 22476-3:2005 does not include any correlations of SPT N-
Values to the shear strength of the respective soils. SPTs are not very common in
Germany in contrast with dynamic probing and separate boring with soil sampling.
Consequently, there are no well established correlations available. After analyzing
various correlations documented in the literature (see e.g. Bowles, 1996), the following
experience values for the shear strength of the boulder clay adopted from von Soos &
Engels (2008) are finally assumed to be applicable for this problem:
c u ,k 300 kN m² cck 15 kN m² tan Mck 30q (5)
The analysis shows that the bearing resistance determines the relevant dimensions
of the square pad foundation. Table 7 includes the resultant pad widths and the overall
factors of safety for the different design approaches. Accordingly, the pad width in case
of undrained conditions determined with DA2* is only minimally smaller compared to
the other design approaches. The overall factor of safety achieved in this case is
K 1,90 , which is less than the global safety factor formerly required in DIN
1054:1976 for LC1 of K 2 and is therefore not acceptable. However, undrained
conditions often represent transient situations (e.g. during construction) which were
formerly attributed to LC2 with a minimum global safety factor of Kmin 1,5 (see
@seismicisolation
@seismicisolation
Table 2). Hence, the design according to the new generation of standards leads to a
sufficient safety level.
Table 7. Results of the square pad foundation design (Example 2.2)

DA1 (Comb. 1) DA1 (Comb. 2) DA2* DA3
Pad width [m]
Undrained
2,53 2,63 2,89 2,53
conditions
Drained
3,44 3,45 4,69 3,44
conditions
K = Rd/Nd or K= Rk/Nk)
Overall factor of safety (K
Undrained
1,01 1,02 1,90 1,01
conditions
Drained
1,02 1,03 2,11 1,02
conditions
For drained conditions the pad width designed with DA2* is much greater than the
one according to the other design approaches which result in more or less the same
widths. This can be attributed to the comparably high partial factor for bearing
resistance of JR,v = 1,4. The design represents an overall factor of safety of 2,11 which
is slightly greater than the global safety factor in LC1 of Kmin 2 (see Table 2).
In general, the safety level achieved for this example is indeed comparable to the
former global safety concept. However, the analysis shows that the evaluation and
interpretation of site investigation results depends on national design traditions as well
as on the experience of the designer. These uncertainties are not included in the partial
factors of DIN 1054:2010 derived from the former global safety factors in the way
illustrated in section 3.1. They have to be considered within the determination of the
characteristic soil parameters instead.
Several recommendations how to evaluate site investigation results (e.g. Ruppert,
2012) as well as statistical methods to determine characteristic soil parameters from
test results which implicitly account for such uncertainties are available (e.g. Baudin,
2001; Pohl, 2011). However, especially statistical methods are often not applied in the
daily design practice. The main reason for this can be found in the usually small sample
sizes together with a lack of experience in the use of statistical methods. Hence, design
experience along with knowledge of the local subsoil conditions is usually essential for
the determination of the characteristic soil parameters by qualitatively taking into
account uncertainties related to the interpretation of site investigation results.
4.2. Example of a Pile Foundation
The pile group of ETC10 Example 2.6 is depicted in Figure 4. The pile length is
designed according to the procedure described in DIN 1054:2010 and the
recommendations of the committee for piles (EAP, 2012) using DA1 combination 1
and DA2*. DA1 combination 2 and DA3 are not considered as the calculation model
used here is based on experience values for the resultant characteristic pile base and
shaft resistances tabulated in EAP (2012), see section 3.3. So, the material factor
approach of DA1, combination 2 and DA3 cannot be used.
@seismicisolation
@seismicisolation
bored piles:
D = 450 mm, a = 2 m
characteristic loads for each pile:

Gk = 300 kN
Qk = 150 kN
soil:
pleistocene fine and medium sand
covered by Holocene layers of loose
sand, soft clay and peat
site investigations:
1 CPT performed and evaluated
according to DIN 4094:2002, 1 bore
profile
ground water level:

1,4 m below subsurface
Figure 4. Example 2.6 – pile group of bored piles (www.eurocode7.com/etc10)
As the soil in the upper 15 m was very inhomogeneous and shows comparably
small CPT tip resistances, only the soil deeper than 15 m is considered as the bearing
layer. An average value of the cone tip resistance of this layer which is considered as
fine sand of q c 11,18 MN m² is derived from an evaluation of the CPT results. The
results of the pile design are summarized in Table 8.
According to EAP (2012) the minimum embedment depth in the competent layer
(fine sand) is 2,5 m. This results in a final pile length of 17,5 m. According to Table 8,
the overall factor of safety is only 1,55 which is significantly less than the global safety
factor of K 2 required in DIN 1054:1976 for LC1. Hence, the old safety level is not
maintained in this example, which was the prerequisite of implementing Eurocode 7 in
Germany.
Table 8. Results of the pile design (Example 2.6)

DA1 (Comb. 1) DA1 (Comb. 2) DA2* DA3
Total pile length
17,45 - 17,25 -
L [m]
Overall factor of
1,01 - 1,55 -
safety K>@
On the other hand, some uncertainty is related to the evaluation of the CPT results,
i.e. in defining representative CPT tip resistances for the bearing layer in this example.
As already mentioned in section 4.1 such uncertainties are not covered by the partial
factors defined in DIN 1054:2010. In this case they have to be accounted for by a
@seismicisolation
@seismicisolation
reasonable definition of the characteristic values of the base and shaft resistance from
the tabulated values in EAP (2012). These values themselves are based on mean values
of the cone tip resistance for a realistic soil profile indicating relevant sublayers to be
derived from the CPT measurements (see also Ruppert, 2012).
In the end, it is obviously difficult to give a general assessment of the success of
the implementation of Eurocode 7 as it depends very much on the type of example, its
boundary conditions as well as on the calculation methods used for comparison.
5. Conclusions
The implementation of the Limit State Design with Eurocode 7 represented a radical
change in the German design philosophy which was based on long-term experience and
which was commonly considered to be very reliable.
Engineers had to adjust to the new concept of limit states and partial factors and
the new terminology with the introduction of DIN 1054:2003/2005 parallel to
Eurocode 7 (the “German way”). However, after some years of experience it can be
concluded that the LSD is broadly accepted among geotechnical engineers.
With the deadline of July 2012 DIN 1054:2005 can no longer be used and
engineers again need to adjust to changes accompanied by the final implementation of
Eurocode 7. These changes not only include new and sometimes unclear regulations
but again new definitions and terminologies as well as a completely new structure of
the basic design codes.
In the near future three codes - Eurocode 7, its National Annex plus a fully revised
DIN 1054:2010 as a supplement to Eurocode 7 - will be used in geotechnical design
together with other design and construction codes. A standard-handbook has been
published to make the practical use of these design codes easier.
The safety level included in these codes is not based on probabilistic calculations,
but has been derived from the former global safety concept. It can be assumed that the
achieved safety level is at least comparable to the traditional design concept in most
cases, but the actual reliability of geotechnical structures remains unknown.
References
Baudin, C. (2001): Determination of Characteristic Values, in: Geotechnical Handbook - part 1, ed. U.
Smoltczyk, 6th edition, Ernst & Sohn, Berlin
Bowles, J.E. (1996): Foundation analysis and design, 5th Edition, McGraw-Hill, New York
DIN 1054 (1976): Baugrund; Zulaessige Belastung des Baugrunds, Normenausschuss Bauwesen im DIN
e.V., Beuth Verlag, Berlin (in German)
DIN 1054 (2003): Baugrund - Sicherheitsnachweise im Erd- und Grundbau, Normenausschuss Bauwesen im
DIN e.V., Beuth Verlag, Berlin (in German)
DIN 1054 (2005): Baugrund - Sicherheitsnachweise im Erd- und Grundbau, Normenausschuss Bauwesen im
DIN e.V., Beuth Verlag, Berlin (in German)
DIN 1054 (2010): Baugrund - Sicherheitsnachweise im Erd- und Grundbau - Ergaenzende Regelungen zu
DIN EN 1997-1, Normenausschuss Bauwesen im DIN e.V., Beuth Verlag, Berlin (in German)
DIN 4014 (1990): Bohrpfaehle - Herstellung, Bemessung und Tragverhalten, Normenausschuss Bauwesen
im DIN e.V., Beuth Verlag, Berlin (in German – withdrawn in 2003)
DIN 4017 (2006): Baugrund - Berechnung des Grundbruchwiderstands von Flachgruendungen,
Normenausschuss Bauwesen im DIN e.V., Beuth Verlag, Berlin (in German)
DIN 4084 (2009): Baugrund – Gelaendebruchberechnungen, Normenausschuss Bauwesen im DIN e.V.,
Beuth Verlag, Berlin (in German)
@seismicisolation
@seismicisolation
DIN EN 1990 (2010): Eurocode: Grundlagen der Tragwerksplanung, Normenausschuss Bauwesen im DIN
e.V., Beuth Verlag, Berlin (in German)
DIN EN ISO 22476-3 (2005): Geotechnische Erkundung und Untersuchung – Felduntersuchungen, Teil 3:
Standard Penetration Test, Normenausschuss Bauwesen im DIN e.V., Beuth Verlag, Berlin (in
German)
EAB (2006): Empfehlungen des Arbeitskreises Baugruben der DGGT – EAB, 4th edition, Ernst & Sohn,
Berlin (in German)
EAP (2012): Empfehlungen des Arbeitskreises Pfaehle der DGGT – EA-Pfaehle, 2nd edition, Ernst & Sohn,
Berlin (in German)
EAU (2004): Recommendations of the Committee for Waterfront Structures, Harbours and Waterways of the
DGGT – EAU, 8th edition, Ernst & Sohn, Berlin (translated from German)
ETC10 (2010): Proceedings of the 2nd International Workshop on the Evaluation of Eurocode 7, April 12-14
2010, Eucentre, Pavia, Italy, http://www.eurocode7.com/etc10/Pavia/proceedings.html
Eurocode 7 (2009): German version DIN EN 1997-1: Entwurf, Berechnung und Bemessung in der
Geotechnik - Teil 1: Allgemeine Regeln, Normenausschuss Bauwesen im DIN e.V., Beuth Verlag,
Berlin (in German)
Eurocode 7/NA (2010): German version DIN EN 1997-1/NA: Nationaler Anhang - National festgelegte
Parameter - Eurocode 7: Entwurf, Berechnung und Bemessung in der Geotechnik - Teil 1: Allgemeine
Regeln, Normenausschuss Bauwesen im DIN e.V., Beuth Verlag, Berlin (in German)
Gaessler, G.; Gudehus, G. (1983): Das neue statistische Sicherheitskonzept am Beispiel der Standsicherheit
verankerter Waende und vernagelter Waende, Fraunhofer IRB Verlag (in German)
Kempfert, H.-G. (2012): Pfahlgruendungen, in: Kommentar zum Handbuch Eurocode 7 – Geotechnische
Bemessung: Allgemeine Regeln, chapter B7, ed. B. Schuppener, Ernst & Sohn, Berlin (in German)
Normenhandbuch Eurocode 7 (2011): Handbuch Eurocode 7 - Geotechnische Bemessung, Band 1:
Allgemeine Regeln, 1st edition, Beuth Verlag, Berlin (in German)
Orr, T.L.L. (2005): Evaluation of Eurocode 7, Proceedings of the 1st International Workshop on the
Evaluation of Eurocode 7, Dept. of Civil, Structural and Environmental Engineering, Trinity College,
Dublin
Orr, T. (2011): Experiences with the Implementation of Eurocode 7 in Europe, Proceedings of the Workshop
on Safety Concepts and Calibration of Partial Factors in European and North American Codes of
Practice, Nov. 30 to Dec. 1 2011, Delft, The Netherlands, pp. 175-185
Orr, T.L.L.; Bond, A.J.; Scarpelli, G. (2011): Findings from the 2nd Set of Eurocode 7 Design Examples,
Proceedings of the 3rd International Symposium on Geotechnical Safety and Risk (ISGSR 2011), eds. N.
Vogt, B. Schuppener, D. Straub, G. Braeu, Bundesanstalt fuer Wasserbau, pp. 537-547
Pohl, C. (2011): Determination of Characteristic Soil Values by Statistical Methods, Proceedings of the 3rd
International Symposium on Geotechnical Safety and Risk (ISGSR 2011), eds. N. Vogt, B. Schuppener,
D. Straub, G. Braeu, Bundesanstalt fuer Wasserbau, pp. 427-434
Ruppert, F. (2012): Geotechnische Unterlagen, in: Kommentar zum Handbuch Eurocode 7 - Geotechnische
Bemessung: Allgemeine Regeln, chapter B3, ed. B. Schuppener, Ernst & Sohn, Berlin (in German)
Schuppener, B. (2010a): Eurocode 7 Geotechnical Design – Part 1: General rules and its latest developments,
Georisk: Assessment and Management of Risk for Engineered Systems and Geohazards, 4 (1), 32-42
Schuppener, B. (2010b): Das Normenhandbuch zu DIN EN 1997-1 und DIN 1054, 7. Kolloquium der
Technischen Akademie Esslingen, „Bauen in Boden und Fels“, eds. H. Schad and C. Vogt
Schuppener, B.; Heibaum, M. (2011): Reliability Theory and Safety in German Geotechnical Design,
Proceedings of the 3rd International Symposium on Geotechnical Safety and Risk (ISGSR 2011), eds. N.
Vogt, B. Schuppener, D. Straub, G. Braeu, Bundesanstalt fuer Wasserbau, pp. 527-536
Schuppener, B.; Ruppert, F. (2007): Zusammenfuehrung von europaeischen und deutschen Normen –
Eurocode 7, DIN 1054 und DIN 4020, Bautechnik, 84 (9), 636-640
Schultze, E.; Pottharst, R. (1981): Versagenswahrscheinlichkeit und Sicherheit von Flachgruendungen als
Grundlage für Bauvorschriften, parts 1-3, Fraunhofer IRB Verlag (in German)
Vogt, N.; Schuppener, B.; Weißenbach, A. (2008): Implementation of Eurocode 7-1 in Germany – Selection
of Design Approach and Values of Partial Factors, Proceedings of the 11th Baltic Sea Conference –
Geotechnics in Maritime Engineering, September 15-18, Gdansk, Poland, pp. 1035-1042
Von Soos, P.; Engels, J. (2008): Eigenschaften von Boden und Fels – ihre Ermittlung im Labor, in:
Grundbau-Taschenbuch, Teil 1: Geotechnische Grundlagen, ed. K. J. Witt, 7th edition, Ernst & Sohn,
Berlin
Weißenbach, A. (2012): Die Entwicklung von DIN 1054, in: Kommentar zum Handbuch Eurocode 7 -
Geotechnische Bemessung: Allgemeine Regeln, chapter A2, ed. B. Schuppener, Ernst & Sohn, Berlin
(in German)
@seismicisolation
@seismicisolation
IOS Press, 2013
doi:10.3233/978-1-61499-163-2-60
Implementation of Eurocode 7 in French

practice by means of national additional
standards
Jean-Pierre MAGNAN a,1 and Sébastien BURLON a
a
Université Paris–Est, IFSTTAR, GER, F–75015, Paris, France
Abstract. This paper deals with the implementation of Eurocode 7 in French

practice by means of national additional standards. The architecture of French
geotechnical design standards is presented. In particular, principles of safety
management and of determination of characteristic value are explained. Then, for
each geotechnical structure type, the values of the partial factors used are
summarized and the methodology of their calibration is briefly discussed.
Keywords. Eurocodes, safety factor, geotechnical structures, calibration
Introduction
French engineers in charge of the standardization process for geotechnical design

consider Eurocode 7 (2004 and 2006) as an umbrella code which defines the
fundamental principles of geotechnical structures design. It cannot be used on a daily
basis for projects. Some sections of it must be developed to reach the level of standards
and regulations which existed before (Documents Techniques Unifiés DTU which were
included in the series of French standards for buildings and Fascicule 62-V for public
works such as roads, bridges, etc.) and which were more detailed.
French engineers are facing a situation where all parts of geotechnical design are
not covered by an existing standard (in particular slopes). Moreover, different rules can
be applied according to whether the structure is a bridge or a building, especially in the
case of foundations, whereas Eurocode 7 does not make such distinction.
The first part of this paper deals with the new architecture of French geotechnical
standards and discusses the management of safety and the determination of
characteristic values as recommended by Eurocode 7. The second part of the paper
presents the seven national standards defining the requirements and the
recommendations to design each type of geotechnical structures (slopes, spread
foundations, deep foundations, etc.)
1
Corresponding Author: E-mail: jean-pierre.magnan@ifsttar.fr.
@seismicisolation
@seismicisolation
J.-P. Magnan and S. Burlon / Implementation of Eurocode 7 in French Practice 61
1. The new architecture of French geotechnical design standards
1.1. Principles
The new French design standards include:

• Eurocode 7 (NF EN 1997-1 and 2) (2004 and 2006) and the French National
Annex of Eurocode 7 – Part 1(NF EN 1997-1/NA) (2006);
• seven application standards for shallow foundations (NF P94-261, under
preparation), pile foundations (NF P 94-262, published in 2012), retaining
walls (NF P 94-281, under preparation), embedded retaining walls (NF 94-282,
published in 2009), reinforced earth structures (NF P 94-270, published in
2009) and earth structures (NF P 94-290, still not undertaken).
Other recommendations have been published or are being prepared as the final
result of “national projects”, funded by the ministry in charge of construction
(buildings and infrastructures) and by the National Research Agency. They deal with
micropiles (FOREVER), raft foundation design in terms of ultimate capacity and
displacements (ASIRI) and the effects of cyclic loads on piles (SOLCYP).
The most important French choices are as follows:
1. Adopt Design Approach 2 (or 2*) as a quasi general rule, some situations
being referred to a distributed safety factor (Design Approach 3).
2. Do not rely on statistical concepts and analyses, but rather on experience and
case histories.
3. Yet accept methods strongly referred to in Eurocode 7, such as the model pile
approach to pile design, but try to make them equivalent to concurrent
(competing) alternative methods in all cases when they should be equivalent
(typically only one or two tests or test profiles).
4. Rely on proven methods used in our country (such as the use of
pressuremeter).
5. Open the list of accepted design methods to classical methods used in other
countries.
1.2. Safety management

The general format for safety in Design Approach 2 is based on actions, resistances and
methods of evaluation. Partial factors are applied to actions A (γAA), to resistances R
(R/γR) and to models (γMA for action model or γMR for resistance model) (see e.g., Frank
et al., 2004 and Bond & Harris, 2008). In the basic case of one action and one
resistance, this leads to an equation such as:
R
γ MAγ A A < Eq. (1)
γ R γ MR
This enables to account for different (more or less satisfactory) ways of estimating
actions (and effects of actions) and resistances. In the past, stability checks in
geotechnical engineering were expressed in terms of one global safety factor F, such as:
@seismicisolation
@seismicisolation
62 J.-P. Magnan and S. Burlon / Implementation of Eurocode 7 in French Practice
R
A< Eq. (2)
F
Provided that modern characteristic values are equal to the former design values,
this implies that:
F = γ Aγ MAγ R γ MR Eq. (3)
Since French engineers did not intend to change the result of design and decided to
rely on former experience, they tried to find combinations of partial factors which
would yield the same usual global value for each type of structure and mode of failure.
There is no relationship between the usual global safety factor and any theoretical
probability of failure. Engineers think that certain values are adequate for the design of
a given type of structure. Experience confirmed this and owners, constructors and
insurance companies accept this as a normal situation. Most research publications were
unable to estimate a likely probability of failure for reference cases, most of the time
because the soil model did not depict reality and the calculation model was too
simplified.
Design Approach 3 is allowed for slope stability problems, reinforced slopes and
overall stability where it is easy to use partial factors on the effective angle of shearing
resistance and cohesion. Moreover this approach is also used for numerical methods.
The calibration of safety factor values in Design Approach 3 is still discussed. In order
to keep the same levels of safety as before, it was decided, for series of simple
geotechnical structures, to compare numerical analyses performed with finite element
method with traditional design which has been calibrated against reality. For the time
being, for example, undrained analyses are performed to study stability of
embankments on soft soils and this will be made compulsory for finite element method,
whereas drained characteristics of soil are used for slope stability problems in the
framework of effective stress. Besides, model factors will be put on the calculation
method, if necessary.
The partial factors used in France are defined in two types of documents: the
French national annex of Eurocode 7 – Part 1 (2006) and the application standards. In
the French national annex of Eurocode 7 – Part 1 (2006), it is mentioned that the partial
factors in the annex A of Eurocode 7 – Part 1 (2004) can be used with the values
proposed, unless the application standards provide different values. In practice, in order
to avoid the production of national documents with different values for the same partial
factor, those proposed by the annex A of Eurocode 7 (2004) are never modified and
national standards provide additional model factors to achieve a sufficient level of
safety.
The principles of partial factor calibration for different types of geotechnical
structures are presented in detail later.
1.3. Characteristic values
Section 2.4.5.2 of Eurocode 7 (2004) introduces the concept of characteristic value and
proposes two methods to estimate it.
@seismicisolation
@seismicisolation
The first method is based on the existing situation in which each geotechnical
engineer evaluates his best careful estimate of the soil properties to be used for design.
Since it was felt impossible to explain what should be changed in this process to get
characteristic values differing from the previous design values, as mentioned above, no
change can be required in everyday practice. The differences between individual
evaluations by engineer are accepted and they are covered by the existing safety factors.
The second method includes a statistical analysis of soil properties to estimate a
value with a 95 % confidence level. Note that in this case, uncertainties originate not
only from soil properties but also from calculation models; 95 % on characteristic
values of soil properties yields higher confidence levels on the values of resistance.
Since most of the resistance calculation models have been calibrated using mean values
of soil properties then it is too conservative to estimate soil properties with a 95 %
confidence level. The methods which lead to the characteristic value are not just an
automatic process but have to be related both to the spatial scatter of soils and to the
strain mechanism affecting the geotechnical structure. For instance, the characteristic
value of the friction angle which is used to study a landslide or a shallow foundation
submitted to an axial load is not the same because the soil volume which can be
mobilized is quite different. The method used to estimate the characteristic value of a
soil property has to handle these two problems. The characteristic value must be
obtained by considering the soil volume which can really be mobilized.
In France, both methods are allowed but as there is no obvious methodology to
define the soil volume affected by the strain mechanism of a geotechnical structure, the
first method is preferred.
2. French national standards
2.1. Slopes and overall stability
This type of problem can be analysed using Design Approach 2 or 3 and is common to
all national standards. The choice between Design Approaches 2 and 3 depends on the
calculation method. In order to use Design Approach 2, the calculation method has to
show distinctly the actions and the resistances. For instance, the Bishop method does
not provide such a distinction since the calculation of the safety factor is implicit.
However, a standard is expected to deal with these problems in detail. In general, the
global safety factor used in France is between 1.3 and 1.5 according to the structure
sensitivity to deformations. The partial factors used are those proposed by Eurocode 7
(2004): for resistances, with Table A.14, γR=1.1 for Design Approach 2 and γR=1.0 for
Design Approach 3. In Design Approach 2, Table A.3 is used with a partial factor
equal to 1.35 for the action. A model factor of 0.9 to 1.0 is introduced, in order to
account for the sensitivity of the structure to deformations. In Design Approach 3,
Table A.4 is used for soil parameters (c’, ϕ’, cU) and Table A.14 provides partial
factors for resistance. In this case, the model factor varies between 1.1 and 1.2.
A recapitulation of the partial factors used is presented in Table 1.
@seismicisolation
@seismicisolation
2.2. Reinforced soil: soil nailing and reinforced soil (NF P 94–270)
The design of these structures is based on Design Approach 3 for stability analyses
using the Bishop or Fellenius method. Partial factors are applied to (Table 2):
• the undrained and drained shear strength parameters for the soils;
• the reinforcement resistance: axial tensile and bending moment;
• the lateral pressure contact or the axial friction between the soil and the
reinforcement (the lateral pressure is deduced from the Menard limit pressure
and the axial friction is estimated from full-scale tests performed before the
works).
The other modes of failure, such as sliding or bearing capacity failures, are
checked according to the principles given in the corresponding national application
standards.
Table 1. Partial factors used for slopes and overall stability
Design Approach 2
Annex A National Standard
Actions 1.35 1.35

Resistances 1.0 1.0 or 0.9
Design Approach 3
Angle of shearing
1.25 1.25
resistance
Effective cohesion 1.25 1.25
Undrained shear strength 1.4 1.4
Resistances 1.0 1.2 or 1.1
Table 2 Partial factors used for reinforced soil
Design Approach 3
Angle of shearing
1.25 1.25
resistance
Effective Cohesion 1.25 1.25
Undrained shear strength 1.4 1.4
Lateral pressure Not defined 1.4
Axial friction Not defined 1.4
@seismicisolation
@seismicisolation
2.3. Spread foundations (NF P 94-261)
The design of spread foundations is based on Design Approach 2 both for bearing and
sliding resistances. It is interesting to mention that similar design principles have been
used in France for many years before the application of Eurocodes.
Bearing resistance

The set A1 of Table A.3 (Eurocode 7, 2004) is used for the actions and the set R2 of
the same table for the resistance. For the actions, the values defined by our former
standards match very well to those presented in the Eurocodes. They were respectively
equal to 1.35 and 1.5 for the permanent actions and the variable actions. For the
resistances, the partial factor defined in Eurocode 7 (1.4) is not sufficient.
According to the French experience of shallow foundation design, which is
nowadays essentially based on the use of pressuremeter test results, a global safety
factor on the resistance equal to 2 was used. Nevertheless, during the development of
the national standard, this value was decreased to 1.68 including a model factor equal
to 1.2. For sands, this value was determined from a statistical analysis of 29 static load
tests. Finally, the design resistance Rc;d can be expressed as:
k p ple* iδ i β
Rv;d = Aie + q0' Eq. (4)
γ R ; d γ R ;v
A is the area of the foundation;

q0' is the vertical effective stress at the foundation level;
ple* is the equivalent limit pressure estimated using Menard pressuremeter values;
k p is the experimental bearing factor;
iδ , iβ and ie are the factors which take into account the load inclination, the surface
inclination β of a slope near the foundation and the load eccentricity;

γ R;d =1.4 is the partial factor defined by Eurocode 7 and γ R;v =1.2 is the model factor
defined by the French national standard.
Methods using the cone penetration test are also allowed, using the same partial
factors. The equivalent limit pressure is then replaced by the equivalent cone resistance.
The calculation of bearing resistances using classical soil parameters (c’, ϕ’, cU) is
not very much used in France but it was kept within Design Approach 2. The value of
the model factor was set to 1.2 for undrained shear strength parameters and to 2.0 for
effective shear strength parameters. The value of 1.2 has been chosen because it has
been considered that the correlation between the undrained cohesion and the Menard
limit pressure is well calibrated whereas the correlation of effective shear strength
parameters with the French reference pressuremeter or penetrometer approach is less
precise. Moreover the calculation methods used for (c’, ϕ’) design refer mainly to
homogeneous soil layers.
The problem of stress distribution under the foundation (rectangular stress
distribution according to Meyerhof’s theory (Meyerhof, 1953) or triangular stress
@seismicisolation
@seismicisolation
distribution according to Navier’s theory) does not appear any longer in the design
since only forces are considered. It is taken into account by means of parameter ie
which is defined as:
ie = (B − 2eB )(L − 2eL ) Eq. (5)
B and L are the dimensions of the foundation;

eB and eL are the load eccentricities in the direction of the width B and of the length
L.
Sliding resistance
As previously, the set A1 of Table A.3 (Eurocode 7, 2004) is used for the actions and
the set R2 of the same table is used for the resistance. The value proposed by the annex
A for sliding resistance within Design Approach 2 is equal to 1.1. Former standards
considered a global safety factor of 1.2 for the resistance and a model factor equal to
1.1 was introduced in order to obtain the same design.
Summary of partial factors used

The different partial factors both for bearing and sliding resistance are summarized in
Table 3.
Table 3. Partial factors used for spread foundations
Design Approach 2
Annex A National Standards
Actions 1.35 1.35
Bearing resistance (in-situ tests, undrained
1.4 1.4 x 1.2 = 1.68
shear strength)
Bearing resistance (drained shear strength) 1.4 1.4 x 2 = 2.8

Sliding resistance 1.1 1.1
2.4. Deep foundations (NF P 94-262)
The design of deep foundations is analysed within the framework of Design Approach
2 both for axial and lateral loadings.
Axial loads

As for spread foundations, the partial factors for the action are defined in Table A.3
(Eurocode 7, 2004). Concerning the resistance, the French standard uses both the
method with factors ξ and the alternative method with the partial factors presented in
Table A.7 or A.8 (γt= γb=γs=1.1). For the two methods, a model factor γRd;1 has been
calibrated from the statistical analysis of 150 static load tests and new rules including
recent pile technologies have been defined. It is important to mention that these new
@seismicisolation
@seismicisolation
rules are less conservative than previous French codes: bearing capacity increases to
10 %. The model factor γRd;1 on bearing capacity is equal to 1.15 and corresponds to a
15% probability of overestimation of the resistance, which is the same as for our
former rules both for buildings and bridges (Figure 1).
(Rc;cal is the calculated value of the ultimate compressive resistance of the pile according to the model, Rc;mes
is the measured value of the ultimate compressive resistance of the pile and F(Rc;cal/Rc;mes) is the distribution
function of the random variable Rc;cal/Rc;mes)
Figure 1. Distribution function of ratio Rc;cal/Rc;mes
These rules include different parameters:

• the base resistance qb estimated from a bearing capacity factor (respectively
k p and kc for pressuremeter tests and cone penetration tests) which is related
to five types of soils (clay, sand, chalk, marl, altered rock) and to twenty types
of piles:
qb = k p p *LM ;e Eq. (6)
p *LM ;e is the equivalent limit pressure (i.e., the mean of the limit pressure values under
the base of the pile);
q b = k c q c ;e Eq. (7)
@seismicisolation
@seismicisolation
q c;e is the equivalent cone resistance (i.e., the mean of the cone resistance values
under the base of the pile);
• curves for unit shaft friction qs , for the five same categories of soils:
q s ( X ) = (a + bX ) 1 − e − cX ( ) Eq. (8)
(a, b and c are calibration parameters)

X may be the Menard limit pressure pLM or the cone resistance qc .
• the installation parameter α which acts on the unit shaft friction values.
For the method with correlation factors ξ, some changes were made to account for
the investigation area S. The suggested values are assumed to correspond to an
investigation area Sref equal to 2500 m2 and the new ones are calibrated to be equal to 1
for a zero investigation area. The minimal value for S has been set to 100 m 2. The new
expression of the correlation factor ξ is the following:
S
ξ i ( N , S ) = 1 + [ξ i' ( N ) − 1] Eq. (9)
S ref
(i=3 or 4 according to the mean or the minimum value)

ξ is the new correlation factor;
ξ ' is the correlation factor defined by Eurocode 7 (2004) (Tables A.9, A.10 and A.11);
N is the number of test profiles;
S is the investigation area, with a minimum value equal to 100 m2;
S ref is the reference area equal to 2500 m2.
Instead of using correlation factors, it is possible to implement a statistical
approach based on the principles described in Annex D of Eurocode 0 (2003). The

estimation of a characteristic value can be performed using the average and the
standard deviation of N values.
For this method, three factors are necessary:
• ξ which takes into account the variability of the soil properties and the extent
of the field – the annex D of Eurocode 0 is an alternative method;
• γRd;1 which takes into account the accuracy of the model defined by the new
rules
(these first two factors allow to obtain the characteristic value);
• γt which is the partial factor defined by Eurocode 7, which enables to deduce
the design value from the characteristic value.
The global safety factor for the resistance is approximately equal to 1.4 (ξγRd;1γt).
For the alternative method, it is necessary to define characteristic values for soil
properties in the various strata. Nevertheless, as the concept of characteristic values is
not commonly used in geotechnical engineering and as the calibration of the calculation
model was not completely done, French engineers preferred to use representative
@seismicisolation
@seismicisolation
values, which is consistent with the past practice. In order to account for the fact that
representative values rather than characteristic values are used, a model factor γRd;2
equal to 1.1 is defined and enables to obtain the same global safety factor as before
(γRd;1γRd;2γt=1.4). It is expected that the value of this model factor decreases to 1.0 when
the procedures to assess characteristic values will have been developed.
For both methods, the model factor γRd;1 is set to other values in the case of tensile
actions (γRd;1=1.4) and grouted piles (γRd;1=2.0) where the technologies seem to be very
sensitive as the results of static load tests are very scattered.
Lateral loads

The design of piles with respect to lateral loading is performed following Design
Approach 2. Partial factors are applied to the actions according to the values of Table
A.3. For the resistance, no coefficient is considered. The contact pressure between the
soil and the pile is limited to the creep pressure, as measured during a pressuremeter
test.
Summary of partial factors used

The different partial factors both for bearing and lateral resistance are summarized in
Table 4.
Table 4. Partial factors used for deep foundations
Design Approach 2
Actions 1.35 1.35
ξ(N,S)
ξ(N)
Annex D of EC0
Method with factors ξ
γRd;1: not defined γRd;1 = 1.15
Compression
γRd;2: not defined γRd;2 : not needed
γt = 1.1 γt = 1.1
γRd;1: not needed γRd;1 = 1.15

Alternative method
γRd;2: not needed γRd;2 = 1.1
Compression
γt = 1.1 γt = 1.1
Lateral loadings Not defined 1.0 (creep pressure)
2.5. Retaining structures – Gravity walls (NF P 94–281)
The design of these structures is performed using Design Approach 2. In fact, this
standard has many common aspects with the standards for shallow and deep
foundations. For gravity walls based on shallow foundations, the model factor with
respect to the bearing resistance is decreased to 1.0 or 1.7 according to the calculation
model to take into account the fact that for these structures the ultimate state is reached
for more important deformations than for foundations supporting buildings. In fact,
@seismicisolation
@seismicisolation
gravity walls on spread foundations are considered to be less sensitive to deformations

than buildings or bridges on spread foundation. Concerning the sliding resistance, the
model factor is decreased to 0.9 for the same reasons.
A list of the partial factors used is presented in Table 5.
Table 5. Partial factors used for gravity walls
Design Approach 2
Actions 1.35 1.35
Bearing resistance (in-situ tests,
1.4 1.4 x 1.0 = 1.4
undrained shear strength)
Bearing resistance (drained shear
1.4 1.4 x 1.7 = 2.4
strength)
Sliding resistance 1.1 1.1 x 0.9 = 1.0
2.6. Retaining structures – Embedded walls (NF P 94–282)
The design of these structures is performed using Design Approach 2 or 2* (the partial
factors are applied to the effects of the action). Design Approach 2 is required for
cantilever walls whereas design approach 2* is required for anchored walls.
For cantilever wall, active and passive states can be clearly identified so that the
Design Approach 2 can be easily performed: the characteristic values of active forces
are multiplied by the partial factor concerning the actions and equal to 1.35 and the
characteristic values of passive forces are divided by the partial factor concerning the
resistances and equal to 1.4. The values obtained at the end of the calculation in terms
of bending moments are the design values.
Anchored walls are usually designed using a soil-structure interaction method
including subgrade reaction modulus. For these calculations it is complex to determine
if the soil reaction must be considered as a resistance (passive state) or an action (active
state). In practice, the calculation is performed without any partial factor. The results
concern on the one hand the characteristic values of the bending moment of the wall
and the tensile resistance of the anchor and on the other hand the stress state around the
wall. The characteristic values of the bending moments and of the shear forces are
multiplied by the partial factor concerning the actions and equal to 1.35. The analysis
of the stress state around the wall enables to determine the ratio between the stresses
which could be mobilized and the stresses which are mobilized. The value of this ratio
has to be larger than 1.9 (≈1.35 x 1.4).
For ultimate states induced by hydraulic forces (UPL – loss of equilibrium of the
structure or the ground due to uplift by water pressure (buoyancy) or other vertical
actions and HYD – hydraulic heave, internal erosion and piping in the ground caused
by hydraulic gradients) which are very common for these structures, the values
proposed in Annex A of Eurocode 7 (2004) were adopted.
A list of the partial factors used is presented in Table 6.
@seismicisolation
@seismicisolation
Table 6. Partial factors used for embedded walls
Design Approach 2*
Annex A National Standards
Effects of the actions 1.35 1.35
Earth resistance 1.4 1.4
Conclusions
The application of Eurocode 7 in France requires engineers to develop standards

compatible with the imposed calculation format. Seven application standards will be
published in a few months. Various partial factors had to be calibrated in addition to
those recommended by the Annex A of Eurocode 7. This work was presented in this
paper. It raised many questions about the reliability of empirical and theoretical rules,
which are still not completely solved.
Besides, the creation of geotechnical models which are linked to the estimation of
characteristic values is considered more as an art than a repeatable process which is
independent of the person doing the job. As a consequence, it seemed too complex to
explain to geotechnical engineers how they should modify their practices of
interpreting geotechnical information to obtain a geotechnical model. The new
characteristic values had therefore to be equal to the former values that the engineers
used for design. The main task was not to try to improve design but only to change the
format of the design process.
References
AFNOR (to be published). NF P 94-261 – Shallow foundation design, AFNOR, Paris

AFNOR (2012). NF P 94-261 – Deep foundation design, AFNOR, Paris
AFNOR (2009). NF P 94-2170 – Reinforced earth structures, AFNOR, Paris
AFNOR (to be published). NF P 94-281 – Gravity wall design, AFNOR, Paris
AFNOR (2009). NF P 94-282 – Retaining wall design, AFNOR, Paris

AFNOR (to be published). NF P 94-290 – Earth structures, AFNOR, Paris
Bond, A. & Harris, A. (2008). Decoding Eurocode 7, 1st edn. New-York: Taylor Francis.
Eurocode 0 (2004) (NF EN 1990) Basis of structural design, European Committee for Standardization,
Brussels.
Eurocode 7 – Part 1 (2004) (NF EN 1997 – 1) Geotechnical design – Part 1: General rules, European
Committee for Standardization, Brussels.
Eurocode 7 – Part 1 – National Annex (2006) Geotechnical design – Part 1: General rules, AFNOR, Paris.
Eurocode 7 – Part 2 (2006) (NF EN 1997 – 1) Geotechnical design – Part 2: Ground investigation and
testing, European Committee for Standardization, Brussels.
Frank, R., Bauduin, C., Driscoll, R., Kavvadas, M., Krebs Ovesen, N., Orr, T. & Schuppener, B. (2004).
Designer's Guide to EN 1997-1 Eurocode 7: Geotechnical Design – General Rules, 1st edn. London:
Thomas Telford.
Meyerhof, G.G. (1953). The bearing capacity of foundations under eccentric and inclined loads. Proceedings,
3rd International Conference on Soil Mechanics and Foundation Engineering, Zürich, vol. 1, 440–445.
@seismicisolation
@seismicisolation
IOS Press, 2013
doi:10.3233/978-1-61499-163-2-72
Implementing Eurocode 7 to achieve

reliable geotechnical designs
Trevor ORR
Trinity College Dublin, Ireland
Abstract. This paper examines how the different measures to achieve safe and
reliable geotechnical designs are implemented in Eurocode 7. These measures
include: use of the concept of geotechnical categories to take account of the
complexity of geotechnical designs, procedures for managing the design and
execution processes, requirements for the selection of characteristic parameter
values, and recommended partial factor values associated with the three Design
Approaches. How the consequences and reliability classes, set out in the head
Eurocode, EN 1990, are implemented in Eurocode 7 for geotechnical designs is
also examined. In additions, the paper presents the results of surveys to investigate
how Eurocode 7 is being adopted in the different European countries and the some
experiences in those countries with its implementation.
Keywords. Eurocode 7, geotechnical design, partial factors, reliability,

consequences classes, implementation
1. Introduction
In 1975 the European Commission decided to prepare set of technical rules for the
design of construction works. The objectives of this program included the elimination
of technical obstacles to trade and the harmonization of technical specifications within
Europe. Arising from this, 10 Eurocodes consisting of 58 parts have been prepared and
published by CEN, the European Committee for Standardization, with EN 1997:

Eurocode 7 Geotechnical Design, consisting of two parts (CEN 2004, CEN 2007). One
purpose of the Eurocodes is to serve as reference documents to prove compliance of
building and civil engineering works with the essential requirements for mechanical
resistance and stability and for safety in case of fire set out in European Construction
Products Directive, 89/106/EEC of 21st December 1988, which is due to be replaced by
the Construction Products Regulation on 1st July 2013. These essential requirements are
performance based and specify that building and construction works must be fit for
their intended use, account being taken of economy and subject to normal maintenance.
These requirements have been implemented in the EU countries through national
building regulations, which refer to the Eurocodes as demonstrating compliance of a
structural or geotechnical design with the building regulations and hence with the
Construction Products Directive. Eurocode 7 has now been implemented in most
European countries, but the way it has been implemented in each country through the
National Annexes has differed to some extent. The reasons why Eurocode 7 has been
implemented differently in each country are because of different design traditions,
different soil and climatic conditions, and, most importantly, because each country is
responsible for setting the level of safety and reliability required for its structures.
@seismicisolation
@seismicisolation
T. Orr / Implementing Eurocode 7 to Achieve Reliable Geotechnical Designs 73
2. Eurocode 7 and reliability
The head Eurocode, EN 1990 Basis of design (CEN, 2002) provides the design method
adopted in all the Eurocodes, including Eurocode 7 for geotechnical design (CEN,
2004), which is the limit state concept used in conjunction with the partial factor
method. EN 1990 {2.1(1)P}, where the { } brackets refer to the paragraph in the code,
states that “a structure shall be designed and executed in such a way that it will, during
its intended life, with appropriate degrees of reliability and in an economical way:
- Sustain all actions and influences likely to occur during execution and use,
and
- Remain fit for the use for which it is required.”
(Note that an action is a set of forces (loads) or an imposed deformation or
acceleration). As shown by the first bullet point above, the aim of the Eurocodes is to
design structures that have appropriate degrees of reliability. Reliability is defined in
EN 1990 {1.5.2.1.7} as “the ability of a structure or structural member to fulfill the
specified requirements, including the design working life, for which it has been
designed, and it is noted that reliability is usually expressed in probabilistic terms.”
Calgaro (2011) notes that structural reliability covers four aspects: safety,
serviceability, durability and robustness. He explains that robustness is the ability of a
system to resist damage while maintaining its important functions and that robustness
is not limited to physical structures but robustness principles can also be applied to
management systems to reduce the effects of unknown risks. This is in accordance with
EN 1990 {2.1(1)P} which states that “the required reliability of geotechnical designs
shall be achieved by design in accordance with EN 1990 and EN 1997-1 together with
appropriate execution and quality management measures”. Application rule {2.2(5)} in
EN 1990 on reliability management says “the required levels of reliability relating to
structural resistance and serviceability can be achieved by suitable combinations of the
following:
− Preventative and protective measures
An example in the case of geotechnical design is the placing of foundations at
appropriate depth to avoid damage due to frost heave

- Measures relating to design calculations
Examples mentioned in {2.2(5)} are the representative values of actions and
the choice of partial factors
- Measures relating to quality management
Examples given are measures aimed to reduce errors in design and execution
of the structure, and gross human errors
- Other measures relating to other design matters
According to {2.2(5)} these include: the basic requirements; the degree of
robustness (structural integrity); the durability; including the choice of design
working life; the extent and quality of preliminary investigations of soils and
possible environmental influences; the accuracy of the mechanical models
used; and the detailing.
- Efficient execution, e.g. in accordance with execution standards referred to in
EN 1990 to EN 1999
- Adequate inspection and maintenance according to procedures specified in
the project documentation.
@seismicisolation
@seismicisolation
74 T. Orr / Implementing Eurocode 7 to Achieve Reliable Geotechnical Designs
Since soil is a natural material, not manufactured, the reliability of geotechnical

designs does not just depend on selecting appropriate actions, soil parameter values,
partial factor values and calculation models but also depends on ensuring that all
aspects of the design, from geotechnical investigations to execution, are carried out
appropriately. This is reflected in the form and the content of Eurocode 7, which
provides a limit state framework for geotechnical design based on the application of
partial factors to characteristic parameters. However, it does not provide any
calculation models in the core text, only in Informative Annexes; instead it provides a
design system that includes many of the measures listed above to achieve the required
reliability level. Hence Eurocode 7 is not the same as the Eurocodes for manufactured
materials, which are generally more prescriptive, being based on agreed calculation
models and specified material parameters. The measures to achieve reliability included
in Eurocode 7 are:
1. The need for good communication. One of the assumptions in Eurocode 7,
which is given in {1.3(2)}, that is particularly pertinent to the reliability of
geotechnical designs is that there is “adequate continuity and communication
between the personnel involved in data collection, design and construction.”
Eurocode 7 aims to achieve compliance with this assumption through
requiring, as a principle {2.8(1)P}, that the details of a geotechnical design
shall be recorded in a Geotechnical Design Report which, according to
{2.8(4)P}, shall include a plan of supervision and monitoring that clearly
identifies “items which require checking during construction or which require
maintenance after construction”. Furthermore, it requires that items which
have been checked “shall be recorded in an addendum to the Report” and
{2.8(6)P} requires that “an extract of the Geotechnical Design Report
containing the supervision, monitoring and maintenance requirements for the
completed structure, shall be provided to the owner/client”.
2. The need to identify the complexity of a geotechnical design. In order to
establish the minimum requirements for the extent and content of
geotechnical investigations, calculations and construction control checks,
{2.1(8)P} requires that “the complexity of each geotechnical design together
with the associated risks shall be identified”. Hence Eurocode 7 requires that
a risk assessment, as discussed in the next section, is carried out.
3. Many lists are provided of items to be considered or taken into account in
design. Examples include the different limit states that need to be considered
for any particular design situation, for example the ultimate limit states for
spread foundations in {6.2(1)P} and for pile foundations in {7.2(1)P}.
4. A special definition for the characteristic value of a geotechnical parameter.
Since soil is a natural material, not manufactured, and since its properties
need to be determined, rather than specified, the definition of the
characteristic value of a geotechnical parameter is given in Eurocode 7
{2.4.5.2(2)P} as the value selected “as a cautious estimate of the value
controlling the occurrence of the limit state”. This definition differs from the
general, statistical definition of the characteristic value of structural material
given in EN 1990 {1.5.4.1 and 4.2(3)} as the 5% fractile value attained in a
hypothetical unlimited test series The Eurocode 7 definition demonstrates that
selecting characteristic geotechnical parameter values is part of the design
process and makes it clear that how they are selected is important for ensuring
@seismicisolation
@seismicisolation
the reliability of geotechnical designs. Although it is not anticipated that

statistical methods will normally be used to select characteristic values for
geotechnical designs, EN 1997-1 {2.4.5.2(11)} states that “if statistical
methods are used, the characteristic value should be derived such that the
calculated probability of the occurrence of a worst value governing the
occurrence of the limit state under consideration is not greater than 5%”.
5. The importance of relevant experience. The importance of having and using
relevant experience is stressed throughout EN 1997-1. For example, in
relation to the selection of characteristic values of geotechnical parameters,
{2.4.5.2(1)} states that the selection “shall be based on results and derived
values from laboratory and field tests, complemented by well-established
experience”. {1.5.2.2} defines comparable experience as “documented or
other clearly established information related to the ground being considered
in design, involving the same types of soil and rock and for which similar
geotechnical behavior is expected, and involving similar structures.
Information gained locally is considered is considered to be particularly
relevant”. How to assess who has the relevant experience and is appropriately
qualified to carry out a geotechnical design is discussed in Section 3.
Soil has the following features that have caused Eurocode 7 to differ from the
other Eurocodes codes for manufactured material; it is (a) a natural material; (b)
composed of 2 or 3 phases; (c) non homogeneous and highly variable; (d) frictional; (e)
dilatant; (f) ductile; (g) compressible and with a low shear stiffness; and (h) non-linear
with a complex stress-strain behavior. These features have affected the way in which
the limit state method in EN 1990 has been adopted for geotechnical design in
Eurocode 7, resulting in Eurocode 7 having requirements for geotechnical
investigations, a special definition for the characteristic value of geotechnical
parameters, reduced partial factors on permanent actions, and an emphasis on
serviceability limit states (Orr, 2011).
Simpson (2011) has examined the features affecting the reliability of geotechnical
designs and noted the following five features of the designer’s situation: (a) his specific
knowledge of the site; the ground conditions and their possible variability; (b) the
importance of extreme variations in the parameters causing failure; (c) the large
number of variables usually involved in a design situation; (d) the need for robustness,
i.e. providing adequate margins for secondary actions that are not related to the primary
parameters, including moderate human errors; and (e) the significance of human errors.
The execution and quality management measures included in Eurocode 7 address most
of the features identified by Simpson as affecting the reliability of geotechnical designs.
3. Design complexity and Geotechnical Categories
The first stage in achieving geotechnical designs to Eurocode 7 that have the
appropriate degree of reliability is to assess the complexity of the design. According to
Eurocode 7 {2.1(8)}, this involves establishing “the minimum requirements for the
extent and content of geotechnical investigations, calculations and construction control
checks … together with the associated risks”. The factors that need to be considered
when assessing the level of risk in a geotechnical design can be divided into the
geotechnical hazards, which include:
@seismicisolation
@seismicisolation
Table 1. Geotechnical Categories, geotechnical complexity and risk (Orr and Farrell, 1999)
Geotechnical
GC1 GC2 GC3
Category (GC)
Geotechnical Complexity and Risk Levels
Geotechnical
Low Moderate High
hazards
Ground conditions Known from comparable Ground conditions and Unusual or exceptionally
experience to ne properties can be difficult ground
straightforward. Not determined from routine conditions requiring non-
involving soft. Loose or investigations and tests routine investigations
compressible soil, loose and tests
fill or sloping ground
Groundwater No excavations below No risk of damage without High groundwater
situation water table, except where prior warning to structures pressures and exceptional
experience indicates this due to groundwater groundwater conditions,
will not cause problems lowering or drainage. No e.g. multilayered strata
exceptional water with variable permeability
tightness requirements
Regional seismicity Areas with low or very Moderate earthquake Areas of high earthquake
low earthquake hazard hazard where seismic hazard
design code (EC8) may be
used
Influence of the Negligible risk of Environmental factors Complex or difficult
environment problems due to surface covered by routine design environmental factors
water, subsidence, methods requiring special design
hazardous chemicals, etc. methods
Vulnerability Low Moderate High
Nature and size of Small and relatively Conventional types of Very large or unusual
the structure and its simple structures or structures with no structures involving
elements construction, insensitive abnormal risks abnormal risks. Very
structures in seismic areas sensitive structures in
seismic areas
Surroundings Negligible risk of damage Possible risk of damage to High risk of damage to
to or from neighboring neighboring structures or neighboring structures or
structures or services and services due, for example, services
negligible risk for life to excavations or piling
Geotechnical risk Low Moderate High
− The ground conditions

− Groundwater situation
− Regional seismicity, and
− Influence of the environment.
and the vulnerability factors, which include:
− The nature and size of the structure and its elements, and
− The surroundings.
Eurocode 7 offers {2.1(10)} the concept of three geotechnical categories, referred to as
Geotechnical Categories 1, 2 and 3 (GC1, GC2 and GC3), as a means to take account
of the different levels of complexity in geotechnical design. The use of these categories
is not a principle requirement but they provide a framework for categorizing the
different levels of risk in geotechnical design. The distinction between the categories
@seismicisolation
@seismicisolation
Table 2. Geotechnical Categories, minimum requirements, expertise required and examples of designs (Orr
and Farrell, 1999)
Geotechnical Categories
GC1 GC2 GC3
Geotechnical Qualitative investigations, Routine investigations Additional more
investigations and including trial pits involving borings, filed sophisticated
tests and laboratory tests investigations and
laboratory tests
Design procedures Prescriptive measures and Routine calculations for More sophisticated
and calculations simplified design stability and deformations analyses
procedures, e.g. design based of design
bearing resistance based procedures in Eurocode 7
on experience or published
presumed bearing
pressures. Stability or
deformation calculations
may not be necessary
Expertise required Personnel with appropriate Experienced qualified Experienced geotechnical
comparable experience personnel specialist
Examples of - Simple 1 and 2-storey Conventional types of: - Very large buildings
structures structures and - Spread and pile - Large bridges
agricultural buildings foundations - Deep excavations
having maximum - Walls and other - Embankments on soft
design column loads of retaining structures ground
250kN and maximum - Bridge piers and - Tunnels in soft and
wall load of 100kN/m abutments highly permeable
- Retaining walls and - Embankments and ground
excavation supports earthworks
where ground level - Ground anchors and
difference does not other support systems#
exceed 2m - Tunnels in hard non-
- Small excavations for fractured rock
drainage and pipes
only lies in the level of expertise required and in the nature and extent of the
geotechnical investigations and calculations to be carried out, as shown in Table 1 (Orr
and Farrell, 1999). Hence to achieve the required reliability of geotechnical designs
involving different categories, Eurocode 7 does not provide for any variation in the
partial factor values but rather requires that greater attention is given to the quality of
the investigations and design calculations.
Having assessed the complexity and level of risk of a geotechnical design, and the
geotechnical category, if this is being used, the next stage is to establish the minimum
requirements for the geotechnical investigations, design procedures and calculations
related to the categories and these are presented in Table 2 from Orr and Farrell (1999).
Therefore, as noted in Section 3, achieving the minimum requirements is dependent on
having and using appropriate experience and therefore on the personnel involved in all
aspects of a geotechnical design being appropriately qualified and having appropriate
skill and experience. Hence Eurocode 7 assumes that all aspects of geotechnical
designs are carried out by appropriately qualified and experienced personnel as it states
in {1.3(2)} that:
“− Data required for design are collected, recorded and interpreted by
appropriately qualified personnel
@seismicisolation
@seismicisolation
− Structures are designed by appropriately qualified and experienced personnel

− Execution is carried out … by personnel having the appropriate skill and
experience.”
Eurocode 7 does not define what is meant by people having appropriate
qualifications, skill and experience. This will depend on the nature of the project and
the ground conditions and has been left for national determination. Simpson and
Driscoll (1998) state that to work in a particular category a designer must be competent
to judge that the situation is not more complex than allowed within that category. An
indication of the sort of person who may be competent for the different categories is
shown in Table 2 A formal measure to address this issue and identify those with
appropriate qualifications, skills and experience has been taken in the UK by the
Institution of Civil Engineers, together with the Geological Society and the Institute of
Materials, Minerals and Mining, by the establishment of a Register of UK Ground
Engineering Professionals (RoGEP).
4. Ultimate limit states, Design Approaches and partial factors
Eurocode 7 {2.4.7.1(1)P} lists the following five different ultimate limit states, each
with a set of partial factors, that the designer needs to verify are not exceeded:
“− Loss of equilibrium of the structure or the ground, considered as a rigid body
in which the strength of structural materials and the ground are insignificant
in providing resistance (EQU)
− Internal failure or excessive deformation of the structure or structural
elements, including e.g. footings, piles or basement walls, in which the
strength of structural members is significant in providing resistance (STR)
- Failure or excessive deformation of the ground, in which the strength of the
soil or rock is significant in providing resistance (GEO)
- Loss of equilibrium of the structure or the ground due to uplift by water
pressure or other vertical actions (UPL)
- Hydraulic heave, internal erosion and piping in the ground caused by
hydraulic gradients (HYD).”

Eurocode 7 provides in Annex A only four sets of partial factor values for use in
calculations to check that none of these five ultimate limit states is exceeded since the
same set is used for both STR and GEO. The reliability of geotechnical design
calculations depends on: the characteristic values provided for the actions, the
characteristic values selected for the geotechnical parameters, as described in Section 2,
the values of the partial factor adopted, and on the calculation model used. The partial
material factors in Eurocode 7 are γM rather than γm indicating that, in accordance with
the definitions in {1.6}, they account for model uncertainties as well as uncertainties in
the material property. As no specific calculation models are provided in the core text of
Eurocode 7 and the same γM or γR values are used for all types of GEO design
situations, proper reliability calibration of geotechnical designs to Eurocode 7 is
therefore not possible.
When verifying STR and GEO ultimate limit states, partial factors may be applied
either to material properties or to resistances using one of three Design Approaches
@seismicisolation
@seismicisolation
Table 3.Design Approaches adopted by 31 CEN Member States

No or DA1,
Design
incomplete DA1 DA2 DA2* DA3 DA2 or
type
answers DA3
A, EST, D, GR, H, CH, DK, F**,
Spread BG, CR, E, B, IT, LT, EST. F CZ, IRL
IT, PL, SK, SLO NL, NO, S
foundations IS, LV, MT PT, RO, UK
CY, L, SF
A, CH, CY, D, EST, F, GR,
Pile BG, CR, E, B, IT, LT, DK, F**, NL,
H, IT, L, PL, SF, SK, SLO, CZ, IRL
foundations IS, LV, MT PT, RO, UK NO, S
SWE
Retaining BG, CR, E, B, IT, LT, A, CH, CY, D, EST, F, GR, A**, DK, F**,
CZ, IRL
structures IS, LV, MT PT, RO, UK H, IT, L, PL, SF, SK, SLO NL, NO, S
A, CH, CY, D, DK,
BG, CR, E, B, EST, IT, F**, GR, H, IT, L,
Slopes F CZ, IRL
IS, LV, MT LT, PT, UK NL, NO, PL, RO,
SF, SK, SLO, S
Total 6 6 1 - 15 5 - 18 2
* Partial action factor applied to action effect rather than to the actions
** For numerical analyses
(DA): DA1 with two Combinations, Combination 1 where partial factors are only
applied to the actions while, in Combination 2, partial factors are applied to the
variable actions and the soil parameters; DA2 with partial factors applied to both the
actions and the resistances rather than to the soil parameters; and DA3 with partial
factors applied to both the actions and the soil parameters in a single calculation. Using
the most recent information that was available to the author, the Design Approach or
Approaches adopted in the 31 CEN member states are shown in Table 3, where the
countries are represented by their conventional abbreviations. Although Table 3 is a
simplification of the actual situation in some countries, it shows that, while some
countries have adopted DA1 for all designs, most countries have adopted DA2 for
spread foundations, pile foundations and retaining structures, and DA3 for slopes.
Some countries allow either of two Design Approaches for all designs, while two
countries, the Czech Republic and Ireland, allow the use of any one of the three Design
Approaches for all designs. Also some countries that have adopted DA2 are using DA3
for numerical analyses.
The EN versions of the Eurocodes, including Eurocode 7, contain recommended

values for the partial factors. The recommended partial factor values in Annex A of
Eurocode 7 for STR/GEO ultimate limit states in persistent and transient design
situations, except those for bored piles, CFA piles and prestressed anchorages, are
presented in Table 4. Eurocode 7 {2.4.8(2)} states that the “values of partial factors for
serviceability limit states should normally be taken equal to 1.0”. However, each CEN
member state retains the exclusive competence and responsibility for setting the safety
levels of works constructed in its state in order to take account of different design
traditions, different regulatory systems and different environmental conditions. For this
reason the selection of the particular Design Approach (DA) and the values of partial
factors and other parameters in the Eurocodes, referred to as Nationally Determined
Parameters (NDP), for use in each country have been left for national choice. These
NDPs are published in National Annexes (NAs) that accompany the full texts of the
Eurocodes, which are published, unchanged, as national standards in each country; for
example in Ireland the Eurocodes are published as Irish Standards. CEN and the
European Commission would like more harmonization regarding the Design
@seismicisolation
@seismicisolation
Table 4. Recommended partial factor values for STR and GEO ultimate limit states
Parameter Symbol Design Approaches

DA1 DA2 DA3
DA1.C1 DA1.C2 Structural Geotechnical
actions actions
Partial factors on actions (γF) or the
effects of actions (γE)
Permanent unfavorable action γG 1.35 1.0 1.35 1.35 1.0
Permanent favorable action γG 1.0 1.0 1.0 1.0 1.0
Variable unfavorable action γQ 1.5 1.5 1.5 1.5 1.3
Variable favorable action γQ 0 0 0 0 0
Accidental action γA 1.0 1.0 1.0 1.0 1.0
Partial factors for soil parameters (γM)

Angle of shearing resistance γtanφ 1.0* 1.25 1.0 1.25
(this factor is applied to tanφ')
Effective cohesion γc' 1.0* 1.25 1.0 1.25
Undrained shear strength cu γcu 1.0* 1.4 1.0 1.4
Unconfined strength qu γqu 1.0* 1.4 1.0 1.4
Weight density of ground γc' γγ 1.0* 1.0 1.0 1.0
Partial resistance factors (γR)
Spread foundations, retaining
structures and slopes
Bearing resistance γR;v 1.0 1.0 1.4 1.0
Sliding resistance γR;h 1.0 1.0 1.1 1.0
Earth resistance, incl. slopes γR;e 1.0 1.0 1.4 1.0
Pile foundations – Driven pile (γR)
Base resistance γb 1.0 1.3 1.1 1.0
Shaft resistance (compression) γs 1.0 1.3 1.1 1.0
Total combined (compression) γt 1.0 1.3 1.1 1.0
Shaft in tension γs;t 1.25 1.6 1.15 1.1
* These partial material factors are used with DA1.C2 to avoid double factoring
Approaches and partial factor values and hence an Evolution Group of CEN
TC250/SC7, the CEN sub-committee responsible for the development of Eurocode 7, is

investigating ways to reduce and simplify the number of DAs and the number of NDPs.
5. Partial factor calibration
The numerical values of the partial factors given in the Eurocodes to achieve the
required reliability level assume that appropriate levels of workmanship and quality
management apply. EN 1990 {C3(2)} states in an Informative Annex that “in principle
numerical values for partial factors … can be determined in either of two ways:
“− On the basis of calibration to a long experience of building tradition, or
– On the basis of statistical evaluation of experimental data and field
observations. (This should be carried out within the framework of a
probabilistic reliability theory.)”
EN 1990 also notes that the recommended partial factor values in the Eurocodes
and also the values published in the National Annexes have mostly been chosen on the
@seismicisolation
@seismicisolation
basis of calibration with experience, i.e. with practices which have proven to be
reliable, and not on the basis of probabilistic methods. However, if a statistical
approach is adopted, EN 1990 {C3(3)} states that the “ultimate limit states partial
factors for different materials and actions should be calibrated such that the reliability
levels of representative structures are as close as possible to the target reliability
index”. The target reliability indexes, i.e. β values, given in EN 1990 {C6(1)} for
ultimate and serviceability limit states are 4.7 and 2.9, corresponding to 1-year
probabilities of failure, Pf of 1x10-6 and 2x10-3, respectively. It is anticipated that, when
sufficient statistical data are available, semi-probabilistic and full-probabilistic design
methods will be used for code calibration purposes and for the further development of
the Eurocodes.
6. Reliability differentiation and management
Reliability differentiation occurs first between ultimate and serviceability limit states.
Higher levels of reliability are required for ultimate limit states, which are concerned
with structural failure and hence risk to human life, compared with serviceability limit
states, which involve excessive deformations of structures. The required higher levels
of reliability are achieved through the use of partial factors greater than unity for
ultimate limit states and partial factors equal to unity in deformation calculations for
serviceability limit states.
EN 1990 {B1(2)} offers the following procedures for managing and differentiating
the reliability of structures with regard to ultimate limit states:
− The introduction of classes based on the assumed consequences of failure and
on the exposure of the construction works to hazard. The bases of these
classes are similar to the bases of the Eurocode 7 geotechnical categories,
referred to in Section 3, where the vulnerability of the structure and the
surroundings correspond to the consequences of failure.
− A procedure for differentiating between various types of construction based on
the quality levels of the design supervision and inspection during execution.
The classes referred to in the first bullet point above are three consequences classes
(CC) and three reliability classes (RC). The consequences classes relate to the
consequences of failure or malfunction of the structure with regard to the consequences
for loss of human life and the economic, social and environmental consequences. These
range from low for CC1, through medium for CC2, to high for CC3, as shown in Table
5 which includes the examples given in EN 1990 {B3.1(1)}. The consequences classes
as defined in EN 1990 differ from the Eurocode 7 geotechnical categories in that they
focus only on the consequences of failure, i.e. on the vulnerability factors in Table 1,
and do not refer to the hazards related to the ground conditions that contribute to the
complexity of geotechnical designs.
Eurocode 7 introduces a further method for reliability differentiation, not given in
EN 1990, that is based on the design situation. Design situations are related to the
duration of the action and are defined in EN 1990 {1.5.2.2} as “sets of physical
conditions representing the real conditions occurring during a certain time interval for
which the design will demonstrate that relevant limit states are not exceeded.” The
following four design situations are given in EN 1990:
@seismicisolation
@seismicisolation
Table 5.Consequences classes and associated reliability classes
Consequences class CC1 CC2 CC3

Description Low consequence for loss Medium consequence for High consequence for loss of
of human life , or loss of human life, human life and economic,
economic, social or economic, social or social or environmental
environmental environmental consequences very great
consequences small or consequences
negligible considerable
Examples of Agricultural buildings Residential and office Grandstands, public
buildings or civil where people do not buildings, public buildings where
engineering works normally enter (e.g. buildings where consequences of failure are
storage buildings), consequences of failure high (e.g. a concert hall)
greenhouses are medium (e.g. an office
building)
Reliability class RC1 RC2 RC3
β 4.2 4.7 5.2
Pf 10-5 10-6 10-7
“− Transient design situation that is relevant during a period much shorter than
the design working life of the structure and has a high probability of
occurrence, … temporary conditions … e.g. during construction or repair
{1.5.2.3}
− Persistent design situation that is relevant during a period of the same order
as the design working life of the structure … generally it refers to conditions
of normal use {1.5.2.4}
− Accidental design situation involving exceptional conditions of the structure
or its exposure, including fire, explosion, impact or local failure” {1.5.2.5}
− Seismic design situation involving exceptional conditions of the structure
when subjected to a seismic event” {1.5.2.7}.
Eurocode 7 {2.4.7.1(2)P} notes that the recommended partial factor values in
Annex A for ultimate limit states shall be used in transient and permanent design
situations. {2.4.7.1(4)} states that “More severe values than those recommended in
Annex A should be used in cases of abnormal risk or unusual or exceptionally difficult
ground or loading conditions”, while {2.4.7.1(5)} states that “less severe values …
may be used for temporary structures or transient design situations, where the likely
consequences justify it.” For accidental design situations, Eurocode 7 {2.4.7.1(3)}
states that “All values of partial factors for actions or the effects of actions … should
normally be taken equal to 1.0. All values of partial factors for resistances should then
be selected according to the particular circumstances of the accidental situation.”
Example of CEN member states that have introduced reliability differentiation
involving consequences classes and design situations are Austria and Germany. The
different partial material factor values for the three consequences classes given in the
Austrian NA for the design of slopes and anchorages are related to the design situation
(BS) and are shown in Table 6. The highest partial factor values are given for persistent
design situations, BS 1, intermediate values for transient design situations, BS 2, and
lowest values for accidental situations, BS 3. In the German Handbook to Eurocode 7,
the partial action factor values are related to the design situation for persistent, transient
@seismicisolation
@seismicisolation
Table 6. Partial factor values in Austrian NA for different consequences classes and design situations
Partial material factor values γM for consequences classes

Symbol CC 1 CC 2 CC 3
Common soil parametersa Design Situation(BS) Design Situation(BS) Design Situation(BS)
BS 1 BS 2 BS 3 BS 1 BS 2 BS 3 BS 1 BS 2 BS 3
Effective angle of frictionb γtanϕ’ 1.10 1.05 1.00 1.15 1.10 1.05 1.30 1.20 1.10
Effective cohesion γc’ 1.10 1.05 1.00 1.15 1.10 1.05 1.30 1.20 1.10
Weight density γγ’ 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
Undrained shear strength γcu 1.20 1.15 1.10 1.25 1.20 1.15 1.40 1.30 1.20
Uniaxial compressive γqu 1.20 1.15 1.10 1.25 1.20 1.15 1.40 1.30 1.20
strength
a
Certain failure mechanisms according to ONORM EN 1997-1:2006, 11.5.2 (for example a block on an
inclined surface) require other safety elements
b
This value is applied to tanϕ’
Table 7. Partial action factor values in German Handbook to Eurocode 7 for different design situations
Action Partial action factor values γF for design situation
Symbol BS-P BS-T BS-A
(persistent) (transient) (accidental)
Unfavorable permanent action γG 1.35 1.20 1.00
Favorable permanent action γG, inf 1.00 1.00 1.00
Unfavorable permanent earth pressure γG, EO 1.20 1.10 1.00
Unfavorable variable action γQ 1.50 1.30 1.10
Favorable variable action γQ 0 0 0
and accidental situations, BS-P, BS-T and BS-A, for all types of geotechnical designs
as shown in Table 7. As in the case of the Austrian partial material factors, the highest
partial action factors are for the persistent design situation, BS-P.
Greece is considering introducing, in a revised version of its NA, reliability classes
with different partial material/resistance factors for each reliability class in order to
provide less severe partial factor values than the recommended values for temporary
structures and transitory design situations.

In France, a form of reliability differentiation has been introduced that does not
involve changes to the partial factor values but links the consequences classes to the
geotechnical categories and hence to the level and quality of the geotechnical
investigations and design calculations as shown in Table 8. This form of reliability
differentiation is similar to the quality management differentiation outlined in EN l990
Annexes B4 and B5, based on the three levels of supervision during design, DSL1,
Table 8. Linkage in France of geotechnical categories to consequences classes, site conditions and basis for
design
Geotechnical category Consequences class Site conditions Basis for design

GC1 CC1 Simple and known Experience and qualitative
investigations
CC1 Complex
GC2 CC2 Simple or complex Site investigations and design
GC3 GC3 Simple or complex Site investigations and detailed design
@seismicisolation
@seismicisolation
Table 9. Design supervision and inspection levels and minimum requirements

Design Characteristics Minimum recommended requirements for checking of
Supervision Level calculations, drawing and inspections
DSL1 Normal supervision Self-checking: Checking performed by person who has
relating toRC1 prepared the design
DSL2 Normal supervision Checking by different persons than those originally responsible
relating to RC2 and in accordance with the procedure of the organization
DSL3 Extended supervision Third party checking: Checking performed by an organization
relating to RC3 different from that which has prepared the design
Inspection Levels Characteristics Requirements
IL1 Normal inspection Self inspection
relating toRC1
IL2 Normal inspection Inspection in accordance with the procedures of the
relating to RC2 organization
IL3 Extended inspection Third party inspection
relating to RC3
DSL2 and DSL3, and the three levels of inspection during construction, IL1, IL2 and
IL3, that are related to the three reliability levels, as shown in Table 9. These are linked
to the minimum requirements in the different geotechnical categories given in Table 2
for design procedures and calculations and for the expertise involved.
7. Experiences with the application of Eurocode 7
Following the implementation of Eurocode 7 and the other Eurocodes in 2010, a series
of Workshops was held as part of the 15th European Conference on Soil Mechanics and
Geotechnical Engineering (XV ECSMGE) in Athens in 2011 to share experiences with
the application of Eurocode 7. A report on these Workshops has been prepared by Orr
(2012). The Workshop presentations showed the different ways in which Eurocode 7 is
being implemented in ten countries, not only with different Design Approaches, as
explained in Section 4, but also with other differences introduced to improve the
reliability of geotechnical designs, such as consequences classes as described in
Section 6.
Prior to the XV ECSMGE, a questionnaire with the following questions was
circulated to geotechnical engineers in some countries to obtain information about the
experiences of applying Eurocode 7 in their countries:
1. Do you find Eurocode 7 easy to use?
2. Do you use it for all geotechnical design situations?
3. Does it provide enough detail? If not, what more detail would you like?
4. Do you as the geotechnical designer choose the characteristic value or does
the ground investigation report provide you with a characteristic value?
5. Are you confident choosing characteristic soil strengths?
6. Are you happy with the partial factor values in your National Annex (NA) and
confident of the safety of the resulting designs?
The responses received to the questions above revealed that:
@seismicisolation
@seismicisolation
− While initially engineers found Eurocode 7 more difficult to use than existing
standards, once familiar with it they now find it easy to use and are generally
happy with it and are confident in the safety of the designs obtained.
− Eurocode 7 is generally used for all geotechnical designs, although in some
countries, such as Ireland and the UK, it is not used for some designs, such as
for the design of ground anchors.
− Many countries do not think it provides enough information and hence have
considered it necessary to produce supporting national documents and
standards with additional non-conflicting complementary information (NCCI),
such as calculation models.
− In most countries the characteristic values of geotechnical parameters are
selected by the designer. However in some countries, mainly for liability
reasons, the characteristic values are selected by the person who carries out
the geotechnical investigation.
− Many engineers said they would like more guidance on how to select the
characteristic values of geotechnical parameters.
− Most engineers are happy with the partial factor values adopted in their
National Annexes.
The presentations on Eurocode 7 during the Workshops in Athens revealed that, in
addition to different Design Approaches being adopted in the different countries, many
differences still remain in the way pile resistances are determined. Most countries have
introduced model factors in the design of piles to increase the level of safety above that
provided by the recommended partial factors so that the overall factor of safety is close
to that used previously and to ensure that the occurrence of a serviceability limit state
as well as an ultimate limit state is sufficiently unlikely.
8. Conclusions
The aim of the Eurocodes is to design structures that have the required level of
reliability. The various reliability measures given in EN 1990 have been adopted in an
appropriate manner in Eurocode 7 in order to achieve reliable geotechnical designs.

This has meant that, compared with the other Eurocodes for manufactured materials,
Eurocode 7 focuses more on the management of the design process than on the details
of the design calculations. Although recommended partial factors are provided for all
design situations, since no calculation models are provided in the core text, only in
Informative Annexes, proper reliability calibration is not possible. Lists of factors to be
considered and taken into account during the different aspects of a geotechnical design,
from geotechnical investigations to design calculations, execution procedures and
maintenance, are given. Due to the different conditions and traditions in the European
countries, different Design Approaches and different partial factors have been adopted.
Many countries have adopted model factors to increase the overall factors of safety for
pile designs so that they are closer to those used previously. The survey of experiences
with the implementation of Eurocode 7 has revealed that engineers, once they are
familiar with Eurocode 7, are generally happy with it and do not have much difficulty
using it. However they would like more guidance on its use and hence many countries
have updated or are updating their national standards to provide complementary non-
conflicting information.
@seismicisolation
@seismicisolation
References
Calgaro J.-A. (2011) Safety philosophy of Eurocodes, Proceedings Geotechnical Risk and Safety,
Proceedings of the 3rd International Conference on Geotechnical Safety and Risk, Munich 2011, eds. N.
Vogt, B. Schuppener, D. Straub and G. Bräu, Bundesanstadt für Wasserbau, Karlsruhe, Germany, 29-
36
CEN (2002) EN 1990:2002 Eurocode basis of structural design, CEN Comité Européen de Normalisation,
Brussels
CEN (2004) EN 1997-1:2004 Geotechnical design: Part 1: General rules, CEN Comité Européen de
Normalization, Brussels
CEN (2007) EN 1997-2:2007: Eurocode 7 - Geotechnical design – Part 2: Ground investigation and testing,
European Committee for Standardization, Brussels
Orr T.L.L. (2011) Codes and standards and their relevance, Manual of Geotechnical Engineering, Institution
of Civil Engineers, London, eds. Burland J., Powrie W. and Chapman T., Volume 1
Orr T.L.L. (2012) Experiences with the application of Eurocode 7, Proceedings XV European Conference on
Soil Mechanics and Geotechnical Engineering, Athens, eds. A. Anagnostopoulos, M. Pachakis and C.
Tsatsanifos, IOS Press, Amsterdam, Vol. 4 (in print), 2012
Orr T.L.L. and Farrell E.R. (1999) Geotechnical design to Eurocode 7, Springer, London
Simpson B. (2011) Reliability in geotechnical design – some fundamentals, Proceedings Geotechnical Risk
and Safety, Proceedings of the 3rd International Conference on Geotechnical Safety and Risk, Munich
2011, eds. N. Vogt, B. Schuppener, D. Straub and G. Bräu, Bundesanstadt für Wasserbau, Karlsruhe,
Germany, 393:399
Simpson B. and Driscoll R. (1998) Eurocode 7 – a commentary, BRE, London
@seismicisolation
@seismicisolation
IOS Press, 2013
doi:10.3233/978-1-61499-163-2-87
Dealing with uncertainties in EC7 with

emphasis on determination of characteristic
soil properties
Hans R. SCHNEIDER, Mark A. SCHNEIDER
HSR University of Applied Sciences, Rapperswil, Switzerland
Abstract. The performance of geotechnical structures is governed by the spatial

average soil properties, such as the average shear strength over an appropriate
length or area or volume of soil. This is considered in Eurocode by defining the
characteristic value as being “a cautious estimate of the value affecting the
occurrence of the limit state” and stating that the selection of this value should take
account of, among other factors, “the extent of the zone of ground governing the
behavior of the geotechnical structure at the limit state being considered”.
To quantify the characteristic values, 3 statistical properties are required: The
arithmetic mean, the variance and the scale of fluctuation. The mean and the
variance (or equivalently the standard deviation or the coefficient of variation) are
known to most geotechnical engineers. The scale of fluctuation may be interpreted
as the distance within which soil properties are largely correlated, whereas for
greater distances the soil properties are largely uncorrelated or statistically
independent. This degree of spatial correlation considerably influences the
variance of the soil properties averaged over a certain geometrical zone.
This paper provides guidelines for the assessment of the characteristic soil
properties for the use in geotechnical designs, taking into account explicitly soil
variability, as well as the soil volume affected by the design.
Examples to illustrate the practical application of the proposed procedures for
typical geotechnical routine work are provided.
Keywords. Characteristic values, geotechnical parameters, uncertainties,

autocorrelation, scale of fluctuation, Eurocode 7, spatial soil variability
Introduction
Civil engineering design is largely a matter of decision making under uncertainty. The
loads and reactions are not known exactly and the accuracy of the design method is
uncertain. At all stages from the initial concept to the final completion, decisions must
be made using either incomplete or conflicting information. Inherent uncertainty is
particularly true for geotechnical engineering since natural soils are extremely variable
in their properties and the rational selection of suitable design parameters is generally
one of the most difficult parts of a design. Furthermore little to no guidance is offered
by Eurocode or standard soil mechanics textbooks to quantitatively determine these
design parameters for practical applications.
In this paper, practical definitions of the fundamental soil parameters as used in
Eurocode 7 “Geotechnical design”, such as the characteristic value xk and the design
@seismicisolation
@seismicisolation
88 H.R. Schneider and M.A. Schneider / Dealing with Uncertainties in EC7
value xd are presented. It is the intention to provide a simplified practical approach on

how to determine the characteristic soil properties according to Eurocode. The
simplified approach is derived from fundamental statistics and probability concepts
taking into account test data as well as the influence of the spatial variability of the soil.
In addition a procedure to include subjective designer-experience is also presented.
1. Definition and meaning of characteristic and design values in soil
1.1. Definition of geotechnical design parameters
Geotechnical analyses according to Eurocode EC 7 are carried out with design

parameters xd, determined for each homogeneously treated soil layer. These design
parameters are based on characteristic soil properties xk and defined in the Eurocodes
as:

(1)
The partial safety factors ߛெ are nationally determined and ensure a nationally
specified safety level or reliability of geotechnical structures. They are independent of
any soil properties and independent of the governing zone of influence. It is the
characteristic value xk only which has to completely describe the behavior of the soil
for a given design situation. The combined knowledge of field- and laboratory test
results as well as any previous experience produce one single deterministic value of xk.
Therefore a rational selection of suitable characteristic soil values is generally one of
the most difficult and most important tasks of a geotechnical design.
1.2. Definition of characteristic soil parameters
The zone of ground governing the behavior of a geotechnical structure at a limit state is
usually much larger than a test sample or the zone of ground affected in an in situ test.
Consequently the characteristic soil parameter is often the mean of the values covering
a large surface or volume of ground. If there is an ability of load transfer from weak to
strong zones in the ground, the characteristic value xk is the best estimate of the
“unknown” statistical mean xm of a soil layer. According to Eurocode EC 7 (2.4.5.2
(11)): “If statistical methods are used, the characteristic value should be derived such
that the calculated probability of a worse value governing the occurrence of the limit
state under consideration is not greater than 5%” (Figure 1). In other words we want to
be approximately 95% confident that the real statistical mean is superior to the selected
characteristic value xk.
To ensure a safe and economical design the selection of characteristic values for
geotechnical parameters shall take account of the following:
• Information from experience as well as field and lab test results
• The variability of the measured property values
• The extent of the zone of ground governing the behavior of the geotechnical
structure at the limit state being considered
@seismicisolation
@seismicisolation
H.R. Schneider and M.A. Schneider / Dealing with Uncertainties in EC7 89
Figure 1. Definition of characteristic value xk in soil
The true properties of a soil layer remain unknown and the estimates often have to be
made on the basis of experience or only a small number of tests. These estimations
always involve uncertainties which need to be accounted for accordingly.
2. Variability and uncertainties in soil properties
2.1. Sources of uncertainties and variability

Geotechnical engineering requires realistic in situ soil properties, which control the
performance of the structure to be designed. Typically, field or laboratory
investigations are carried out on a very small portion of the soil volume influencing the
geotechnical structure. The measured laboratory or field data therefore represent the
soil properties of a very small volume. Those measured properties are not necessarily
the ones governing the performance at the site. In fact, the pertinent soil properties
controlling the performance of a geotechnical structure often involve a much larger
volume of soil. Hence the domain-average property is needed instead of the property of
discrete soil specimens.
In general the characteristic value xk is the average soil property over an appropriate
spatial domain, large or small, explicitly taking into account the soil variability and
uncertainties.
Excellent contributions [Vanmarcke (1977); Harr (1987); Rethati (1988); Tang (1993);
Thorne & Quine (1993); Christian, Ladd & Baecher (1994); Lacasse & Nadim (1997);
Jaska, Brooker & Kaggwa (1997); Phoon & Kulhawy (1999); Uzielli (2008); Bond &
Harris (2008); Wu (2009); among others] dealing with soil variability and uncertainties
show the following primary sources of geotechnical uncertainty to be relevant for the
assessment of soil properties (Figure 2):
Figure 2. Sources of uncertainties and variability in soils
@seismicisolation
@seismicisolation
2.2. Natural or inherent soil variability

This uncertainty is inherent to the soil layer and cannot be reduced by collecting more
samples or performing more tests. It is the natural variation in a soil layer.
2.3. Measurement errors

Measurement uncertainties arise from inaccurate measuring devices, poor testing
procedures, low quality workmanship as well as random testing effects. Strictly
adhering to testing standards with sufficiently accurate measuring devices should
largely reduce or even eliminate these uncertainties.
2.4. Statistical errors

Statistical uncertainty is introduced due to limited information or a small number of
tests or observations.
2.5. Transformation uncertainty resulting from empirical correlations

Typically, measured field or even laboratory test values have to be modified by some
transformation model such as empirical correlations or theoretical models to find the
soil properties suitable for design.
3. Conceptual framework to assess characteristic values of soil parameters
Generally, the Student’s t-distribution is used when estimating the mean of a normally
distributed soil property when there are only a small number of samples available and
the population’s standard deviation is unknown. The t-distribution approaches the
normal distribution as more samples become available. In geotechnical engineering
both the mean value and the standard deviation are usually known with reasonable
accuracy from experience with similar soil types.
3.1. Normal distribution or log-normal distribution

Since the log-normal distribution cannot reach negative values it often provides a better
model of the variability of soil parameters. Usually the results of the two distributions
are very similar for small coefficients of variation (COVtotal < 10 ÷ 15%) and the
possibility of negative (physically impossible) values is close to zero. For larger
coefficients of variation, especially for COVtotal > 30%, the log-normal distribution
should be used.
COVtotal < 0.3 → normal distribution is possible

COVtotal ≥ 0.3 → log-normal distribution should be used
The log-normal distribution can be used in any case but, if applicable the normal
distribution is much easier to handle for practical applications. Table 1 shows the range
of typical coefficients of variation for different soil properties. Therefore the cohesion,
the undrained shear strength and the compressibility modulus of a soil layer often
should be evaluated using the log-normal distribution. Due to variance reduction in
case of large soil volumes involved in the governing limit states, it is often possible that
@seismicisolation
@seismicisolation
the above mentioned soil properties can also be determined using the normal
distribution. The angle of internal friction and the density on the other hand can usually
be derived by using the normal distribution.
Table 1. Range of typical values of coefficients of variation for soil
Soil property ࢏࢔ࢎࢋ࢘ࢋ࢔࢚
Density 1 10 %
Angle of internal friction 5 15 %
Cohesion 30 50 %
Undrained shear strength 30 50 %
Compressibility modulus 20 70 %
3.2. Mathematical concept for COVtotal < 0.3 (normally distributed soil values)
Used abbreviations:
xm = mean value
Sx = standard deviation
β = factor for 5%-fractile

(β ≈ 1.65 for normally
distributed soil values)
COV = coefficient of variation
Figure 3. Definition of normal distribution and used abbreviations
The characteristic values of soil properties can be derived with the following equation.
(Note: Equation (2) is also valid for any other man-made-material, for example
concrete, if its values are normally distributed)
· · 1 · (2)
The total coefficient of variation COVtotal accounts for the combined overall variability
and uncertainties as well as the spatial extent of the governing failure or deformation
mechanism. Tang (1984) and Phoon & Kulhawy (1999) proposed an additive model to
take account of the combined effects of aleatory and epistemic uncertainties.
COV Γ, COV COV COV COV

(3)
The variance reduction function Γ, considers the influence of the spatial extent
(averaged length L, area A or volume V) of the governing failure mechanism. Although
@seismicisolation
@seismicisolation
there exist several approaches to mathematically describe the variance reduction

function, Vanmarcke (1977) proposes to use the following simple form:
Γ Γ Γ Γ Γ Γ Γ Γ Γ Γ Γ Γ Γ Γ · Γ
೔
with Γ ೔ · 1 if L δ and Γ 1 ೔
if L # δ $ % , ', ( (4)
೔ · ೔ ·೔
in which δ )*+ is the scale of fluctuation and L )*+ is the length of the governing failure
mechanism in the considered direction x, y or z. The scale of fluctuation estimates the
distance within which soil properties show relatively strong correlation. Instead of the
scale of fluctuation often the autocorrelation distance is used in literature. The
autocorrelation distance R defines the separation distance at which the covariance
function decays to a value of Sx/e and the correlation between the soil properties can be
considered relatively weak (Sx stands for the standard deviation and e for the base of
the natural logarithm). Generally the scale of fluctuation is about twice the value of the
autocorrelation distance (exponential correlation model: δ 2 · R; Gaussian correlation
model δ √/ · R 0 1.8 · R). In the following considerations only the scale of fluctuation
is used.
3.3. Mathematical concept for COVtotal ≥ 0.3 (log-normally distributed soil values)
If the total coefficient of variation becomes larger than about 0.3 the log-normal
distribution should be used to avoid unrealistic negative values. Equation (6) to
determine the characteristic values can be derived as follows:
9
3 .· 456 7 ln 456 9 :;5 1 <
2
(5)
. ^
·

(6)
4. Simplification of general equations according to section 3
4.1. Simplifications for soil properties with COVtotal < 0.3 (normally distributed)
Substitution of the coefficient of total variation COVtotal leads to equation (7).
x x 1 β Γ, COV COV COV COV

(7)
With a 5%-fractile value of β ≈ 1.65 (normal distribution) and the assumption that the
coefficients of variations covering the influences of measurement, transformation and
statistical errors can be neglected (see Table 2) compared to the dominating COV inherent,
equation (7) can be rewritten as:
x > x ?1 1.65 COV !"!# · :Γ , % E

(8)
@seismicisolation
@seismicisolation
Table 2. Assumptions referring to coefficients of variation of different sources of uncertainty and variability

Measurement errors
Strictly adhering to testing standards with
sufficiently accurate measuring devices small 0
should largely reduce these uncertainties.
Transformation errors
Applying well established transformation
! small ! 0
models should reduce these uncertainties to
fairly small values.
Statistical errors
In geotechnical engineering the mean value
as well as the standard deviation is usually "# small "# 0
known with reasonable accuracy from
experience with similar soils.
The values of the proposed variance reduction function Γ, depend, as previously
seen on the spatial extent of the governing failure mechanism (averaged length L or
area A or volume V) as well as on the scale of fluctuation δ. Fenton and Vanmarcke
(1991) reasoned that the scale of fluctuation is largely dependent on the geotechnical
process of layer deposition rather than on the specific soil property being studied.
Therefore it is not surprising that the scale of fluctuation in horizontal direction (in
general: parallel to soil layering) is generally more than one order of magnitude larger
than in vertical direction (in general: perpendicular to soil layering).
Table 3. Summary of typical values of scale of fluctuations in horizontal and vertical direction within layers
of relatively uniform soils
Vertical direction Horizontal direction Source of data

δv = δz [m] δh = δx = δy [m]
1÷6 40 ÷ 60 Phoon and Kulhawy (1999)

Typical range
for different soil
properties ≈2÷6 ≈ 20 ÷ 80 El-Ramly et al. (2003)
(R = 1 ÷ 3) (R = 10 ÷ 40)
In Rackwitz et al. (2002) typical values for in situ soil properties from many different authors have been
summarized. They fall all within the above mentioned range and are therefore not listed separately.
Recommended
design values 2 50 Schneider and Fitze (2011)
With the given information a general approach to determine characteristic soil values
can be worked out. In the following equations the x- and y- directions are always
referred to as the horizontal ones and the z-direction as the vertical one.
@seismicisolation
@seismicisolation
General approach for soil properties with COVtotal < 0.3 (normally distributed)
x > x F1 1.65 COV !"!# · :Γ&, · Γ&, · Γ', G (9)
: from experience or test results ,see section 5-
!". from experience or test results ,see section 5-
Γ$,% Γ$ 4
Γ",&
Γ" 5
Γ",'
Figure 4. Variance-Reduction-Function in vertical direction

Figure 5. Variance-Reduction-Function in horizontal direction
@seismicisolation
@seismicisolation
4.2. Simplifications for soil properties with COVtotal ≥ 0.3 (log-normally distributed)
In case the log-normal distribution should be used (COVtotal ≥ 0.3) the characteristic
value of a certain soil property can be determined according to equation (10).
General approach for soil properties with COVtotal ≥ 0.3 (log-normally distributed)
() *+,-, ·-, ·-, ·./ 01

0.2^
0 ·
:1 < Γ&, · Γ&, · Γ', · *&2%2*
(10)
: from experience or test results ,see section 5-
!". from experience or test results ,see section 5-
Γ$,% Γ$ 4
Γ",&
Γ" 5
Γ",'
5. Simplified approach for design practice
In the following section a diagram that allows a fast and easy determination of
characteristic soil values is presented.
5.1. General considerations

Spatial averaging is only possible if the governing failure mechanism is capable of
redistributing forces or stresses along the failure surfaces. True cohesion is often not
ductile, i.e. brittle (is lost after small strains) and is mobilized on the failure surface
depending on stress history and stress level. Cohesion can therefore not generally be
averaged because the strains acting on the failure surface generally vary along the
failure surface. Many engineers are well aware of this fact and will consequently
neglect cohesion in most stability calculations.
To estimate the characteristic values of cohesion and the angle of internal friction for
strain-softening soils it is advised to use the critical state angles of friction and no
cohesion along the entire failure surface.
5.2. Diagram for practical applications

For typical geotechnical problems the horizontal extension of the governing failure
mechanism is usually small compared to the scale of fluctuation in its direction.
Furthermore the scale of fluctuation in vertical direction is much smaller than in
horizontal direction and therefore dominates the averaging process. Within reasonable
accuracy it can be assumed that the variance reduction function is only a function of the
vertical extension of the failure mechanism and its corresponding scale of fluctuation.
The assumption of perfect correlation in horizontal direction Γ& Γ Γ 1 and the
@seismicisolation
@seismicisolation
recommended design value for the vertical scale of fluctuation according to Table 3 (δ v
= 2m) leads to the diagram presented below (Figure 6). To evaluate the characteristic
value of a certain soil property only the mean (xm), the coefficient of variation
(COVinherent) as well as the vertical extension of the considered failure mechanism (L z)
are required. The required input values xm, COVinherent and Lz are outlined in the next
section. Since the lognormal distribution and the normal distribution are almost
identical for small coefficients of variation it is advised to use the black lines (log-
normal distribution) to determine characteristic values.
Figure 6. Diagram to evaluate characteristic soil properties
In case the governing failure mechanism extents only in horizontal direction, for
example sliding of a foundation, the characteristic values can be obtained using Figure
6 for Lz = 0m. The black dots assume a log-normal distributed and the grey dots a
normal distributed soil property.
Note: For failure mechanisms with large horizontal extensions (Lh > 30m) it is possible
to further reduce the variance reduction function resulting in higher characteristic
values and more economical design. In such a case the design values can be determined
using the general design approach as presented in section 4.
5.3. Required input parameters xm, COVinherent and Lz

As seen by inspection of Figure 6 only three input parameters are required for the
assessment of the characteristic value xk. These input parameters are the mean value xm,
the coefficient of variation COVinherent (alternatively the standard deviation) and the
vertical extent of the considered failure mechanism Lz. The mean value and the
coefficient of variation of a soil property are either known from experience or can be
evaluated trough field- or laboratory testing. It has been found on a worldwide data
@seismicisolation
@seismicisolation
basis (e.g. that the coefficient of variation is approximately constant for a certain soil
property. Table 4 presents typical and recommended values for different soil properties.
Table 4. Range of typical values and recommended values of coefficients of variation

Range of typical Recommended
Soil property values COVinherent values COVinherent
Density 0.01 0.10 0.0
Angle of internal friction 0.05 0.15 0.1
Cohesion 0.30 0.50 0.4
Undrained shear strength 0.30 0.50 0.4
Compressibility modulus 0.20 0.70 0.4
If a priori knowledge (experience, personal judgment) as well as a certain number n

measured test values is available Bayes` theorem can be used to combine the
information.
Table 5. Bayes theorem to combine a priori knowledge (experience, personal judgment) with test values
Estimated a priori values
Test values (2) Combined information (3)
(1)
!௠ଵ &௫ଶ ଶ
∑ !௜ !௠ଶ ·% '
# &௫ଵ
!௠௜ !௠ଵ !௠ଶ !௠ଷ
# 1 &௫ଶ ଶ
1 ·% '
# &௫ଵ
1
∑)!௜ !௠ଶ *ଶ &௫ଷ &௫ଶ ·
&௫௜ &௫ଵ &௫ଶ ( + & ଶ
#1 # % ௫ଶ '
&௫ଵ
௫௜
௫ଵ &௫ଵ ⁄!௠ଵ ௫ଶ &௫ଶ ⁄!௠ଶ ௫ଷ &௫ଷ ⁄!௠ଷ
6. Illustrative examples
6.1. Example 1: Slope stability

Slope stability - Input values:
JK 30°; 3 0.1
O 30*; O 20*; O 10*
General approach, Equation (9)

JK P 28.2° (normal distribution)
Simplified approach, Figure 6

Figure 7. Example 1: Slope stability JK P 27.9° (normal distribution)
@seismicisolation
@seismicisolation
6.2. Example 2: Pile foundation
a) Skin friction – Input values

Note: For illustration purposes only the
friction angle is considered
JK 30°; 3 0.1
O 20*


b) Tip resistance – Input values:

Note: For illustration purposes only the
cohesion is considered
T 4 20 UV4; 5 0.4
O 4*; O 2*

TK P 10.9 UV4 (log-normal distribution)
Figure 8. Example 2: Pile foundation
TK P 10.6 UV4 (log-normal distribution)
6.3. Example 3: Earth pressure and sliding
a) Earth pressure – Input values:

JK 30°; 3 0.1
O P 6*; O 10*


b) Sliding – Input values:

JK 30°; 3 0.1
O P 10*; O 40*
Figure 9. Example 3: Earth pressure and sliding

@seismicisolation
@seismicisolation
7. Summary
Geotechnical performance of a structure is usually governed by spatial average soil

properties, such as the average shear strength along a potential slip surface or the
average compressibility of a volume of soil beneath a footing. It becomes clear that the
characteristic value cannot be quantified by a field- and laboratory investigation alone,
but is dependent on the spatial extent of the governing failure mechanism to be
designed for.
In EC 7 the characteristic value xk is the fundamental soil value. Despite its important
safety relevance, the definition and determination of the characteristic value is far from
clear. An attempt is made to develop a simplified approach to determine characteristic
soil values. It is based on the mean value, the standard deviation (or coefficient of
variation) of a certain soil property and the extent of the governing failure mechanism.
It has been proposed that for most geotechnical problems it is sufficient only to
consider the vertical extent of the failure mechanism. Based on this assumption a
simplified approach for practical applications has been developed, and the examples
included in section 6 confirm that this approach is valid.
The presented simplified approach is easy to use and the results correspond well with
more rigorous methods. Examples to illustrate the approach as well as guidelines of
typical input parameters are provided.
References
Baecher, G.B. (2003). Reliability and Statistics in Geotechnical Engineering. John Wiley and Sons. Ltd,
England.
Baker, J. and Calle, E. (2006). JCSS Probabilistic Model Code, Section 3.7: Soil Properties, Joint Committee
on Structural Safety, Updated version Aug. 2006, Zürich, Switzerland.
Bond, A. and Harris, A. (2008). Decoding Eurocode 7, London: Taylor and Francis, 616 pp.
Brinch Hansen, J. (1956). Limit State and Safety Factors in Soil Mechanics, Danish Geotechnical Institute,
Copenhagen, Bulletin No. 1.
Butler, H.C. (2001). Spatial autocorrelation of soil electrical conductivity, M.Sc.-Thesis, Iowa State
University, Ames, Iowa.

Christian, J.T. and Ladd, C.C. and Baecher, G.B. (1994). Reliability Applied to Slope Stability Analysis, J
Geotech Eng-Asce 120, 2180-2207.
Craig, R.F. (1992). Soil Mechanics, Chapman and Hall.
Denver, H. and Ovesen, N.K. (1994). Assessment of Characteristic Values of Soil Parameters for Design,
Proc. XIII ICSMFE. New Delhi, India.
Elkateb, T. and Chalaturnyk, R. and Robertson, P.K. (2003). An overview of soil heterogeneity:
quantification and implications on geotechnical field problems, Can. Geotech. J. 45, 1-15.
El-Ramly, H. (2001). Probabilistic analyses of landslide hazards and risks: Bridging theory and practice,
Ph.D. thesis, University of Alberta, Edmonton, Alta.
El-Ramly, H. and Morgenstern, N.R. and Cruden, D.M. (2003). Probabilistic stability analysis of a tailings
dyke on presheared clay-shale, Can. Geotech. J. 40, 192-208.
El-Ramly, H. and Morgenstern, N.R. and Cruden, D.M. (2006). Lodalen slide: a probabilistic assessment,
Canadian Geotechnical Journal 43, 956-968.
Eurocode 7. (1994). Part 1: Geotechnical Design, General Rules, Final Version of ENV 1997-1, Oct 3.,
produced by CEN.
Fellenius, W. (1936). Calculation of the stability of earth damn, Transactions of the 2nd Congress on Large
Dams, Washington, D.C., 445-462.
Fenton, G.A. and Vanmarcke, E.H. (1991). Spatial variation in liquefaction risk assessment, Geotechnical
Engineering Congress 1991, Boulder, Colorado, 10-12 June. Geotechnical Special Publication No. 27,
American Society of Civil Engineers, Reston, Va. Volume 1, 594-607.
@seismicisolation
@seismicisolation
Harr, M.E. (1987). Reliability-based design in civil engineering”, McGraw Hill.

Kay, J.N. and Krizek, R.J. (1971). Estimation of the Mean for Soil Properties. Int. Conf. on Applications of
Statistics and Probability to Soil and Strucutral Engineering, Hong Kong.
Lacasse, S. and Nadim, F. (1997). Uncertainties in characterizing soil properties, NGI Publ. No. 201.
Lumb, P. (1974). Application of Statistics in Soil Mechanics, Chapter 3 in Soil Mechanics-New Horizons, ed.
by I.K. Lee.
Mantoglou, A. (1987). Digital simulation of multivariate two- and three-dimensional stochastic processes
with a spectral turning bands method, Mathematical Geology 19, 129-149.
Mostyn, G.R. and Li, K.S. (1993). Probabilistic slope analysis – state-of-play. Probabilistic Methods in
Geotechnical Engineering, Li and Lo (eds), 1993 Balkema, Rotterdam, ISBN 90 5410 303 5, pp. 89 –
109.
N.R.C. (1995). Probabilistic methods in geotechnical engineering, National Academy Press, Washington,
D.C.
Orr, T. (1993). Use of Partial Factors in Eurocode 7. Paper presented at 6th meeting of SC7, Berlin.
Orr, T.L.L. and Breysse, D. (2008). Eurocode 7 and reliability based design, in: Phoon, K.-K. (Ed.),
Reliability-based design in geotechnical engineering; computations and applications, Taylor and
Francis, London, 2008, 298-343.
Phoon, K.K. and Kulhawy F.H. (1999). Evaluation of geotechnical property variability. Canadian
Geotechnical Journal 36, 625 – 639.
Phoon, K.K. and Kulhawy F.H. (1999). Characterization of geotechnical variability. Canadian Geotechnical
Journal 36, 612 – 624.
Rackwitz, R. and Denver, H. and Calle, E. (2002). JCSS probabilistic model code, section 3.7: Soil properties,
5th (final) version. Joint Committee on Strucutral Safety, Zürich, Switzerland.
Rethati, L. (1988). Probabilistic solutions in geotechnics. in: Developments in geotechnical engineering 46,
Elsevier, Amsterdam.
Schneider, H.R. (1990). Die Wahl der Baugrundkennwerte. in: Anwendung der neuen Tragwerksnormen des
SIA im Grundbau – Referate der Studie Schweizerische Gesellschaft für Boden- und Felsmechanik,
Mitteilungen der Schweizerischen Gesellschaft für Boden und Felsmechanik, Zürich.
Schneider, H.R. (1997). Definition and determination of characteristic soil properties, XIV ICSMFE,
Hamburg, Balkema.
Schneider, H.R., (2010). Characteristic Soil Properties for EC7: Influence of quality of test results and soil
volume involved, Proc. 14th Danube-European Conference on Geotechnical Engineering, Bratislava.
Schneider, H.R., (2011). Safety Concepts and Calibration of Partial Factors in European and North
American Codes of Practice, Workshop Nov. 30 – Dec. 1, Delft University of Technology, Delft, The
Netherlands.
Schneider, H.R. and Fitze, P. (2009). Charakteristische Baugrundwerte: Erfahrung, Versuchswerte und
Statistik, Herbsttagung SBGF, 6. Nov., EPFL Lausanne.
Schneider, H.R. and Fitze, P. (2011). Characteristic shear strength values for EC7: Guidelines based on a
statistical framework, XV European Conference on Soil Mechanics & Geotechnical Engineering,
Athens, Greece, Sept. 2011
Schneider, H.R. and Tietje, O. and Fitze, P. (2010). Charakteristische Werte nach Swisscode: Definition,
Bestimmung und Anwendung in der Geotechnik-Praxis, Jan.18, HSR Hochschule für Technik,
Rapperswil, Switzerland.
Sivakumar Babu, G.L. and Dasaka, S.M. (2007). The Effect of Spatial Correlation of Cone Tip Resistance on
the Bearing Capacity of Shallow Foundations, Department of Civil Engineering, Indian Institute of
Science, India
Soulié, M. and Montes. P. and Silvestri. V., (1990). Modeling spatial variability of soil parameters. Canadian
Geotechnical Journal 27, 617 – 630.
Tang, W.H. (1993). Recent developments in geotechnical reliability, Balkema, ISBN 9054103035.
Tang, W.H. (1971). A Bayesian Evaluation of Information for Foundation Engineering Design, Int. Conf. on
Applications of Statistics and Probability to Soil and Structural Engineering, Hong Kong.
Taylor, D.W. (1948). Fundamentals of soil mechanics, John Wiley and Sons, Inc., New York.
Terzaghi, K. (1940). Sampling, Testing and Averaging. Proceedings, Purdue Conference on Soil Mechanics
and its Applications, Purdue University, West Lafayette, USA.
Thorne, C.P. and Quine, M.P. (1993). How reliable are reliability estimates and why soils engineers rarely
use them. Probabilistic Methods in Geotechnical Engineering, Li and Lo (eds), (1993). Balkema,
Rotterdam, ISBN 90 5410 303 5, 325 – 332.
Tietje, O. and Richter, O. (1992). Stochastic modeling of the unsaturated water flow using auto-correlated
spatially variable hydraulic parameters, Modeling Geo-Biosphere processes 1, 163-183.
Tietje, O. and Fitze, P. and Schneider, H.R. (2011). Slope stability based on autocorrelated shear strength
parameters, XV European Conference on SMGE. Sept., Athens.
@seismicisolation
@seismicisolation
Uzielli, M. (2008). Statistical analysis of geotechnical data, in: Geotechnical and Geophysical Site
Characterization, Huang and Mayne (eds), Taylor and Francis Group.
Vanmarcke, E.H. (1977). Probabilistic modeling of soil profiles. Journal of the Geotechnical Engineering
Division, ASCE, 103 (GT11), 1227 – 1246.
Vanmarcke, E.H. (1977). Reliability of earth slopes. Journal of the Geotechnical Engineering Division,
ASCE, 103 (GT11), 1247 – 1265.
Vanmarcke, E.H. (1983). Random fields: analysis and synthesis, MIT Press, Cambridge.
White, W. (1993). Soil variability: characterization and modeling. Probabilistic Methods in Geotechnical
Engineering, Li and Lo (eds), 1993 Balkema, Rotterdam, ISBN 90 5410 303 5, 111 – 120.
Wu, T.H. (2009). Reliability of geotechnical predictions, in: Geotechnical risk and safety, Honjo et. al (eds),
Taylor and Francis Group.
Zhang, L.L. and Zhang, L.M. and Tang, W.H. (2008). Similarity of soil variability in centrifuge models, Can.
Geotech. J. 45, 1118-1129.
@seismicisolation
@seismicisolation
IOS Press, 2013
doi:10.3233/978-1-61499-163-2-102
The Safety Concept in German

Geotechnical Design Codes
Bernd Schuppener
Federal Waterways Engineering and Research Institute, Karlsruhe, Germany
Abstract. The Eurocodes were officially adopted in Germany on 1st July 2012.
Since then, application of the partial safety concept has been mandatory in all
areas of structural engineering. The partial safety concept creates the impression
that partial factors based on probability theory are applied to actions or effects of
actions and resistances depending on the degree of uncertainty in each case. This is
hardly possible in geotechnical engineering. European geotechnical engineers have
therefore decided that the partial factors for the permanent and variable actions
from the ground should be the same as those used in other areas of structural
engineering for the sake of consistency in the field of construction. In Germany,
the partial factors for the resistances from the ground have been selected so that
the safety level is more or less the same as the tried-and-tested global safety level.
In other words, the application of the partial safety concept results in
approximately the same dimensions for foundations and geotechnical structures as
those obtained with the global safety concept in the past. Thus, on closer
inspection, the partial safety concept in its present form is, in geotechnical
engineering at least, a modified global safety concept. Users are faced with the
challenge of having to apply a concept which in some respects is new and more
complex. However, this is reasonable when one considers the important
contribution that has been made to placing European construction standards on an
urgently needed common basis, thus promoting the unification of Europe.
Keywords. Safety, reliability, partial factor, geotechnical design, standards

1. The probabilistic safety concept
The traditional global safety factor concept has the serious disadvantage that the actual
variability of the soil strength is not directly taken into account, and consequently a
particular conventional safety factor does not necessarily have the same meaning for all
soils. It is not easy to compare different designs with different soil types or even
different designs with the same soil type. A probabilistic approach instead of the
traditional global concept is therefore a fascinating vision for geotechnical engineers as
it not only provides a rational basis for the quantification of geotechnical safety but
also a meaningful and consistent basis for comparison (Lumb, 1970). By defining a
probability of failure, a direct comparison is possible whereas global safety factors are
related to every single verification format which cannot be compared with each other.
The probabilistic safety concept, i.e. a safety concept based on probability theory,
which was meant to replace the global safety concept upon implementation of the
Eurocodes in Europe, is based on the following considerations:
• If it is assumed that the actions on a structure and the resistances of that structure
are randomly dispersed quantities and that both the actions on a structure and the
@seismicisolation
@seismicisolation
B. Schuppener / The Safety Concept in German Geotechnical Design Codes 103
resistances of structural elements and components can be described in a rational

manner by means of statistics,
• then it must also be possible to use probability theory to define a common safety
level which is independent of the design and construction materials.
The probabilistic safety concept is thus based on the assumption that the actions
and resistances can be described statistically by an appropriate statistical distribution,
in simple cases by a normal distribution. As it is generally not the actions themselves
but the internal forces, moments or stresses caused by the actions that are of relevance,
the term “effects of actions” as defined in current standards is used below instead of
“actions”.
Figure 1. Distribution densities of effects of actions and resistances.
If, as shown in Figure 1 (left), the probability distributions of the effects of actions
S and the resistances R overlap (i.e. the resistance R is lower than the effects of actions
S) failure may be expected to occur. If the difference Z between the resistance R and
the effects of actions S is then calculated, the result is another probability distribution
describing the random variable Z. The area under the distribution curve in the negative
part of Z corresponds to the probability of failure pf.
The probability of failure pf, where the resistance R is lower than the effects of
actions S, is shown by the hatched area to the left of the ordinate. By contrast, the
reliability of a structure is indicated by the area to the right of the ordinate (p s=1 - pf)
where the resistance R is greater than the effects of actions S.
The greater the mean value mz and the lower the standard deviation σz of the
random variable Z, the greater the reliability of the structure will be. A safety index β
was therefore defined as mz/σz as a measure of reliability. At that time, the aim was to
achieve a safety index β equal to 4.7 for a reference period of one year. This
corresponds to a probability of failure of around 10-6 for the one-year reference period,
in other words the likelihood of similar structures failing in a single year is one in a
million.
So far the pure theory, which was very unfamiliar to structural engineers. However,
it was also clear to convinced statisticians at the time that it would not be possible to
perform such time-consuming and complex statistical safety analyses when designing
structures in practice. The theory therefore had to be radically simplified in order to
facilitate its application.
The partial safety concept, in which it is demonstrated that the design value of the
effects of actions to which a structural element or component is subjected does not
@seismicisolation
@seismicisolation
104 B. Schuppener / The Safety Concept in German Geotechnical Design Codes
exceed the design value of the resistances, was shown to be highly suitable. Eurocode 7
Geotechnical design - Part 1: General rules (CEN, 2004) (EC 7-1) proposes 3 design
approaches as options which are later presented in detail. For Design Approach 2 of EC
7-1 the design value Ed of the effects of actions is calculated in one of two ways:
• either by multiplying the characteristic value Fk of the actions by the partial
factor γF to obtain the design value Fd of the actions which is then used to
determine Ed,
• or by determining the characteristic value Ek of the effects of actions from the
characteristic value Fk of the actions and multiplying the resulting value by the
partial factor γF to obtain Ed.
The design value Rd of the resistances (as used in Germany) is obtained by
dividing the characteristic value Rk of the resistances by the partial factor γR. The
resulting limit state condition is therefore:
Ed ≤ Rd = Rk /γR
The principle of the simplified method, referred to by various authors as the “semi-
probabilistic method”, consisted of determining the partial factors in extensive
statistical studies performed with the verification methods usually applied in soil
mechanics. The values of the partial factors had to be such that the required safety level
β of 4.7 would be achieved.
The standards writers who began drafting the Eurocodes in the 1970s were
fascinated by this combination of probability theory and the partial safety concept for
two reasons:
• On the one hand, the theory provided an opportunity to define a common safety
level irrespective of the design and construction materials, in other words, for all
types of construction;
• on the other hand, the theory could serve as a common European safety concept
for the design of structures. It was a new approach with which each of the
Member States would have to familiarize themselves and which would need to
be a compromise to enable agreement on it to be reached.
The reliability theory had been adopted as the common safety concept for the
future Eurocodes (Joint Committee on Structural Safety, 1976). As a consequence, a
special committee was established in Germany which lay down the principles for the
application of the reliability theory in future structural standards (Arbeitsausschuss
“Sicherheit im Bauwesen” (Committee for Safety in Structural Design), 1981).
2. The scientific discussion on the safety concept among geotechnical engineers
For geotechnical design, the reliability theory using a probabilistic approach was
officially introduced at the German National Geotechnical Conference in 1978. The
concept was presented by G. Breitschaft (Breitschaft and Hanisch, 1978) who was
president of the DIBt (Deutsches Institut für Bautechnik, an institute of the German
Federal and Regional (Laender) Governments for a uniform fulfilment of technical
tasks in the field of public law) and later became chairman of the technical committee
of CEN in charge of the structural Eurocodes (TC 250). Their lecture intended to
@seismicisolation
@seismicisolation
promote the ideas among German geotechnical engineers and to lay the basis for future
standardization work in Europe and Germany.
In the following years numerous research studies on the application of the
reliability theory were conducted and published for the various geotechnical
verifications (e.g. Rackwitz and Peintinger, 1981). Moreover, a revised guidance paper
was drawn up by the Committee for Safety in Structural Design (Arbeitsausschuss
“Sicherheit im Bauwesen”, 1981) which was intended to serve as a mandatory basis for
all future structural design standards.
The probabilistic safety concept was brought up again in a special session held
during the National Geotechnical Conference in 1982. Five papers dealt with the
subject and its application to
• the verification of the safety of spread foundations against failure by heave
(Pottharst, 1982),
• anchored or nailed walls (Gässler, 1982)
• the evaluation of test results (Peintinger, 1982; von Soos, 1982) and
• the application of previously available information (Rackwitz, 1982).
The most interesting part of the special session was a panel discussion. Most of the
arguments for and against the probabilistic safety concept, which have been repeated
over and over again in discussions since then, were put forward there. They were as
follows:
• The probabilistic approach does not take account of human error in design and
execution although it is one of the main causes of damage (Blaut, 1982).
• The possibilities of collecting statistical data on soil are severely limited in
practice (Vollenweider, 1982).
• The differences between geotechnical engineering and other areas of structural
engineering are not only the higher coefficients of variation in the former – soil
cannot be produced with clearly defined characteristics according to a set
formula – but also that the geotechnical engineer only ever sees a limited part of
the structure he is designing (Vollenweider, 1982).
Most prominent German geotechnical engineers took rather a critical view of the
probabilistic approach (in favour: 3; undecided: 5, against: 4). However, it was
generally agreed that greater effort was required during soil investigations, there was a
definite need for databases for information on soil to be set up and that more extensive
checks and inspections of geotechnical engineering work were necessary.
In the following years the probabilistic approach was a research topic at nearly all
geotechnical engineering departments at German universities and nearly all analyses in
geotechnical design were examined to establish whether they were suited to the
application of the probabilistic approach. Eder recalculated the failure of a rock slope
(Eder, 1983), Heibaum examined the stability of anchored retaining walls at deep slip
surfaces (Heibaum, 1987), Genske and Walz (Genske and Walz, 1987) as well as
Smoltzcyk and Schad (Smoltczyk and Schad, 1990) considered the application of the
probabilistic safety philosophy to calculations of the bearing capacity of soil, Reitmeier
researched the possibility of applying a stochastic approach to quantifying differential
settlements (Reitmeier, 1989) while Hanisch and Struck applied the method to evaluate
pile loading tests (Hanisch and Struck, 1992).
In addition, there were a number of publications dealing with the evaluation of soil
investigations in terms of how the results could be used in connection with the
probabilistic approach (Hanisch and Struck, 1985, von Soos, 1990, and Alber, 1992) as
@seismicisolation
@seismicisolation
well as papers in which the new concept was clearly set out and explained to colleagues
with the intention of promoting it (Gudehus, 1987 and Franke, 1990).
Even though the future direction of standardization work in geotechnical
engineering seemed to have been firmly established by a decision of the steering
committee of the national committees in charge of drafting geotechnical engineering
standards in 1982, the “Principles for the specification of safety requirements for
structures” (Arbeitsausschuss “Sicherheit im Bauwesen”, 1981) were repeatedly the
subject of fundamental criticisms in the years that followed. Thus Franke (Franke,
1984) demonstrated the problems that occur when the probabilistic safety concept is
applied to piles, commenting scathingly that the possibility (of applying the
probabilistic approach) was viewed most optimistically by those colleagues who were
least involved in conducting soil and rock investigations and describing soil and rock
on a daily basis in practice. He went on to say that, in his view, the observation method
was a far superior aid even though it is not mentioned in the “Principles for the
specification of safety requirements for structures”. Furthermore, it was also shown
that, for a constant safety level, the partial factors depend on the magnitude and number
of parameters involved and in particular on the coefficient of variation (Heibaum,
1987). For Germany at that time it was seldom possible to obtain more than only a
rough estimate of the coefficient of variation of geotechnical parameters.
Fundamental criticism of the new safety concept was voiced above all by Swiss
colleagues. After analysing 800 cases of structural damage that had been described by
means of the same criteria and evaluated in different ways, Matousek and Schneider
concluded that random deviations of the material properties, the resistances of
structures or the loads on structures from the expected values are evidently well
covered by the conventional safety concept. The vast majority of cases of damage
occur during execution. Matousek and Schneider went on to state that while every care
is taken at the design stage, the construction conditions are often viewed as of
secondary importance although they require greater attention (Matousek and Schneider,
1976). Schneider considered the probability of serious errors generally to be far greater
(ten- to a hundredfold) than the theoretical probabilities of failure (Schneider, 1994).
Vollenweider expressed similar doubts about the safety goal of a very low probability
of failure. He questioned whether the statistical data for this range of values, if
available at all, was sound and whether the correct distribution laws were applied.
Vollenweider spoke out in favour of applying the hazard scenario approach instead to
enable the risk potential to be managed more reliably (Vollenweider, 1983 and 1988).
Summing up the scientific studies and the debate up until around 1990 it can be
seen that the probabilistic approach in geotechnical engineering yielded a great number
of interesting scientific research results and findings in Germany but that it was not yet
possible to develop a convincing standardization concept for application in everyday
practice. Although the partial safety concept had won through, the probabilistic
approach no longer had any part to play during discussions between standards writers
on the issue of which parameters partial safety factors should be applied to and what
the values of those factors should be. There were only a few isolated voices who
continued to advocate taking the probabilistic approach into account in geotechnical
engineering standards (Hanisch, 1998).
Although the probabilistic approach was finally abandoned in German
geotechnical design standards, the subject continued to be attractive in scientific
research. Thus Hartmann and Nawari attempted to discover new ways of evaluating
uncertainty and risk with the aid of fuzzy logic and the fuzzy set theory (Hartmann and
@seismicisolation
@seismicisolation
Nawari, 1996), Pöttler et. al. examined the application of the probabilistic approach to
tunnel construction (Pöttler et. al., 2001), Ziegler considered the possibilities of risk
simulation calculations (Ziegler, 2002), Katzenbach and Moormann used the data
collected for Frankfurt clay over many decades to examine the structural performance
of piled raft foundations (Katzenbach and Moormann, 2003), Stahlmann et. al.
employed probabilistic methods to simulate the inhomogeneities in the soil properties
of a railway embankment (Stahlmann et. al., 2007) and Russelli compared various
probabilistic methods as applied to investigations of the bearing capacity of soil,
demonstrating the great influence of the combination of friction and cohesion and their
correlation (Russelli, 2008). So far, none of these studies has been taken into account in
geotechnical standards or recommendations.
3. Eurocode 7 Geotechnical design
Work on the Model Code for Eurocode 7 “Geotechnical design” started in 1981 and
was headed by Krebs Ovesen (Orr, 2007) who chaired the subcommittee (SC 7) in
charge of the work for 18 years. One of the fundamental ideas was that the Eurocode
should only contain qualifying rules, in other words, should require the bearing
capacity to be verified but would not specify which method of calculation should be
used. Naturally, this improved the likelihood of reaching a consensus on the rules.
There were intense discussions on the applicability of the statistical safety concept in
the European committee as the original enthusiasm for the probabilistic approach had
vanished. It was agreed that, should the probabilistic safety concept be introduced in
geotechnical engineering, a great number of difficulties would still need to be
overcome and that the partial factors would initially have to be based on experience but
would have to be confirmed by probabilistic analyses at a later date (Sadgorski, 1983).
The drafts of the Eurocode differentiated between the core text and supplementary
comments. Initially there was no intention of specifying numerical values for either the
loads or the partial factors in the core text of the Eurocodes (Sadgorski, 1983); the
values were to be set in National Annexes instead.
In 1987, the “Draft Model for Eurocode 7 – Common unified rules for
Geotechnics, Design” was published (Representatives of the Geotechnical Societies

within the European Countries, 1987) as a report prepared for the European
Communities. The annex of the draft model specified partial factors after all. Reference
was made to the relevant loading codes for structures above ground level for variable
actions while a partial factor, γg, of 1.0 was specified for permanent actions from the
structure, ground and groundwater. The following partial factors were given for
geotechnical parameters: γϕ = 1.2 on the tangent of angle of internal friction, γc1 = 1.8
on the cohesion when verifying the load-bearing capacity of foundations and γc2 = 1.5
on the cohesion when verifying the stability and earth pressure. Moreover, partial
factors were stated for the load bearing resistance of piles and anchors and for
structures under construction.
In 1989 “Eurocode 7 Geotechnics” was published as a Preliminary Draft for the
European Communities on the basis of the December 1987 version of the Model Code
produced by the ISSMFE (EC 7 Drafting Panel, 1989). A chapter 7 for piles and a
chapter 8 for retaining structures had not yet been prepared. This version now gave
numerical values for partial factors in the core text and it was emphasized in the
@seismicisolation
@seismicisolation
preface that “they represented the best estimate of the drafting panel. In geotechnical
engineering limited experience has been gained until now on a European basis on the
use of limit state design and partial safety factors. Consequently there is a strong need
for calibration of all safety elements introduced into the draft before it is issued for
use”. In Section 2 “Basis of Design” it was stated as a fundamental requirement that:
“A structure shall be designed and constructed in such a way that with acceptable
probability, it will remain fit for the use …., and with appropriate degrees of reliability,
it will sustain all actions ....”. However, neither principles nor application rules were
given for the derivation of partial factors on actions and ground parameters by means of
reliability theory.
A first complete version of Eurocode 7 was published in 1994 as pre-standard
ENV 1997-1:1994 “Geo-technical design - Part 1 General rules” (EC 7-1). Three cases
were introduced. Case A covered the loss of static equilibrium of a structure as a rigid
body where the strength of the construction material of the ground is not governing.
For the verification of ultimate limit states in the ground two combinations of partial
factors had to be investigated: Case B and Case C.
• Case B aimed to provide safe design against unfavourable deviations of the
actions from their characteristic values. Thus, in Case B, partial factors greater
than 1.0 were applied to the permanent and variable actions from the structure
and the ground, the factors being the same as those used in other fields of
structural engineering. By contrast, the calculations of the ground resistance were
performed with characteristic values, i.e. the partial factors for the shear
parameters, γϕ, γc and γcu, were all set at 1.00.
• Case C in the pre-standard aimed to provide safe design against unfavourable
deviations of the ground strength properties from their characteristic values and
against uncertainties in the geotechnical calculation model. It was assumed that
the permanent actions corresponded to their expected values and the variable
actions deviated only slightly from their characteristic values. Thus, the partial
factors for the characteristic values of the ground strength parameters were γϕ=
1.25 γc = 1.6 and γcu = 1.4 while the characteristic values of the permanent
actions from the structure (with γG set at 1.00) were used in the verification.
This concept for the verification of two cases, B and C, was strongly opposed in
Germany. The philosophy for Cases B and C was not convincing because it could not
guarantee a sufficient safety level for the combination or superposition of the
uncertainties of the material properties (soil and other material) and the actions.
Furthermore, there were strong objections to the mandatory application of partial
factors to the ground strength properties ϕ´, c´ and cu in order to determine the design
values of the resistances of the soil. Although this corresponded to German practice for
the verification of slope stability, in which the Fellenius method was applied, it was not
the case for the verification of the design of shallow foundations and retaining walls.
The application of partial factors to the ground strength properties would have resulted
in some cases in larger dimensions and in others in smaller dimensions than would
have been obtained if the former global safety concept had been applied (Weißenbach,
1991). Moreover, with factored shear strength parameters, the relevant verification
would be based on failure geometries in the ground which might not be realistic. A
more detailed critical review and a proposal for an improvement of the pre-standard of
EC 7-1 can be found in Schuppener et. al. (1998) and Weißenbach (1998).
@seismicisolation
@seismicisolation
These fundamental criticisms were shared by many other European countries. As a

compromise, the final version of EC 7-1 of 2004 (CEN, 2004) gives three design
approaches (DA) as options. Each Member State has to establish in its National Annex
to EC 7-1 which of the three DAs is mandatory for which limit state verification.
In Design Approach 1 (DA 1), two combinations of partial factors have to be
investigated. Combination 1 aims to provide safe design against unfavourable
deviations of the actions from their characteristic values. Thus, in that combination,
partial factors greater than 1.0 are applied to the permanent and variable actions from
the structure and the ground. The recommended factors are γG = 1.35 for unfavourable
permanent actions, γG;inf = 1.00 for favourable permanent actions and γQ = 1.50 for
variable actions. The factors are the same as those used in other fields of structural
engineering and they are consistent with those specified in EN 1990: Basis of structural
design. By contrast, the calculations for the ground resistance are performed with
characteristic values, i.e. the partial factors γϕ, γc and γcu, which are all set at 1.00, are
applied to the shear parameters; the partial factor of the ground resistance, γR, is also
1.00. Combination 2 of DA 1 aims to provide safe design against unfavourable
deviations of the ground strength properties from their characteristic values and against
uncertainties in the calculation model. It is assumed that the permanent actions
correspond to their expected values and the variable actions deviate only slightly from
their characteristic values. Thus, in this verification, the partial factors γϕ´, γc' and γcu
with numerical values of 1.25 or 1.40 are applied to the characteristic values of the
ground strength parameters whereas the characteristic values of the permanent actions
from the structure (with γG set at 1.00) are used. The partial factors are applied to the
representative values of the actions and to the characteristic values of the ground
strength parameters at the beginning of the calculation. Thus the entire calculation is
performed with the design values of the actions and the design shear strength. Of the
two combinations, the one resulting in the larger dimensions of the foundation will be
relevant for designs according to Design Approach DA 1.
In Design Approach 2 only one verification is ever required unless different
combinations of partial factors for favourable and unfavourable actions need to be dealt
with separately in special cases. The partial factors applied to the geotechnical actions
and effects of actions are the same as those applied to the actions on or from the
structure, i. e. γG = 1.35, γG;inf = 1.00 and γQ = 1.50. There are two ways of performing
verifications according to Design Approach 2. In the design approach referred to as
“DA 2”, the partial factors are applied to the characteristic actions at the very start of
the calculation and the entire calculation is subsequently performed with design values.
By contrast, in the design approach referred to as “DA 2*”, the entire calculation is
performed with characteristic values and the partial factors are not introduced until the
end when the ultimate limit state condition is checked. As characteristic internal forces
and moments are obtained in the calculation, the results can generally also be used as a
basis for the verification of serviceability.
Similarly, only one verification is required for Design Approach 3 (DA 3). The
partial factors applied to the actions on the structure or coming from the structure are
the same as those used in Design Approach DA 2. However, for the actions and
resistances of the ground, the partial factors are not applied to the actions and
resistances but to the ground strength parameters, ϕ´, c´ or cu instead. The
recommended values of γϕ', γc' and γcu are 1.25 and 1.40. The partial factors are applied
to the representative values of the actions at the beginning of the calculation and to the
@seismicisolation
@seismicisolation
characteristic values of the ground strength parameters. Thus, in DA 3, the entire

calculation is performed with the design values of the actions and the design shear
strength.
4. DIN 1054 Safety in Earthworks and Foundation Engineering
The steering committee of the German national committees in charge of drafting

geotechnical engineering standards decided in 1982 to gradually incorporate the new
safety concept into the standards for that field (Gudehus, 1987). It was even decided to
prepare a Guidance Paper on Reliability in Geotechnical Design. A new standardizing
committee “Safety in Earthworks and Foundation Engineering” was established. Its
aim was to act as a mirror committee for the European subcommittee drafting Eurocode
7 “Geotechnical Design” and revise the German standard DIN 1054 with the new title
“Safety in Earthworks and Foundation Engineering” to make it compatible with the
principles and application rules of the future Eurocode 7.
The idea behind revising DIN 1054 at the same time as Eurocode 7 was to
familiarize German geotechnical engineers with the new design concepts as early as
possible and to enable them to make technically sound contributions to the discussions
held during the process of writing the Eurocode. With the implementation of EC 7-1 in
the Member States all conflicting national standards had to be withdrawn after a
coexistence period. DIN 1054 was a conflicting standard as it partly covered the same
items as EC 7. So the current version of DIN 1054 (2010) therefore only contains
specific German rules, which are not given in EC 7-1 (CEN, 2004). The title of the
DIN standard has been amended accordingly to “Subsoil – Verification of safety of
earthworks and foundations – Supplementary rules to DIN EN 1997-1”.
During the process of revising DIN 1054 a design concept was developed which
was later adopted as Design Approach 2 in the final draft of EC 7-1 (CEN, 2004). In
order to eliminate the discrepancies in EC 7-1 described above, the following proposals
were included in the revised version of DIN 1054-100 (1996):
• as a first step, the characteristic values of the actions and the resistances are
determined with the aid of the characteristic soil parameters,
• as a second step, the characteristic effects of actions, such as the reactions at

supports and bending moments, are determined and
• only at the end of the verification are the resulting characteristic effects of
actions increased and the characteristic resistances reduced by applying partial
factors in order to obtain the design values. These are then used to demonstrate
that the limit state conditions have been satisfied.
This design approach, referred to as “the Weißenbach Approach” during the
discussion phase, was then incorporated into the drafts of 2004 and 2005. It was also
included in EC 7-1 of 2004 as Design Approach DA 2*. The approach is used in
DIN 1054:2010-12 for the structural analysis of retaining walls, spread and pile
foundations and anchors.
@seismicisolation
@seismicisolation
5. Partial safety factors
Germany has a tradition of standards for geotechnical engineering that dates back
more than 70 years. The first edition of DIN 1054, entitled “Guidelines for the
permissible loads on ground in building construction”, was published in 1934. Since
then, geotechnical standards have continuously been optimized and have reached an
outstanding quality. The safety level of the former global safety concept proved
successful and the specified safety factors made safe and economic geotechnical
designs possible. The Advisory Board of the Standards Committee for Building and
Civil Engineering of the German Standards Institute, DIN, therefore decided in 1998
that any increase in cost as a result of new standards had to be justified. As the existing
standards were well tried and tested, it was decided that the safety level of the former
global safety concept should be maintained when the geotechnical standards were
adapted to accommodate the partial safety factor concept of the Eurocodes. This meant
that the design approaches and the partial factors had to be selected in such a way that a
foundation designed according to EC 7-1 would have roughly the same dimensions as a
design in accordance with the previous standards. This was a prerequisite as serious
problems regarding the acceptability of the Eurocodes would otherwise have arisen.
For example, a structure undergoing modification might need strengthening or even
underpinning according to the new safety concept, although this may not have been
necessary under the previous one. As reliability theory was not considered to provide
partial factors for ground resistance and ground properties, maintaining the safety level
of the former global safety concept was also a necessary assumption for the deter-
mination of the partial factors for geotechnical actions and resistances. In order to
maintain that safety level in the concept of partial factors in Design Approach 2
(DA2*) of EC 7-1 the equation
γR ⋅ γG,Q ≈ ηglobal
must be fulfilled, where γR is the partial factor for the resistance of the ground, γG,Q
is a weighted mean partial factor for the effects of permanent and variable actions and
ηglobal is the global safety factor used hitherto. The values recommended in Annex A of
“Eurocode - Basis of structural design” (CEN, 2002), which are γG = 1.35 and γQ = 1.50
for the permanent and variable effects of actions respectively, were adopted in EC 7-1
and in German geotechnical design standards as they had been in the other fields of
structural engineering. As the permanent actions are generally greater than the variable
actions in geotechnical engineering, a weighted mean value, γG,Q, of 1.40 was used to
calculate the partial factor for the ground resistance, γR, for the various verifications.
Thus the following partial factor, γR, for the resistance is obtained from
γR ≈ ηglobal / γG,Q.
For the ground bearing resistance, where a global safety factor ηglobal, of 2.00, was
used in Germany we then arrive at a partial factor of γR,v = 2.00/1.40 ≈ 1.40. The partial
factors for the ground resistance in each limit state and design situation were
determined in this way.
The numerical values of the partial factors for actions have been specified by
structural engineers and it is therefore certainly debatable whether they provide a
@seismicisolation
@seismicisolation
realistic description of the uncertainties in geotechnical engineering. Yet EC 7 and the

national German standards committee for geotechnical engineering considered it more
important for common partial factors to be used in all fields of civil engineering in
future than for specific partial factors to be laid down for geotechnical design,
especially as selecting the values would also have given rise to endless discussions.
Design Approach 3 (DA 3) of EC 7-1 is used in Germany for the verification of
overall and slope stability. The approach specifies that partial factors must be applied
to the shear strength of the ground as well as to the loads from the structure and the
variable loads. Reducing the shear strength conforms to the Fellenius method which
was already an option for verifying overall stability in Germany. The reduction in shear
strength leads to an increase in the actions from the ground and a decrease in the
ground resistance. In order to maintain the safety level of the global safety concept it
was decided in Germany that the characteristic values of the permanent actions from
the structure would be used and only the variable actions would be increased by a
partial factor greater than unity. Generally speaking, their effect on the safety of slopes
is very slight anyway as it is the self weight of the soil that is the predominant factor.
Further details of the implementation of EC 7-1 in Germany can be found in Lesny
(2012).
6. Summary and outlook
The introduction of the partial safety concept provided a common format for
analysis in structural design for different types of construction and construction
materials. However, a common safety level, in terms of a common probability of
failure, has not been achieved, even if very similar partial factors have been introduced
for the actions in all areas of structural design. As explained above, these partial factors
have also been adopted in geotechnical engineering, with no attempt being made to
develop separate partial factors for geotechnical actions. Thus they are not – as was
originally planned – a measure of the reliability with which the magnitude of
geotechnical actions can be determined. The same applies to the partial factors for the
resistances as they were derived on the basis of the condition that approximately the
same dimensions for foundations should be obtained for designs in accordance with the
partial safety concept as for those performed with the former global safety concept.
Thus, in fact the partial safety concept in German geotechnical design is a global safety
concept. The incorporation of the new concept into all German geotechnical
engineering standards and recommendations has meant that these have been
harmonized and thus become more user-friendly. Any technical progress was only an
indirect consequence owing to the fact that the German standards and
recommendations were, of course, brought up to date and improved as they were being
revised to include the partial safety concept.
Eurocodes do not take account of human errors, nor are such errors mentioned in
the definitions of the partial factors. Instead, all Eurocodes have a list of assumptions
which define and make sure that everything is planned, executed, supervised and
maintained according to the plans by personnel having the appropriate skill and
experience. Although human error was never explicitly referred to in the standards
based on the global safety concept it was implicitly assumed that it was covered, at
least to a certain extent, by the safety factors. The objective was always to achieve a
robust yet economic design that would not fail just because of a few minor errors. The
@seismicisolation
@seismicisolation
adoption of the safety level of the previous standards has thus meant that “minor”
human errors are now included in the partial factors.
There is general agreement amongst geotechnical experts that human errors and
insufficient knowledge of the soil conditions present the greatest risk in geotechnical
engineering. It is for this reason that, from now on, greater attention should be paid
above all to raising the requirements for soil investigations and introducing stricter
controls during execution instead of refining the stability analyses. The author therefore
believes that, in future, the incorporation of the hazard scenario approach
(Vollenweider, 1983, SIA 260:2003 and SIA 267:2003) or risk simulation calculations
(Ziegler, 2002) into geotechnical engineering standards would be more appropriate,
especially as the theories behind them are closer to engineering practice.
In Germany, there are various views on the technical benefits of applying the
partial safety concept in geotechnical engineering. This issue was discussed at length at
a meeting of the Steering Committee of the Geotechnical and Earthworks Engineering
Section of the Building and Civil Engineering Standards Committee (NABau) of DIN.
Irrespective of the technical points of view, the Steering Committee agreed that the
partial safety concept should be retained, not least for political reasons. It presents a
common language for structural design in Europe which now needs to be simplified
and developed further in order to work towards the political aim of eliminating
technical barriers to trade and harmonizing technical tendering procedures. A return to
the global safety concept would therefore not be desirable.
References
Alber, D. (1992): Überlegungen und Verfahren zur Schätzung statistischer Parameter von Bodenkennwerten,
Bauingenieur 67 (1992), S 39 - 45.
Arbeitsausschuss „Sicherheit im Bauwesen“, (1981): Grundlagen für die Festlegung von
Sicherheitsanforderungen für bauliche Anlagen (Principles for the establishment of safety requirements
for structures), Beuth Verlag, Berlin
Blaut, H. (1982): Diskussionsbeitrag bei der Podiumsdiskussion zur Spezialsitzung „Sicherheit im
Grundbau“ Vorträge der Baugrundtagung, Braunschweig

Breitschaft, G. and Hanisch, J. (1978): Neues Sicherheitskonzept im Bauwesen aufgrund
wahrscheinlichkeitstheoretischer Überlegungen – Folgerungen für den Grundbau unter Einbeziehung
der Probennahme und der Versuchsauswertung, Vor-träge der Baugrundtagung in Berlin, Deutsche
Gesellschaft für Erd- und Grundbau, Eigenverlag,
CEN (2002) Eurocode: Basis of structural design. European standard, EN 1990: 2002. European Committee
for Standardization: Brussels.
CEN (2004) Eurocode 7 Geotechnical design - Part 1: General rules. Final Draft, EN 1997-1:2004 (E), (F)
and (G), November 2004, European Committee for Standardization: Brussels, 168 pages (E).
DIN 1054 (2010): Subsoil – Verification of the safety of earthworks and foundations – Supplementary rules
to DIN EN 1997-1, Beuth Verlag, Berlin
EC 7 Drafting Panel (1989): Eurocode 7 Geotechnics, Preliminary Draft for the European Communities,
Geotechnik 13 (1990), S 1-40
Eder, F. (1983): Erläuterung des Statistischen Sicherheitskonzepts am Beispiel einer Rutschung, Buchkapitel
in: Mitteilungen des Instituts für Bodenmechanik, Felsmechanik und Grundbau, TU Graz, Nr.6
Franke, E. (1984), Einige Anmerkungen zur Anwendbarkeit des neuen Sicherheitskonzepts im Grundbau,
geotechnik, Jg.7, Nr.3, S.144-149
Franke, E. (1990): Neue Regelung der Sicherheitsnachweise im Zuge der Europäischen Bau-Normung - Von
der deterministischen zur probabilistischen Sicherheit auch im Grundbau? Bautechnik 7/1990
Gässler, G. (1982): Anwendung des statistischen Sicherheitskonzeptes auf verankerte Wände und vernagelte
Wände, Vorträge der Baugrundtagung, Braunschweig, S 49 – 81
@seismicisolation
@seismicisolation
Genske, D.D. und Walz, B. (1987): Anwendung der probabilistischen Sicherheitsphilosophie auf
Grundbruchberechnungen nach DIN 4017, geotechnik 10, S 53 – 66
Gudehus, G. (1987): Sicherheitsnachweise für Grundbauwerke , Geotechnik 10, S. 4-34
Hanisch, J. und Struck, W. (1992): Betrachtungen zur Ermittlung der Sicherheitsbeiwerte für
Pfahlbelastungen aus Stichprobenergebnissen und zusätzlichen Informationen, Geotechnik 1992, S 138
ff
Hanisch, J. und Struck, W. (1985): Charakteristischer Wert einer Boden- oder Materialeigenschaft aus
Stichprobenergebnissen und zusätzlicher Information, Bautechnik 10/1985
Hanisch, J. (1998): Ist der EUROCODE 7 noch zu retten? - Wird der EC 7 zur Hilfe oder zur Bremse bei der
Beurteilung der Zuverlässigkeit neuer Bauarten? Bautechnik 9/1998
Hartmann, R., Nawari, O. (1996): Ansatz der Fuzzy-Logik und –Set Theorie in der Geotechnik – neue Wege
zur Unsicherheits- und Risikobewertung. Vorträge der Baugrundtagung in Berlin, Deutsche
Gesellschaft für Erd- und Grundbau, Eigenverlag
Heibaum, M. (1987): Zur Frage der Standsicherheit verankerter Stützwände auf der tiefen Gleitfuge,
Darmstadt; Mitteilungen des Instituts für Grundbau, Boden- und Felsmechanik, Heft 27
Joint-Committee on Structural Safety (1976): Common unified Rules for Different Types of Construction
and Material, Comité-Euro-International du Beton (CEB), Bulletin d´ínformation No 116 E
Katzenbach, R., Moormann, C. (2003): Überlegungen zu stochastischen Methoden in der Bodenmechanik am
Beispiel des Frankfurter Tons, Heft 16 der Gruppe Geotechnik, Technische Universität Graz, pp 255-
282
Kramer, H. (1982): Diskussionsbeitrag bei der zur Spezialsitzung „Sicherheit im Grundbau“ Vorträge der
Baugrundtagung 1982, Braunschweig
Lesny, K. (2012): Implementation of Eurocode 7 within German Geotechnical Design Practice, Geotechnical
Special Publication „Modern Geotechnical Design Codes of Practice – Development, Calibration &
Experiences“
Lumb, P. (1970): Safety factors and the probabilistic distribution of soil strength, Canadian Geotechnical
Journal, 7, 225-242
Matousek, M. und Schneider, J. (1976): Untersuchungen zur Struktur des Sicherheitsproblems bei
Bauwerken, Bericht Nr. 59 aus dem Institut für Baustatik und Konstruktion ETH Zürich, Basel und
Stuttgart: Birkhäuser Verlag
Orr, T. (2007): The Story of Eurocode 7, Spirit of Krebs Ovesen Session, European Conference on Soil
Mechanics and Geotechnical Engineering, Madrid, 2007
Peintinger, B. (1982): Auswirkung der räumlichen Streuung von Bodenkennwerten, Vorträge der
Baugrundtagung Braun-schweig,1982, S 105 – 117
Pöttler, R.; Schweiger, H.F.; Thurner, R. (2001): Probabilistische Untersuchungen für den Tunnelbau –
Grundlagen und Berechnungsbeispiel, Bauingenieur - Ausgabe 03-2001, S. 101
Pottharst, R. (1982): Erläuterung des statistischen Sicherheitskonzepts am Beispiel des Grundbruchs,
Vorträge der Baugrundtagung, Braunschweig, S 9 – 47
Rackwitz, R. und Peintinger, B., (1981), Ein wirklichkeitsnahes stochastisches Bodenmodell mit unsicheren
Parametern und Anwendung auf die Stabilitätsuntersuchung von Böschungen, Bauingenieur 56, 215 –
221
Rackwitz, R. (1982): Können Vorinformationen über den Baugrund quantifiziert werden? Vorträge der
Baugrundtagung 1982, Braunschweig, S 83 – 104
Reitmeier, W. (1989): Quantifizierung von Setzungsdifferenzen mit Hilfe einer stochastischen
Betrachtungsweise, Lehrstuhl und Prüfamt für Grundbau, Bodenmechanik und Felsmechanik der
Technischen Universität München, Schriftenreihe Heft 13
Representatives of the Geotechnical Societies within the European Countries (1987): Draft Model for
Eurocode 7 – common unified rules for Geotechnics, Design.
Russelli, C. (2008): Probabilistic methods applied to the bearing capacity problem, Mitteilung 58 - Institut für
Geotechnik, Universität Stuttgart
Sadgorski, W. (1983): Neues vom Eurocode 7, Geotechnik 6, S 107 – 110
Schneider, J. (1994): Sicherheit und Zuverlässigkeit im Bauwesen, B. G. Teubner Verlag, Stuttgart
Schuppener, B., Walz, B., Weißenbach, A. and Hock-Berghaus K. (1998): EC7 – A critical review and a
proposal for an improvement: a German perspective, Ground Engineering, Vol. 31, No. 10
Smoltzcyk, U. und Schad, H. (1990): Zur Diskussion der Teilsicherheitsbeiwerte für den
Grundbruchnachweis, Geotechnik 13 (1990) S 41-43
Stahlmann, J.; Schmitt, J.; Fritsch, M. (2007): Anwendung probabilistischer Methoden zur Simulation der
stofflichen Inhomogenitäten des Untergrunds in der Geotechnik, Bauingenieur - Ausgabe 5/2007, S.
214-223
Vollenweider, U. (1982): Diskussionsbeitrag bei der zur Spezialsitzung „Sicherheit im Grundbau“ Vorträge
der Baugrundtagung, Braunschweig
@seismicisolation
@seismicisolation
Vollenweider, U. (1983): Denkanstöße im Grundbau oder die Lösung grundbaulicher Probleme mittels
Gefährdungsbildern, Schweizer Ingenieur und Architekt, 7/83
Vollenweider, U. (1988): Gedanken zur Sicherheit im Grundbau, Schweizer Ingenieur und Architekt Nr. 39,
S 1069-1075
von Soos, P. (1982): Zur Ermittlung der Bodenkennwerte mit Berücksichtigung von Streuung und
Korrelationen, Vorträge der Baugrundtagung 1982, Braunschweig, S 83 – 104
von Soos, P.: (1990), Die Rolle des Baugrunds bei der Anwendung der neuen Sicherheitstheorie im
Grundbau, geotechnik 13 (1990), S, 82-91
Weißenbach, A. (1991): Diskussionsbeitrag zur Einführung des probabilistischen Sicherheitskonzepts im
Erd- und Grundbau, Bautechnik 68, Heft 3 S 73-83 (1991)
Weißenbach, A. (1998): Umsetzung des Teilsicherheitskonzepts im Erd- und Grundbau, Bautechnik 9/1998
Ziegler, M. (2002): Risikosimulationsrechnung – eine Möglichkeit zur Quantifizierung von Sicherheit und
Risiko in der Geotechnik, Vorträge der Baugrundtagung in Mainz, Deutsche Gesellschaft für Erd- und
Grundbau, Eigenverlag
@seismicisolation
@seismicisolation
IOS Press, 2013
doi:10.3233/978-1-61499-163-2-116
British choices of geotechnical design

approach and partial factors for EC7
Brian SIMPSON 1
Arup Geotechnics
Abstract. The Eurocode system allows each nation to specify, for designs to be
constructed on its territory, which of its three geotechnical “design approaches”
are to be used and what values are to be given to partial factors. These are
published in the National Annex to Eurocode 7 (EC7). This paper reviews the
choices made by British engineers working under the auspices of BSI to produce
the UK National Annex. The UK has chosen to use Design Approach 1, judging
that it has the potential to provide the best balance between safety and economy
over a wide range of design types, while allowing broad compatibility with
previous designs. It also facilitates use of finite element analysis, a feature not
shared with Design Approach 2. At this stage in development, values for the
partial factors have been selected on the basis of calibration against past
experience; probabilistic calculations have not been used to a significant extent in
this process. Particular difficulties have been encountered in providing for pile
design to EC7, for which the code gives separate requirements for design based on
load testing and design based on calculation using ground properties. In practice
in the UK, these are often used in combination, and it was found necessary to vary
the factors offered for calculations by EC7 as a function of the amount and type of
testing undertaken.
Keywords: Eurocode 7; National Annex; Design Approach; piling

Introduction
The Eurocodes have been developed as a compatible system of standards for the design
of buildings and civil engineering structures. The Basis of design is provided by
EN1990 (BSI 2002). EN1991 Actions on structures is a loading code and EN1992 to
EN1996 are concerned with design of structures using various types of materials.
Eurocode 7 is concerned with Geotechnical design; Part 1 (EN1997-1, BSI 2004) gives
General rules for design and Part 2 (EN1997-2) covers Ground investigation and
testing. This paper is concerned with EN1997-1, which will be referred to here as EC7.
The Eurocodes are first prepared by international committees for the European
standards body Comité Européen de Normalisation (CEN). The system allows each
nation to provide “National Annexes” giving values for partial factors, and some other
“nationally determined parameters”, to be used for construction on its territory. For
EC7, one of three “Design Approaches” can be chosen, which indicates the system of
partial factors to be applied for ultimate limit state (ULS) design. The United Kingdom
1
Corresponding Author: Brian Simpson, Arup Geotechnics, 13 Fitzroy Street, London W1T 4BQ, UK; E-
mail: Brian.Simpson@Arup.com.
@seismicisolation
@seismicisolation
B. Simpson / British Choices of Geotechnical Design Approach and Partial Factors for EC7 117
National Annex (UKNA, BSI 2007) for EC7 requires the use of Design Approach 1
(DA1); the reasons for this are discussed in this paper.
For design of piled foundations, the code gives separate requirements for design
based on load testing and design based on calculation using ground properties. In
practice in the UK, these are often used in combination, and it was found necessary to
vary the factors offered for calculations by EC7 as a function of the amount and type of
testing undertaken.
1. Design approaches for ULS calculations
Table 1 shows the values of partial factors for the three design approaches, which can
be varied nationally. Table 1 shows the UK values for DA1 and the CEN values for
DA2 and DA3. Values for piled foundations will be considered later.
Table 1. Values for the partial factors in EC7.

DA1 DA2 DA3
Comb 1 Comb 2
Actions Permanent unfav 1.35 1.35 1.35(1.0)*
fav
Variable unfav 1.5 1.3 1.5 1.5(1.3)*
Soil tan φ' 1.25 1.25
Effective cohesion 1.25 1.25
Undrained strength 1.4 1.4
Unconfined strength 1.4 1.4
Spread Bearing 1.4
footings Sliding 1.1
Retaining Bearing capacity 1.4
walls Sliding resistance 1.1

Earth resistance 1.4
Note: Blanks in the table indicate that factors are 1.0.
* 1.35 and 1.5 for structural loads (1.0 and 1.3 for loads derived from the ground).
Design Approach 1 requires two separate calculations using two “combinations” of

factors. The entire design, geotechnics and structure, has to accommodate both
combinations. The action factors in Combination 1 of DA1 are generally applied to the
actions themselves, but in some cases EC7 2.4.7.3.2(2) is followed, applying the
factors to action effects; this applies particularly to the structural action effects
(bending moments etc) caused by earth and water pressures.
Design Approach 2 (DA2) includes factors to be applied to actions; it is similar to
the LRFD method favoured in the USA. Originally the factors were to be applied to
actions themselves, meaning the basic pre-defined loads acting on a structure at the
start of the equilibrium calculations; this form of DA2 is used by some countries.
However, some developments, particularly in Germany, have specified that equilibrium
and compatibility calculations are carried out in terms of unfactored characteristic
@seismicisolation
@seismicisolation
118 B. Simpson / British Choices of Geotechnical Design Approach and Partial Factors for EC7
values, applying the factors to derived action effects (such as bending moments,
bearing pressures or active earth forces). This approach, called DA2*, is considered by
its developers to follow EC7 2.4.7.3.2(2).
In Design Approach 3, factors are generally applied to actions, not to action effects.
The calculations are performed using design values for loads and material strengths,
rather than characteristic values.
It is generally recognised that DA2 is not suitable for problems such as slope
stability and for the use of finite element analysis. As a result, most countries that have
adopted DA2 as their basic approach use it in combination with DA3 for specific cases.
The use of DA1 is not very different from the combined use of DA2 and DA3, except
that DA1 requires checking of two calculations, whereas combined use of DA2 and
DA3 could imply acceptance of a design that passes according to one DA but fails
according to the other. The legal implications of such a situation might be debatable.
2. British choice of Design Approach 1
2.1. UK National Annex
The British choice, represented by the UK National Annex published by BSI, is to use
DA1. This is the original approach published in the earlier ENV version of EC7, and it
is considered to have important advantages noted below.
The main objections to DA1 have been (a) that it requires two calculations and (b)
that it is not consistent with previous practice using global factors of safety. It is
sometimes also argued (c) that strength factoring leads to the “wrong” failure
mechanism.
In response to objection (a), it is argued that carrying out a second calculation is in
most cases trivial in comparison with other tasks required in the process of design, such
as ground investigation and determining the characteristic values of soil strengths and
other parameters to be used. This is particularly the case when computer software is
used for the calculations.
2.2. Compatibility with previous practice
In relation to objection (b), the partial factor format of the Eurocodes was set up with
the intention that it could give a more rational basis of design than the former global
safety factors. While the benefits of past experience must not be lost, in the author’s
view it is regrettable that attempts are made simply to replicate the past, as has been
done to a large degree by the development of approach DA2*.
Previous UK practice used single factors of safety, which in some cases were
applied in calculations as factors on soil strength. For example, in design of embedded
retaining walls factors on soil strength have been in common use since they were
introduced as an option in CIRIA Report 104 (Padfield and Mair 1984). Although the
values were little changed, these were interpreted as “mobilisation factors”, associated
with both serviceability and ULS in the British Standard on retaining structures,
BS8002 (BSI 2001). CIRIA Report C580 (Gaba et al 2003) recommended the use of
the factor values from BS8002 but considered them as strength factors; this approach
differed little from DA1 Combination 2 of EC7, which is usually the dominant
combination for embedded retaining walls.
@seismicisolation
@seismicisolation
For slope stability, UK practice, in common with most others, uses factors of
safety applied to soil strength. For both spread and piled foundations, however,
resistance factors were commonly used. Use of factors on pile resistance is retained in
DA1.
It can be seen, therefore, that approaches using factors on soil strength were
already familiar in the UK, so its adoption on DA1 was not inconsistent with previous
practice. The factors adopted in DA1 are not identical with those of previous UK
practice and therefore designs may differ slightly. Nevertheless, comparisons have
shown broad compatibility with previous designs, and this is considered acceptable (eg
Gaba et al 2003, Simpson 2005, Bond and Simpson 2010-11).
2.3. The wrong failure mechanism?
In relation to objection (c), the author questions whether there is such a thing as the
“right” failure mechanism. Failure is not the “right” outcome of design, and so EC7 is
concerned more about proving success than studying failure. The purpose of partial
factoring, therefore, is to show that even in extreme circumstances failure will not
occur, or at least that the system could only be on the very point of failure, by the most
critical mechanism.
3. Consistent levels of reliability
DA1 is intended to provide a reasonable balance of safety and economy over the full
range of geotechnical problems, and to provide a system consistent with structural
design. This means, for example, that if a retaining wall is used to stabilize a slope that
supports foundations, the wall, slope and foundations can all be designed in consistent
geotechnical calculations, which pass seamlessly into the structural designs of the wall
and foundations. The connection between geotechnical and structural designs was a
noted weakness of existing UK codes, particularly for retaining structures.
These issues were considered by Simpson (2007) and by Simpson et al (2009).
One of the aims of design is to achieve roughly constant reliabilities irrespective of
how actions, strengths and resistances combine in particular situations. In Annex C of

EN1990, reliability is represented by the target reliability index β, which represents the
number of standard deviations between the characteristic state and the ULS design state.
EN1990 discusses how the values of partial factors might be selected in order to
achieve a target reliability index, proposing that factors could be applied
simultaneously to actions and strengths (or action effects and resistances). In effect it
proposes that the action effects for ULS design should be 0.7β standard deviations from
their characteristic values, and the margin on resistances should be 0.8β (the symbol
used for the factors 0.7 and 0.8 is α). But EN1990 places an important limit on this
approach: it is only applicable if the ratio of the standard deviations of the action effect
and resistance, σE/σR, lies within the range 0.16 to 7.6. The implication of this is that a
different approach is to be used if the uncertainty of one of variables – actions or
resistances – is much more important to the design than is the other one. For such a
situation, the margin on the more critical variable is required to be 1.0β, with a lower
margin, 0.4β, on the less critical variable.
@seismicisolation
@seismicisolation
Ratio of β achieved to β required

1.2
E R = 0.16
E R = 7.6
Less
economic
1.1
σ /σ
SAFETY RATIO.
σ /σ
1
0.9
αE=-0.7, αR=0.8
0.8
Less
safe
0.7
0.6
0 0.2 0.4 0.6 0.8 1
σE/(σR+σE)
Figure 1. Reliability achieved using (0.7, 0.8) combination for α.
1.2
Typical
1.1 foundations
.
1
SAFETY RATIO
αE=-0.7, αR=0.8
0.9
0.8
Slope Tower
0.7 stability foundations
0.6
0 0.2 0.4 0.6 0.8 1
σE/(σR+σE)
Figure 2. Reliability for some typical geotechnical situations.
The result of this approach is shown in Figure 1, in which the reliability achieved
(in terms of number of standard deviations of the design point from the characteristic
state) is plotted against the ratio of the standard deviations expressed as σE/(σE+σR).
The result is plotted as a “safety ratio” by dividing by the required reliability,
β standard deviations, so that the desired value is 1.0. Over the range in which both σE
and σR are of similar, significant magnitude, the result is reasonably close to the desired
value. However, as either σE and σR becomes small compared to the other, the
reliability achieved drops substantially, indicating an unsafe design with inadequate
reliability. This explains why EN1990 limits the range of applicability of the approach
to σE/σR = 0.16 to 7.6.
Figure 2 shows that in geotechnical design it is important to consider the full range
of σE/σR values. Conventional foundations may have σE and σR of similar magnitude,
but other situations may be dominated by either σE or σR. For example, in slope
stability problems there is often very little uncertainty about the loading and
@seismicisolation
@seismicisolation
uncertainty of soil strength is dominant, as shown by the fact that factors of safety are
normally applied to soil strength. At the other extreme, designs for foundations of tall
towers may have loading as the dominant uncertainty. In geotechnical design, these
problems often occur together, so the approach adopted must be able to accommodate
the full range of σE/σR.
Figure 3 shows the result of an approach using two “combinations” in which the
margin on the more critical variable is required to be 1.0β, with 0.4β on the less critical
variable. Much greater consistency is achieved, with none of the resulting values
falling substantially lower than required (ie 1.0). This illustrates the benefit of the use
of two combinations: a very wide range of design situations can be covered without
change in the design approach.
In common with other design approaches, the factors used in DA1 have not been
deduced by probabilistic calculation. Nevertheless, they do reflect the principles
propounded in EN1990, illustrated in Figure 3, and the lessons that may be learnt by
considering a probabilistic framework.
Although the concept of “combinations” is relatively new to geotechnics, it is
familiar to structural engineers who frequently design for several combinations of
actions. The background to DA1 is essentially the same as that of combinations of
actions, giving a severe value to the lead variable in combination with less severe
values of other variables, but in DA1 the method is extended to include resistances or
material strengths, as suggested by EN1990. The fundamental principle of DA1 is that
“All designs must comply with both combinations in all respects, both geotechnical and
structural”. I this context, the word “design” means “that which will be built”.
Ratio of β achieved toβ required
1.2
Uneconomic
αE=-0.4, αR=1.0 αE=-1.0, αR=0.4
1.1
.
1
SAFETY RATIO
0.9
0.8
Unsafe
0.7
0.6
0 0.2 0.4 0.6 0.8 1
σE/(σR+σE)
Figure 3. Reliability achieved using (1.0, 0.4) combinations for α
4. Apply the factors where the uncertainties lie
EN1990 [6.3.2(4)] refers to “non-linear analysis”, by which it means situations where

an action effect changes disproportionately as the action changes. Figure 4 shows the
requirements of EN1990 when there is a non-linear relationship between actions and
@seismicisolation
@seismicisolation
Resistance Factor the

Ac tio n effect
Factor the resistance
action effect
Factor the
material
strength
Factor the
action
Ac tio n Material
strength
Figure 4. Non-linear relationship between actions Figure 5. Non-linear relationship between material
and action effects strengths and resistance
action effects. If the changes in the effect are disproportionately large compared with
those in the action, then it is important that safety factors are applied to the action, not
the action effect.
EN1990 does not consider the possibility of a similar disproportionate relationship
between material strength and resistance, and this may not be very important in
structural engineering. In soils, however, where the strength is essentially frictional,
such disproportionality is often significant. For example, passive resistance and
bearing capacity both increase disproportionately with angle of shearing resistance ϕ′
(Simpson et al 2009); in some cases, when ϕ′ is large, a small change in ϕ′ has a very
large effect on the resistance. The author suggests that similar thinking should therefore
be applied to strengths and resistances as to actions and action effects.
Figure 5 shows situations where there is a non-linear relationship between material
strength and resistance. The author submits that if the changes in the resistance are
disproportionately large compared with those in the strength, then it is important that
safety factors are applied to the strength, not the resistance.
5. Apply factors before combining variables
Disproportionate effects may occur simply due to the addition of actions which tend to
cancel. Interest in partial factoring methods in the United Kingdom was encouraged by
the study of the collapse of the Ferrybridge cooling towers (for more detail see
Simpson et al 2009). The stress in the concrete was derived from the addition of
compression due to the weight of the towers and tension due to wind loading, which
tended to cancel each other. Using a working state approach, this resulting stress was
then compared with a factored strength. Unfortunately, the wind loading was
underestimated and this led to a disproportionately large increase in the resulting
tension, which caused a very serious collapse. The disaster might have been avoided if
the two actions, weight and wind load, had been factored separately before being
combined into a single action effect.
In view of this, parameters are factored before they are combined in DA1, as far as
it is reasonably possible, and generally factors are applied to soil strength rather than to
“resistances”. Difficult situations for combining actions occur in earth pressure
calculations for design of retaining walls, affecting DA1 Combination 1 and also DA2.
@seismicisolation
@seismicisolation
These are considered in EN1997-1 in 2.4.7.3.2(2) which says “In some design
situations, the application of partial factors to actions coming from or through the soil
(such as earth or water pressures) could lead to design values which are unreasonable
or even physically impossible. In these situations, the factors may be applied directly to
the effects of actions derived from representative values of the actions.”
The author submits that this approach is acceptable for DA1 Combination 1, where
factors are applied to actions, provided that Combination 2 is also checked, factoring
the soil strength before using it to calculate resistances. The absence of an equivalence
of this in DA2 is considered by the author to be a shortcoming of DA2. One additional
benefit of this approach is that it coincides with conventional structural design,
indirectly applying a factor to ground water pressure for the calculation of structural
forces and bending moments.
6. Compatibility with numerical analysis
Numerical methods can be used relatively easily for ULS computations if this merely
requires using factored values for the input to the program, or simply factoring the
structural action effects resulting from the geotechnical program. Design Approach 2
requires factors to be applied to quantities that are internal to the geotechnical analysis
such as active and passive forces or pressures, and bearing resistance for spread
foundations. So it is generally accepted that full numerical analyses of ultimate limit
states, in which resistances are internal to the numerical computation, cannot be
undertaken for DA2. Most countries that use DA2 require use of DA3 for numerical
analysis.
Design Approach 1 was the only approach in the ENV version of EC7 published in
1995. In its development, the possible use of numerical methods was considered, so
they can be used relatively easily with DA1.
7. Distinction between ULS and SLS

As adopted in the UKNA, DA1 is used to prevent exceedance of ultimate limit states
and it is assumed that serviceability will be considered separately. An exception occurs
for piled foundations, as discussed below.
For spread foundations, it will often be the case that design is governed by
settlement criteria, and no attempt has been made to adjust the DA1 factors for ULS so
as to cover this. It could well be that spread foundations designed only checking DA1
factors for ULS would have unacceptably high bearing pressures in service and so
would settle too much. For spread foundations on clays, EC7 allows the possibility
that settlement could be limited by providing a larger overall factor of safety
(6.6.2(16)), but this is seen clearly as a serviceability requirement.
For retaining structures, if displacement is critical a separate assessment of it is
required, using a combination of experience, case histories and calculation. In practice,
this assessment is most likely to affect the choice of type of wall rather than the details
of its geometry or strength. It is usually found that walls that are adequate for ULS
using DA1 factors do not need modification for serviceability, apart from the
reinforcement needed for crack width control in reinforced concrete.
@seismicisolation
@seismicisolation
8. Reliability calculations and “calibration” in deriving values for partial factors
It was noted above that principles of reliability theory underlie the general form of
DA1. However, values of partial factors have been checked by calibration with past
experience, and generally reliability calculations have not been attempted. It is not
clear whether use of reliability analysis could provide a more rational basis for
selection of factor values, but it is thought that an extremely high level of expertise
would be needed to handle the very wide variety of data, often quite sparse, typically
involved in geotechnical design. This issue was considered further by Simpson (2011)
in relation to the use of reliability theory in the design process itself.
The Eurocodes provide partial factors on a small number of leading parameters.
However, in order to achieve robust designs sufficient to protect public safety, it is
necessary to accommodate unexpected events and variations in other, secondary
parameters; these could include, for example, minor accidents or vandalism, small
errors or changes in construction, minor deterioration, etc. Calibration of new formats
against existing experience is seen to be the best way to achieve this.
Derivation of factor values from existing experience is problematic, however,
when there is a desire to separate ULS and SLS calculations. Successful existing
designs have usually satisfied both criteria, but it may be unclear which of them
governed the need for particular factors of safety. The process followed has therefore
been to check the use of the factor values proposed in the CEN version of EC7 against
design results from previous experience, noting the relevance of serviceability
calculations and assessments. These values have been found to be appropriate for
general use, but for piling the factors in the UK national annex have been changed
significantly from teh CEN values.
9. Piling
Design of pile foundations is the subject of Section 7 of Eurocode 7. Clause 7.4.1 says
that design shall be based on:
a) the results of static load tests, which have been demonstrated, by means of
calculations or otherwise, to be consistent with other relevant experience; or

b) empirical or analytical calculation methods whose validity has been
demonstrated by static load tests in comparable situations.
Design may also be based on dynamic tests or observed behaviour of other piled
foundations. However, in the UK most piles are designed using a combination of load
testing and calculation.
Method (a) above emphasises load tests and method (b) emphasises calculations,
but both methods rely to some extent on both load tests and calculations. However, in
the remainder of Section 7 design procedures based on load testing and calculation are
dealt with separately, and it is not clear how they are to be combined.
In the default values of partial factors given in the CEN version of EC7, factors
greater than unity are allocated to pile resistances in Combination 1, but these are all
taken as unity in the UKNA, thus providing a simple system. Combination 1, with load
factors as shown in Table 1, may govern structural design of piles, but Combination 2
generally governs the geotechnical design. Table 2 shows the values of resistance
factors for ULS in the CEN version of EC7 and in the UKNA. These are to be applied
to characteristic resistances, defined by EC7 (2.4.5.2(2)) as “a cautious estimate of the
@seismicisolation
@seismicisolation
value affecting the occurrence of the limit state”. Paragraph 2.4.5.2(11) adds that “If
statistical methods are used, the characteristic value should be derived such that the
calculated probability of a worse value governing the occurrence of the limit state
under consideration is not greater than 5%”. This gives an indication of the meaning of
“cautious estimate” in the definition of characteristic value. It implies that the result of
a single pile loading test or of a “best estimate” calculation is unlikely to be sufficiently
cautious to comply with the definition.
For design based primarily on load testing – method (a) above – EC7 (7.6.2.2)
provides a method of deriving characteristic pile resistance from the results of one or
more tests, using the mean and minimum of the test results and dependent on the
number of tests. In the UK, however, the industry has judged that the method used is
primarily based on calculation – method (b) above – using ground test results as input
and supplemented by load testing. For this, 7.6.2.3 says in a note “the values of the
partial factors γb and γs recommended in Annex A may need to be corrected by a model
factor larger than 1,0. The value of the model factor may be set by the National annex.”
It is considered that the biggest uncertainty in calculating pile resistances is in the
calculation model, rather than in the parameters used to characterize the ground mass.
The UK approach is therefore to calculate shaft and base resistance from measured
ground parameters (“using methods whose validity has been demonstrated by static
load tests in comparable situations”), and to apply a model factor to the result of the
calculation in order to derive characteristic (ie cautious) values of resistances. The
complete process is illustrated in Figure 6. The value of the model factor depends on
whether there has been a trial pile tested at least to a load equivalent to the calculated
unfactored ultimate resistance. If such a trial has been carried out the model factor is
1.2; otherwise it is 1.4.
Characteristic soil strengths

(cu,k, tanφk, etc)
Calculation model – accurate or
erring on the side of safety
Calculated shaft and base

resistance
γRd=1.4 or 1.2 sĂůƵĞĚĞƉĞŶĚƐŽŶ

ƚƌŝĂůƉŝůĞƚĞƐƚƐ
Characteristic shaft and
base resistance
γs and γb sĂůƵĞĚĞƉĞŶĚƐŽŶ
ǁŽƌŬŝŶŐƉŝůĞƚĞƐƚƐ
Design shaft and base
resistance (ULS)
Figure 6. Process of pile design required by the UKNA.
Having derived the characteristic resistance of the pile, the design ULS resistance
is found by dividing the shaft and base resistances by further factors γs and γb;
alternatively, the total resistance may be divided by γt. As for the model factor, the
@seismicisolation
@seismicisolation
values of these partial factors are dependent on the amount of testing undertaken. In
this case, a reduction is allowed if at least 1% of the piles is loaded tested to at least 1.5
times the representative load; these are generally working piles. The UKNA also
allows this reduction in γ-values if settlement in service is explicitly predicted by some
other reliable means, or if settlement at the serviceability limit state is of no concern.
Table 2 shows that the partial factor values adopted in the UK are bigger than the
default values recommended in the CEN version of the code. Most other countries
have also increased these factors, either directly or indirectly through the use of model
factors. The previous British foundations code, BS8004 (BSI 1986), recommended
that piles be designed to an overall factor of safety between 2 and 3, depending whether
“ultimate bearing capacity has been determined by a sufficient number of loading tests
or where they may be justified by local experience”. In a more detailed paper on pile
design to EC7, Bond and Simpson (2009-10) have shown that the factors included in
the UKNA are reasonably consistent with previous practice. For shaft controlled bored
piles, overall factors of safety may be slightly lower than previously; in the case of a
pile with negligible base resistance and carrying only permanent compression loads,
the overall factor of safety could be as low as 1.68 with maximum testing, increasing to
2.24 if the design relies on calculation alone with no testing. Slightly lower values are
allowed for driven piles. The values are higher for base-controlled piles, being close to
3 for base-controlled bored piles with no load testing.
Table 2. Values for the partial factors in piling – DA1 Combination 2.

CEN UKNA UKNA
<1% tested >1% tested
Driven piles Base 1.3 1.7 1.5
Shaft in compression 1.3 1.5 1.3
Total/combined in compression 1.3 1.7 1.5
Shaft in tension 1.6 2.0 1.7
Bored piles Base 1.6 (1.45) 2.0 1.7
(CFA piles) Shaft in compression 1.3 (1.3) 1.6 1.4
Total/combined in compression 1.5 (1.4) 2.0 1.7

Shaft in tension 1.6 (1.6) 2.0 1.7
10. Concluding remarks
The UK National Annex for Eurocode 7 requires the use of Design Approach 1. The
reasons for this choice have been given: essentially it is considered that DA1 covers the
widest range of design types, balancing of economy and safety, and also allows the use
of all appropriate forms of calculation, including the finite element method. The
default values for partial factors given by CEN have generally been accepted, except in
the case of pile design, for which values have generally been increased. A model factor
is included in the derivation of characteristic pile resistances, and all the factors are
adapted to suit the combination of calculation and testing typically used in the design
process.
@seismicisolation
@seismicisolation
References
Bond, AJ and Simpson, B (2009-10) Pile design to Eurocode 7 and the UK National Annex (2 parts).
Ground Engineering, Dec 2009 and Jan 2010.
BSI (1986) BS 8004:1986, Code of Practice for Foundations. BSI, London.
BSI (2001) BS 8002:1994, Code of Practice for Earth Retaining Structures. BSI, London.
BSI (2002) BS EN 1990:2002. Eurocode: Basis of design. BSI, London.
BSI (2004) BS EN 1997:1: 2004. Eurocode 7: Geotechnical design - Part 1: General rules. BSI, London.
BSI (2007) UK National Annex to Eurocode 7: Geotechnical design - Part 1: General rules. BSI, London.
Gaba, AR, Simpson, B, Powrie, W,& Beadman, DR (2003) Embedded retaining walls: guidance for
economic design. CIRIA Report C580.
Padfield, C.J. & Mair, R.J. (1984) Design of retaining walls embedded in stiff clay. CIRIA Report 104.
Simpson, B (2005) Eurocode 7 Workshop – Retaining wall examples 5-7. ISSMGE ETC23 workshop,
Trinity College, Dublin.
Simpson, B (2007) Approaches to ULS design - The merits of Design Approach 1 in Eurocode 7.
ISGSR2007 First International Symposium on Geotechnical Safety & Risk pp 527-538. Shanghai
Tongji University, China.
Simpson, B (2011) Reliability in geotechnical design – some fundamentals. Proc 3rd Int Symp on
Geotechnical Safety and Risk, Munich.
Simpson, B, Morrison, P, Yasuda, S, Townsend, B, and Gazetas, G (2009) State of the art report: Analysis
and design. Proc 17th Int Conf SMGE, Alexandria, Vol 4, pp. 2873-2929.
@seismicisolation
@seismicisolation
IOS Press, 2013
doi:10.3233/978-1-61499-163-2-128
Dutch approach to Geotechnical design by

Eurocode 7, based on probabilistic analyses
Ton VROUWENVELDERa, Adriaan van SETERSb and Geerhard HANNINKc
a
Delft University of Technology, TNO, Delft
b
Fugro Geoservices, Leidschedam
c
Public Works Rotterdam
Abstract. In the Netherlands, due to the large infrastructure and waterway projects
from the 1980’s onwards, much experience was developed using probabilistic
design. Due to the vast extent of e.g. the Eastern Scheldt Storm Surge Barrier,
these structures could not have been designed economically using conventional
engineering. The probabilistic approach is reflected in the building and
geotechnical standards which were introduced in the early 1990’s. These standards
were based on a target probability of failure or reliability index β. Probabilistic
analyses (Monte Carlo) were undertaken to determine the partial load and material
factors. These analyses were also performed for geotechnical structures, leading to
a large set of material factors for soil properties. The old method of Ultimate Limit
State analysis consisting of overall resistance factors was then replaced by partial
factors. Consequently, when Eurocode 7 was established around 2000, Design
Approach 3 adopting partial load and material factors was preferred in the
Netherlands as its philosophy is similar to the 1991 Dutch building codes.
Keywords. Partial factors, Eurocodes, Limit state design, Geotechnics

Introduction
The first Dutch geotechnical design codes appeared around 1990. They have been
developed in the same period as part 1 of Eurocode 7. The need for the geotechnical
codes became imminent by new legislation for the Dutch building industry, that was
approved in 1991.
The new Dutch geotechnical design codes consisted of one general part and two
specific parts for the design calculations of foundations on piles and of shallow
foundations respectively. They were the result of a discussion of more than 15 years
about the use of an overall safety factor or the use of partial factors on loads and
strength of materials. The content of the Dutch code on the design of pile foundations
in the Netherlands has been reported during a seminar of the European Regional
Technical Committee 3 ‘Piles’ in Brussels (Everts & Luger, 1997).
Because of the presence of Holocene peat and clay layers at the top, pile
foundations are, especially in the western part of the Netherlands, installed since the
medieval ages. The Royal Palace in Amsterdam, that was built from 1648 to 1665, has
been founded on 13,659 wooden piles. The length of the piles was mainly assessed by
experience. For important buildings and constructions a test pile was driven to be sure
that the pile base reached the bearing sand layer.
@seismicisolation
@seismicisolation
T. Vrouwenvelder et al. / Dutch Approach to Geotechnical Design by Eurocode 7 129
The experimental way of realizing pile foundations in the Netherlands lasted until
the invention of the Dutch cone penetrometer test (CPT) around 1930. The first
relations between the results of CPT’s and the bearing capacity of a pile were proposed
by Koppejan & Van Mierlo (1952). Since that time the CPT’s were used to predict the
ultimate load on a specific type of pile.
In the Netherlands generally low factors are used for driven prefabricated concrete
piles, and relatively high factors for in situ made piles. This resulted in overall safety
factors varying from 1.7 to 2.5 for displacement piles and from 2.5 to 3.5 for non-
displacement piles.
The current safety for driven piles in the Netherlands is made up of a number of
partial factors: the partial factor on actions γF (1.3 – 1.5), the partial resistance factor γR
on compression (1.2), and the correlation factor ξ3 to derive characteristic values from
ground test results (1.25 – 1.40). This can be compared with an overall safety factor of
1.4 * 1.2 * 1.3 ~ 2.2.
The ultimate bearing capacity of a shallow foundation is found when a potential
slip pattern or a continuous plastic zone has developed. In the past Prandtl’s approach
was followed in the Netherlands with an overall safety factor between 2 and 3.
Sheet pile walls are usually installed in undisturbed soils and afterwards
excavation at one side takes place. A number of failure modes needs to be considered,
such as rotation of the wall, slip of the anchor, buckling of the strut, excessive bending
of the wall, excessive settlement behind the wall, bottom heave and slip failure. In the
past Blum’s design approach was followed in the Netherlands, where the required
quality of the wall is determined by the maximum bending moment. Usually a safety
factor of 2 was applied.
Slopes were generally analyzed according to the Bishop method of analysis using
low characteristic values for the soil parameters. The resulting overall factor of safety
should be at least 1.3 for the permanent situation and 1.1 for the building phase.
The present paper will deal with the general theory behind the development of
partial factors. Examples will be given for shallow foundations and sheet pile walls.
1. The partial factor method

In partial factor design a structure is considered as safe enough if, for all relevant limit
states, the following inequality is fulfilled:
Z ( Xd1,Xd2, ..) > 0 (1)
Here Z( X1, X2, ..) is the so called limit state function, indicating for which values of
the basic variables Xi (loads, material properties, etc) the limit state is exceeded and for
which it is not. Negative values of Z correspond to the failed situation and positive
values of Z to the non-failed situation. The design Xdi values are obtained from:
Xdi = γi Ski (for loads) and Xdi = Rk / γm (for resistance) (2)
where Ski is the representative value of load number i, Rk is the representative value of
the resistance and γ are the partial factors. In the American standards 1/γm is normally
referred to as φ, but that is no essential difference.
In practice it may be difficult to choose the characteristic values and partial factors.
Obviously design values equal to mean values will lead to too many failures. But which
@seismicisolation
@seismicisolation
130 T. Vrouwenvelder et al. / Dutch Approach to Geotechnical Design by Eurocode 7
margins should be taken? Two basic notions are of importance: the scatter or
uncertainty present in the variables on the one hand and the desired reliability on the
other. The latter one may depend on the consequences of failure and the costs
necessary to obtain more safe solutions.
Using the theory of reliability (ISO/TC98, 1994) the following formula for γm
may be derived:
(1 − 1 . 64 V ) (3)
γm =
(1 − αβ V )
where α is a sensitivity coefficient (0<α<1), β is the reliability index, V the coefficient
of variation (V=σ/μ). Formula (3) holds only for Gaussian distributed resistance
variables where the representative value has been taken equal to 5% fractile. Let us
discuss the symbols α, β and V in some detail. The coefficient of variation V in (3)
should follow from observations in nature or in the laboratory. The coefficient should
also include the uncertainties due to the limited amount of data (statistical uncertainty),
measurement errors and the lack of accuracy due to schematisation in the calculations
(model uncertainty).
The index β is related to the level of reliability that is required; roughly, one
might say that the failure probability PF equals about 10 to the power minus β:
PF = Φ(-β)≅ 10-β or β ≅ -log(PF) (4)
This relation is not valid for β < 0.5 and becomes inaccurate for β > 4. From the
theoretical point of view, this β-value should follow from optimisation of the sum of
construction costs and expected damage (risk). However, in practice, one often
calibrates the reliability level β to current design levels. As long as the current design
practice is not obviously unsatisfactory, either from the safety or from the economical
point of view, the average reliability is not changed. An interesting consequence of this
calibration principle is that the choice of the probabilistic models seems to be less
critical.
In Table 1 target reliabilities are presented for ultimate limit states according to the
JCSS Probabilistic Model Code (JCSS, 2001). The values are obtained from a
theoretical cost benefit analysis for representative but simple example structures
(Rackwitz, 2000). The values are reasonably compatible to the calibration studies
mentioned in the previous paragraph.
Table 1. Target reliability indices β for ultimate limit states related to a one year reference period (JCSS,
2000)
Relative cost of safety Consequences of failure
measure Minor Moderate Large
Large β=3.1 β=3.3 β=3.7
Normal β=3.7 β=4.2 β=4.4
Small β=4.2 β=4.4 β=4.7
Finally, the coefficient α in (3) is the sensitivity coefficient. This coefficient should in
principle follow from a probabilistic FORM calculation. FORM stands for first order
@seismicisolation
@seismicisolation
reliability method; in this method nonlinear relations are linearized in a well-chosen

design point or beta point. As the value of a partial factor should hold for a large set of
design situations, all having different sensitivity factors, the value of α may be
conceived as an "average" of a great number of cases. In the course of the time certain
patterns in those average α values have been discovered; according to ISO-2394
(1994) and EN-1990 (1994). The values are presented in Table 2.
Table 2. Standardized α values (ISO/TC98, 1994 and EN 1990, 1994)
Load Resistance
Dominant variable αi = 0.70 αi = 0.80
All other variables αi = 0.28 αi = 0.32
Which variable is the dominant one can be found out by giving all variables the
dominant status one after the other and looking which combination is governing the
design. By the way, for the Ultimate Limit State verification of offshore and other
coastal engineering structures it might be more appropriate to take higher α values for
loads and lower α values for the resistance. The reason is that hydraulic loads often
have slowly decreasing tails in the distribution functions.
One of the immediate advantages of the probability-based method is the possibility
to solve load combinations in a rational way. The essence of load combinations is that
there is no need to check the structure for improbable (or impossible) load
combinations. The usual method in partial factor design is to consider one load as
dominant (as indicated in Table 2) and the other ones as non-dominant or
accompanying. Take as an example the combination of self-weight G and wind loading
W, with V(G) = 0.10 and V(W) = 0.30 respectively. Take the target for β equal to 4. In
that case we have for loads (where usually k ≅ 0):
Combination "G dominant": γG = 1 + 0.70 * 4.0 * 0.10 = 1.28

γW = 1 + 0.28 * 4.0 * 0.30 = 1.34
(5)
Combination "W dominant": γG = 1 + 0.28 * 4.0 * 0.10 = 1.11
γW = 1 + 0.70 * 4.0 * 0.30 = 1.84
The point is that it is not necessary to combine the 1.28 for the dead load and the 1.84
for the wind load in a single verification. Note that in Eurocodes the reduced loads
factors are presented as γΨo. If two loads are fluctuating in time, there is even a further
reduction. The reason is that the maximum values of the two loads need not necessarily
occur at the same point in time.
When calibrating partial factors also the characteristics of the structural system
should be taken into account: What happens after a certain part of the structure has
become instable? A proper safety analysis should address these questions. One should
consider the reliability for the structure as a whole. Next, one might follow two ways of
analysis. One way is to perform a full system reliability analysis. On the market there
are an increasing number of software codes that can handle this type of analyses (see,
Waarts, 2000 and Karadeniz et al, 2003). A less advanced but more practicable
approach is to take the system behaviour into account when stating the target reliability
for the individual element or mechanism in (3).
@seismicisolation
@seismicisolation
As an example, given a simple series systems (see Figure 1) of n completely

uncorrelated members, one might use:
PF member = PF system / n (6)
Using again PF = 10 –β we arrive at:
βmember = βsystem + log(n) (7)
A series system occurs for instance in a long stretched slope. In a similar way, for a
simple parallel system of n completely uncorrelated members, we have for the
coefficient of variation of the resistance:
VR system = VR member / √n (8)
Neglecting still the variability of the loading side one may derive:
βmember = βsystem / √n (9)
These reductions may be used to simulate the averaging effect of large slip planes or a
large number of foundation piles under a stiff superstructure. One may use it to reduce
the partial factors or increase the representative value. Of course, in practice, one needs
also to consider the correlations between members and the possible mixture of systems.
Figure 1. Simple series and parallel systems
2. Application to Eurocode 7
2.1 Shallow foundations
In (Van der Meer et al, 1991a,b) partial safety factors for shallow foundations were
derived, based on FORM type probabilistic analyses. In the Netherlands shallow
foundations are generally built on sands or very stiff, very sandy clays or silts. In these
soils the influence of the effective cohesion is minimal. Design based on undrained
shear strength cu is therefore not common in the Netherlands, as the long term
behaviour is often governing.
@seismicisolation
@seismicisolation
q' q'
B
volume weight
strip footing
Figure 2. Shallow foundation
The bearing capacity R for a strip footing in cohesionless soils (Figure 2) per m’ can be
computed for vertical loading according to the well known bearing capacity equation
(Terzaghi, 1943) as:
R = f m ( N q q '+ 0 . 5 N γ B γ ' ) B (10)
Where:
R = bearing capacity for a strip footing per m’ [kN/m’]
q’ = vertical effective stress at foundation level next to the foundation [kN/m 2]
B = width of the foundation [m]
γ’ = effective volumetric weight below the footing [kN/m3]
Nq, Nγ = bearing capacity factors, dependent of friction angle ϕ’ (Prandtl, 1921 and
Terzaghi, 1943) (see Figure 3)
fm = model factor
Bearing capacity factors
100
90
80
70
Bearing capacity factors Nq, Nγ ' [-]
60
50 Nq
Ng'
40
30
20
10
0
0 5 10 15 20 25 30 35 40 45 50
Angle of internal friction φ ' [degrees]
Figure 3. Bearing capacity factors
@seismicisolation
@seismicisolation
The probabilistic analyses were based on a cohesionless soil (drained conditions,

effective cohesion c’= 0). Three cases were analysed: (1) Family house on silt with
deep groundwater level, (2) Office building on silt with deep groundwater level and (3)
Family house on sand with a shallow groundwater. The parameters used in the analyses
are given in Table 3.
Table 3. Parameter values – Case 1, 2 and 3

Parameter Mean value Coefficient of Characteristic
Case 1 / 2 / 3 Variation V or value
Standard dev σ Case 1 / 2 / 3
Vertical load S [kN/m] 120 / 333 / 120 V = 0.10 139 / 396 / 139
Silt, Sand – friction angle ϕ’[o] 32,4 / 32,4 / 37,4 V = 0.10 27.5 / 27.5 / 32.5
Effective Vol Weight γ’ [kN/m3] 20.8 / 20.8 /10.8 V = 0.05 20.0 / 20.0 / 10.0
Stress q’ [kPa] 15 / 15 / 6 V = 0.20 10 / 10 / 4
Footing width B [m] variable σ = 0.1 1.0 / 2.0 / 1.2
Model factor [-] 1.0 σ = 0.1 1.0 / 1.0 / 1.0
The value of the footing width B was determined in such a way that the reliability
index β was in the realistic and acceptable range. From the FORM probabilistic
analyses the sensitivity coefficients α were derived, as shown in Table 4.
Table 4. Sensitivity coefficient α

Sensitivity coefficient α
Case 1 Case 2 Case 3 Average
Vertical load - 0.20 - 0.20 - 0.19 - 0.20
Friction angle ϕ’ 0.84 0.91 0.87 0.87
Effective Volumetric Weight γ’] 0.02 0.03 0.06 0.04
Stress q’ 0.28 0.20 0.24 0.24
Footing width B 0.35 0.17 0.31 0.28
Model factor 0.23 0.24 0.22 0.23
From the probabilistic analysis, it can be seen, that the α-factor is fairly constant for all
cases, except for the α-factor for footing width B. Table 4 also shows that the friction
angle ϕ’ is the dominant parameter with an α-factor of approximately 0.9. Note that
this is slightly higher than the standardised ISO values in Table 2. This may partly be
the result of the fact that only vertical permanent load is considered. From the α-factors
resulting from the probabilistic analyses, partial factors can be derived. For the partial
factor γϕ associated with the characteristic 5% value can be determined according (3).
For Eurocode the β-values for consequences classes CC1, CC2 and CC3 are 3.3, 3.8
and 4.3 respectively, resulting in the following partial factors:
γϕ = 1.17, 1.25 and 1.34.
For cohesion, assuming the same the α-factors and a coefficient of variation V=0.2 we
may arrive in a similar way at:
γc = 1.58, 1.98 and 2.67.
We compare these results with the values in the Eurocode and the Dutch national annex
(Table 5).
@seismicisolation
@seismicisolation
Table 5. Partial factors – Shallow foundation

Parameter Eurocode recommendation Dutch National Annex
β - level β - level
3.3 3.8 4.3 3.3 3.8 4.3
Permanent Load γG 1.23 1.35 1.49 1.23 1.35 1.49
Volumetric Weight γγ 1.00 1.10
Friction angle γϕ’ 1.25 1.15
Effective cohesion γc’ 1.25 1.60
Note first that in the Eurocode system a change in consequences class only leads to a
change in the load factors, not in material factors (EN 1990, Annex B). Taking β = 3.8
as the reference case, the proposed differences are 10% upward and downward. For the
friction angle this is a bit too little, for the cohesion it might have been more. This is
the result of the examples chosen.
For the central level β = 3.8 itself we observe a material factor in Eurocode
that is too low (in particular for the cohesion due to a greater variability of cohesion
assumed in the Netherlands) and a load factor that is too high (according to equation
(3) it could have been 1.00 + 0.20 * 3.8 * 0.10 = 1.08). The two deviations seem to
cancel each other in most cases, except possibly for the partial factor 1.25 for cohesion
according to the Eurocode recommendation. This value seems on the low side.
Finally note that the above analysis is far from complete as variable loads are
lacking, conservative normal distributions have been taken and no cases with both
friction and cohesion active have been considered.
Sheet pile walls
Three anchored sheet pile wall structures were analysed (Calle et al, 1991 and Van der
Meer et al, 1991b):
• Case 1 Canal bundwall: the outline of the sheet pile wall analysed is shown in
Figure 4. The parameters, including the mean value and the variation, are
given in Table 6.
• Case 2 Building pit: case 2 concerns a building pit in sand, including a small
clay layer (Figure 5). Here the water table is lowered on the low side of the
sheet pile wall. The parameters are shown in Table 7.
• Case 3 Harbour quay: the cross-section of the Harbour Quay is outlined in
Figure 6. The parameters are given in Table 8.
The probabilistic evaluation focusses on the cohesion c’, the friction angle ϕ’, the
excavation level and the groundwater levels at both sides of the sheet pile wall. In the
Netherlands for clays generally drained parameters c’ and ϕ’ are used instead of the
undrained strength cu.
@seismicisolation
@seismicisolation
Figure 4. Sheet wall Case 1 - Canal Bundwall
Figure 5. Configuration of the building pit

Figure 6. Harbour Quay
@seismicisolation
@seismicisolation
Table 6. Parameter values – Sheet wall Case 1

Parameter Mean value Coefficient of variation V /
Standard deviation σ
Clay-layer – cohesion c’ 3 kPa V = 0.2
Clay-layer – friction angle ϕ’ 22.5 o V = 0.1
Clay-layer - roughness δ/ϕ’ 0.5 V = 0.16
Variable load q 10 kPa V = 0.25
Groundwater level – active side 0.5 m - Groundlevel σ=0m
Water level – canal – passive side 0.5 m - Groundlevel σ = 0.125 m
Excavation depth 2.75 m σ = 0.25 m
Table 7. Parameter values – Sheet wall Case 2

Sand – friction angle ϕ’ 30 o V = 0.1
Sand - roughness δ/ϕ’ 0.67 V = 0.125
Groundwater level – active side 1.5 m - Groundlevel σ=0m
Groundwater level – passive side 5.5 m - Groundlevel σ = 0.05 m
Table 8. Parameter values –Sheet wall Case 3

Sand – friction angle ϕ’ 32.5 o V = 0.1
Sand - roughness δ/ϕ’ 0.67 V = 0.125
Groundwater level – active side 1.5 m - Groundlevel σ = 0.20 m
Water level – passive side 1.5 m - Groundlevel σ = 0.20 m
The analyses were executed using a conventional finite element program, where the
sheetpile is modelled as an elastic beam supported by linear elastic plastic soil springs.
Three mechanisms were considered: (1) Loss of passive resistance, (2) Exceeding of
the bending resistance in the steel wall and (3) Failure of the anchor. Using
probabilistic methods the sheet walls were designed in such a way that for all
individual mechanisms the reliability level was equal to β = 2.75, 3.75 and 4.5, which,
using equations (4) and (6) corresponds to 2.5, 3.4 and 4.2 for the system failure. Note
that the value 2.5 is outside the Eurocode range. From these analyses also followed
values of the sensitivity coefficients α. The α values have been averaged over the three
cases and the three reliability levels (Table 9).
@seismicisolation
@seismicisolation
Table 9. Typical averaged values for the sensitivity coefficient α

Parameter α-value
Cohesion c’ 0.41
Friction angle ϕ’ 0.69
Roughness δ/ϕ’ 0.17
Variable load q - 0.24
Excavation level 0.35
Ground water level active (high) side 0.20
Ground water level passive (low) side 0,30
Based on these results (and some rounding off procedure) the partial factors for ϕ’ and
c’ as well as geometrical offsets Δ= αβσ of Table 10 were derived.
Table 10. Partial factors on c’ and ϕ’ and geometrical offsets

Parameter β-level – total structure
2.5 3.4 4.2
Cohesion c’ 0.9 1.0 1.1
Friction angle ϕ’ 1.05 1.15 1.20
Excavation level Δ = 0.20 m Δ = 0.30 m Δ = 0.35 m
(Ground)water level active (high) side Δ = 0.05 m Δ = 0.05 m Δ = 0.05 m
(Ground)water level passive (low) side Δ = 0.15 m Δ = 0.20 m Δ = 0.25 m
It is interesting to see the low partial factor for the cohesion c’, despite its large
uncertainty. This is probably due to the fact that the cases concern anchored sheet pile
walls, where the cohesion is a less dominant parameter than in unanchored walls.
Compare the α =0.41 to α =0.87 used at the shallow foundation cases. Note also
that the partial factor for c’ would be higher (1.3-1.5) if not applied on the
characteristic value but on the mean.
The resulting partial factors adopted in the Eurocode (default values in EN 1997-
1, Table A4) and in the Dutch National Annex are shown in Table 11. In the Eurocode
typically values have been given for CC2/RC2 structures only (FR stands for
Reliability Class).
Table 11. Partial factors – Eurocode 7 and Dutch National Annex

Parameter Eurocode 7 Dutch National Annex

(default)
β-level
3.3 3.8 4.3
All CC/RC- CC1/RC1 CC2/RC2 CC3/RC3
Classes
Volume weight γ 1.0 1.0 1.0 1.0
Cohesion c’ 1.25 1.15 1.25 1.40
Friction angle ϕ’ 1.25 1.15 1.175 1.20
Excavation level 10 % height, idem idem idem
max Δ = 0.5 m
(Ground)water level active No value Δ = 0.05 m Δ = 0.05 m Δ = 0.05 m
high) side
(Ground)water level No value Δ = 0.20 m Δ = 0.25 m Δ = 0.25 m
passive (low) side
In the Netherlands it was decided to increase the partial factor for the cohesion c’ in
order to avoid a low reliability if the cohesion would be dominant and α
correspondingly higher. Furthermore, the margin on the excavation level was taken
according to the specifications in Eurocode 7, art 9.3.2.2 (2).
@seismicisolation
@seismicisolation
3. Conclusions
The authors have illustrated the merits of Design Approach 3. Also the probabilistic
design and determination of partial factors from probabilistic analyses have been
illustrated by examples of geotechnical structures.
In the Dutch approach of ULS the highest safety is attributed to the most
uncertain parameter with the largest influence on the design. For a sheet pile wall,
therefore, a higher partial factor was determined for the effective cohesion than for the
friction angle.
Furthermore, statistics play a major role in analysing geotechnical structures. The
basis of the statistics has also been given in the Dutch National Annex with Eurocode 7.
The authors have illustrated its use in design.
In the Eurocodes distinction has been made depending on the importance of the
structure in Consequence Classes having a different probability of failure. For
geotechnical structures as piled and raft foundations, the difference in Consequence
Class can be taken into account by variation of the load factor as given in Eurocode
EN-1990. However for slopes and retaining walls, where the soil acts as a load and as a
resistance at the same time and where the external loads are small, the difference in
Consequence Class can only be established by different material factors. The Dutch
National Annex makes this distinction based on probabilistic analyses.
References
Calle, E.O.F. and Spierenburg, S.E.J. (1991), Veiligheid van damwandconstructies – Onderzoeksrapportage
deel 1, for CUR-committee C69, Deltares report CO-316980/12
EN1990 (1994), Eurocode Basis of Design, CEN, TC250, Brussels
Everts, H.J.and H.J. Luger (1997), Dutch national codes for pile design; Proc. Of the ERTC3 Seminar,
Brussels, pp. 243-265.
ISO 2394 (1994), General Principles on Reliability for Structures - Revision of IS 2394, ISO/TC98
JCSS (2001), Probabilistic Model Code, JCSS internet web page: http://www.jcss.byg.dtu.dk/
Karadeneniz, H, Vrouwenvelder A, Rackwitz R, Sudret B, Devictor A. (2003), Overview of reliability
methods, European Projet Saferelnet, Taks 5.1
Koppejan, A.W. and W.C. van Mierlo (1952), Lengte en draagvermogen van heipalen; Bouwmachines
Meer, M.T. van der, Smits, M.Th.J.H. (1991a), Veiligheid van damwand constructies –
Onderzoeksrapportage deel 2, for CUR-committee C69, Fugro report M-0053
Meer, M.T. van der, Smits, M.Th.J.H. (1991b), Studie compatibiliteit Partiële Factoren voor NEN 6740–
serie Geotechniek, Fugro report M-0044
Prandtl, L. (1921), Uber die Eindringungsfestigkeit plastischer Baustoffe und die Festigkeit von Schneiden,
Zeitschrift für Angewandte Mathematik und Mechanik, 1, No.1
Rackwitz, R. (2000), Optimisation – the basis of code making and reliability verification, Structural Safety,
22, pp 27-60.
Terzaghi. K. (1943), Theoretical Soil Mechanics, Wiley
Waarts, P H, Structural Reliability using Finite Element Analysis – An appraisal of DARS, Directional
Adaptive Response Surface Sampling, TU Delft, 2000
@seismicisolation
@seismicisolation
@seismicisolation
@seismicisolation
Code Application
@seismicisolation
@seismicisolation
@seismicisolation
@seismicisolation
IOS Press, 2013
doi:10.3233/978-1-61499-163-2-143
Limit state design of the foundations of

concrete gravity dams - A case study
Laura CALDEIRAa,1, Maria Luísa B. FARINHAa, Emanuel MARANHA das NEVES b,
and José V. LEMOSa
a
LNEC – National Laboratory for Civil Engineering, Lisboa, Portugal
b
IST – Technical University of Lisbon, Portugal
Abstract. The use of the limit state design (LSD) is not common practice among
dam engineers. However, it has been applied in the design of structures since the
middle of the last century, and its application in the geotechnical area has
increased over the last two decades, mainly as a result of both the European pre-
norm and European codes concerning geotechnical design. While it is accepted
that, due to the hazard involved in dam operation, extreme caution must be
exercised in the process of dam conception and safety assessment, there are no
significant reasons why dam safety and serviceability cannot be analyzed using the
LSD approach. This case study presents an example of the application of the LSD
of the foundation of a large concrete gravity dam, in static conditions.
Keywords. LSD, sliding, gravity dam, numerical analysis
Introduction
The application of the limit state design (LSD) to dams has been scarce in recent
decades. Reports by the European Club of the International Commission on Large
Dams (ICOLD) regarding sliding safety of gravity dams and uplift pressures under
dams (European Club of ICOLD, 2004a and 2004b) do not mention the LSD.
Nevertheless, the final report on sliding safety of existing gravity dams (opus cite
2004a) refers, in its Appendix dedicated to Regulatory Rules, Guidelines and Normal
Practice in different countries, to the Chinese Technical Standards related to the design
of hydraulic engineering structures, namely the Design Specification for concrete
gravity dams. According to this standard, hydraulic structures must be designed for the
ultimate limit states (ULS) and the serviceability limit states (SLS), the latter called
normal operation limit states. The estimation of the safety against sliding is identified
as a ULS. Deformations which affect the normal operation or condition of the structure,
local damage involving the condition and durability of the dam as well as
impermeability of the watertight elements, correspond to SLS.
The different partial factors are combined according to given rules for the
following design situations – sustained (corresponding to persistent situations
according to Eurocodes), transient, and occasional (corresponding to accidental
situations) status. In general terms, the ULS assessment is expressed by
1
@seismicisolation
@seismicisolation
144 L. Caldeira et al. / Limit State Design of the Foundations of Concrete Gravity Dams
Ψ , , 1/ / (1)
where is the importance factor of the structure ( = 0.9, 1.0, 1.1), Ψ the factor of
design situation (Ψ =1.0, 0.95, 0.85) and the structure coefficient, equal to 1.2 for
the sliding limit state. The remaining symbols are those used in Eurocodes. For the
assessment of the ULS of sliding, a rigid body limit equilibrium approach is used,
evaluating the total driving force and the overall strength along the surface of the
failure mechanism under examination. The overall shear strength is evaluated by means
of the Mohr-Coulomb criterion based on cohesion and friction angle in terms of
effective stresses. The procedure for evaluating the characteristic values is not included
in the above-mentioned Appendix of the European Club report.
More recently, in France, recommendations have been published regarding the
safety and serviceability verification for both gravity dams (CFBR, 2006) and
embankment dams (CFBR, 2010). The latter enumerates the ULS and SLS to be
verified on embankment dams and their foundations, the design situations to be
considered and the partial factors to be used on actions and material properties, and can
be seen as a proposal of which the main objective is the use of the LSD method.
However, the document concerning gravity dams (CFBR,2006) falls short as an
attempt to move in the direction of using the LSD method as it only mentions limit
states (with no distinction between ULS and SLS) and is not based on a semi-
probabilistic approach.
The present study aims to widen the use of the LSD application for large dams, in
particular concrete gravity dam foundations, although it should be stressed that the
immediate objective is not to question the acceptable safety level of concrete dams. On
the contrary, based on prescribed values of EC7 for the partial factors, a safety
verification of the foundation design of Pedrogão concrete gravity dam (Portugal) will
be presented, in static conditions. This example only envisages the ULS corresponding
to the overall stability concerning sliding.
The potential failure modes regarding sliding which are analyzed involve the dam
foundation rock mass, taking into account the orientation of the sets of discontinuities
within the dam foundation. The role that the grout and drainage curtains plays in the
foundation safety level is discussed and quantified.
Instead of an a priori definition of the mechanisms corresponding to ultimate limit

states of sliding, the critical mechanism was determined through the analysis of the
foundation displacements, considering characteristic values of strength affected by an
increasing partial factor. This was done until a failure surface was determined, which
was assumed as the critical mechanism for the overall sliding. To ensure that such
ultimate limit state will not occur, or, rather, to check its sufficiently low probability of
occurrence, analysis was carried out only for the identified mechanism.
Comments are made regarding the criteria adopted in order to determine the
characteristic values of the material mechanical properties, with an emphasis on
discontinuities. The values used for the partial factors are discussed. Dam safety
assessment of sliding follows the EN 1997 – Eurocode 7: Geotechnical Design, and the
partial factor method (EN 1997-1, 2004).
@seismicisolation
@seismicisolation
L. Caldeira et al. / Limit State Design of the Foundations of Concrete Gravity Dams 145
1. Pedrogão Dam
1.1. General Characteristics
Pedrógão dam (Figure 1) is a gravity dam located on the River Guadiana, in the
southeast of Portugal, with a maximum height of 43 m and a total length of 448 m, of
which 125 m are of conventional concrete and 323 m of roller compacted concrete
(RCC). Figure 1 shows a cross-section of this dam, of the central area of the dam, in
which the base length in the upstream-downstream direction is 44.4 m and the dam
height is 33.8 m. It should be noted that the cross section of this dam is unusual, as in
the majority of gravity dams the base length is around 20% lower than their height.
1.2. Characteristics of the Foundation Rock Mass
The sliding verification of Pedrogão dam is carried out with a 2D hydromechanical

discontinuum model, using the discrete element method (DEM), involving the cross
section of the central area of the dam. The most relevant geotechnical aspects of the
foundation rock mass in this area of the river bed are described below.
After the removal of the alluvial and residual soils, the dam foundation consists of
fine to medium-grained granite. Five sets of sub vertical and three of sub horizontal
discontinuities were identified. One of those sub horizontal joint sets must be pointed
out due to its substantial extension at the valley bottom, near the riverbed and
downstream from the dam.
The main characteristics of the discontinuities at the surface, observed in dry
conditions, are the following: (i) average spacing between 0.20 and 0.60 m and
roughness between 2 and 5 mm; (ii) sub horizontal – persistence between 10 and 20 m,
medium weathered joint walls with unfilled apertures between 0.1 and 5 mm; (iii) sub
vertical – persistence between 3 and 10 m, very weathered joint walls, filled with
plastic material (with a diameter less than 5 mm).
The hydro-geological studies showed that the ground-water table was controlled
by the River Guadiana. The results of the Lugeon tests show a low permeability rock
mass at depth, except in localized zones or along discontinuities.
The dilatometric deformation modulus in the tests performed in the valley bottom
boreholes varied between 3 and 24 MPa. The uniaxial compression laboratory test
results and the corresponding deformation modulus varied with the granite type (fine
and medium grained) and the weathering stage. For the fine grained granite, the
uniaxial compression strength range between 160 MPa (degree of weathering W1 –
ISRM, 1981) and 10 to 90 MPa (W3), and the deformation modulus between 60 GPa
(W1) and 15 to 50 GPa (W3). For the medium grained granite the same properties
varied between 60 MPa (W2) and 10 MPa (W4) and between 25 GPa (W2) and 5 GPa
(W4), respectively.
From samples collected in boreholes PD5, PD7 and PD12A, located in the valley
bottom at given depths (see Figure 2 and Table 1) five compression and shear tests of
the discontinuities’ planes were carried out at different normal stresses at the
Portuguese National Laboratory for Civil Engineering (LNEC, 2004). The results
presented in Table 1 are the normal stiffness at a normal stress of 1 MPa, shear stiffness
corresponding to a shear displacement of 0.2 mm at the same normal stress and
ultimate friction, dilatancy and residual friction angles. Figure 3 presents the shear test
results of the sample collected in borehole PD5 at a depth of 22.70 m.
@seismicisolation
@seismicisolation
Block 6A-7B
(m)
95
90
RWL = 84.8
85
80
75
70 33.8
65
60
55
50
45 44.4
40
RWL – retention water level
a) b)
Figure 1. Pedrógão dam: a) downstream view from the right side of the uncontrolled spillway and b) cross-
section of the central area of the dam.
Figure 2. Geological longitudinal profile at the central zone of dam. Boreholes location.
Table 1. Compression and shear test results of the discontinuities planes

Borehole Sample Normal Shear Ultimate Dilatancy Residual
Identification depth stiffness1 Stiffness2 friction angle friction
(m) (MPa/mm) (MPa/mm) angle (º) (º) angle (º)
20.15 15.2 2.9 41.3 5.5 35.5
PD5
22.70 25.4 1.9 37.5 6.5 32.9
PD7 14.20 39.4 2.4 38.7 4.2 36.2

37.55 23.6 2.5 41.8 6.3 35.3
PD12A
38.20 13.2 3.0 39.8 7.8 31.4
1
Normal stiffness at normal stress of 1 MPa
2
Shear stiffness corresponding to a shear displacement of 0.2 mm and at a normal stress of 1 MPa
In the design phase, the basic rock mass rating quality index (RMR) was adjusted
(RMRadj) for the unfavourable strike and dip orientations of certain discontinuities as
proposed by Bieniawski & Orr (1976) for dam rock foundations. The parameters of
shear strength of the foundation rock mass were obtained based on RMR classification
and three geotechnical zones (GZ) were proposed (see Table 2). In the valley bottom,
after removing the superficial deposits, the foundation rock mass was, in general, of
type GZ1 and/or GZ2, except at a few localized zones where a GZ3 was considered.
In ultimate limit state verifications the most important ground properties are those
related to strength, the deformability characteristics generally being second in
importance. For the present case, a sensitivity analysis showed that the stiffness
variation has a very small influence on water pressure build up. Therefore, values of 10
GPa/m for the normal stiffness and 1 GPa/m for the shear stiffness were assumed.
@seismicisolation
@seismicisolation
1.6 0.40
1.4 0.20
1.2
Shear stress (MPa)
0.00
Normal displacement (mm)

1.0 0 1 2 3 4 5 6 7
-0.20
0.8
0.6 -0.40
0.4 -0.60
0.2
-0.80
0.0
0 1 2 3 4 5 6 -1.00
Shear displacement(mm) Shear displa cement (mm)
0,25 MPa 0,5 MPa 1,0 MPa 2,0 MPa 0,25 MPa 0,5 MPa 1,0 MPa 2,0 MPa
Figure 3. Results of shear tests of discontinuities’ planes carried out at different normal stresses.
Table 2. Geotechnical zones of the design and respective classification and properties.
Geotechnical Degree of RMR89 RMRadj Cohesion Friction E
zones weathering (MPa) angle (º) (GPa)
W4-5 <38 <28 <0.14 21 <1
GZ3
W3-4 30-50 20-46 0.10-0.23 15-29 1-4
GZ2 W3 39-67 29-57 0.14-0.29 22-33 2-15
GZ1 W1-2 >61 >51 >0.26 >32 >12
RMR89 as proposed by Bieniawski (1989); RMRadj as proposed by Bieniawski & Orr (1976);
The ground volume involved in this limit state is very large compared to the spatial
correlation length of the ground property, In addition, no predominantly weaker zone
(besides the discontinuities pattern considered) was identified and therefore the
behavior is governed by the mean value of the ground parameter.
As recommended by Eurocode 7 (EN 1997-1, 2004) for those cases, the
characteristic value of the friction angle of the discontinuities was a cautious estimate
of the mean value. It represents an estimated value corresponding to a 95% confidence
level that the mean value of the population governing the occurrence of the limit state
on the ground is more favourable than the calculated characteristic value derived from
the test results and previous knowledge (literature values and local experimental data).
The mean value and the coefficient of variation (COV) of the individual sample
friction angle are equal to 34.3º and 7.4%, respectively. Following the recommendation
of Schneider (1999), a value of 10% was adopted for COV. This COV value seems to
be appropriated taking into account that it is applied to an ultimate limit state where the
shear strength mobilized is the residual one (critical state). The spatial variability is
indirectly considered due to the large distances between borehole samples. The
characteristic value of the tangent of the friction angle was calculated according to
1 (2)
where is a statistical coefficient accounting for parts of the epistemic uncertainty (in
this case equal to 0.74) taking into account the number of test results, in this case of 5
samples. Thus the characteristic friction angle is 32.3º.
@seismicisolation
@seismicisolation
2. Verification of The ULS of Sliding
The use of the limit state design (LSD) for dam safety assessment is based on the
Eurocode 7 (EN 1997-1, 2004), as mentioned in the introduction. This European
standard describes the general principles and requirements for geotechnical design,
primarily in order to ensure safety (strength and stability), serviceability and durability
of supported structures, i.e. buildings and civil engineering works, founded on soil or
rock. As such, it should be used in conjunction with EN 1990 Eurocode: Basis of
structural design (EN 1990, 2002) and EN 1991 Eurocode 1: Action on structures (EN
1991-1-1, 2002) (Frank et al., 2004). The Eurocode 7 consists of two parts. Part 1,
General Rules (EN 1997-1, 2004), which presents the general rules of geotechnical
design; Part 2, Ground Investigation and Testing (EN 1997-2, 2007) discusses the use
of field investigations and laboratory testing for geotechnical design.
The EN 1997-1 may also serve as a reference document for other aspects of
geotechnical design, such as the design of dams, tunnels and slope stabilization, or the
design of foundations of special construction works, such as nuclear power plants and
offshore structures, which require additional provisions to those provided by the
Eurocodes.
Although the Eurocode 7 is intended to be applied to small dams, this study
focuses on large dams, particularly on concrete gravity dams and the safety analysis of
their foundations. According to the current Portuguese Regulations for Dam Safety, a
dam can only be considered as “small” dam if its height is lower than 15 m, the
reservoir volume is not greater than 105 m3 and can be included in the lower
consequence classes defined within the regulations.
2.1. Formulation
The EN 1997-1 (2004) and EN 1990 (2002) include, as options, the following three
different Design Approaches (DA) for ULS design in persistent and transient situations
(Frank et al., 2004).
DA1 implies, in principle, two calculations involving two sets of partial factors.
Where it is obvious that if one these sets governs the design it will not be necessary to
carry out calculations for the other (Annex B.1 (2), EN 1997-1, 2004). This DA may be
termed an action and material factor approach with factors applied at the source, i.e.,
to actions, rather than to the effects of actions, and to shear strength parameters, rather
than to resistances.
DA2 requires a single calculation to either actions or effects of actions and to
resistances. It may be termed an action effect and resistance factor approach. It must
be noted that in this case, the application of partial factors to the effects of actions, does
not deviate significantly from the conventional overall factor safety (OFS) approach.
DA3 requires a single calculation where partial factors are applied to actions or
effects of actions from the structure and to ground strength parameters and may be
termed an action effect and material factor approach.
A National Annex allows each European country to set, within certain limitations,
the safety levels for civil engineering works through the partial factors values, called
Nationally Determined Parameters (NDP). The National Annex may also specify the
procedure to be used when alternative procedures are given in the Eurocode, namely
the design approaches.
@seismicisolation
@seismicisolation
In this study the Portuguese National Annex will be followed, which requires DA1
(combination 1 and combination 2) to be adopted for the ULS design.
The actions as well as the relevant mechanical properties of the materials will be
represented by their characteristic values. The design values result from the use of
partial factors on the characteristic values. Obviously if the partial factor is 1, the
characteristic and design values are equal.
A particular and important aspect of the actions acting on a gravity concrete dam,
when compared with the structures covered by the Eurocode 7, are the large values of
water pressure on the upstream face of the dam, as well as the significant weight of the
dam body itself.
Three design situations are considered (Figure 4), that is, dam without any curtain
(A), dam with grout and drainage curtains (B), and dam with only the grout curtain (C -
drainage curtain assumed to be clogged). For these three situations the water in the
reservoir is considered at its retention water level (RWL = 84.8 m, 33.8 m above
reservoir bottom).
A B C
WP D
φ’k = 32.3° G.C. φ’k = 32.3° G.C. φ’k = 32.3°
NOTE: G.C., grout curtain; D., drainage curtain; WP, water pressure; φ’k, characteristic friction angle.
Figure 4. Design situations.
According to the Eurocode system, the following partial factors must be used:
(i) Combination 1 - partial factors for permanent actions greater or equal to one
( 1 ) and for the material properties equal to 1 ( 1); (ii) Combination 2 -
partial factors for permanent actions equal to one ( 1 ) and for the material
properties greater than 1 ( 1).
Through the use of a discrete element numerical method and adequate constitutive
equations, the strength reduction procedure ( reduction in this case) is applied in
order to assess the critical failure mechanism for both combinations.
In the present case, for Combination 1, two permanent actions (water pressure on
the upstream face of the dam and dam weight) must be combined with different partial
factors, as one is unfavourable and the other favourable. The recommended value of
is 1.35, for unfavourable permanent actions, and 1.0, for favourable permanent
actions.
In partial factor design based on numerical analysis, due to the difficulty in
distinguishing the effects caused by different actions, a particular factor, , , is applied
to the whole effect of the combination of actions, so the following expression must be
used (Gulvanessian et al., 2002):
,
, !, ; ! #,$ 1 (3)
, ,
@seismicisolation
@seismicisolation
where is the design effect, % & the result of the calculation model, !, the
dominant action for the considered limit state associated to , , !, the remaining
actions and , the corresponding partial factors to !, .
The dominant action is the water pressure on the upstream face, therefore
according to equation (3) its partial factor was not initially applied in the numerical
model and the weight of the dam body was multiplied by the following factor:
, /, 1.0/1.35 0.74 (4)

The design value of ( ) is reduced by applying a

strength reduction factor, -! 1 tan /-! and the resulting
displacement, 1, at the crest of dam (failure indicator), is registered.
The model is re-analysed for several -! values until displacement 1 becomes
considerably large, indication that failure has occurred. The -! value corresponding
to the last stable situation is equal to ! (the well-known overall safety factor).
An auxiliary factor, the over-design factor (2!), is then introduced as
2! !/ , (5)
This over-design factor gives an additional margin of safety beyond that required
by Eurocode for the studied limit state.
For Combination 2, the calculations were performed using the design values of the
material parameters, applying the partial factor, to its characteristic value. In this
case only is required and according to the EN 1997-1, its value is 1.25. In a
procedure similar to the one described for Combination 1, an -! value corresponding
to large displacements is obtained. For this combination, the over-design factor
coincides with the -!:
2! -! (6)

For both Combinations 1 and 2, a value of 2! equal to unity indicates that
sliding safety is exactly the adequate and the available margin of safety is exactly that
required by the EN 1997-1. If 2! 1, the margin of safety is more than adequate
and 2! 3 1 means inadequate safety (though not necessarily meaning a failure).
If the stress-strain relationship is stress path dependent the 2! obtained through
the two methods are not necessarily equal.
2.2. Two Dimensional Hydro mechanical Discontinuum Model
The discontinuous model developed to study the ULS concerning sliding is shown in
Figure 5. The model presented here is based on a previously developed model, which
had been calibrated taking into account the quantity of water collected at the dam’s
drainage curtain (Farinha, 2010). In this model, two sets of discontinuities were
simulated: the first joint set is horizontal and continuous, with a spacing of 5.0 m, and
the second set is formed by vertical cross-joints, with a spacing of 5.0 m normal to joint
tracks and standard deviation from the mean of 2.0 m (for the safety assessment it is
not necessary to consider the actual joint spacing; a coarse mesh with an average zone
@seismicisolation
@seismicisolation
size of 5.0 m × 5.0 m is sufficient). Although there is no evidence from site

investigations, an additional rock mass joint was assumed, dipping 11° towards
upstream (see α angle in Figure 5), in order to allow a failure mechanism of sliding
along foundation discontinuities to develop. This hypothetical situation may simulate a
combined mode of failure, where the failure path occurs both along the dam/foundation
interface and through intact rock, in geology where the rock is horizontally or near
horizontally bedded and the intact rock is weak (USACE, 1994).
The foundation model is 200.0 m wide and 80.0 m deep. The dam has the crest of
the uncontrolled spillway 33.8 m above ground level and the base is 44.4 m long in the
upstream-downstream direction, as shown in Figure 1b).
The analysis was carried out with the program UDEC (Itasca, 2004), which allows
the interaction between the hydraulic and the mechanical behaviour to be studied in a
fully-coupled way. Joint apertures and water pressures are updated at every time step,
as described in Lemos (1999, 2008). It is assumed that rock blocks are impervious and
that flow takes place only through the set of interconnecting discontinuities. Total
stresses are obtained inside the impervious blocks and effective normal stresses at the
mechanical contacts. Flow is simulated by means of the parallel plate model, and the
flow rate per model unit width is thus expressed by the cubic law. The medium is
assumed to be deformable and the flow is dependent on the state of stress within the
foundation.
Both dam concrete and rock mass blocks are assumed to follow linear elastic
behaviour, with the properties shown in Figure 5. Discontinuities are assigned the
Mohr-Coulomb constitutive failure criterion, complemented with a tensile strength
criterion. A joint normal stiffness (kn) of 10 GPa/m, a joint shear stiffness (ks) of
1 GPa/m, and a characteristic friction angle of shear strength (φ’ k) of 32.3° were
assumed at the foundation discontinuities and at the dam/foundation interface. In rock
joints, effective cohesion and tensile strength were assumed to be zero, while at the
dam lift joints both parameters were assigned 2.0 MPa. As the aim of this study is the
ULS foundation analysis, the strength reduction method was only applied to the
foundation rock mass properties. In rock joints dilatancy may be relevant for small
displacements, but it should not be considered in failure analysis.
Concrete:
33.8 m
α = 11o Unit weight = 24 kN/m3
Young´s modulus = 30 GPa
Poisson’s ratio = 0.2
Foundation blocks:
Unit weight = 26.5 kN/m3
Young´s modulus = 10 GPa
80 m
Poisson’s ratio = 0.2
Foundation discontinuities:
kn = 10 GPa/m
ks = 0.1 kn
φ’ k= 32.3°
200 m α varying from 8º to 35º
Figure 5. Discontinuum model of Pedrógão dam foundation and material properties.
@seismicisolation
@seismicisolation
Analysis was carried out in two loading stages. Firstly, the mechanical effect of
gravity loads with the reservoir empty was assessed. In the UDEC model, an in-situ
state of stress with K 0.5 was assumed in the rock mass. The water table was
assumed to be at the same level as the rock mass surface upstream from the dam.
Secondly, the hydrostatic loading corresponding to the full reservoir (retention water
level, RWL) was applied to both the upstream face of the dam and reservoir bottom.
Hydrostatic loading was also applied to the rock mass surface downstream from the
dam. In this second loading stage, mechanical pressure was first applied, followed by
hydro mechanical analysis. In both stages, vertical displacements at the base of the
model and horizontal displacements perpendicular to the lateral model boundaries were
prevented. Regarding the hydraulic boundary conditions, joint contacts along the
bottom and sides of the model were assumed to have zero permeability. On the rock
mass surface, the head was 33.8 m upstream from the dam, and 0.5 m downstream. The
drainage system was simulated by assigning domain water pressures along the drain
axis, equivalent to 1/3 of the sum of the reservoir and tailwater hydraulic heads.
In UDEC, the hydraulic aperture of the discontinuities is given by:
5 ∆ (7)
where is the aperture at nominal zero normal stress and ∆ the joint normal
displacement taken as positive in opening. A maximum aperture, , is assumed,
and a minimum value, , below which mechanical closure does not affect the
contact permeability. In the study presented here it was assumed that could vary
between 0.05 mm and 0.2 mm, and that 0.02 88 . The values of were
defined taking into account the results of numerous tests performed at US dam sites in
the depth range 0-60 m which indicated that most conducting apertures were in the
range of 50-150 μm at this shallow depth (Barton et al., 1985). The maximum aperture
was limited to 20 × . It was assumed that the grout curtain was 10 times less
pervious than the surrounding rock mass.
2.3. Identification of the Foundation Failure Mechanism

The failure mechanism was identified carrying out numerical analysis using DEM and
several different geometries, each one of them with the rock mass joint downstream
from the dam dipping towards upstream at a different angle α (α is shown in Figure 5).
Values of α varying from 8° to 35° were considered. For each model, the strength of
the foundation discontinuities was gradually reduced, in a sequence of analysis. In this
study the rock mass discontinuities’ effective cohesion is zero, and thus the reduction
coefficient was applied to the tangent of the effective friction angle only. The critical
mechanism was that for which instability was achieved for the highest friction angle of
shear strength. It was concluded that the critical mechanism involves an inclined rock
mass joint dipping from 10° to 15° towards upstream. In the study presented here a
rock mass joint dipping 11° towards upstream was considered.
2.4. Results Analysis
Analysis was carried out with the reservoir at the RWL, both with constant joint
hydraulic aperture and taking into account the hydro mechanical (HM) interaction.
@seismicisolation
@seismicisolation
Figure 6 shows a detail of dam and foundation deformation due to the simultaneous
effect of dam weight, hydrostatic loading and flow.
The variation in water pressures along the dam/foundation joint and along the rock
mass joint downstream from the dam dipping 11º towards upstream is shown in Figure
7 (the failure surface is highlighted in black). Results obtained with different a0 are
presented. In the hydraulic analysis in which the HM effect is not taken into account
(curves shown in black), variations in uplift pressures along both the interface and the
foundation discontinuities are the same regardless of , because the joint hydraulic
aperture remains constant. Figure 7 shows that variations in water pressures are highly
dependent on the pressure on the drainage line. Upstream from this line, water
pressures obtained increase as the different increase, and are lower than those
obtained with constant joint aperture. Downstream from the drainage line, in this
particular situation, water pressures obtained when the HM interaction is taken into
account are lower than those obtained with constant joint aperture. This is due to the
unusual cross section of this dam, as previously mentioned in section 1.1. Figure 7 also
shows a comparison of water pressures along the base of the dam with both bi-linear
and linear uplift distribution (curves shown in orange), usually used in stability analysis
with and without drainage systems, respectively. Water pressures along the inclined
rock mass joint downstream from the dam vary almost linearly between the pressure at
the toe of the dam and zero.
Table 3 shows the results of the LSD according to the DA1 of Eurocode 7, in the
design situations A, B and C (see Figure 4). It was assumed that 0.1 88 .
From the analysis of this table, it can be concluded that the design situation A, as
expected, is the most critical and Combination 1 the most severe. The inclusion of this
case in the present study envisages assessing the increase of safety associated to the
current practice in dam field of designing grout and drainage curtains.
Due to the very favourable effect of the dam weight in terms of sliding stability
and the application of 1, for the design situations A and C, the 2! is less than 1
for Combination 1, indicating an unsafe design. In design situation B, the 2! is
greater than 1, with a comfortable margin of safety. It must be noted that the design
situation C pretends to represent a rare or accidental occurrence, where the safety levels
are expected to be lower.
Figure 6. Block deformation (magnified 3000 times) due to dam weight, hydrostatic loading and flow.
@seismicisolation
@seismicisolation
Figure 7. Water pressure along the dam/foundation joint and along the rock mass joint downstream from the
dam: a) dam with both grout and drainage curtains, and b) dam with only the grout curtain.
@seismicisolation
@seismicisolation
Table 3. Results of the LSD according to DA1of the Eurocode 7
Combination 1 Combination 2
Design (dam body weight (dam body weight

situation favourable - ீ 1.0) unfavourable - ீ 1.35)
SRF ODF SRF ODF SRF ODF
A 0.85 0.63 1.70 1.26 1.36 1.36
B 1.55 1.14 2.00 1.48 1.60 1.60
C 0.90 0.67 1.80 1.33 1.44 1.44
A large dam is a special structure, the failure of which may have severe
consequences (Class CC3 according to EN 1990). For this type of structure, a special
reliability level is required (class RC3), for which a penalizing factor must be applied
to actions or resistances. A value of 1.1 is suggested in EN 1990 only for actions.
Adopting this procedure, the for the normal design situation B reduces from 1.14
to 1.04.
However, instead of applying factor 1.1 to ߛி for unfavourable actions, EN 1990
states that it is normally preferable to require high levels of quality control during
construction, maintenance and operation. This is the usual procedure in large concrete
dams and thus the ranges presented in Table 3 may be considered as final values.
Additional analysis was carried out for the most unfavourable case (Combination 1,
dam body weight favourable, design situation A), assuming no hydro-machanical
interaction. A SRF of 0.8 was calculated and thus an ODF of 0.59 is obtained. This
result is coherent with the results presented in Figure 7, in which it can be seen that, as
previously mentioned, in this particular situation, water pressures obtained when the
HM is taken into account are lower than those obtained with constant joint aperture.
3. Conclusions
This paper presents an example of the application of the Eurocode system in the LSD
of a ULS of the foundations of a large concrete gravity dam. The discrete element
method and a hydro-mechanical discontinuum model were used to analyse the sliding
safety. The failure mechanism was identified using the strength reduction method. The
same method was used to verify the ULS for both combinations of DA1.
The numerical analysis raises problems regarding the direct application of load
partial factors due to the difficulty in distinguishing the favourable and unfavourable
effects caused by different actions. To overcome this difficulty, the loads used in the
calculation are unfactored and the partial factors are applied to the effects, as proposed
in Eurocode 0.
Results analysis has shown that in this case, despite the geotechnical character of
the problem, the safety is controlled by Combination 1, considering the gravity load as
favourable. This may be explained by the high hydrostatic pressures on the upstream
face of dam and by the high uplift pressures when compared with the structure´s self-
weight. This result can be considered relevant for further applications of the procedure
for concrete gravity dams.
@seismicisolation
@seismicisolation
The study also shows that when the dam weight is affected by a partial factor equal
to one the safety level decreases significantly.
The three different design situations which were analysed showed that, according
to Eurocode 7, safety is only ensured when both the grout and drainage curtains are
operating properly (design situation B). In fact, the clogging of the drainage curtain (a
rare design situation simulated in situation C) does indeed affect the dam safety,
reducing the ODF from 1.14 to 0.67. The main purpose of the grout curtain is to reduce
the quantity of water that flows through the dam foundation, having a very small effect
in terms of uplift pressures, and thus on the ODF (it only increases from 0.63 to 0.67).
This analysis of Pedrogão dam has also shown evidence that a dam’s SLS was not
reached before the sliding ULS is attained. This conclusion is based on the very small
horizontal displacement calculated for the failure indicator at the crest of the dam,
when the foundation ULS was reached.
References
Barton, N.; Bandis, S. and Bakhtar, K. (1985). Strength, deformation and conductivity coupling of rock joints,
International Journal of Rock Mechanics and Mining Science and Geomechanics Abstracts 22 (3),
121–140.
Bieniawski, Z.T. (1989). Engineering Rock Mass Classification, John Wiley, New York.
Bieniawski, Z.T. and Orr, C.M. (1976). Rapid Site Appraisal for Dam Foundations by Geomechanics
Classification, 12th International Congress on Large Dams. Mexico City. Q46.R32.
CFBR (2006). Recommandations pour la justification de la stabilité des barrages-poids, Group de Travail
«Calcul des Barrages-poids», Comité Français des Barrages et Réservoirs.
CFBR (2010). Recommandations pour la justification de la stabilité des barrages et des digues en remblai,
Group de Travail «Justification des Barrages et des digues en Remblai», Comité Français des Barrages
et Réservoirs.
EN 1990 (2002). Eurocode: Basis of Structural Design, CEN/TC 250.
EN 1991-1-1 (2002). Eurocode 1: Actions on Structures – Part 1-1: General Actions – Densities, Self-weight,
Imposed Loads for Buildings, CEN/TC 250.
EN 1997-1 (2004). Eurocode 7: Geotechnical Design – Part 1: General Rules, CEN/TC 250.
EN 1997-2 (2007). Eurocode 7: Geotechnical Design – Part 2: Ground Investigation and Testing, CEN/TC
250.
European Club of ICOLD (2004a). Working Group on Sliding safety of existing gravity dams – Final Report.
European Club of ICOLD (2004b). Working Group on Uplift pressures under concrete dams – Final Report.
Farinha, M.L.B. (2010). Hydromechanical behavior of concrete dam foundations. In situ tests and numerical
modelling, PhD Thesis, IST, Technical University of Lisbon, Portugal.
Frank, R.; Bauduin, C.; Driscoll, R.; Kavvadas, M.; Krebs Ovesen, N.; Orr, T. and Schuppener, B. (2004).
Designer´s Guide to EN 1997-1 Eurocode 7: Geotechnical Design – General Rules, Thomas Telford
Ltd, London.
Gulvanessian, H.; Calgaro, J.-A. and Holichy, M. (2002) Designer´s Guide to EN 1990. Eurocode: Basis of
Structural Design, Thomas and Telford Ltd, London.
ISRM (1981). Suggested Methods for Rock Characterization, Testing and Monitoring. ISRM Suggested
Methods. Ed. E.T. Brown. Pergarmon Press.
Itasca (2004). UDEC – Universal Distinct Element Code. Version 4.0. Itasca Consulting Group, Minneapolis,
USA.
Lemos, J.V. (1999). Discrete element analysis of dam foundations. In Distinct Element Modelling in
Geomechanics, Balkema, Rotterdam, 89–115.
Lemos, J.V. (2008). Block modeling of rock masses, European Journal of Environmental and Civil
Engineering 12 (7-8), 915–949.
LNEC (2004). Geomechanical characterization tests of the foundation rock mass of the Pedrógão dam.
Report 44/04-NFOS (in Portuguese).
Schneider, H.R. (1999). Determination of Characteristics Soil Properties. Proceedings of 12th European
Conference on Soil Mechanics and Foundation Engineering, Amsterdam, Balkema, Rotherdam, Vol. 1,
pp. 273–281.
USACE (1994). Rock foundations. Engineer Manual 1110-1-2908. Washington, DC.
@seismicisolation
@seismicisolation
IOS Press, 2013
doi:10.3233/978-1-61499-163-2-157
Using Numerical Analysis with

Geotechnical Design Codes
Andrew LEES a,1
a
Frederick University, Cyprus and Geofem Ltd.
Abstract. A number of issues concerning the verification of ULS using numerical

analysis are described by means of three simple benchmark problems. There are
pros and cons associated with the two limit state factoring approaches: material
factoring approach (MFA) and load and resistance factoring approach (LRFA).
Based on these simple examples, for the calculation of factored structural forces,
LRFA provides a more consistent level of conservatism. MFA often also provides
satisfactory levels of conservatism, but when there is no soil yield, the factoring
can have a negligible effect, and where there is a lot of yield, the degree of
conservatism can also be too low. For the verification of geotechnical ULS, MFA
is more straightforward because it involves the factoring of input parameters and
identifies the critical failure mechanism, as well as its margin of safety. LRFA
requires particular geotechnical failure forms to be verified but forcing models into
particular forms, even in simple cases, is difficult to achieve. Alternatively, outputs
of mobilised resistance can be compared with independently calculated limiting
values, but this was found to be unworkable for the passive resistance of an
embedded retaining wall where the factored output of horizontal stress exceeded
the factored passive limit, no matter how deep the wall embedment. Consequently,
when using numerical analysis, it is recommended to employ both approaches
LRFA and MFA, which together are equivalent to DA1 or DA2+DA3 in Eurocode
7, in order to benefit from the advantages of each.
Keywords. Numerical modelling, limit state design/analysis, geotechnical design.

Introduction
Numerical analysis is widely used now in practical geotechnical design, primarily to

assess ground deformations (serviceability limit states – SLS) induced by construction
activities such as excavations and tunnelling. However, its use in assessing the safety
(ultimate limit states – ULS) of designs is increasing and there are several advantages
to be gained from the use of numerical analysis in ULS design. These include the
checking of multiple failure forms which are not pre-determined, as well as more
accurate soil modelling with the various constitutive relationships that are available.
Limit state design is now commonplace in geotechnical design codes and the
verification of SLS using numerical analysis is relatively straightforward. There are
several issues to consider, however, when verifying ULS using numerical analysis.
Some of these are modelling issues arising from the difficulties of simulating the large
plastic strains associated with ground failure, e.g. influence of mesh refinement, and
1
Corresponding Author: Director, Geofem Ltd, P.O. Box 14751, Nicosia 2456, Cyprus; E-mail:
Andrew.Lees@geofem.com.
@seismicisolation
@seismicisolation
158 A. Lees / Using Numerical Analysis with Geotechnical Design Codes
the effects on soil strength of stress ratio, deformation parameters and consolidation, as
described by Potts & Zdravkovic (1999, 2001) and summarised in Bauduin et al (2005).
Other issues relate to the design code and arise from the difficulty of simulating what
should be, as defined by the code, an unrealistic occurrence. A well-executed
numerical analysis should provide a reasonably accurate prediction of what will happen,
not what should not happen. Somehow, partial factors that remove the analysis model
from reality need be introduced in order to verify acceptable and consistent levels of
safety. Partial factoring is relatively straightforward when using traditional design
methods when pre-defined failure modes are often being considered, but less
straightforward when using numerical analysis. Academic study in this field has
increased steadily in Europe with the introduction of Eurocode 7 (EC7), with several
publications covering benchmark and case study numerical analyses performed in
accordance with EC7 (Schweiger, 2010a,b; Schweiger et al, 2010; Simpson and
Hocombe, 2010; Bauduin et al, 2003) and wider discussions of the various issues
(Bauduin et al, 2000, 2005; Simpson and Yazdchi, 2003; Heibaum and Herten, 2009,
2010).
This chapter will focus on the key issues associated with partial factoring in
geotechnical numerical analyses, using simple finite element models to help illustrate
the main points.
1. Partial factoring approaches
There are essentially two main approaches to introducing partial factors in geotechnical
limit state design. In the load and resistance factoring approach (LRFA), applied
structural and geotechnical (e.g. earth pressure on a retaining wall) loads are factored.
The resulting increased structural forces (e.g. wall bending moment) are checked
against resistances factored in accordance with the relevant structural code.
Geotechnical loads are difficult to factor in numerical analysis, so the effect of the loads
(e.g. output of wall bending moment) is often factored instead, whose approach is
referred to here as LRFA*. Where there is a differential between permanent and
variable load factors, such a differential can be applied to the input values of variable
loads only (which are not usually geotechnical), before factoring the overall effect
using the permanent load factor, as described by Frank et al (2004). Calculated
geotechnical resistances (e.g. passive earth pressure supporting a retaining wall, or
bearing resistance), rather than soil strength input parameters, are also factored. In
numerical analysis, this requires actual simulation of specific failure forms in order to
calculate the resistances. For simple, externally-loaded structures (e.g. spread
foundation) this might be straightforward, but for structures where geotechnical loads
dominate (e.g. embedded retaining wall) this is far from straightforward. Perturbing
forces can be applied to induce failure, as described by Smith & Gilbert (2011a,b), but
in complex cases it may not be clear where to apply such forces. A further disadvantage
is that resistance factoring requires prior selection of failure mechanisms, thereby
losing the advantage of numerical analyses’ ability to consider all failure forms and to
identify the most critical. In EC7, LRFA is employed in Design Approach 2 (DA2),
while DA1 Combination 1 (DA1/1) employs only load factoring, primarily to obtain
design values of structural forces for comparison with structural resistances. As will be
demonstrated in the numerical analysis examples later, LRFA is more suited to
obtaining factored structural forces than to verifying safety against geotechnical failure.
@seismicisolation
@seismicisolation
A. Lees / Using Numerical Analysis with Geotechnical Design Codes 159
In the material factoring approach (MFA), input parameters of soil strength are
factored, which is more straightforward in numerical analysis. The factoring of soil
strength has two effects: it reduces geotechnical resistance, thereby performing the
function of geotechnical resistance factoring in LRFA, which, in turn, transfers more
stress to structural elements, causing structural forces to increase (usually). If, having
factored the soil strength parameters, geotechnical failure does not occur, the
geotechnical ULS can be said to have been verified. The resulting increased structural
forces are checked against resistances factored in accordance with the relevant
structural code, as for LRFA. Commonly, MFA includes some degree of load factoring
as well, such as on structural loads only (EC7 DA3) or variable loads only (EC7
DA1/2). As will be demonstrated in the numerical analysis examples later, MFA is
more suited to verifying geotechnical ULS and, while often providing satisfactory
values of factored structural forces, the degree of conservatism can be inconsistent and
it is better not to rely on MFA alone to verify ULS in structural elements.
A key decision in numerical analysis is when to apply the partial factors on
material parameters and loads. There are essentially two choices: duration factoring
and staged factoring, as described in Figure 1. In duration factoring, factored soil
parameters are inputted from the start of the analysis (in situ stage) and factored values
of loads are applied from their first application (unless the effect of loads is being
factored, in which case the outputs would be factored at each stage), and these factored
values are used throughout all the following construction stages of the analysis. This is
the easiest method and can be applied in any software, but it may be unclear whether
the factoring has a pessimistic or optimistic effect in later stages (Simpson & Yazdchi,
2003). Staged factoring involves using unfactored parameters throughout all the
sequential construction stages, while at critical stages where a ULS check is considered
appropriate, parallel construction stages away from the normal sequence are simulated
where applied loads are increased to their factored values (or outputs are factored if
factoring the effect of loads) and soil strengths are reduced to their factored values to
obtain the structural forces to verify structural ULS and to check for geotechnical
failure (MFA), or mobilised geotechnical resistance is compared with ultimate values
(LRFA).
Staged factoring has a higher workload and, for material factoring, this may
require two operations: factoring of soil strength to obtain structural forces, then either
checking for soil failure or, preferably, undertaking a second operation to reduce soil
strength in a stepwise fashion to identify the critical soil failure mode and its factor of
safety. The first operation is relatively straightforward but to perform the second
operation, the software must have a function for stepwise reduction of soil strength.
This is not necessarily straightforward, particularly for advanced constitutive models,
and users should be aware of the method, and its shortcomings, employed by the
software. This is covered in more detail by Potts & Zdravkovic (2013) and Bauduin et
al (2005).
@seismicisolation
@seismicisolation
Duration factoring Staged factoring

Unfactored Factored Unfactored
values values values
Initial state Initial state

accurate?
Critical Factor loads

Construction stage 1 e.g. excavation Construction stage 1 ULS check
Factor strength
Non-critical
Construction stage 2 e.g. strut added Construction stage 2
Critical Factor loads

Construction stage 3 e.g. load applied Construction stage 3 ULS check
Factor strength
LRFA or MFA
Direct output of design
values of structural
Last construction stage Last construction stage forces at each ULS
check. Verify geotech
ULS at each ULS
check.
LRFA* LRFA or MFA LRFA*
Multiply outputs Direct output of Multiply outputs
of structural factored values of of structural
forces by load structural forces. forces by load
effect factors at Verify geotech effect factors at
each stage. ULS at each stage. each stage.
Figure 1. Methods of applying partial factors in numerical analyses.
In the simple examples described below, the difference in outputs between the
duration and staged factoring methods was insignificant, so only the results from staged
factoring are shown. However, in more complex analyses with multiple construction
stages, duration factoring may lead to a greater divergence from reality, and not
necessarily in a conservative direction, than with staged factoring. Some analyses in
built-up areas require quite complex simulations of historical construction activities

merely in order to establish the present-day in situ condition, in which case, if
employing duration factoring, it would certainly be desirable to establish the present-
day condition with unfactored parameters prior to performing subsequent stages with
factored parameters. An added advantage of staged factoring is that both LRFA* and
MFA could be employed in a single analysis, whereas duration factoring would require
two complete analyses. Also, SLS verification may be possible from the same single
analysis, provided that its soil parameters were the same as the unfactored ULS values.
This is not always the case because on some occasions different soil parameters may be
appropriate for the ULS and SLS cases (e.g. peak soil strength for SLS and critical state
strength for ULS).
2. Benchmark Analyses
This section describes three simple benchmark analyses that have been used to help
illustrate the issues associated with the application of LRFA and MFA using finite
@seismicisolation
@seismicisolation
element analysis. All the analyses were undertaken in plane strain and the soil was
modelled with a linear elastic-perfectly plastic (LEPP) Mohr-Coulomb model, using
the Plaxis 2011 software. Note that the following observations from these specific
examples are for illustrative purposes only and cannot necessarily be extended to more
general cases.
2.1. Example 1: Vertically-loaded, flexible strip foundation
This is the most simple of benchmark problems, a strip foundation, 2 m wide, placed
on the ground surface. Using the symmetry of the problem, half of the foundation was
modelled, in plane strain, as shown in Figure 2. The foundation was modelled with
linear elastic beam elements, in full contact with the soil elements, with a flexural
stiffness EI of 200 MNm2. The soil model parameters were: Young’s modulus Eu=100
MPa, Poisson’s ratio =0.495 and undrained shear strength cu varied as described
below (undrained total stress case), and E=100 MPa, =0.2, drained cohesion c=3 kPa,
friction angle varied as shown below and stress ratio K0=1-sin (drained effective
stress case).
200 kN/m (permanent, unfactored)
20 kPa surcharge (permanent, drained analyses only)

rotation beam el.
fixity
1m
right hand boundary

at 20 m distance
bottom boundary
at 20 m depth
axis of symmetry
Figure 2. Example 1 benchmark problem.
Without changing the dimensions of the foundation, cu was varied between 60 kPa
and 180 kPa in a series of total stress undrained analyses, as shown in the output of
bending moment in Figure 3. The unfactored plot shows that at lower cu values, as the
soil began to yield, the bending moment reduced due to more uniform bearing
pressures and so less flexure. Consequently, when cu was factored (by 1.4), this also
had the non-conservative effect of reducing bending moment below the unfactored
value, as shown by the MFA plot. In this analysis, no load factoring was applied (as
would be the case with EC7 DA1/2 where only variable loads are factored). DA3
involves factoring of all structural forces, in which case some conservatism was
restored and bending moments outputs were closer to those of the LRFA plot. The
LRFA (applied load factored by 1.35) and LRFA* (bending moment output factored by
@seismicisolation
@seismicisolation
1.35) plots are quite different, even for this simple example, due to yielding of the soil
under the higher factored applied load in the LRFA case, showing that there can be a
significant effect associated with factoring load effects rather than the loads at source.
However, factoring the characteristic output of bending moment in LRFA* naturally
led to a consistent degree of conservatism over the unfactored values as cu varied.
180 1.6
Maximum bending moment (kNm/m)
1.5
Resistance or material factor

160
unfactored 1.4
LRFA*
140 1.3
LRFA
MFA LRFA
1.2
MFA
120
1.1
100 1.0
60 90 120 150 180 0 10 20 30
Undrained shear strength cu (kPa) Maximum settlement (mm)
Figure 3. Example 1: output of bending moment and geotechnical ULS verification for undrained case.
The right hand graph in Figure 3 compares the verification of geotechnical ULS
for the cu=80 kPa case between LRFA and MFA. Both approaches started with a 1.35
factor on applied load, with -c reduction employed until failure for MFA, and the
applied load stepped up for LRFA. Due to the simplicity of undrained bearing
resistance, both approaches reached the same factor of 1.53 at failure (and very close to
the 1.52 value obtained using the classical (2+)cu solution). As can be seen from the
number of points in the LRFA plot, significantly more computation was required for
load control in this analysis, and more settlement was required to cause failure, due to
the effects of soil deformation under the increasing load.

In a similar fashion to the total stress analyses, a reduction in drained shear
strength led to a reduction in outputs of maximum bending moment, and hence MFA
resulted in bending moment values less than the unfactored case, as shown in the left
hand graph of Figure 4. Again, LRFA resulted in lower bending moments than LRFA*
due to yielding in the soil under the higher factored applied load and LRFA* provided
the most consistent degree of conservatism in bending moment over the unfactored
case as was varied.
The geotechnical ULS was verified using LRFA and MFA, in a similar fashion to
the total stress analyses, and the results for a range of starting values of are shown in
the right hand graph in Figure 4. The results are presented in this way to illustrate the
very different effects of LRFA and MFA on drained bearing failure, with increases in
having an exponential effect on bearing resistance. Consequently, except at very low
values, MFA will always predict a lower safe bearing pressure, even with the
typically higher partial factors on bearing resistance than on soil shear strength.
@seismicisolation
@seismicisolation
160 7
unfactored LRFA*
Resistance or material factor at failure

LRFA MFA
6
140
5
120 4
LRFA
MFA
3
100
2
80 1
26 30 34 38 42 26 30 34 38 42
Friction angle (deg.) Friction angle (deg.)
Figure 4. Example 1: output of bending moment and geotechnical ULS verification for drained case.
In both the drained and undrained analyses, manipulation of the soil and
foundation stiffness was found to have only a minor effect on outputs of structural
forces or geotechnical ULS.
2.2. Example 2: Inclined-loaded, flexible strip foundation
The second example was similar to the first, except that an inclined load was applied to
the foundation, which required that the complete width, rather than half of the
foundation be modelled, as shown in Figure 5. Elastic-plastic interface elements were
placed between the foundation beam elements and the soil, with a Coulomb failure
criterion (=13°, c=0) and tension cut-off. The same drained soil properties as in
Example 1 were used, except =25°, c=0. The aim of this example was to apply the
LRFA and MFA techniques to verify geotechnical ULS in a case where the critical
failure mode (bearing resistance failure or sliding) was not immediately obvious.
340 kN/m vert., 30kN/m horiz. (unfactored, permanent)

20 kPa surcharge (unfactored, permanent)
beam el.
1m 1m
left hand boundary at right hand boundary

20 m distance bottom boundary at 20 m distance
at 20 m depth
Figure 5. Example 2 benchmark problem.
@seismicisolation
@seismicisolation
The unfactored inclined load was applied initially, as shown in Figure 5. For
sliding failure, the vertical component of the load is considered favourable and remains
unfactored, so the horizontal load only was increased to determine the sliding
resistance by LRFA. The unfactored result was 78 kN/m (equivalent to Ntan) which,
according to EC7 DA2 (load factor 1.35, sliding resistance factor 1.1), was adequate
(70.9 kN/m > 40.5 kN/m).
Continuing on from the unfactored inclined load application, the inclined load was
then raised by a load factor of 1.35. From this point, in separate stages, the critical
failure mode was determined using -c reduction (MFA) and the bearing resistance
was determined by increasing the vertical component of load to failure (LRFA). The
results of each (material factor and vertical component of applied load, respectively),
plotted against vertical and horizontal displacement of the foundation at the loading
point are shown in Figure 6.
Figure 6. Example 2: verification of geotechnical ULS by MFA and LRFA

The -c reduction identified the first failure mechanism that would occur if the
soil strength were less than expected, which occurred with a material factor of about
1.13, which is inadequate in accordance with EC7 DA3 (1.25 required). The graph
shows that both vertical and horizontal failure occurred simultaneously and, when
viewing the deformed mesh, it was clear that the mechanism was a combination of
bearing and sliding failure. In design codes requiring ULS checks against particular
failure forms (e.g. bearing resistance failure and sliding), with different partial factors
for each, this would raise an issue because the critical failure mode was neither one nor
the other. To try to address this, the foundation can be artificially loaded to determine
the resistance to particular failure modes. Sliding resistance was determined this way,
as described above, and now the vertical load was increased to try to determine bearing
resistance. However, when executing load control, it is difficult to mobilise complete
failure mechanisms in order to identify their mode but the graph in Figure 6 appears to
show simultaneous initiation of failure in both the vertical and horizontal directions at a
vertical load of about 625 kN/m. Even in this simple example, with the application of a
vertical perturbing force, it was not unequivocal whether a bearing resistance failure
had been mobilised, indeed the analysis may have predicted a combined failure again.
@seismicisolation
@seismicisolation
Clearly, the requirement to check adequate safety against particular failure forms in
LRFA is not well suited to numerical analysis where analyses are free to calculate a
single critical case from all failure forms. For this reason, numerical analysis and
LRFA are not very compatible for geotechnical ULS because it is a difficult and
artificial process to force analysis models into particular failure forms. Even in this
simple example, the failure was of a mixed mode, making the application of resistance
factors rather subjective.
2.3. Example 3: Double-propped embedded retaining wall in stiff clay
The third benchmark example concerns a double-propped embedded retaining wall in

stiff clay, as shown in Figure 7, which was modelled in both drained and undrained
conditions. The embedment depth of the wall was verified by limit equilibrium
methods with a single prop at -4m elevation, so that there was no doubt that the double-
propped wall was stable prior to attempting to establish this by numerical analysis.
Geotechnical loads cannot be factored within numerical analyses, only their effect, so
for LRFA* cases, the variable surcharge was factored by 1.11 (=1.5/1.35) and the
outputs of structural forces and mobilised earth resistance by 1.35, which is equivalent
to EC7 DA2*. In MFA cases, the (structural) variable surcharge was factored by 1.5
and the permanent geotechnical loads did not require factoring, which is equivalent to
EC7 DA3 (whereas DA1/2 would require a factor of 1.3 on the surcharge). The soil
strengths were factored directly in order to determine structural forces, while -c
reductions were invoked to verify geotechnical ULS. The soil and wall model
parameters are shown in Table 1. Pore pressures were assumed hydrostatic in saturated
soil and zero in unsaturated. The construction sequence began with wall and upper prop
installation, followed by application of the surcharge, excavation to -4m level,
installation of lower prop and excavation to -8m level. All the results presented below
are from the final construction stage.
variable surcharge 10 kPa

elevation
0m
props spring stiffness 200 kN/mm/m
-4 m
-6 m
-8 m
left hand boundary at
30 m distance
-16 m
bottom boundary at
30 m depth right hand boundary
at 30 m distance
Figure 7. Example 3 benchmark problem
@seismicisolation
@seismicisolation
Table 1. Example 3 model parameters

Parameter Effective stress value Total stress value
Young’s modulus, E (MPa) 60+2/m 75+2.5/m
Poisson’s ratio, 0.2 (drained), 0.495 (undr.) 0.495
Friction angle, (deg.) 23 (varied) 0
Cohesion, c (kPa) 5 55+5/m
Dilatancy angle, (deg.) 0 0
In situ stress ratio, K0 1.0 -
Weight density, (kN/m3) 18 18
Soil/wall friction/adhesion 0.6, 0.6c 0.6cu
Wall stiffness, EI (MNm2/m) 800 800
500 700
unfactored
450
LRFA* 600
Lower prop load (kN/m)

400
MFA
350 500
300
400
250
200 300
23 27 31 35 39 23 27 31 35 39
Friction angle (deg.) Friction angle (deg.)
Figure 8. Example 3: outputs of structural forces in drained analyses on excavation to -8m level.
Figure 9. Example 3: verification of drained geotechnical ULS by MFA and LRFA*.
To compare the outputs of structural forces between LRFA* and MFA in the
drained analyses, maximum bending moment (positive sign convention for tension on
the excavation side) and lower prop load (compression positive) are plotted against the
@seismicisolation
@seismicisolation
varying value in Figure 8. As ever, LRFA* maintains a consistent level of

conservatism over the unfactored values as was varied because of the way it is
calculated, while the MFA-calculated bending moment is quite consistently larger
(about x1.18) than the unfactored values at higher values, but at the lowest value,
where the soil yields more, there was a marked drop in conservatism to only about 1.08
times the unfactored value, which is not desirable. The MFA-calculated upper prop
load (not shown) showed a similar, lesser loss of conservatism at the lowest value,
but not the lower prop load (shown in Figure 8). Here, the degree of conservatism over
the unfactored values was more consistent for all values, but falling gradually to
about only 1.14 times the unfactored value at the highest value.
Now the verification of geotechnical ULS using LRFA* and MFA will be
compared. When the soil strength was factored by 1.25, no failure was evident. In order
to determine the actual safety margin, a -c reduction was invoked (with surcharge
already factored by 1.5 – equivalent to EC7 DA3) and wall toe deflection recorded, as
shown in Figure 9. As can be seen, toe deflection increased more rapidly at a material
factor of about 1.5, which is more than adequate.
Verification of ULS passive resistance using LRFA* is far less straightforward. To
determine the value of passive resistance by forcing the wall to failure with artificial
perturbing forces would be very difficult because the appropriate position, magnitude,
distribution and direction of such forces is unclear. Frank et al (2004) recommend
ensuring that the factored value of mobilised earth resistance does not exceed the
factored value of limiting earth resistance calculated, presumably, according to
Equation C2 in EC7:
ߪ௣ᇱ ൌ ‫ܭ‬௣ ߪ௩ᇱ ൅ ʹܿ ᇱ ඥ‫ܭ‬௣ (1)
Such a limiting passive pressure was calculated for Example 3, using a Kp value of
3.1, and is plotted, factored by 1.4, in Figure 9. On the same graph, the output of
horizontal effective stress h from the numerical analysis, factored by 1.35 is plotted.
The passive resistance appears to be inadequate by a significant margin, in spite of the
resistance being verified by both limit equilibrium methods (for a single prop) and
MFA numerical analysis. In fact, no matter how big the embedment depth, the passive
resistance cannot be verified. Only if both plots were unfactored would the passive
limiting pressure exceed the output of h. This problem was caused by K0 being raised,
though not excessively (K0=1.0), as appropriate for over-consolidated soils, which
caused passive pressure to be mobilised to the full embedment depth quite readily in
the analysis. This resulted in the unfactored output of earth pressure more or less
matching the limiting passive pressure, leaving no adequate safety margin when the
ULS is verified in this way. Even when a minimum K0 (=1-sin) value was specified,
passive pressure was mobilised to some depth, requiring a significantly greater wall
embedment depth compared with the other ULS verification methods. Clearly, LRFA
has some drawbacks when verifying passive resistance by numerical analysis.
Finally for Example 3, the short-term undrained case was analysed in order to
study the effects of undrained modelling with geotechnical design codes. There are
three methods of undrained analysis, using the now commonplace nomenclature of the
Nicoll Highway collapse inquiry, as summarised in Lees (2012):
• Method C (total stress analysis). Undrained values of stiffness and strength.
@seismicisolation
@seismicisolation
• Method B (effective stress analysis, undrained strength criterion). As Method

A but undrained strength specified directly.
• Method A (effective stress analysis). Drained values of stiffness and strength
model parameters, with undrained bulk modulus and Poisson’s ratio.
With MFA, one issue is whether undrained or drained material factors are
appropriate in each case. All three undrained methods were used with the appropriate
material parameters shown in Table 1, using a factor of 1.4 on cu for Methods C and B,
and 1.25 on tan and c for Method A. The outputs of structural forces obtained using
all three undrained methods for LRFA as well as MFA are shown in Figure 10 for the
final construction stage. For maximum bending moment, similar outputs were obtained
between the three undrained modelling methods, which may be fortuitous because
there was more difference between the predictions of upper prop load, and the LEPP
Mohr-Coulomb model is usually a poor predictor of pore pressure, which is critical for
Method A. LRFA* factoring provided consistent levels of conservatism compared with
unfactored values, but MFA provided virtually no margin of safety because there was
negligible yield in the undrained soil and so the strength factoring had almost no effect.
This further highlights the drawbacks of MFA for determining structural forces,
because too little, as well as too much (see Example 1) soil yield can have inconsistent
effects on structural forces.
Figure 10. Example 3: outputs of structural forces in undrained analyses on excavation to -8m level.
Verifications of geotechnical ULS by -c reduction for the three undrained

modelling methods for the final construction stage and with a 1.5 factor applied to the
surcharge are compared in Figure 11. Since Methods B and C are both based on an
undrained failure criterion, the results for these were identical, with failure appearing to
occur with a material factor of about 4.0. This was significantly higher than the value
of about 2.3 obtained from Method A, suggesting that there was some discrepancy in
the analyses. This was most likely caused by a poor prediction of pore pressure in
Method A by the basic LEPP soil model, combined with the specified zero dilatancy,
which resulted in an under-prediction of effective stress in the over-consolidated clay.
However, the danger of specifying any dilatancy is that effective stress can increase
@seismicisolation
@seismicisolation
indefinitely during yield leading to over-prediction of failure loads. These are not only
design code issues, but wider numerical analysis issues, and because of this it is
difficult to draw conclusions on appropriate material factors to be adopted in undrained
analyses. More study is needed and, in the meantime, great care is needed in the
verification of undrained ULS using numerical analysis.
4
Material factor
2 B and C
A
1
0 20 40 60 80 100
Wall toe horizontal deflection (mm)
Figure 11. Example 3: verification of undrained geotechnical ULS by MFA.
3. Conclusions
Limit state design is commonplace in geotechnical design codes but, while the use of
numerical analysis to verify SLS is relatively straightforward, there are several issues
concerning the verification of ULS, some of which have been highlighted here using
three simple FEA benchmark problems. In particular, there are pros and cons
associated with the two limit state factoring approaches: material factoring approach
(MFA) and load and resistance factoring approach (LRFA).
Based on these simple examples, for the calculation of factored structural forces in
soil-structure interaction analyses, LRFA, or LRFA* for factoring the effect of loads,
provide a more consistent level of conservatism compared with unfactored results as
soil strengths were varied. MFA often also provides satisfactory levels of conservatism,
but when there is no soil yield, the factoring can have a negligible effect, and where
there is a lot of yield, the degree of conservatism can fall as well.
For the verification of geotechnical ULS, MFA is more straightforward because it
involves the factoring of input parameters and identified the critical failure mechanisms
in these examples, as well as their margin of safety. LRFA requires particular
geotechnical failure forms to be verified which is not suited to numerical analysis
where analyses are free to calculate the most critical form. Structures can be forced to
fail in particular forms but even in the simple spread foundation example with an
inclined load, this proved difficult to achieve. Alternatively, outputs of mobilised
resistance can be compared with independently calculated limiting values, but this was
found to be unworkable in the retaining wall example where the factored output
exceeded the factored passive limit, no matter how deep the wall embedment.
@seismicisolation
@seismicisolation
Consequently, when using numerical analysis, it is recommended to employ both

approaches LRFA/LRFA* and MFA. Where possible, both structural and geotechnical
ULS should be verified by both approaches, but where it is not possible to verify
geotechnical ULS using LRFA, only MFA should be used. In Eurocode 7, this
combination of LRFA and MFA is equivalent to Design Approach 1 (or, alternatively,
Design Approaches 2 and 3 combined).
A further issue is the selection of appropriate material factors for effective stress
strength parameters in undrained analyses. However, this issue is heavily influenced by
wider modelling issues, and any guidance needs to consider these too. As with many of
the issues described in this chapter, it is an area for further study. In the meantime,
great care must be exercised in the verification of ULSs in undrained numerical
analyses, particularly with effective stress parameters.
References
Bauduin C., De Vos M. & Simpson B. (2000). Some considerations on the use of finite element methods in
ultimate limit state design, LSD2000: Proc. Int. Workshop on Limit State Design in Geotechnical
Engineering, Melbourne, Australia, 18th Nov.
Bauduin C., De Vos M. & Frank R. (2003). ULS and SLS design of embedded walls according to Eurocode 7,
Proc. XIII ECSMGE, Prague, Czech Republic; Vol. 2: 41-46.
Bauduin, C., Bakker, K.J. & Frank, R. (2005). Use of finite element methods in geotechnical ultimate limit
state design, Proc. XVI ICSMGE, Osaka, 2775-2779.
Frank, R., Bauduin, C., Driscoll, R., Kavvadas, M., Krebs Ovesen, N., Orr, T. & Schuppener, B. (2004).
Designers’ Guide to EN 1997-1 Eurocode 7: Geotechnical Design – General Rules, Thomas Telford,
London.
Heibaum, M. & Herten, M. (2009). Geotechnical verifications using the finite-element method?, Bautechnik,
Special Issue, DOI: 10.1002/batc.200910037, 7-15.
Heibaum, M. & Herten, M. (2010). Verifications in accordance with EC7 / DIN1054 using numerical
methods, 14th Danube-European Conf. on Geotechnical Engineering, 2-4th June, Bratislava, Slovakia.
Lees, A.S. (2012). Obtaining Parameters for Geotechnical Analysis, NAFEMS Guidebook R0105, Glasgow,
UK.
Potts, D.M. & Zdravkovic, L. (1999). Finite Element Analysis in Geotechnical Engineering: Theory,
Thomas Telford, London.
Potts, D.M. & Zdravkovic, L. (2001). Finite Element Analysis in Geotechnical Engineering: Application,
Thomas Telford, London.
Potts, D.M. & Zdravkovic, L. (2013). Accounting for partial material factors in numerical analysis,
Géotechnique. In press.
Schweiger, H.F. (2010a). Design of deep excavations with FEM – influence of constitutive model and
comparison of EC7 design approaches, Proc. of the 2010 Earth Retention Conference, Aug. 1-4, 2010,
Bellevue, WA, USA, 804-817.
Schweiger, H.F. (2010b). Numerical analysis of deep excavations and tunnels in accordance with EC7 design
approaches, Int. Conf. Geotechnical Challenges in Megacities, Moscow, 7-10th June, 206 – 217.
Schweiger H., Marcher T. & Nasekhian A. (2010). Nonlinear FE analysis of tunnel excavation – comparison
of EC7 design approaches, Geomechanics and Tunnelling 3 (1), 61-67.
Simpson, B. & Hocombe, T. (2010). Implications of modern design codes for earth retaining structures, Proc.
of the 2010 Earth Retention Conference, Aug. 1-4, 2010, Bellevue, WA, USA, 786-803.
Simpson, B. & Yazdchi, M. (2003). Use of finite element methods in geotechnical limit state design,
LSD2003: Int. Workshop on Limit State Design in Geotechnical Engineering Practice, Cambridge,
MA, USA, 26th June.
Smith, C. & Gilbert, M. (2011a). Ultimate limit state design to Eurocode 7 using numerical methods, Part 1:
methodology and theory, Ground Engineering, 44(10), October, 25-30.
Smith, C. & Gilbert, M. (2011b). Ultimate limit state design to Eurocode 7 using numerical methods, Part 2:
proposed design procedure and application, Ground Engineering, 44(11), November, 24-29.
@seismicisolation
@seismicisolation
IOS Press, 2013
doi:10.3233/978-1-61499-163-2-171
Influence of Ground Water Level on

Shallow Foundation Design.
Application of EC7 Probabilistic and
Deterministic Methods
Carlos PEREIRAa and Laura CALDEIRA a,1
a
LNEC, Lisbon, Portugal
Abstract. The influence of ground water level on shallow foundation design is

presented for the ultimate limit state of the bearing resistance, according to the
formulation presented in annex D of EN 1997:2004. Probabilistic and
deterministic methods were used and compared. Concerning probabilistic methods,
the advanced first-order second-moment method (AFOSM) was applied and the
results were validated by Monte Carlo simulations. For the deterministic
calculation, the partial factors method recommended by the Eurocode and applied
in most practical cases, was implemented. For the assumptions herein made the
width B determined by the probabilistic method is always smaller than the one
obtained deterministically.
Keywords. Shallow foundation, Hasofer-Lind method, bearing capacity, partial

factor method, probabilistic methods.
Introduction
The traditional method used in geotechnical design is deterministic. In this method, the
characteristic values of the random variables are usually considered and the
uncertainties as well as the degree of safety are indirectly taken into account via partial
factors.
In comparison, probabilistic methods are a more rational means for geotechnical
design, allowing direct consideration of the inherent uncertainty of each random
variable of the problem.
The Eurocode philosophy (EN 1990:2002, EC0) prescribes, for simplicity, the
partial factor method as the principal design method. Nevertheless, the possibility of
applying probabilistic methods is also given.
In this paper, the influence of ground water level on the width design, B, of a
square shallow foundation that results from the application of deterministic and
probabilistic methods is presented and compared for the ultimate limit state of the
bearing resistance. Notwithstanding the existence of some important variables, namely
friction angle, this paper is only focused on the influence of water level variation in the
shallow foundation design.
1
Corresponding Author. E-mail: laurac@lnec.pt.
@seismicisolation
@seismicisolation
172 C. Pereira and L. Caldeira / Influence of Ground Water Level on Shallow Foundation Design
1. Design Situation
As stated in Eurocode 7 (EN 1997:2004, EC7), the geotechnical design by calculation

can involve analytical, semi-empirical and numerical models. The design methodology
herein implemented for calculating the bearing resistance belongs to the analytical
group and follows the formulation presented in Annex D of EC7. In drained conditions
and in a homogenous sandy soil (with low spatial variability) with a near horizontal
surface, the soil bearing resistance can be obtained by
R 1
= q' N q s q iq + γ ' B' N γ sγ iγ (1)
A' 2
where N q and N γ are the soil bearing capacity factors, sq and sγ the foundation
shape factors, iq and iγ the load inclination factors caused by a horizontal load, q' the
design effective overburden pressure at the foundation base, γ ' the design effective
weight density of the soil below the foundation base, B' the effective foundation width,
R the design ultimate vertical load and A' (= B' B ) the design effective foundation area.
The expressions of all the previous variables are available in Annex D of EC7. Figure 1
presents the case of this study.
When the ground water level, wl (designated in this paper by water level), is
below the foundation base and affects the bearing resistance, i.e., z < wl < (z + Dm ) , γ '
is replaced by an average design effective weight density of the soil, γ , given by
Eq. (2), where z is the design depth of foundation base, Dm the distance from the
foundation base to the deepest point of the failure mechanism, given by Eq. (3), and γ h
the design weight density of the soil above the water level.
⎛ wl − z ⎞
γ = γ '+⎜⎜ ⎟⎟(γ h − γ ') (2)
⎝ Dm ⎠
B ⎡⎛ π φ ⎞ ⎤
Dm = exp ⎢⎜ + ⎟ tan (φ )⎥ cos(φ ) (3)
⎛π φ ⎞ ⎣⎝ 4 2 ⎠ ⎦
2cos⎜ + ⎟
⎝ 4 2⎠
Qh
4m
wl
Soil: Gv + Qv z
γ sat ; γ h ; φ
B
Figure 1. Case study.
@seismicisolation
@seismicisolation
C. Pereira and L. Caldeira / Influence of Ground Water Level on Shallow Foundation Design 173
2. Random Variables
Table 1 defines the eight random variables considered in this case study. These values
characterise the uncertainties of the loads, soil properties, depth of foundation base and
water level.
The actions uncertainties were evaluated according to the recommendations of the
Joint Committee on Structural Safety (JCSS, 2001). The JSCC (2001) considers that the
live load, represented herein by the vertical variable load, Qv, comprises three parts: (i)
the overall mean load intensity for a particular user category, (ii) a zero mean normal
distributed variable and (iii) a zero mean random field with a characteristic skewness to
the right. For simplicity, only the first two parts were considered herein, assuming a
constant spatial distribution. Taking into account that the horizontal variable load, Qh,
represents the wind action, a Gumbel distribution was considered for this variable.
There are numerous studies that characterise and quantify the uncertainties of the
physical and mechanical properties of soils. Based on studies of other researchers,
Chalermyanont and Benson (2005) reported that a normal distribution is suitable to
describe the weight density and the angle of shearing resistance of soils. According to
Phoon and Kulhawy (1999), the weight density and the angle of shearing resistance
typically have coefficient of variation values, CV, between 3–10 % and 5–11 %,
respectively.
The water level was assumed as a random variable characterised by a normal
distribution function with a range of mean values between 0 and 10 m and four
different standard deviation values, STDV: (i) 0.67, (ii) 1.0, (iii) 1.5 and (iv) 2.0. These
standard deviation values are intended to study the influence of different levels of
uncertainties on the variation of B.
Table 2 presents the correlation matrix assumed between the random variables,
based on assumptions of the authors, after careful consideration of their physical
meaning. The variables that are not indicated in Table 2 were assumed independent.
3. Partial Factor Method

The design calculation of the width B of the shallow foundation was carried out
following the three design approaches presented in EC7 (DA1, DA2 and DA3),
considering the corresponding partial factors (EC7) and the characteristic values shown
in Table 3. These characteristic values are obtained from the distribution functions
defined in Table 1. A comparison of the three design approaches is presented in this
paper to cover the selections of different European countries.
Table 1. Random variables distributions and parameters.

Random variable Distribution m CV
function
Vertical permanent load (kN) Normal 3000 0.10
Vertical variable load (kN) Normal 1000 0.50
Horizontal variable load (kN) Gumbel 250 0.25
Soil saturated weight density (kN/m3) Normal 20 0.05
Soil dry weight density (kN/m3) Normal 18 0.05
Soil shear resistance angle (º) Normal 32 0.07
Depth of foundation base (m) Uniform between z = 1.5 and 2.5 m
Depth of water level (m) Normal Between wl = 0 and 10 m Variable
@seismicisolation
@seismicisolation
Table 2. Soil properties correlation matrix.

Random variable Soil saturated Soil dry weight Soil shear
weight density density resistance angle
Soil saturated weight density 1.0 0.9 0.5
Soil dry weight density 0.9 1.0 0.5
Soil shear resistance angle 0.5 0.5 1.0
Table 3. Characteristic values obtained from the distribution functions of the random variables (EC7).
Random Mean Characteristic Perc. Random Mean Characteristic Perc.
variables values values (%) variables values values (%)
3493.5 Gv,k,sup 95 19.48 γ h ,k ,sup 95
Gv (kN) 3000 γ h (kN/m3) 18
2506.5 Gv,k,inf 5 16.52 γ h ,k ,inf 5
2026.9 Qv,k,desf 98 35.7 φ k ,sup 95
Qv (kN) 1000 φ (º) 32
0 Qv,k,fav - 28.3 φ k ,inf 5
412 Qh,k,desf 98 2.45 zk,inf 95
Qh (kN) 250 z (m) 2
0 Qh,k,fav - 1.55 zk,sup 5
21.65 γ sat ,k ,sup 95 95/98
γ sat (kN/m3) Between
20 wl (m) Variable
18.36 γ sat ,k ,inf 5 0-10 5
For the calculation of the water level characteristic value, the percentiles 95 % and
98 % were considered, taking into account the possibility of the water level state to be
assumed as permanent or variable action. The water level design value, as a geometric
data, was assumed equal to the characteristic value. In the present case, the load
combination of the maximum horizontal load with the minimum vertical load
determines the design width.
Figure 2 presents the variation of width B with the mean water level for the three
design approaches and different standard deviation values. For clarity, the values
associated with the standard deviation of 1.5 are not represented. Figure 3 presents the
relation between the width B and the water level design value for design approaches
DA1-C2 and DA3. From these Figures, the following can be observed:
• The width B of the foundation is a function of the water design value.
Therefore, different mean water, standard deviations and percentiles values

(Figure 2) corresponding to the same design value give the same result in
terms of B (see Figure 3).
• DA3 produces the most conservative design.
• DA1-C1 is always less conservative than DA1-C2, as usual in geotechnical
design.
• In this case, DA1 and DA2 give almost the same results for the two limits of
width B. However, the variation between the two limits occurs at different
mean water levels.
• The variation of B is represented by four different lines:
o a vertical line (with a constant and maximum value of B – Bmax ) for design
water levels above the ground surface (point A1 is at the ground surface);
o an abrupt variation of B for design water levels between the ground
surface and the lower design foundation depth value (zk,inf – point A2);
this variation is essentially induced by the increase of the design effective
overburden pressure (q’);
@seismicisolation
@seismicisolation
B (m) B (m)
4.0 4.5 5.0 5.5 6.0 6.5 4.0 4.5 5.0 5.5 6.0 6.5
0 0
1 1
Mean Water Level (m)

2 2
3 3
4 4
5 5
sd95=0.67 sd95=0.67
6 6
sd98=0.67 sd98=0.67
7 7
sd95=1.00 sd95=1.00
8 8
sd98=1.00 sd98=1.00
9 9
sd95=2.00 sd95=2.00
(i) 10 (ii) 10
sd98=2.00 sd98=2.00
B (m) B (m)
4.0 4.5 5.0 5.5 6.0 6.5 4.0 4.5 5.0 5.5 6.0 6.5
0 0
1 1
2 2
3 3
4 4
5 5
6 sd95=0.67 6
sd95=0.67
7 sd98=0.67 7 sd98=0.67
8 sd95=1.00 8 sd95=1.00
9 sd98=1.00 9 sd98=1.00
sd95=2.00 sd95=2.00
(iii) 10 (iv) 10
sd98=2.00 sd98=2.00
Figure 2. Variation of width B with the mean water level, by the application of the partial factor method and
design approaches: (i) DA1-C1, (ii) DA1-C2, (iii) DA2 and (iv) DA3.
B (m) B (m)
4.0 4.5 5.0 5.5 6.0 6.5 4.0 4.5 5.0 5.5 6.0 6.5
-5.0 -5.0
Water Level Design Value (m)
Water Level Design Value (m)
-2.5 -2.5
A1 A1
0.0 0.0
A2 A2
2.5 2.5
sd95=0.67 sd95=0.67
5.0 sd98=0.67 5.0 sd98=0.67
sd95=1.00 sd95=1.00
7.5 A3 sd98=1.00 7.5 A3 sd98=1.00
sd95=2.00 sd95=2.00
(i) 10.0 (ii) 10.0
sd98=2.00 sd98=2.00
Figure 3. Variation of the width B with the water level design value, for the design approaches: (i) DA1-C2
and (ii) DA3.
@seismicisolation
@seismicisolation
o a small variation of B for water level design values between the lower
design foundation depth value (zk,inf – point A2) and the deepest point of
the failure mechanism (point A3), produced by the variation of weight of
the soil beneath the foundation; the influence of this term in Eq. (1) is
small due to the large load eccentricity caused by the wind action;
o a vertical line (with a constant and minimum value of B – Bmin) for water
level design values beneath the failure mechanism.
• In terms of design, conservatively two limits for B can be established (the
upper limit considering the water level at the surface and the lower limit
ignoring the water table).
For design approach DA3, the limit values are: B = 6.31 and 5.01 m. Some
European countries, like Portugal, adopt design approach DA1. In this case, the width
B varies between 5.76 and 4.63 m.
In terms of the percentile selected (95 or 98%) for the definition of the water level
characteristic value, B limit values are independent of this quantity, as well as the
assumed standard deviation. In contrast, the transition between those values is greatly
affected by those selected values and reflected in the design values. With a percentile
and standard deviation increase, an abrupt variation of the width B occurs at the deeper
water table levels.
4. Probabilistic Design
According to EC0, the structural safety verification, for a particular reliability level, is
carried out by applying the limit state concept. For each structural system, the relevant
ultimate limit state must not be exceeded during the lifetime of the structure, for any
design situation with no negligible probability of occurrence.
Reliability is the probability that a structure fulfils the specified requirement for
which it has been designed during its design working life. In a simplified way, using a
first order reliability method (level II methods), namely the Hasofer-Lind method, the
structural reliability, Ps , can be normally evaluated using two measures, related by
Ps = 1 − Pf = 1 − Φ(− β ) (4)
where β is the reliability index and Pf is the failure probability. Φ represents the
cumulative distribution of a standard normal variable. For current structures, as defined
in EC0, with an expected lifetime of 50 yr, the EC0 sets a minimum reliability index of
3.8 for the ultimate limit states design, concerning RC2 class associated with CC2
consequences (medium consequence for loss of human life and economic, social or
environmental consequences considerable). In the following, the shallow foundation
was considered a current structure, so a minimum 3.8 reliability index was adopted.
4.1. Hasofer-Lind method
In its original form, the Hasofer-Lind method (an advanced first-order second-moment
method - AFOSM) is applicable to problems with uncorrelated normal random
variables. The corresponding reliability index is defined as the minimum distance from
@seismicisolation
@seismicisolation
the origin of the normalised coordinate system to the performance function, g (x ) ,

herein considered the limit state function, and can be expressed as:
β HL = min β = min ⎛⎜ (x'*)T (x'*) ⎞⎟ (5)

⎝ ⎠
where (x'* ) is the point of the performance function closest to the origin in normalised
coordinates, named the calculation or design point. In this definition, the original
coordinate system X = (x1 , x2 ,..., xn ) is transformed into a normalised coordinate
system X ' = (x'1 , x' 2 ,..., x' n ) according to Eq. (6). Thus, the zeroing of the performance
function is carried out in the normalised coordinate system, g ( X ' ) = 0 .
X i − μ Xi
X 'i = (6)
σ Xi
For nonlinear performance functions, the minimum distance calculation is an

optimisation problem, defined by β minimisation with the constraint condition
g (x ) = g (x' ) = 0 . This calculation procedure was implemented in the Mathcad program.
According to Low and Tang (1997), it is possible to consider the correlation between
random variables in the value of the reliability index using Eq. (7), where ρ −1 is the
inverse matrix of correlation coefficients.
β HL = min β = min ⎛⎜ (x'*)T ρ −1 (x'*) ⎞⎟ (7)

⎝ ⎠
For random variables with non-normal distributions, the Rackwitz and Fiessler
(1976) method can be used to transform the variables distribution into an equivalent
normal distribution. The estimation of the equivalent normal distribution parameters,
μ XNi and σ XNi , is performed by imposing equality of the cumulative distribution
functions, F, and probability density functions, f, at the design point,
X '* = (x'*1 , x'*2 ,..., x'*n ) of the non-normal variables and the equivalent normal variables.
The ultimate limit state verification of bearing resistance took into account four
load cases, presented in Table 4. For problems with two variable actions, the EC0 states
that, for each load combination, one of these actions shall be selected as leading
variable and the other as accompanying action. The latter shall be affected by the factor
ψ 0 to take into account the reduced probability of both variable actions simultaneously
having extreme values. In the probabilistic approach, the EC0 defines the ψ 0 value for
normal distributions according Eq. (8), where V is the CV of the accompanying action
for the reference period, T1 the greatest basic period of combined variable actions and
T the reference period (50 yr). For the basic period for the vertical and horizontal
variable actions, values of 7 yr (typical for imposed loads on building floors) and 1 yr
(associated to climate actions) were adopted, respectively.
Table 4. Load combinations.

Load combinations
I G v + ψ 0 Qv + Q h III Gv + Qh
II G v + Qv + ψ 0 Q h IV G v + Qv
@seismicisolation
@seismicisolation
1 + (0.28β - 0.7 ln (T T1 ))V

ψ0 = (8)
1 + 0. 7 β V
4.2. Probabilistic Design Results
Figure 4 presents the variation of B with the mean water level for the load
combinations indicated in Table 4, four different standard deviation values of the water
level and β HL equal to 3.8. The results obtained by this method were validated through
comparison with Monte Carlo simulations with, approximately, 5% of difference in
β HL for the same width B. These results are not presented here due to space limitations.
In the probabilistic approach, the percentiles associated with the design values are
not predefined as in the partial factor method. This approach gives the design values of
each random variable (in general, corresponding to different percentiles) associated
with a particular reliability level.
As previously mentioned, the design width B is determined by the load
combination III, characterised by: (i) no vertical variable load, (ii) the lower value of
the vertical permanent load and (iii) the upper value of the horizontal variable load. For
a better understanding of the influence of the water level in the variation of B, Figure 5
presents some complementary results for load combination III.
As observed in Section 3, B decreases with the increase of the water level depth.
For each load combination, this variation takes place between two limit values
corresponding to the water level design point at the ground surface and below the
failure mechanism. For combination III, the values of B vary between 4.06 and 4.81 m.
As before, the limit values are not affected by the water level uncertainty (here
included by the consideration of different standard deviation values).
However, the variation of B between the limit values differs significantly in shape
with the depth and the uncertainty. For smaller standard deviation values, the variation
of B occurs virtually entirely when the mean water level is between the ground surface
and just below the foundation base. For higher standard deviation values, the design
value of B can be affected by mean water tables at larger depths.
Figure 5 presents the variation of B and of the mean water level with the water
level design point (Figures 5 (i) and (ii)) and the variation of the percentile associated
with the water level design point and of foundation depth design point with the mean
water level (Figures 5 (iii) and (iv)).
According to Figure 5, the relation between B and the water level design point
presents a distinguishing pattern, dependent on the water level standard deviation. In
this Figure, the following can be verified:
• Whenever the water level has no influence on B, the water level design point
coincides with the mean water level and a percentile of 50% can be found.
• For lower standard deviations the water level influence is almost confined
between the ground surface and the foundation base (Figures 5 (iii)).
• In the zone of water level influence the percentile associated with the water
level design point the latter presents a significant variation, dependent on
standard deviation values (Figures 5 (ii) and 5 (iii)).
• The design percentile value decreases from the surface to a certain depth and
then has a very instantaneous increase, bringing the value back to the
percentile 50 %. This depth varies with the assumed standard deviation.
@seismicisolation
@seismicisolation
B (m) B (m)
3,0 3,4 3,8 4,2 4,6 5,0 3,0 3,4 3,8 4,2 4,6 5,0
0 0
1 1

2 2
3 3
4 4
5 5
6 6
Stdev=0.67 Stdev=0.67
7 7
8 8
9 9
(i) 10 (ii) 10
B (m) B (m)
3,0 3,4 3,8 4,2 4,6 5,0 3,0 3,4 3,8 4,2 4,6 5,0
0 0
1 1
2 2
3 3
4 4
5 5
6 6
Stdev=0.67
7 7
8 8
9 9
(iii) 10 (iv) 10
Stdev=2.00
Figure 4. Evolution of the width B with the water level mean designed by the probabilistic method to
combinations: (i) I, (ii) II, (iii) III and (iv) IV.
• The ultimate limit state of bearing capacity, according to Eq. (1), is the sum of
two terms: the first term due to the effective overburden pressure and the
second term mainly due to the effective width of the foundation (dependent of
the load eccentricity caused by the wind action). The first one decreases with
the increase of foundation depth, z, due to the increase of the
eccentricity/moment caused by the wind action. The other term increases with z
increment. Reflecting these opposite effects the design value of the foundation
base remains near the percentile of 50 %. In some situations it presents higher
percentile values to increase the influence of the horizontal load.
Figure 6 presents the relation between the percentile of the water level design point
and foundation base depth with the mean water level, for load combination II. Load
combinations II and IV have vertical loads (of large intensity) with small or no
eccentricity. The displayed curves present the same trend between them for all standard
deviation values. Compared with the results of Figure 5, as expected due to the
favourable effect of the effective overburden pressure, the design value of the
foundation depth in Figure 6 is smaller than in Figure 5 (around 1.7 m opposed to
around 2.0 m).
@seismicisolation
@seismicisolation
B (m) Mean Water Level (m)

4.0 4.2 4.4 4.6 4.8 5.0 0 2 4 6 8 10
0 0
Water Level Design Point (m)

1 1
2 2
3 3
4 4
5 5
6 6
7 7
8 8
9 9
(i) 10 (ii) 10
Perc. W.L. Design Point (%) z (m)

0 20 40 60 80 100 1.5 1.7 1.9 2.1 2.3 2.5
0 0
1 1
2 2
3 3
4 4
5 5
6 6
7 7
8 8
9 9
(iii) 10 (iv) 10
Figure 5. Complementary results of the load combination III.
5. Comparison between probabilistic and deterministic approaches

Figure 7 (i) presents the values of B, calculated with the different deterministic
approaches and the probabilistic method (identified as HL – Comb 3) as a function of
the mean water level. The width B found for different mean water levels using the
deterministic method (between 6.31 and 5.01 m, for DA3, and between 5.76 and
4.63 m, for DA1-C2, respectively) is always higher than the one obtained using the
probabilistic method (between 4.81 and 4.06 m). These results show that different
design approaches using the partial factors method produce different design values with
different failure probabilities. On other hand, assuming that the uncertainties taken into
account by the deterministic and probabilistic methods are the same, these results could
mean that a partial factors design is more conservative and has a lower probability of
failure than one based on the Hasofer-Lind method.
Considering DA1-C2 with STDV = 95 % as a reference case, Figure 7 (ii) displays
the difference (in percentages) between the results obtained in various calculations and
the reference case. Beyond the water level influence, these differences are almost
+10 % for DA3 and around -15 % for the probabilistic approach. In the remaining zone
these differences are larger for both cases.
@seismicisolation
@seismicisolation
Perc. W.L. Design Point (%) z (m)

0 20 40 60 80 100 1.5 1.7 1.9 2.1 2.3 2.5
0 0
1 1
Stdev=0.67

2 2
Stdev=1.00
3 3
Stdev=1.50
4 4
Stdev=2.00
5 5
6 6
Stdev=0.67
7 7
Stdev=1.00
8 8
Stdev=1.50
9 9
Stdev=2.00
(i) 10 (ii) 10
Figure 6. Complementary results of the load combination II.
B (m) DB/B (%)

3.5 4.0 4.5 5.0 5.5 6.0 6.5 -30 -20 -10 0 10 20 30
0 0
1 1
2 2
3 3
4 4
5 5
6 6
7 HL-Comb3 7
DA1-C2-95
8 8
DA1-C2-98
9 9
DA3-95
(i) 10 (ii) 10
DA3-98
B/Bmax B/Bmax
0.75 0.80 0.85 0.90 0.95 1.00 1.05 0.75 0.80 0.85 0.90 0.95 1.00 1.05
0 -4
1
-2
2
3 0
4 2
5 HL-Comb3-0.67
6 4 HL-Comb3-1.00
7 HL-Comb3 HL-Comb3-1.50
6
DA1-C2-95 HL-Comb3-2.00
8
DA1-C2-98 8 DA3-95
9
DA3-95 DA3-98
(iii) 10 DA3-98 (iv) 10
Figure 7. Comparison of the deterministic and probabilistic methods, in terms of (i) B as function of the
mean water level; (ii) ΔB/B as function of the mean water level; (iii) B/Bmax as function of the mean water
level (STDV=1.00); (iv) B/Bmax as function of the water level water level design point for distinct STDV.
@seismicisolation
@seismicisolation
From Figure 7 (iii) representing the B/Bmax associated with each design approach, it
can be seen that in the partial factor method proposed by EC7, the most significant
variation of B occurs in a restricted range of mean water level, due to the previous
selection of characteristic values.
The difference between the variations of water level design point obtained by both
methods is clear in Figure 7 (iv). It can be verified that the EC7 deterministic method
does not have the ability to represent variations of the random variables percentile
values, restraining in that way the natural adaptation of these random variables to the
specific design problem.
6. Conclusions
This paper presents and compares the influence of the ground water level on the design
of the width B of a square shallow foundation, applying deterministic and probabilistic
methods to the verification of the ultimate limit state of the bearing resistance.
Considering only the uncertainties of the eight random variables, the results show
that the width B determined by the probabilistic method is always smaller than the one
obtained by the partial factor method, in the entire range of the mean water level
analysed.
The Eurocodes partial safety factors have been adjusted accordingly to the
knowledge gained over time, reflecting past experience. In this way, they constitute the
reference case with which other results must be compared.
The comparison allows the conclusion that perhaps there are some important
uncertainties, for instance related to construction activities and soil variability, not
included in probabilistic methods. Therefore, these probabilistic methods should be
applied with caution due to the fact that they can produce unsafe designs. Their
utilisation should be restricted to the study of the evolution of the random variables
design values with the objective of improve the selection of partial factors.
As future developments of the present work, other sources of uncertainties namely,
spatial variability, construction activities and calculation model accuracy, will be
incorporated into the probabilistic methods.
References
Chalermyanont, T. and Benson, C. H. (2005). Reliability-based design for external stability of mechanically
stabilized earth walls. International Journal of Geomechanics, 5(3), 196-205.
EN1990 (2002). Eurocode: Basis of design, Comité Européen de Normalisation.
EN1997 (2004). Eurocode 7: Geotechnical design – Part 1: General rules, Comité Européen de Normalisation.
JCSS (2001). Probabilistic model code. Part II – Load models. JCSS Internet Publication.
Phoon, K.-K. and Kulhawy, F. H. (1999). Characterization of geotechnical variability. Canadian
Geotechnical Journal, 36, 612-624.
Rackwitz, R. and Fiessler, B. (1976). Structural reliability under combined random load sequences.
Computers and Structures, 9(5), 489-494.
@seismicisolation
@seismicisolation
IOS Press, 2013
doi:10.3233/978-1-61499-163-2-183
Reliability Based Design of Drilled Shafts:

LRFD and Performance Based Design
Lance A. ROBERTSa,1 and Anil MISRA b
a
Vice President, RESPEC Consulting & Services, 3824 Jet Drive, Rapid City,
SD 57702, U.S.A., Lance.Roberts@respec.com
b
Professor, Civil, Environmental and Architectural Engineering Department,
The University of Kansas, Lawrence, KS 66045, U.S.A., amisra@ku.edu
Abstract. The Load and Resistance Factor Design (LRFD) methods are being
widely adapted for deep foundations design. The benefits of the LRFD method are
two-fold: (1) that uncertainties related to the design of a deep foundation are
rationally incorporated into the engineering process, and (2) that the foundation
design is harmonized with the superstructure design. The key issue in the wide
application of the LRFD method for foundation design is the specification of
resistance factors that are appropriate for various geotechnical conditions and
design criteria. In recent years, a number of agencies have been calibrating
resistance factors using regional/local load test data and design methods. These
calibrations have to be used carefully since the design methods are not always in
consonance with the load test data interpretation methods. In addition and more
significantly, little effort has been made to address the dichotomy of design at the
strength and service limit states. This paper first describes the current LRFD
practice and resistance factor calibration approach. A performance based design
(PB-LRFD) methodology that permits concurrent calibration and design of deep
foundations at both the strength and service limit states is then presented. In the
PB-LRFD approach, project specific resistance factors are obtained based upon the
available site assessment and load-test data. Depending upon the scope of the
subsurface investigation and field load testing/verification program, the proposed
methodology may be considered in three-levels. The applicability of the PB-
LRFD methodology is illustrated through foundation designs that satisfy strength

and service limit state criteria.
Keywords. Load and Resistance Factor Design, performance-based design, drilled

shafts, AASHTO
Introduction
The axial design of drilled shafts has traditionally followed a strength limit state
approach in which the ultimate capacity, or resistance, of the drilled shaft is determined
using a number of static capacity prediction methods. In order to account for
uncertainties that exist in the design from a number of sources, the Load and Resistance
Factor Design (LRFD) method is being specified. The basis of the LRFD method is
the application of load factors, with values greater than unity, to the loads and the
application of resistance factors, with values less than unity, to the calculated resistance.
1
@seismicisolation
@seismicisolation
184 L.A. Roberts and A. Misra / Reliability Based Design of Drilled Shafts
The load and resistance factors are “calibrated” using field observations and statistical
measures. This makes the LRFD method advantageous over the application of a global
factor of safety as uncertainties related to the design of the drilled shaft are rationally
incorporated into the engineering process.
Although the AASHTO Bridge Design Specifications provide values for the
resistance factors to be used in design of many different types of deep foundations,
including drilled shafts, many states in the US are calibrating their own resistance
factors based on local field data and design methodologies. Within the resistance factor
calibration process, it is important for the designer to understand the wide-range of
assumptions that must be made and how these assumptions could lead to potential
limitations that should be applied when using the calibrated resistance factors in design.
In addition, very few, if any, states have addressed the dichotomy of design at the
strength and service limit states and in fact, most calibrated resistance factors are for
design at the strength limit state only.
In this paper, the current LRFD practice and resistance factor calibration approach
used by most states are presented. A performance based design (PB-LRFD)
methodology that permits concurrent calibration and design of deep foundations at both
the strength and service limit states is then presented. In the PB-LRFD approach,
project specific resistance factors are obtained based upon the available site assessment
and load-test data. Depending upon the scope of the subsurface investigation and field
load testing/verification program, the proposed methodology may be considered in
three-levels. The applicability of the PB-LRFD methodology is illustrated through
foundation designs that satisfy strength and service limit state criteria.
1. LRFD and Resistance Factors
Traditionally, foundation designs were completed using the Allowable Stress Design
(ASD) method with a global factor of safety applied to the calculated capacity of the
deep foundation. The following equality was to be satisfied:
Capacity
Loads t (1)
FS
The global factor of safety, FS, attempts to account for variability in the foundation
capacity and the loads and is typically based on experience. However, the factor of
safety can vary from one design to another and can vary between design engineers. In
LRFD, the design equality is revised to:
¦Ji
i Qi d IR R or ¦Ji
i Qi d ¦ M R , S j RS j M R , B RB
j
(2)
where: Ji = load factors, Qi = loads (dead load, live load, etc.), IR = resistance factor,
and R = total foundation resistance. For deep foundations, the resistances and the
resistance factors for the side and the tip can be considered separately as noted in Eq. 2,
where: IR,Sj = resistance factor applied to the side resistance for layer j, RSj = ultimate
side resistance for layer j, IR,B = resistance factor applied to the tip resistance, and RB =
@seismicisolation
@seismicisolation
L.A. Roberts and A. Misra / Reliability Based Design of Drilled Shafts 185
ultimate tip resistance. The load factors have been extensively investigated for
structural design; however, resistance factors, IR, for foundation design have received
attention only recently.
In current practice, the resistance factors for drilled shafts are typically calibrated
using a combination of reliability theory with adjustments made to coincide with
historical practice and judgment. For example, resistance factors can be calibrated to
fit previous design practices (e.g. ASD) as follows:
J D DL
J L
IR LL (3)
§ DL ·
FS ¨ 1¸
© LL ¹
where: JD = dead load factor, JL = live load factor, DL = dead load, LL = live load, and
FS = global factor of safety. Since the allowable stress design method uses a historic
factor of safety in the resistance factor calibration; the uncertainty in resistance is not
explicitly considered. Table 1 gives resistance factors calibrated in this manner.
Table 1. Resistance factors using the calibration to fit method

Global Resistance Factor, IR
Factor of
Safety DL/LL = 1 DL/LL = 2 DL/LL = 3 DL/LL = 4
1.0 1.50 1.42 1.38 1.35

1.5 1.00 0.94 0.92 0.90
2.0 0.75 0.71 0.69 0.68
2.5 0.60 0.57 0.55 0.54
3.0 0.50 0.47 0.46 0.45
4.0 0.38 0.35 0.34 0.34
Ideally, the resistance factors should be determined using reliability theory

integrated with field load test data and calibrated to a specific target probability of
failure (1 in 1,000 for example) that is appropriate to the design limit states, such as (1)
Strength Limit States – ultimate resistance, (2) Service Limit States – limitation of
deformations, and (3) Extreme Limit States – earthquake, impact, etc. The reliability
based methods currently used for resistance factor calibration can be classified as
follows: (1) the AASHTO approach (Paikowsky 2004), and (2) the performance-based
design approach or PB-LRFD (Roberts and Misra 2010). In both of these methods, the
resistance factor, IR, is computed following the First Order, Second Moment (FOSM)
method using the probability density functions of the load and the resistance as
depicted in Figure 1 (Baecher and Christian 2003):
§ J D E (QD ) · 1 :QD :QL

2 2
OR ¨ J L ¸
© E (QL ) ¹ 1 :R2
IR (4)
§ E (QD ) · ET ln ª 1: R 2 1:QD 2 :QL 2 º
¨ OQD OQL ¸ e ¬ ¼
© E (QL ) ¹
where: JD, JL = dead and live load factors, E(QD)/E(QL) = dead to live load ratio, OQD,
OQL, OR = bias and :QD, :QL, :R =coefficient of variation (COV) for the dead load, live
load and resistance, respectively, and ET = target reliability index.
@seismicisolation
@seismicisolation
JQ=IRR
Load, Q Resistance, R
Frequency
Load/Resistance
Figure 1. Resistance factor calibration.
The COV and bias of the dead and live load, along with the dead load and live load
factors have been extensively investigated by the structural engineering community and
are widely available. The ratio of the expected value of the dead load to the expected
value of the live load does not significantly affect the value of the resistance factor and
can be assigned a magnitude based on the characteristics of the structure, with a value
typically between 2 and 4. The target reliability index, ET, relates the desired
performance level and in deep foundation design, a ET of between 2.0 and 3.5 is
generally utilized (Kulhawy and Phoon 2006) and can be adjusted in the design to
reflect different limit states and/or design criteria. It is observed from Eq. 4 that the
only unknown parameters are the COV of the resistance, :R, and the bias of the
resistance, OR. The distinction between the AASHTO approach and the PB-LRFD
stems from how the COV and bias of the resistance are defined and computed.
2. Calibration of Resistance Factors – AASHTO Approach

In the United States, AASHTO has developed the LRFD specifications for bridge
design. Those specifications include resistance factors for several types of deep
foundation elements based on calibrations conducted during the NCHRP Project 24-17
(Paikowsky 2004). The basic resistance factor calibration procedure employed during
this NCHRP project is described herein for completeness with particular focus on
potential shortcomings of the procedure during each step of the process.
As with most resistance factor calibration methods, static load testing must be
conducted on deep foundation elements installed in the field. The data collected during
the static load test may consist of head settlement versus load only or may contain
strain gauge readings or telltale measurements. Eventually, numerous static load tests
are conducted on similar foundation elements in similar subsurface conditions to
develop a database of test results. This database is analyzed to calibrate a resistance
factor. Essentially, the calibration process is divided into three steps: (1) determination
of the failure load from the load test data, termed as the “actual” resistance, (2)
determination of the theoretical failure load using static capacity calculation techniques,
@seismicisolation
@seismicisolation
termed as the “predicted” resistance, and (3) computation of bias as the ratio of the
“actual” capacity to the “predicted” resistance for each load test.
With respect to the head settlement data, there are several methods to determine
the failure load or “ultimate” resistance of the deep foundation element using that data:
(1) Davisson’s offset limit method, (2) Chin-Kondner’s method, (3) DeBeer’s method,
(4) Brinch-Hansen’s criterion, or (5) assume the failure load occurs when the
movement of the foundation head exceeds a certain value such as 5% or 10% of the
diameter of the foundation element. The exact details of each of these methods will not
be discussed here; rather, the idea is that the reader be aware that many methods do
exist, that each of these methods will result in a different failure load from the data, and
lastly, that this list is certainly not all inclusive!
Now, for each test site where a static load test is conducted, it is possible to
compute a theoretical failure load or “ultimate” resistance for the deep foundation
element through the use of static capacity prediction techniques. Of course, there are
many static capacity prediction techniques reported in general foundation engineering
textbooks. For example, some of the common methods include: (1) Reese and O’Neill,
(2) Horvath and Kenney, (3) E method, (4) D method, and (5) O method (Das 2007). It
should be noted that although many methods do exist, most methods are recommended
for certain foundation types (i.e., driven piles, drilled shafts, etc.) and within certain
soils (i.e., clay, sand, etc.).
If one were to develop a matrix of all of the possible methods to compute the
“actual” failure load versus all of the possible methods to compute the “predicted”
failure load, it would be observed that the number of combinations would be quite high.
In addition, not all “actual” failure load methods are compatible with all “predicted”
failure load methods (i.e., variable underlying assumptions within each method).
Therefore, to proceed with the resistance factor calibration procedure, it is necessary to
reduce the combinations by selecting one method to interpret the load test data and
calibrate a resistance factor for a single static capacity prediction method (i.e.,
Davisson’s method combined with the Horvath and Kenney method, for example).
50000
FHWA - Drilled Shaft CFEM - Drilled Shaft
side side
tip tip
40000 total total shaft 3
Actual (kN)
30000
shaft 1
4 shaft 2 4
20000
frequency
frequency
3 3
shaft 4
shaft 4 2 2
1
10000 1
0 0
0.0 0.6 1.2 1.8 2.4 3.0 0.0 0.6 1.2 1.8 2.4 3.0
total resistance ratio total resistance ratio

0
0 10000 20000 30000 40000 50000 0 10000 20000 30000 40000 50000
Predicted (kN)
Figure 2. Plot of “actual” resistance versus “predicted” resistance for drilled shaft.
As stated previously, once all load tests are analyzed, the ratio of the “actual”
failure load to the “predicted” failure load for each test is computed. This results in a
set of ratios. The mean, standard deviation and COV for the set of ratios is computed.
@seismicisolation
@seismicisolation
A value of the mean that is greater than unity indicates that on average, the actual
failure load of the deep foundation systems is greater than what is predicted. The
magnitude of the COV provides an indication of how variable the static capacity
prediction method can be with respect to predicting the “actual” failure load. Figure 2
provides a plot of “actual” versus “predicted” resistance assuming that the failure load
from the static load test corresponds to a shaft head movement equal to 5% of the shaft
diameter, while the predicted failure load was computed using a E-method approach.
Figure 2 indicates that on average, the E-method tends to under predict the failure load
of a drilled shaft and is relatively consistent with respect to the magnitude of the under
prediction (i.e., the variability within the E-method is low). Although the bias
computed in this manner is treated as a random variable, it is noteworthy that the
statistical parameters of the predicted bias (i.e., the mean and standard deviation) are
not only a function of the data variability but also the choice of load-test data
interpretation method and the prediction method. In other words, a part of the bias is
truly random and contributed from the inherent variability of the subsurface, the
construction methods, and the load-displacement response measurements. However, a
part of the bias also arises from the mismatch between the highly nonlinear load-
displacement response and the relatively simplistic load-test data interpretation and
static capacity prediction methods. This second part is not a true random contribution
to the bias, although it often appears as one and in the calibration it is treated as such.
For further understanding of the process, Figure 3 presents the ratio data in the
form of a histogram. The mean of the histogram is assumed to be equal to ORand the
coefficient of variation of the histogram is assumed to be equal to :R for substitution
into Equation 4. The current AASHTO specifications provide calibrated resistance
factors for design at the Strength Limit State only. As of the date of this manuscript, no
Service Limit State resistance factors have been calibrated and the AASHTO
specifications report the resistance factors as unity for the Service Limit State.
Figure 3. Histogram of ratios and pertinent statistics to compute a resistance factor.
There are potentially several problems with the approach used to calibrate
resistance factors for the Strength Limit State in this manner. First and foremost, the
process results in the calibration of a single resistance factor that must be applied in all
designs. It is certainly known that uncertainty varies from site to site and thus project
to project and so it is questionable as to how a single resistance factor would be
@seismicisolation
@seismicisolation
appropriate to cover all design scenarios. Secondly, as stated previously, there are
many different methodologies to compute the failure load for a deep foundation
element from load test data, along with several static capacity prediction methods. The
calibration of resistance factors results from a single combination of these methods and
it is possible that the static calculation method and load test interpretation method may
not be compatible (i.e., similar assumptions with respect to how each method was
developed).
To address some of these issues, numerous state agencies have started the process
(or have progressed quite significantly) of calibrating their own resistance factors using
local load test data and in-house design methods. In addition, some state agencies
have performed calibrations using different uncertainty levels, thereby resulting in a
series of resistance factors that could theoretically vary from project to project.
However, very little effort has been made to address the dichotomy of design at the
Strength and Service Limit States. To that end, the PB-LRFD approach presented in
the next section will permit resistance factor calibration and subsequent design of deep
foundations at both the Strength and Service Limit States.
3. Calibration of Resistance Factors – Performance Based Design
The PB-LRFD method considers the strength and service limit states in terms of the
limiting tolerable settlement and the serviceability settlement (Roberts and Misra 2010,
Roberts et al. 2011). The limiting tolerable settlement is defined as the settlement at
which the stresses within a structure become greater than allowable or where the
settlement causes the structure to become inoperable. The strength limit state design
then corresponds to the limiting permissible settlement for the deep foundation under
the factored load. In addition, the serviceability settlement is defined as the settlement
that could result in aesthetic or functional impediments or less-than-optimal
performance. The service limit state design corresponds to the permissible settlement
for the deep foundation under the working load. The design of deep foundations using
this approach can be accomplished by utilizing a soil-structure interaction model, such
as the t-z model, which allows for the development of a load-settlement curve for the
foundation under axial loads.
3.1. Load-settlement determination using t-z model
In the t-z model, the soil-pile interface resistance along the length and at the tip of
the pile is represented by a series of non-linear spring-slider elements as shown in
Figure 4. This mechanical model can be solved using numerical techniques that can be
easily implemented using calculations tools, such as spreadsheets, typically available in
design offices. The authors have extensively applied the t-z method to model the load-
settlement and load transfer relationships for piles subjected to axial load (Misra and
Chen 2004; Misra and Roberts 2006). The t-z method is particularly useful for
developing a comprehensive design methodology that incorporates the service and
strength limit states as well as reliability analysis of deep foundations (Misra et al.
2007; Misra and Roberts 2009; Roberts and Misra 2009; Roberts and Misra 2010,
Roberts et al. 2008, 2011, Park et al. 2012).
The force-displacement behavior of the soil-structure interface and tip spring-slider
elements can be assumed to be appropriate nonlinear functions such as the hyperbolic
@seismicisolation
@seismicisolation
function shown
s in Figuure 4. The reelevant param meters for hyp perbolic functiion for the
interface include
i the innitial stiffnesss, Kinit, and thhe ultimate strrength, qo, wh hich is the
product of the deep foundation periimeter and thee ultimate sheear strength of o the soil-
structure interface,
i u, inn drained or undrained
u connditions. Simiilarly for the tip
t soil, the
relevant parameters
p incclude the initiaal tip soil stifffness, Kti, and
d the ultimate strength of
the tip soiil, qt, in draineed or undraineed conditions (Misra and Roberts R 2006).. The soil-
structure interface strength and stifffness parameeters are relatted to the typpe of deep
foundationn, constructioon techniquess, and propeerties of the geomaterial along the
interface. The tip soil sstrength and sttiffness param meters are geneerally only rellated to the
type of deeep foundation and propertties of the geomaterial at the t tip and ten nd to have
less depen ndency on coonstruction method m but a strong depenndency on co onstruction
quality (i..e. presence of o loose mateerial at the boottom of a driilled hole). The T model
parameterrs can be estim mated using eiither: (1) the subsurface
s expploration and laboratory
test data, or
o (2) back-caalculations from field load ttest data.
Figure 4. Schematic deppiction of drilled shaft-soil interacttion and non-lineear spring-slider behavior.
b
3.2. Modeel parameter variability

v andd design-levells
In most
m deep fooundation app plications, thhe subsurfacee conditions are non-
homogeneeous and the construction techniques

t forr deep foundaations are varriable from
one installlation to the next. Conseequently, the ultimate
u stren
ngth and initiaal stiffness
model parrameters can vary v along thee length of thhe deep foundaation elementt as well as
with the foundation
f loccation at the site due to “w within-site varriability” (Zhaang 2008).
Thus, the strength and stiffness
s parammeters can bee assumed to be b random varriables that
follow a probability
p disstribution funcction (PDF) and are characcterized by a mean m and a
COV. Noominal or meean values of the t-z modeel parameters can be definned using a
parameterr database andd/or site speciffic load test data. In addition, the variabbility of the
t-z modell parameters can be defineed using subsurface investigation, in-siitu testing,
laboratoryy test data, andd site specific load test dataa.
Devellopment of a local or regio onal model parrameter databbase for speciffic types of
deep founndation system ms can lead too better defineed mean and COV C for speccific design
applications. However, it is notew worthy that thhe extent and d quality of innformation
available to obtain thee model paraameters can hhave wide varriation from project-to-
project deepending on factors such as, project sccale, time, fuunding, and im mportance.
These facctors control the scope off the subsurfaace site invesstigation and laboratory
testing proogram, foundaation design methods,
m and field
f load testiing/verificatioon program.
@seismicisolation
@seismicisolation
In addition, for larger projects, it is often desirable to optimize the efficiency of the
design (and thus the economics) while ensuring an adequate level of inherent safety,
while for smaller projects, the economics of the design may be less of a concern.
Therefore, it is prudent to view the design process in terms of design levels that reflect
the project goals and scope factors yet yield a site specific design and resistance factor
that satisfy the LRFD strength and serviceability limit states. To this end, we have
proposed the following three design-levels in PB-LRFD that can be suitably modified
to incorporate project variances (Roberts and Misra 2010):
1. Level “A” Design – In this design-level, the nominal values and the variability of
model parameters are obtained from back-calculation of several field load tests
conducted at the construction site along with the subsurface investigation data and
any existing regional or local database. Since the resultant parameters are site
specific, this design-level is expected to yield the most efficient design, albeit at
the cost of relatively extensive field load-test program.
2. Level “B” Design - In this design-level, the nominal values and the variability of
model parameters are based on regional or local database of model parameters,
supplemented by limited number of field load tests (one or two) and subsurface
investigation data.
3. Level “C” Design - In this design-level, the nominal values and the variability of
model parameters are based on subsurface investigation data and any available
regional or local database of model parameters. In this case, the default magnitude
of variability for different assumed levels – “high”, “medium”, and “low”
described by AASHTO may be used as given in Table 2.
Table 2. Model parameter COV for site/construction variability based upon AASHTO C10.5.5.2.4
Site + Construction Range of COV
Variability for model parameters
Low < 25%
Medium 25% to 40%
High > 40%
3.3. Resistance factor and exceedance probability for strength and service limit design
For all the three design-levels, the COV of the resistance, :R, can be obtained
using Monte Carlo or Latin Hypercube sampling techniques. To this end, the strength
and stiffness parameters are assumed to follow a PDF, such as the lognormal or normal,
whose value of the mean and COV are specified as described for the three design-levels.
A large number of deep foundation load-settlement curves are then computed using the
t-z method and randomly sampled model parameters based upon a given (or assumed)
probability distribution function. Figure 5 gives a selection of the computed load-
settlement curves (Roberts and Misra 2010).
The load-settlement curves obtained in this manner provide an extensive sampling
of the possible deep foundation behavior at the site. Consequently, these curves can be
analyzed to find: (1) the distribution of resistances at a specified limiting tolerable
settlement (i.e., strength limit state), and (2) the distribution of settlement at a specified
working load (i.e., service limit state). The analysis procedure is illustrated in Figure 5
for a limiting tolerable settlement of 40 mm and a working load of 1000 kN. The
resultant distributions of resistance and settlement are given in Figure 6.
@seismicisolation
@seismicisolation
Figure 5. Schematic depiction of drilled shaft-soil interaction and non-linear spring-slider behavior.
Figure 6. Distributions of (a) resistance and (b) settlements at tolerable limiting settlement and working
loads, respectively.
As shown in Figure 6(a), the COV of the resistance, :R, is easily obtained from the
distributions of the resistance. Thus, from Eq. 4, the resistance factors can be readily
obtained assuming that the bias of resistance OR | 1 given that the t-z method replicates
closely the load-settlement behavior. Similarly, the probability of exceeding a service
@seismicisolation
@seismicisolation
limit state or serviceability settlement at a given working load can be obtained as

shown in Figure 6(b). Thus, the developed approach explicitly permits the
probabilistic integration of design uncertainties for both limit states, thereby resulting
in a safe and efficient design that is site specific. A detailed example application of the
described methodology for design of drilled shaft is given in Roberts et al. (2011).
Furthermore, the procedure for obtaining resistance distributions can be repeated by
changing the design parameters, such as the shaft geometries (length and diameter), the
t-z model parameter nominal values and magnitudes of uncertainty to assess the design
the reliability. Figure 7 gives example reliability indices, ET, obtained by varying the
drilled shaft length, diameter (36”, 42” and 48”), and the COV of the t-z model
parameters (10%, 20% and 30%).
12
Increasing diameter (36” – 48”)–
11 Decreasing model parameter COV (30%-10%)
10
Index ET
9
ReliabilityIndex
7
Reliability
1
30 35 40 45
Total
TotalShaft Length
Shaft Length (ft)(ft)
Figure 7. Reliability index variation with shaft geometry and model parameter COV.
4. Summary and Conclusions

In conclusion, the LRFD approach is a powerful method that can rationally incorporate
the design uncertainties known to exist in deep foundation engineering. The LRFD
approach requires the systematic calibration of load and resistance factors for
application in the design of a deep foundation. Calibration of these factors utilizing
reliability theory allows for explicit integration of load and resistance uncertainties that
are otherwise “lumped” into a global factor of safety using the historic ASD procedure.
In this paper, the approach used for the calibration of resistance factors within the
AASHTO bridge design specifications was described. This approach compares the
interpreted “actual” resistance from a field load test to the “predicted” resistance
computed in design to arrive at the bias. Many states are calibrating their own
resistance factors using this type of approach based on local load test data and design
methods. It is noteworthy that the resultant bias obtained in this manner is not only due
to uncertainty, but a part of this bias also arises from the incompatibility between the
static capacity “prediction” method and the load test interpretation method. This aspect
of the bias is not a true random contribution, although it is treated as such in the
resistance factor calibration process.
@seismicisolation
@seismicisolation
A calibration approach that utilizes performance-based design criteria (e.g.

settlement criteria) was also presented. This approach allows for concurrent design of a
deep foundation at the Strength and Service Limit States. In addition, the approach
results in a project specific reliability-based design and thus a project specific
resistance factor. The advantage of the approach is that it allows for explicit
incorporation of the numerous design uncertainties, permits engineering judgment, and
can be conducted within different design levels in response to project demands and/or
economic conditions. Since the approach is project specific, the calibration of
resistance factors is only one component of the approach and refinement and
improvement of the deep foundation design process is also possible by revising the
foundation geometry and reanalyzing the design using the reliability-based techniques.
Lastly, it is important to note that the integration of strength and serviceability design is
critical as the service limit state tends to attract less attention in the design process;
however, even minor serviceability issues (i.e., vertical post-construction settlements)
can create maintenance, durability, functionality, or aesthetic problems with a bridge
structure and therefore can require significant funding on the part of the owner agency
to repair or maintain.
References
AASHTO. (2009) LRFD Bridge Design Specifications. 4th Edition with Interim Revisions. American
Association of State Highway and Transportation Officials, Washington, D.C.
Baecher, G.B. and Christian, J.T. (2003) Reliability and Statistics in Geotechnical Engineering. John Wiley
& Sons, West Sussex, UK.
Kulhawy, F.H. and Phoon, K.K. (2006). Some critical issues in Geo-RBD calibrations for foundations.
Geotechnical Engineering in the Information Technology Age, Proc. GeoCongress, ASCE, Reston.
Misra, A., and Chen, C.-H. (2004). Analytical Solutions for Micropile Design Under Tension and
Compression. Geotechnical and Geological Engineering, 22(2), 199-225.
Misra, A., Roberts, L., and Levorson, S. (2007). Reliability analysis of drilled shaft behavior using finite
difference method and Monte Carlo simulation. Geotechnical and Geological Engineering, 25(1), 65-
77.
Misra, A., and Roberts, L. A. (2006). Probabilistic analysis of drilled shaft service limit state using the "t-z"
method. Canadian Geotechnical Journal, 43(12), 1324(9).
Misra, A., and Roberts, L. A. (2009). Service Limit State Resistance Factors for Drilled Shafts. Geotechnique,
59(1), 53-61.
Paikowsky, S. G., Birgisson, B., McVay, M., Nguyen, T., Kuo, C., Baecher, G., Ayyub, B., Stenersen, K.,
O'Malley, K., Chernauskas, L., and O'Neil, M. (2004). "NCHRP Report 507: Load and Resistance
Factor Design (LRFD) for Deep Foundations." Transportation Research Board, Washington, D.C.
Park, S., Roberts, L. A., and Misra, A. (2012). Design Methodology for Axially Loaded Auger Cast-In-Place
(ACIP) and Drilled Displacement (DD) Piles. Journal of Geotechnical and Geoenvironmental
Engineering, (in print).
Roberts, L. A., and Misra, A. (2009). Reliability-Based Design of Deep Foundations Based upon Differential
Settlement Criterion Canadian Geotechnical Journal, 46(2), 168-176.
Roberts, L. A., and Misra, A. (2010). Performance-based design of deep foundation systems within the
LRFD framework. Transportation Research Record, 2186, 29–37.
Roberts, L. A., Misra, A., and Leverson, S. (2008). Practical Method for Load and Resistance Factor Design
(LRFD) of Deep Foundations at the Strength and Service Limit States. International Journal of
Geotechnical Engineering, 2(4), 355-368.
Roberts, L.A., Fick, D. and Misra, A. (2011) “Performance-Based Design of Drilled Shaft Bridge
Foundations,” Journal of Bridge Engineering, 16(6), 749-758.
Zhang, L.M. (2008). Reliability Verification Using Pile Load Tests. Reliability-Based Design in
Geotechnical Engineering: Computations and Applications, K.-K. Phoon, ed., Taylor & Francis, New
York.
@seismicisolation
@seismicisolation
IOS Press, 2013
doi:10.3233/978-1-61499-163-2-195
Application of Computational Limit

Analysis in Ultimate Limit State Design
Colin SMITH
Department of Civil and Structural Engineering, University of Sheffield,
Sheffield, S1 3JD
Abstract. Computational Limit Analysis is a numerical tool that allows

the user to rapidly and directly determine the ultimate limit state for
a general problem geometry without the need to iterate or timestep to
a solution. This paper summarises recent work placing Computational
Limit Analysis within a rigorous theoretical framework that can be used
in the context of modern limit state design codes such as Eurocode 7.
It then illustrates how a Computational Limit Analysis approach may
be carried out for a range of example foundation and retaining wall
problems within the Eurocode 7 context.
A particular focus of the paper is to examine the key issue of ac-
tion/resistance factoring as used in Eurocode 7 Design Approach 1 Com-
bination 1 and Design Approach 2.
Keywords. Eurocode 7, limit analysis; ultimate limit state, plasticity;
cohesion; friction; upper bound; discontinuity layout optimisation.
1. Introduction
Computational Limit Analysis (CLA) is a numerical tool that allows the user to
rapidly and directly determine the ultimate limit state (ULS) for any problem
geometry without the need to iterate or timestep to a solution. As such it is an
attractive tool for carrying out ULS design.
Since conventional ULS design is typically based on limit analysis (LA) or
limit equilibrium (LE) methods, it should in principal be possible to apply CLA
seamlessly in ULS design. However conventional LA and LE analysis often contain
implicit assumptions, or ‘loosening’ of the principles of mechanics, something that
is not usually permitted in numerical methods. Thus the primary aim of this
paper is to illustrate how ULS design can be carried out using rigorous CLA/LA
methods. Specific objectives are:
i. review the key issues involved in undertaking a ULS design assessment
using numerical methods in such a way that such a methodology will work
consistently across all ULS design approaches e.g. DA1, DA2 and DA3 in
Eurocode 7.
ii. illustrate through several examples how a CLA analysis can be employed
to achieve a ULS assessment.
@seismicisolation
@seismicisolation
196 C. Smith / Application of Computational Limit Analysis in Ultimate Limit State Design
iii. examine the issues around factoring actions or action effects (the so called
starred (*) approach in Eurocode 7).
In this paper the Discontinuity Layout Optimization technique as imple-
mented in the LimitState:GEO software program (LimitState 2009) is utilised to
illustrate the issues associated with CLA, however the approach is applicable to
most generic numerical software packages. In addition discussion of ultimate limit
state design will be undertaken in the context of Eurocode 7 and its terminology.
It is assumed that the reader is familiar with the Eurocode Design Approaches.
2. ULS Design
2.1. General
Ultimate Limit State Design may be viewed as a process whereby a design is

assessed in terms of how ‘close’ it is to an ULS condition with the aim of rendering
the probability of it actually reaching an ULS extremely small.
Such an assessment can be achieved either through the assessment of the
system at a state of collapse (the ULS) or at a pre-collapse state (Smith 2012).
The focus of this paper is the direct use of the ULS to undertake ULS design.
This use of the ULS is typically implicit in may conventional design calcula-
tions, e.g. retaining wall design is usually undertaken assuming active and pas-
sive Rankine earth pressure distributions which in turn imply soil failure (i.e. an
ULS) either side of the wall. Likewise conventional bearing capacity assessment
commonly uses Terzaghi’s bearing capacity formula to determine the soil resis-
tance at the collapse state. Such calculations are based on the theorems of limit
analysis, however they are limited to a narrow range of problem geometries.
Modern limit state design codes typically aim to achieve safety by ensuring
that a factored action or action effect that is driving a system to failure is always
less than or equal to the available factored resistance at the same failure condition.
Factoring of actions/action effects and resistances may be achieved through the

direct application of partial factors to these parameters or indirectly through the
factoring of material properties. While conceptually straightforward, it is at this
stage of detail that significant difficulties can arise in determining when, what
and how to factor. There is a wide range of viewpoints on this issue that can lead
to confusion. Smith & Gilbert (2011a, 2011b) and Smith (2012) have discussed
these issues in detail and proposed a theoretically consistent approach suitable
for both hand calculations and numerical methods which will be summarised in
the following section.
2.2. Analysis levels
Smith & Gilbert (2011b) proposed the following different stages of a design cal-
culation at which actions/action effects and resistances may be determined prior
to application of the partial factors. These were described as ‘analysis levels’, and
will typically occur at 3 possible stages:
@seismicisolation
@seismicisolation
C. Smith / Application of Computational Limit Analysis in Ultimate Limit State Design 197
1. At source i.e. factors applied to the originating action such as soil self
weight or external load. This precedes the ‘ULS mechanics’ calculation.
2. At a soil/structure interface i.e. factors applied to an action, action effect
or resistance such as an earth pressure. This follows the ‘ULS mechanics’
calculation.
3. Within the structure i.e. factors applied to an action effect/resistance pair
such as a bending moment/plastic moment of resistance. This follows the
‘ULS mechanics’ calculation.
The ‘analysis level’ 2 and 3 assessments can also be referred to as the ‘starred’
(*) approach in Eurocode 7 (e.g. DA2*) in which factors are applied to action
effects.
The designer must specify at which level(s) they are choosing to assess the
system. It is noted that the current Eurocode BSI (2004) has left it up to the
designer to decide which of these are relevant to the design in question, but this
potentially adds a significant additional calculation burden that might be reduced
by additional guidance from the code. It can be argued (Smith & Gilbert 2011b)
that all levels should be checked to allow for any non-linearity in the system.
In this paper, actions will be assumed to be quantities known in advance of
a calculation (e.g. self weight, drained water pressures), while action effects and
resistances can only be derived after a mechanics calculation since, in general,
they will depend on the collapse mode. This is slightly at variance with Eurocode
7 where e.g. earth pressures are considered as actions.
2.3. Action and material factoring
Any ULS design method that requires the direct factoring of an action and/or
material properties can be straightforwardly achieved by first applying factors
to the actions (which are by definition known in advance) and then assessing
whether the resulting design state remains stable or is outside the (factored) ULS
envelope for the system.
2.4. Action, action effect and resistance factoring
Any ULS design method that requires the direct factoring of an action effect
and/or a resistance requires those quantities to be first determined by a ‘mechan-
ics’ calculation (e.g. equations of equilibrium and yield) and then to be factored.
Most designs should be quite stable under characteristic loadings. However
as stated earlier, designs are to be assessed at their collapse state (ULS). Two
issues then arise: (i) which collapse states should be investigated (there will be
many) and, (ii) how can these states be induced?
Conceptually (ii) can be achieved by pushing (perturbing) a structure in a
certain direction until it fails. If it requires a ‘large’ push then it is safe, a ‘small’
push and it is unsafe (in that direction). The size of push can be quantified by
comparing the magnitude of other forces acting on the structure at failure; those
assisting the push (actions or action effects) and those opposing the push (resis-
tances). This process is precisely what is done implicitly in many conventional
design calculations. Technically the collapse states to be investigated should be all
@seismicisolation
@seismicisolation
possible modes. However conventional calculations typically only check: (i) a few
failure modes (e.g. sliding, overturning, bearing for a gravity wall), (ii) a single
specific mechanism for each failure mode that in general will be an upper bound
to the actual optimal collapse mode.
This application of this external perturbing force in an ‘equilibrium direction’
e.g. horizontal for sliding of a gravity wall was described by Smith & Gilbert
(2011a), and has also been presented in a theoretical framework using a stress
probe approach coupled with the use of a yield surface (Smith 2012). This allows
both action/resistance factoring and material factoring to be described within a
single consistent and rigorous framework.
The actions, action effects and resistances are those forces excluding the
probe, resolved in the direction of the force process. The factors on actions will
always be unfavourable in the direction of the stress probe.
2.5. Avoiding double factoring and/or double analysis
It is theoretically invalid to factor action effects and resistances as part of a ‘me-

chanics’ calculation since the application of partial factors implicitly breaks New-
ton’s 3rd law. However some suggested calculation procedures implicitly involve
double factoring and double analysis e.g. gravity wall design against sliding fail-
ure. In this case, the first ‘mechanics’ calculation is often implicit through using
Rankine earth pressures. This is followed by factoring of the earth pressures and
wall weight (an action) and then a second ‘mechanics’ calculation is carried out
to determine the basal shear resistance which may itself be subject to factoring.
Splitting the calculation this way may be attractive in its simplicity but is the-
oretically invalid and impossible to achieve in a general way. The perturbation
approach outlined above avoids these problems.
2.6. Single source principle in action/action effect/resistance factoring
In the above methodology, action effects can only be determined and factored after
the ‘mechanics’ calculation. However in contrast actions could be factored before
or after the ‘mechanics’ calculation without violating any fundamental principles

of mechanics. Smith & Gilbert (2011a) argue that for consistency, actions should
also be factored after the ‘mechanics’ calculation. However for some problems (e.g.
a sliding check on a foundation subjected to inclined loads), this can conflict with
the single source principle (BSI 2004) in that it may effectively lead to different
factoring of orthogonal components of an action.
Thus it can also be argued that actions (as opposed to action effects) should
be factored before the ‘mechanics’ calculation. This choice is left to the engineer.
The Eurocode does not make the single source principle mandatory.
3. Computational Limit Analysis
3.1. General
CLA typically uses optimization to directly derive the ULS state rather than it-
erating to the ULS as can be done using conventional elasto-plastic FE. It thus
@seismicisolation
@seismicisolation
offers a straightforward, robust, rapid method for determining the ULS for any
problem geometry. Methods of computational limit analysis include Finite El-
ement Limit Analysis (Lysmer 1970, Sloan 1988, Makrodimopoulos & Martin
2006) and Discontinuity Layout Optimization (Smith & Gilbert 2007). The latter
method, for example, automates the classical Coulomb wedge type analysis, but
instead of using just one or two wedges, it can effectively model any number of
wedges and handle any problem geometry in a simple robust manner. For exam-
ple it can generate the classic Prandtl failure mechanism for a footing on clay as
shown in Fig. 1a with corresponding collapse load 5.14Bcu , but can also be used
for more challenging problems e.g. the layered footing problem in Fig. 1b, with
corresponding accuracy (Smith 2011).
(a) Example of CLA slip line mecha- (b) Example of CLA slipline mechanism for footing
nism solution to classic Prandtl problem on sand layer over clay.
(footing of width B on clay of undrained
strength cu .)
Figure 1. Examples of CLA analysis
3.2. Basic principles
CLA utilises either the lower bound or upper bound theorems to determine bounds
on the collapse load for a system that displays perfect plasticity. By discretising
either stress or velocity fields in combination with optimization, CLA can be used
to find the following bounds on specified load or loads in the system:
(i) Lower bound: maximise a load (or loads) in the system while not violat-
ing yield and satisfying stress boundary conditions, (ii) Upper bound: find the
smallest magnitude of a load (or loads) in the system that will generate a collapse
mechanism and satisfy kinematic boundary conditions.
These bounds bracket the true plastic solution. Modern CLA methods will
typically deliver one or both bounds close to the true solution. If these loads have
a nominal specified magnitude then the multiple on this specified magnitude that
generates the upper or lower bound collapse load may be termed the ‘adequacy
factor’.
In general any stress states determined by a CLA analysis are uniquely defined
within a yielding zone. However in a non-yielding zone, any non-yielding stress
state that satisfies equilibrium is valid. Thus it is preferable to use stresses and
forces from yielding zones only for assessment. This will normally happen in most
cases.
@seismicisolation
@seismicisolation
3.3. Determination of ULS (material factoring)
In order to determine if a given state is inside or outside the ULS, the adequacy
factor should be applied to any unfavourable load or combination of loads. If
the factor exceeds 1.0 then the system is inside the ULS (an unfavourable load
had to be increased to cause collapse). If less than 1.0 then the system is unsafe.
Note that adequacy factor could be applied to all loads in the system since in
combination they will be inherently unfavourable by definition.
If a lower bound approach is being used, then the above process is conservative
for determining if the state is outside the ULS. If upper bound then the process is
non-conservative and it will be necessary to use a suitable technique to estimate
how close to the true solution the upper bound solution is.
3.4. Application of perturbing forces/stress probes (action/resistance factoring)
Perturbing forces (or stress probes) are straightforward to apply in CLA. An

additional unit load is applied to the system in the direction of the perturbing
force. This and only this load has an adequacy factor applied to it.
3.5. Model factors
Any analysis method is an approximation of reality, and thus will have an inherent
level of accuracy associated with it. Limit analysis or related limit equilibrium
methods have been used in conventional design calculations for many years and
do not typically attract model factors. The methods may either be assumed (i)
to be accurate or, more likely (ii) to have the model factor implicitly included in
the recommended partial factors.

Such conventional methods are usually based on simple kinematic collapse
mechanisms for simple problems, and are often therefore unconservative ‘upper
bounds’ in limit analysis. Upper bound (kinematic) CLA methods will typically
generate more critical collapse mechanisms than are assumed in conventional
calculations and are thus more conservative. It may therefore be argued that
no model factor is required. (In this paper, all CLA analyses will be taken as
effectively giving the true plastic collapse load. )
Alternatively, any user of CLA might wish to determine the ‘true solution’
at least to a given degree of accuracy (e.g. 1% or 5%). This may be achieved
either through analysing both upper and lower bounds at progressively higher
numerical mesh or nodal resolution until they are within 1% of each other or by
using extrapolation methods. Such results would require model factors of 1.01 or
1.05.
A lower bound method, by definition would not attract a model factor but
may be overconservative.
@seismicisolation
@seismicisolation
4. Worked Examples
4.1. Inclined loaded footing on layered soil
4.1.1. Problem Description

Consider a 1m wide by 0.2m thick strip footing founded on the surface of a 0.5m
layer of granular soil, φ = 40o , unit weight γ = 20kN/m3 , underlain by a clay soil
with undrained shear strength cu =28kPa. The characteristic loading applied to
the footing surface is made up of a vertical permanent load of 75kN/m (including
footing weight), and a variable 20 kN/m load inclined at 45 degrees to the vertical.
The soil/footing interface friction is taken as 0.5 tan φ .
The footing is to be assessed at ‘Level 1’ using DA1 and at ‘Level 2’ using DA1
and DA2 against ‘sliding’, ‘bearing’ and failure in direction of applied variable
loading.
4.1.2. DA1 Analysis Level 1 - factor at source

For any method with no resistance factoring e.g. DA1/1, DA1/2, DA3 and which
have no action effects, the actions can be factored prior to an analysis to obtain
the design state and then a check made that the design state lies inside the yield
surface. This will be done simply by undertaking the CLA analysis with adequacy
factor applied to the factored permanent and variable loads, and checking that it
exceeds 1.0.
Fig. 2 shows the mechanism found from the above analysis undertaken using
the DA1/1 factor set (assuming all actions are unfavourable). It was found nec-
essary to increase the applied loads (factored vertical and inclined actions) by an
additional factor of 1.12 (i.e. adequacy factor = 1.12) to cause an ULS with the
depicted failure mechanism. Since this is greater than 1.0 it means the system
complies with the DA1/1 criterion. It is also necessary to check other combina-
tions of favourable and unfavourable partial factors. The results are given in Table
1 and show that the case with both actions unfavourable is most critical as would
be expected from inspection.
DA1/1 Factor sets

Vertical load 1.35 1.35 1 1
Inclined load 1.5 0 1.5 0
Adequacy factor 1.12 1.74 1.3 2.36
DA1/2
Vertical load 1 1 1 1
Inclined load 1.3 0 1.3 0
Adequacy factor 0.89 1.51 0.89 1.51
Table 1. DA1 stability assessment for footing undergoing combined loading at ‘Level 1’.
For DA1/2 (factor of 1.3 applied to variable unfavourable actions, 1.25 to

tan φ and 1.4 to cu ) it was found necessary to reduce the applied loads by a factor
of 0.89 for the system to be on the point of a ULS. Since this is less than 1.0 it
means the system does not comply with the DA1/2 criterion.
Therefore, overall the design does not comply with the DA1 criteria.
@seismicisolation
@seismicisolation
113.0
33.5
23.7
136.7
Figure 2. Simple strip footing showing optimal pattern of slip-lines at collapse under combined
loading with DA1/1 unfavourable factors applied at ‘analysis level 1’ (at source, before numerical
analysis). A additional factor of 1.12 on all actions was required to cause an ULS. The resulting
values (in kN/m) of the actions and resistances are indicated.
4.1.3. DA1 and DA2 Analysis Level 2 - factor at soil/structure interfaces
For any method that factors resistances e.g. DA2, it is necessary to consider
specific failure modes in turn. In this example perturbation directions in the
vertical, horizontal and 45 degrees will be considered. It is also necessary to specify
the structure on which actions and resistances will be determined. In this case
it is simply the footing itself. The actions are clearly the external load on the
footing and the weight of the footing itself. The resistance is the resistance of the
soil and is experienced by the footing at the soil/footing interface. There are no
action effects in this problem. The resistances can be determined as an output of
the CLA analysis.
To assess failure in the three specified directions, then stress probes will be
applied in these three directions: horizontal, 45 degrees and vertical. In this ex-
ample the actions will be factored before the ‘mechanics’ calculation. Their status
as favourable or unfavourable will depend on the stress probe direction. These

were checked separately and in all cases were both unfavourable.
The results are presented in Table 2 and the failure mechanism and applied
forces for the horizontal perturbation case given in Fig. 3 as an example. In all
cases the loadings were found to be unfavourable.
Since DA1/1 passed at ‘analysis level 1’, as expected it passes at ‘analysis level
2’ with full pre-factoring for all directions as it is essentially the same analysis,
since it is the same actions that are factored (though ‘analysis level 1’ is simpler
in that it automatically checks all directions). The same holds for DA1/2 though
specific ‘analysis level 2’ calculations are not shown here.
DA2, however fails in the vertical and inclined directions. The 45 degree
analysis in DA2 raises an interesting issue of factoring a resistance which is neither
pure sliding (which would attract a factor of 1.1) or pure bearing (which would
attract a factor of 1.4). In this case the normal and shear components have been
separately factored as appropriate and then their components resolved in the 45
degree direction and added to give the design resistance.
@seismicisolation
@seismicisolation
.
Stress probe direction Horizontal 45 degrees Vertical

Characteristic actions
Permanent 0 53.03 75
Variable 14.14 20 14.14
Design actions (pre-factored)
Permanent factor 1.35 1.35 1.35
Permanent action 0 71.59 101.25
Variable factor 1.5 1.5 1.5
Variable action 21.21 30 21.21
Total design actions 21.21 101.59 122.46
Results from ULS analysis
ULS Perturbation 7.35 7.94 19.96
Resistance 28.56 109.53 142.42
DA1/1 assessment
Actions 21.21 101.59 122.46
Resistance 28.56 109.53 142.42
Pass/Fail PASS PASS PASS
DA2 assessment
Actions 21.21 101.59 122.46
Resistance 25.97 88.91 101.73
Pass/Fail PASS FAIL FAIL
Table 2. DA1/1 and DA2 stability assessment in various directions for footing undergoing com-
bined loading at ‘analysis level 2’. All units are in kN/m. Values given are those of the stated
parameters resolved in the direction of the stress probe.
101.25
30
7.35
28.6
122.5
Figure 3. Simple strip footing showing pattern of slip-lines at collapse under horizontal pertur-
bation (or horizontal stress probe) with DA1/1 unfavourable factors pre-applied. A horizontal
perturbation of 7.35kN/m (indicated by curve headed arrow) was required to cause an ULS,
and the resulting values (in kN/m) of the actions and resistances are indicated.
4.2. Stem wall
4.2.1. Problem Description

Consider the concrete stem wall depicted in Fig. 4, base width and thickness
4.5m x 0.8m, vertical wall height and thickness 6m x 0.6m with 1m of the base
extending in front of the vertical wall section. The wall retains and is founded
@seismicisolation
@seismicisolation
upon a granular soil, φk = 32o , unit weight γ = 15kN/m3 . The retained soil slopes
back at an angle of 14o . The soil/wall interface friction is taken as 0.5 tan φ on
the upper parts of the wall but as tan φ on the base. The vertical wall section
bending strength is taken as 216 kNm/m
The wall is to be assessed at ‘analysis level 1’ using DA1, and at ‘analysis
level 2’ using DA2 against ‘sliding’ and ‘bearing’ failure (corresponding to the
perturbation directions: vertical and horizontal). In these assessments the wall is
modelled as fully rigid. Finally the wall is also to be assessed at ‘analysis level 3’
against bending failure at the base of the stem using DA1.
4.2.2. DA1 Analysis Level 1 - factor at source

In this model, the retained soil self weight will be assumed to be providing the
unfavourable source action and this will be pre-factored prior to the analysis to
obtain the design state and then a check made that the design state lies inside
the yield surface. This will be done simply by undertaking the CLA analysis with
adequacy factor applied to the unfavourable source action i.e. the retained soil
self weight.
Fig. 4 shows the above analysis undertaken using the standard DA1/1 factor
set. The adequacy factor obtained was 1.95. Since this is greater than 1.0 it means
the design state is inside the yield surface and the system complies with the
DA1/1 criterion. For DA1/2 the adequacy factor obtained was 0.84. Since this is
less than 1.0 it means the design state is inside the yield surface and the system
does not comply with the DA1/2 criterion.
Figure 4. Stem wall analysis showing pattern of sliplines at collapse under combined loading
and DA1/1 factors applied at ‘analysis level 1’ (at source, before numerical analysis).
4.2.3. DA2 Analysis Level 2 - factor at soil/structure interfaces

In this approach it is necessary to specify the structure on which actions, ac-
tion effects and resistances will be determined. For CLA analysis, the appropriate
structure is taken as the stem wall. This differs from common conventional anal-
yses, where the structure is taken as the wall and the soil immediately overlying
the base (block abcd in Fig. 5) and a ‘virtual’ back cd is employed to delineate this
new ‘structure’. However CLA analyses of this type of wall typically indicate that
yield will take place in block acd while block abc remains a rigid, non-yielding
@seismicisolation
@seismicisolation
zone. In this zone a CLA analysis will identify one of many possible equilibrium
stress states that does not violate yield, and thus will generate solutions that
distribute shear and normal stresses along ab and bc in a non-unique way. This in
turn can lead to differing interpretation of these stresses as either action effects or
resistances depending on the direction in which they act. To avoid this all forces
acting along boundary abc will be combined into resultant vertical and horizontal
forces (VB and HB ).
For horizontal perturbation (applied through the wall centroid), there are no
actions acting in the direction of interest. The action effects are HB and NC The
resistance is the shear force TA .
d
a
VB
HB
b c TC
WW NC
TA
NA
Figure 5. Free body diagram for stem wall and associated wedge of soil.
For vertical perturbation (also applied through the wall centroid), the action
is the wall weight WW , the action effects are VB and TC and the resistance is the
normal force on the underside of the base NA .
Since this problem has an action, it might be considered that this could
be factored before or after the ‘mechanics’ calculation. However for the vertical
perturbation case, it makes no difference. For the horizontal perturbation case,
the wall weight could be factored before the ‘mechanics’ calculation. However for
consistency with action effects all actions will be factored after the ‘mechanics’
calculation. In practice it makes no difference as the wall weight would normally
be factored by 1.0 as it is favourable.
The analysis results are presented in Table 3 and failure mechanism for the
vertical perturbation case given in Fig. 6, together with the forces acting on the
wall at failure. The bearing failure mechanism is clearly significantly influenced
by the lateral soil pressure on the wall.
4.2.4. Analysis Level 3 - internal structural stability

If the bending strength at the base of the wall section is required, then this
requires investigation of the action/resistance pair at this point, i.e. bending
@seismicisolation
@seismicisolation
Stress probe direction Horizontal Vertical

Characteristic forces from ULS analysis
ULS Perturbation 95.4 3289.1
NA 500.6 3787.1
TA 230.1 133.7
VB 325.2 322.2
HB 126.2 123.9
NC 8.6 9.8
TC 2.7 3.1
Ww 172.8 172.8
Values in stress probe direction
Total characteristic actions (excluding probe) 134.8 498
Total characteristic resistance 230.1 3787.1
DA2 assessment
Actions 181.9 672.4
Resistance 209.2 2705.1
Pass/Fail? PASS PASS
Table 3. DA2 Stability assessment in horizontal and vertical directions for stem wall problem
assessed at ‘Level 2’. All units are in kN/m.
3289
322
124
3
173 10
134
3787
Figure 6. Stem wall showing pattern of sliplines at collapse under vertical perturbation. A ver-
tical perturbation of 3289kN/m (indicated by curve headed arrow) was required to cause an
ULS, and the resulting values (in kN/m) of the action effects and resistances acting on the wall
are indicated.
moment/plastic moment of resistance. The stress probe is an imposed moment at

this point. This can be modelled by an applied moment to the wall section and
by modelling the wall as a rigid element with a single plastic hinge at the wall
base of design ULS strength 216 kNm/m. The result of the analysis is shown in
Fig. 7.
The moment (stress) probe for failure was 59.8 kNm/m with no soil strength
material (DA1/1) factoring. The actions due to the soil pressures are thus 216 −
@seismicisolation
@seismicisolation
59.8 = 156.2 kNm/m. For DA1/1 (and DA2) the factored action is 1.35 x 156.2
= 210.9 kNm/m, which is smaller than the resistance (216 kNm/m) and thus the
design passes this criterion
For DA1/2 soil strength material factoring, the moment (stress) probe for
failure was 5.6 kNm/m . The actions due to the soil pressures are thus 216 −5.6 =
210.4 kNm/m. For DA1/2 the factored action is 1.0 x 210.4 = 210.4 kNm/m,
which is smaller than the resistance (216 kNm/m) and thus the design passes this
criterion. Note that the same assessment result (in terms of wall failure) would
have been obtained at ‘analysis level 1’ if the wall bending strength had been
modelled in that analysis.
59.8
Figure 7. DA1/1 structural analysis for stem wall. External moment perturbation applied at
nominal 1m height on wall section. Plastic hinge is modelled at the vertical wall section base.
Moment perturbation of 59.8 kNm/m required to cause ULS.
4.3. Propped sheet pile wall

4.3.1. Problem description

The stability of the smooth propped sheet pile wall (prop resistance 25 kN/m)
depicted in Fig. 8 will now be considered. This was analysed by Smith (2012) using
conventional hand calculation methods. The overall stability will be analysed here
using DA1 at ‘analysis level 1’, and the design prop force and bending moment
will be examined using DA1 at ‘analysis level 3’. In the following analyses, the
resistance factor γR for the wall will be taken as 1.0.
4.3.2. DA1 Analysis Level 1 - overall stability, factor at source

As with the stem wall, in this case the analysis is undertaken with the adequacy
factor applied to the retained soil weight, using DA1/1 and DA1/2.
Fig. 9 shows the above analysis undertaken using the DA1/1 factor set and a
factor of 1.35 applied to the retained soil weight. The adequacy factor obtained
was 1.18 with a plastic hinge forming 1.6m above the excavation level. Since this
@seismicisolation
@seismicisolation
Prop
4m Wall, Mp = 35 kNm/m
Sand φ = 35o ,
γ = 15 kN/m3
1.8m
Figure 8. Propped (smooth) retaining wall.
is greater than 1.0 it means the design state is inside the yield surface and the
system satisfies the DA1/1 criterion.
For DA1/2 the adequacy factor obtained was 1.09 with a plastic hinge forming
1.48m above the excavation level. The system also satisfies the DA1/2 criterion.
It is note that design values of wall bending moment and prop force could be
determined from these ‘analysis level 1’ calculations, however these will tend to
be over-estimates unless the adequacy factor is equal to 1.0.
Figure 9. Propped (smooth) retaining wall analysed according to DA1/1 at analysis level 1.
4.3.3. DA1 Analysis Level 3 - determination of design prop force

In this assessment the ‘stress probe’ to be used is a force applied against the prop
in the direction of the prop using characteristic soil strength and resistance. The
analysis determines that a load of 12.2 kN/m is required to cause an ULS, with
no plastic hinge formation (Fig. 10). Since the prop resistance is 25kN/m, then
@seismicisolation
@seismicisolation
the actual action effects (due to the earth pressures) at ULS are 12.8kN/m (since
action effects + probe = resistance). For DA1/1 (and DA2), the design action
is 12.8 x 1.35 =17.3 kN/m and the design resistance is 25 kN/m, therefore the
design complies with DA1/1.
For DA1/2 (factor of 1.25 on tan φ ), a load of 4.1 kN/m is required to cause
a ULS, and therefore the design action is 25 - 4.1 = 20.9 kN/m which is less than
the design resistance so the design complies with DA1/2.
12.2 kN/m
Figure 10. Propped (smooth) retaining wall analysed at ‘analysis level 3’ for prop failure. Hori-
zontal perturbation force of 7.2 kN/m applied at prop level required to generate ULS.
4.3.4. DA1 Analysis Level 3 - factor on internal structural action

effects/resistances
In this analysis, the design bending moment for the wall should be determined at
all points along the wall. This is achieved by applying a stress probe in the form
of a matched pair of opposing couples at the specified point (as shown in Fig. 11)
with adequacy factor applied to these couples. For the purposes of illustration
this will be carried out st 1.8m above excavation level. The analysis shows that
an additional moment of 21.2 kNm/m is required to induce bending failure at
this point.
Given the design ULS bending resistance of 35 kNm/m then the action effects
(moment) induced by the earth pressures at this ULS is 35-21.2 = 13.8 kNm/m.
For DA1/1 (and DA2) the design moment is 13.8 x 1.35 = 18.6 kN/m. This is
less than the resistance 35 kNm/m and the design therefore complies with the
DA1/1 criterion.
For DA1/2 an additional moment of 5.3 kNm/m is required to induce bending
failure at 1.8m above excavation level. Thus the design moment is 35-5.3 = 29.7
kNm/m which is less than the resistance and so the design also complies with the
DA1/2 criterion.
@seismicisolation
@seismicisolation
Figure 11. Close up of wall showing how opposing couples are applied.
4.4. Special case: action effects and fixed resistances
4.4.1. Single hinge problem

For the wall bending problem considered above, it was necessary to apply a mo-
ment probe at each point down the wall and check that the action/resistance pair
equation is satisfied ( γA A ≤ R/γR , where γA and γR are the corresponding ac-
tion effect and resistance partial factors). Clearly this is an involved process and a
more automatic method would be desirable. In many cases the plastic moment of
resistance can be consider fixed (unless large tensile, compressive or shear stresses
in the pile are anticipated) and so a special case can be considered.
The above equation can be rewritten as follows: A ≤ R/(γR γA ). Thus an
equivalent approach would be to locally modify the wall resistance at each point
down the wall in turn and check that the above equation is satisfied. If, how-
ever only one hinge is anticipated in the collapse mechanism, then the resistance
factoring can be applied globally to the wall as an additional material factoring
(of the wall material) and will result in a new contracted yield surface for the
system. A material factoring approach can then be used to determine safety and
automatically identify the hinge position.

Taking γR = 1.0, a DA1/1 analysis similar to that carried out at ‘analysis
level 1’ could thus be carried out with with a wall plastic moment of resistance of
MP =35/γR γA =35/1.35=25.9 kNm/m, and unit partial factors elsewhere. With
an adequacy factor applied to the weight of the retained soil, an adequacy factor
of 1.34 is obtained with a single plastic hinge at 1.67m above excavation level.
The design therefore passes the DA1/1 criterion. To determine the minimum
acceptable design strength for the wall, the value of MP is adjusted until an
adequacy factor of 1.0 is obtained. For this example this occurs at a MP = 14.0
kNm/m with a hinge forming at 2.0m above excavation level. This gives a design
value of MP of 1.35×14.0 = 18.9 kNm/m.
A check may be carried out of this design at ‘analysis level 3’ by applying the
moment stress probe at 2.0m above excavation level for a wall with MP = 18.9
kNm/m. This requires a moment of 4.9 kNm/m to cause collapse which translates
to a design moment of 1.35 × (18.9 − 4.9) = 18.9kN/m. This is exactly equal to
the design moment of resistance as expected.
@seismicisolation
@seismicisolation
4.4.2. Multi-hinge problem
The above method is valid for the formation of one hinge only. If more than one
hinge forms, then the method is no longer equivalent to the stress probe approach,
since it is the whole wall strength that is being factored, rather than just the local
probe position. However it does provide a more conservative assessment.
This is easily demonstrated by examining a deeper wall with embedment
depth 2.8m. Using a DA1/1 ‘analysis level 1’ approach and no factoring, a bending
strength of 13.5 kNm/m is required to avoid an ULS. With this strength the wall
is at the point of collapse (corresponding to an adequacy factor of 1.0) with two
hinges forming, one at 1.15m below, and one at 2.03m above excavation level.
The corresponding wall design strength would be 1.35 × 13.5 = 18.2kNm/m.
The corresponding DA1/1 ‘analysis level 3’ approach requires a moment per-
turbation at either of the hinge positions. It is found that a perturbation of 6.8
kNm/m is required at the hinge height of 2.03m to cause a collapse with another
hinge forming at 1.15m below excavation level, as shown in Fig. 12. This means
that the actual design moment action at 2.03m height is 1.35 × (18.2 − 6.8) = 15.4
kNm/m. This is smaller than the design strength as expected and the wall is
overdesigned (due to the fact that the strength of the wall at the lower hinge
position is 18.2 kNm/m rather than 13.5kNm/m).
4.24 kNm/m
Figure 12. Propped (smooth) retaining wall analysed at ‘analysis level 3’ for bending failure.
Moment perturbation of 6.8 kNm/m applied at height 2.03m above excavation level required to
generate ULS with additional hinge generated at depth 1.15m below excavation level.
@seismicisolation
@seismicisolation
5. Discussion
The worked examples presented above have been designed to illustrate specific
design approach assessments at specific analysis levels. As stated earlier it is up
to the designer to decide which analysis level is appropriate.
When action/resistance factoring is involved, the analyses presented here as-
sess the system at the specific ultimate limit state relevant to the action/resistance
pair being tested. Adoption of a limit analysis approach means that the full plas-
tic resistance of all materials relevant to the specific ULS are mobilised. This in
turn means that the method will give more favourable results than a calculation
in which, for example, a design wall bending strength is determined on the basis
of an analysis of a rigid wall failure, which neglects the benefits of the bending
failure to redistribute stresses due to arching, and to possibly mobilise additional
soil strength.
It should also be noted that because a limit analysis assumes that the given
yield strength being mobilised in all system components simultaneously, there
may be issues with strain compatibility. For example for prop failure, if significant
movement of the wall top has to take place to fully mobilise the soil strength,
then this should be accounted for by ensuring the prop resistance is available at
this displacement.
Finally it is observed that all the assessments discussed are ULS assessments
and SLS assessments will also be required.
6. Conclusions
1. Computational Limit Analysis is a numerical method that uses optimiza-

tion to directly derive the ULS state rather than iterating to the ULS.
It thus offers a straightforward, robust, rapid method for determining the

ULS for any problem geometry.
2. In a general CLA based ULS design assessment, it is necessary to drive the
system to failure by some means e.g. by increasing a load or a moment.
(a) For material factoring, this can be any unfavourable load (or all loads).
(b) For action/resistance factoring an additional force/moment probe
must be applied in a specified direction to test any given failure mode.
3. Design calculations may be undertaken at several different ‘analysis levels’,

with different results obtained at each level.
4. Examples illustrating the application of CLA at various analysis levels
for a footing, stem wall and sheet pile wall have been given. Care may
be required for defining the ‘structure’ or forces on that ‘structure’ to be
assessed, for example in the design of a stem wall.
5. A convenient direct approach for determining bending strength design for
a sheet pile wall has been outlined.
@seismicisolation
@seismicisolation
References
BSI (2004), BS EN 1997-1:2004 Geotechnical design. General rules.

LimitState (2009), LimitState:GEO Manual VERSION 2.0, September 2009 edn,
LimitState Ltd.
Lysmer, J. (1970), ‘Limit analysis of plane problems in soil mechanics’, Journal
of the Soil Mechanics and Foundations Division ASCE 96(4), 1311–1334.
Makrodimopoulos, A. & Martin, C. (2006), ‘Lower bound limit analysis of
cohesive-frictional materials using second-order cone programming’, Int. J.
Num. Meth. in Eng. 6(4), 604–634.
Sloan, S. (1988), ‘Lower bound limit analysis using finite elements and linear
programming’, Int. J. Num. Anal. Meth. in Geomech. 12(4), 61–77.
Smith, C. (2011), ‘Rapid stability analysis’, The Geotechnica (4), 4–6.
Smith, C. (2012), ‘Theoretical basis of ultimate limit state design’, Geotechnique
Letters (in submission) .
Smith, C. & Gilbert, M. (2007), ‘Application of discontinuity layout opti-
mization to plane plasticity problems’, Proc. R. Soc. A 463, 2461–2484.
doi:10.1098/rspa.2006.1788.
Smith, C. & Gilbert, M. (2011a), ‘Ultimate limit state design to Eurocode 7
using numerical methods Part I: methodology and theory’, Ground Engineering
pp. 25–30.
Smith, C. & Gilbert, M. (2011b), ‘Ultimate limit state design to Eurocode 7
using numerical methods Part II: proposed design procedure and application’,
Ground Engineering pp. 24–29.
@seismicisolation
@seismicisolation
IOS Press, 2013
doi:10.3233/978-1-61499-163-2-214
Probabilistic assignment of design strength

for sands from in-situ testing data
Marco UZIELLIa,1, Paul W. MAYNE b and Mark J. CASSIDY c
a
Georisk Engineering S.r.l., Florence (Italy)
b
Geosystems Group, Georgia Institute of Technology, Atlanta, Georgia (USA)
c
Centre for Offshore Foundation Systems, Uni. of Western Australia, Perth (Australia)
Abstract. This paper proposes a simple design code format for assigning
design values of effective friction angle of clean sands from in-situ tests
incorporating all sources of geotechnical uncertainty, including model uncertainty.
New transformation models are proposed using a high-quality database of
undisturbed sands. The quantitative characterization of uncertainty is pursued
using a Bayesian hybrid Markov Chain Monte Carlo simulation framework.
Design factors are calibrated using the posterior distributions output by the
Bayesian framework. The effects of the aleatory and epistemic components of
geotechnical uncertainty are discussed and assessed quantitatively. The paper
concludes by demonstrating how to apply the results practically in geotechnical
design.
Keywords. Geotechnical uncertainty, effective friction angle, Bayesian modeling,

probability, factor calibration, in-situ tests, sand strength.
Introduction
Modern codes of practice for geotechnical design are increasingly framed within a limit
state design approach relying on probabilistic and reliability concepts. However, this
development has been more evolution than revolution, with load and resistance partial
factors calibrated, in the main part against historical cases and to ensure consistency
with the traditional total factor of safety approach. In many cases, the previous total
factor (defined as the ratio of overall resistance to applied load) has been simply split
into partial load and resistance factors (Meyerhof 1994; Kulhawy & Phoon 2002). The
historical development of load and resistance factor design codes for geotechnical
engineering is well summarized in Phoon (2008) and will not be repeated here.
Part of the motivation for the development of geotechnical codes has been to
harmonize across national boundaries and provide compatibility with the probabilistic
framework established within structural engineering codes (Phoon & Kulhawy 2005).
The introduction of the Eurocode (EC1 1994; EC7 1995) represents a prominent
example of this. Harmonization has caused considerable debate as to how the design
values of soil parameters should be defined. Design codes, on the whole, have not
provided explicit or even clearly defined procedures for selecting design values of soil
properties (such as strength). To align with the structural sections of the Eurocode
(EC1 1994), for instance, “characteristic” values should be defined as the 5% fractile of
the distribution of properties when a low value is unfavorable. It has been extensively
noted that practicing engineers do not follow consistent methods, with some choosing
@seismicisolation
@seismicisolation
M. Uzielli et al. / Probabilistic Assignment of Design Strength for Sands from In-Situ Testing Data 215
the mean whilst others taking the more conservative test results (Whitman 1984; Goble
1999; Green & Becker 2001; Kulhawy & Phoon 2002).
As partial factors of safety have been calibrated with conventional geotechnical
design and to give the same margin of safety as previously obtained using a total factor
safety approach (Meyerhof 1994; Kulhawy & Phoon 2002; amongst others), no
guidance is available on how resistance factors are affected by measurement techniques
and the models used to derive the design values (Phoon et al. 2003).
The rigorous implementation of reliability-based methods requires the preliminary
characterization of geotechnical uncertainty. Geotechnical uncertainty stems from
multi-sources including aleatory uncertainty, resulting from inter-site macro-diversities
and intra-site spatial and temporal variability of soil properties; and epistemic
uncertainty, comprising: (a) measurement uncertainty in the laboratory and/or in-situ
testing data which serve as inputs to engineering models; (b) statistical estimation
uncertainty, which stems from the finite number of available observations; and (c)
transformation uncertainty, stemming from the inevitable degree of approximation and
simplification of the physical world by any theoretical, empirical or experimental
model. Among the sources of geotechnical uncertainty, model uncertainty has received
the least attention, even though studies have shown that its magnitude is often greater
than those of the other components (Honjo 2011).
Although the reliability literature stresses the importance of incorporating model
uncertainty (e.g. Kulhawy et al. 1992; Gilbert & Tang 1995), and some design codes
provided cursory reference to this effect (e.g. Bureau Veritas 2004), no design codes
give guidance as to the level of model uncertainty to explicitly use. An improved
characterization of model uncertainty would enhance the objectivity and quality of
reliability-based codes.
Empirical correlations for estimating peak effective friction angle using the
standard penetration test (Kulhawy & Mayne 1990; Hatanaka & Uchida 1996; Mayne
et al. 2002) and cone penetration test results (Kulhawy & Mayne 1990; Ghiassian &
Jalili 2008) have been developed. These studies do not provide estimates of total
geotechnical uncertainty associated with the proposed models. To the extent of the
Authors’ knowledge, no pairwise correlations between stress-normalized shear wave
velocity and effective friction angle are available in the geotechnical literature.
Little direct guidance is given in design codes as to how to transform in-situ

measurements to drained parameters values. Within the offshore industry the API
RP2A (2011) provides some recommendations on in-situ and laboratory test selection
and within the commentaries provides figures converting relative density and unit
weight to design friction angles. The recently published InSafeJIP Guidelines for the
Prediction of Geotechnical Performance of Spudcan Foundations during Installation
and Removal of Jack-up Units (Osborne et al. 2010) provides the most specific advice,
reporting the coefficient of determination R2 associated with the model proposed by
Kulhawy & Mayne (1990) for the peak friction angle from cone resistance.
The Bayesian approach to geotechnical uncertainty quantification is well suited for
geotechnical engineering discipline (e.g., Christian & Baecher 2011), and is receiving
increasing attention at research level. Among other notable contributions, Ching et al.
(2012) presented a set of equations to update the mean and standard deviation of
friction angles of reconstituted clean sands using multivariate data from intra-site
laboratory and in-situ testing through Bayesian analysis. However, the outputs of these
studies are not available in general forms suitable for inclusion in design codes.
@seismicisolation
@seismicisolation
216 M. Uzielli et al. / Probabilistic Assignment of Design Strength for Sands from In-Situ Testing Data
This paper addresses the probabilistic definition of design values of drained

strength of cohesionless soils as obtained from in-situ testing data, specifically in a
design code format. The quantitative characterization of uncertainty is pursued in a
Bayesian framework. Though the framework focuses operationally on the
quantification of model uncertainty, all components of geotechnical uncertainty are
accounted for: aleatory uncertainty and statistical estimation uncertainty are addressed
through the inter-site character of the source datasets and its finite dimensions, while
epistemic measurement uncertainty is accounted for by modeling in-situ and laboratory
testing data as random variables.
1. Presentation of datasets
In this study, the characterization of uncertainty relies on the comparison between

model-predicted and observed values of effective strength of cohesionless soils.
Predicted values are output by purposely derived transformation models relating
strength to field testing results obtained from direct in-situ measurements; observations
derive from triaxial testing on undisturbed sands.
The reference dataset comprises a total of 16 sands sampled from 12 sites using
special undisturbed freezing techniques resulting in high-quality samples. The sands
are located in Canada (4 sands from 4 sites), Japan (6 sands from 4 sites), Norway (1
sand), China (1 sand), Italy (1 sand), and North Atlantic offshore (3 sands from 1 site).
The sites are described in greater detail in Mayne (2006; 2007), Plewes et al. (1993)
Pillai and Stewart (1994), Thompson & Long (1989) and Taylor et al. (1993). In
general, the sands can be considered as clean sands to slightly silty sands of primary
particle constituency of quartz to quartz-feldspar (silica), with additional minor
amounts of other rock mineralogies. In terms of grain size distributions, the materials
include 10 fine sands, 5 medium sands, and one coarse sand (Italy). The sands from
Canada and the North Atlantic were slightly silty or clayey having fines contents:
5<FC<15%, whereas the other sands were all relatively clean with FC<4%. Mean
values of index parameters (with plus and minus one standard deviation) of these sands
indicated: specific gravity (Gs=2.66±0.03), fines content (FC=4.36±4.49%), particle
size (D50=0.35±0.23 mm), and uniformity coefficient (UC=D60/D10=2.80±1.19). The

freezing process ensures precise and accurate laboratory measurement of effective
friction angle (I) as well as relative density and void ratio.
At most sites, field testing included CPTu, SPT, and seismic testing. Piezocone
penetrometer soundings provided continuous readings of corrected cone tip resistance
(qt) as well as sleeve friction (fs), and porewater pressures taken at the shoulder position
(u2). The SPT N-values were obtained in rotary drilled soil borings and these were
adjusted to 60% energy efficiency (N60). Shear wave velocity (Vs) values were obtained
primarily via downhole testing (DHT) with seismic piezocone testing (SCPTu) at one-
meter intervals, although crosshole test (CHT) data were available to confirm the DHT
results in some cases. Exceptions include the following: (a) only a CPT profile was
available at the Duncan Dam site (with no u2 data obtained); (b) no Vs data were
available for Kowloon or Hibernia sites; (c) SPT were not performed at the Hibernia
site. In-situ testing data were stress-normalized by vertical effective stress Vvo' and
atmospheric pressure Vatm=100 kPa according to conventional means:
@seismicisolation
@seismicisolation
qt1 (qt / V atm ) / V vc 0 / V atm 0.5

(1)
N1 60 N 60 / V vc 0 / V atm 0 .5
(2)
Vs1 Vs / V vc 0 / V atm 0.25

(3)
Stress-normalized in-situ testing parameters and effective friction angles from triaxial
compression tests for the sands in the database are listed in Table 1.
Table 1. Stress-normalized in-situ testing parameters and effective friction angles from triaxial compression
tests for the sands in the database (note: values of I’ are mean values)
Site Type I' [°] qt1 Vs1 [m/s] (N1)60

W. Kowloon (China) Hydr. Fill 38.1 74.5 0.0 21.3
Yodo (Japan) Alluvial 42.4 175.9 197 27.4
Yodo (Japan) Alluvial 38.4 102.2 213 34.6
Yodo(Japan) Alluvial 39.1 105.7 195 31.3
Natori (Japan) Alluvial 40.9 207.3 218 49.9
Tone (Japan) Alluvial 41.7 127.2 203 32.0
Edo (Japan) Alluvial 39.7 110.6 164 22.2
Massey (Canada) Alluvial 36.7 54.1 170 10.3
Kidd (Canada) Alluvial 37.3 67.8 179 13.4
J-pit (Canada) Hydr. Fill 32.7 22.2 129 3.4
Holmen (Norway) Alluvial 33.2 24.3 144 1.5
Gioia Tauro (Italy) Natural 41.5 174.1 219 29.3
Duncan Dam (Canada) Natural 34.8 30.9 - 7.7
Hibernia (N. Atlantic) Marine 43.6 396.0 - -
2. Definition of reference models
The first step of uncertainty characterization is the definition of the reference models,
to which uncertainty is subsequently to be associated. New models are proposed in this
study. Polynomial, logarithmic and power models were fitted comparatively to the data
using nonlinear least squares regression. Among these, power models of the form
I c k1 [ k 2 (4)
in which [ is the in/situ testing parameter (qt1, (N1)60, Vs1) provided superior fits. Table
2 contains the number of dataset cases available for each model, as well as the best-fit
regression coefficients, the determination coefficient R2, the standard error of
@seismicisolation
@seismicisolation
Table 2. Sample numerosity, best-fit regression coefficients, determination coefficient, standard error of
regression and suggested domain of applicability for the qt1, (N1)60 and Vs1 models
model N k1 k2 R2 st.error [°] domain
qt1-I’ 16 25.0 0.10 0.92 1.27 qt1 25
(N1)60-I’ 13 30.8 0.08 0.77 1.57 (N1)60>0
Vs1-I’ 12 3.9 0.44 0.67 1.98 125 m/s Vs1 225 m/s
regression and the suggested domain of applicability. Figure 1 shows the fits of the
power model to the three data sets.
It should be noted that the Bayesian approach adopted herein would allow the
characterization of model coefficients as random variables. For the sake of practical
applicability in engineering practice, it is deemed convenient here to define model
coefficients deterministically and characterize model uncertainty with respect to
deterministic models.
3. Quantitative characterization of uncertainty
The Bayesian approach allows the systematic, simultaneous quantitative estimation of

the multiple components of uncertainty. Model uncertainty can be parameterized by an
“uncertainty factor”, which is modeled as a random variable. Model inputs (i.e., in-situ
testing measurements) and observations (i.e., effective friction angles from triaxial
testing) are also modeled as random variables, which distributions can also be updated
simultaneously and consistently with the characterization of model factors.
3.1. Uncertainty modeling
To account for the uncertainty in the relations between in-situ and laboratory
measurements of I, the model uncertainty factor H is applied to the deterministic power
models defined in the previous section. Moreover, due to the presence of epistemic
measurement uncertainty in triaxial testing, the observed values of I are generally not
equal to the “real” values. To consider the effect of measurement uncertainty in triaxial
testing, an additive observational uncertainty factor 'I is defined for each of the
observations (n=1,…,N). Such uncertainty factor parameterizes the difference between

the “real” (unknown) value Ir and the observed value I of the friction angle:
(a) (b) (c)

50 50 50
45 45 45
40 40 40
I[°]
I[°]
I[°]
35 35 35
30 30 30
25 25 25
20 20 20
0 200 400 0 20 40 60 140 160 180 200 220
qt1 (N1)60 Vs1[m/s]

Figure 1. Fits of power models to database cases for: (a) qt1, (b) (N1)60 and (c) Vs1
@seismicisolation
@seismicisolation
Irc I c 'I (5)
By combining Eqs. (4) and (5), and assuming an additive uncertainty factor H (data sets
appear to be homoscedastic, see Fig. 1), the expression for the effective fiction angle of
the n-th pair of model predictions and observations is (n=1,…,N):
Inc k1 [ n k 2 H 'In (6)
The model uncertainty factor H is assumed to be normally distributed; hence, it is

completely defined by a mean and a standard deviation. It is also assumed that H is
unbiased. In assigning informative prior distributions, this implies the use of zero-mean
Gaussian prior for the mean of H, i.e. PPH=0°. The standard deviation of the mean (VPH)
is assigned subjectively. The standard deviation of H is modeled as a lognormal variable
with mean PVH assigned on the basis of the calculated standard error in the
deterministic regressions, and standard deviation VVH obtained by assuming a
coefficient of variation for VH of approximately 0.20, indicating intermediate
confidence in the assignment of the mean value. The values assigned to PVH are 1.3°,
1.6° and 2.0° for qt1, (N1)60 and Vs1, respectively; the corresponding standard deviations
VVHare 0.25°, 0.30° and 0.40°, respectively.
It can be expected that the magnitude of epistemic parameter uncertainty would
affect the posterior statistics of H, since the latter represents operationally a model
uncertainty factor which contributes a quota of the total uncertainty. Zhang et al. (2012)
showed that while the mean of the model factor is relatively insensitive to variations in
the magnitude of epistemic uncertainty in input parameters, the standard deviation of
the model factor displays an initial decrease for increasing epistemic uncertainty in
model inputs [, reaching a minimum value and subsequently increasing. By assuming
no epistemic uncertainty in [ the model uncertainty factor H parameterizes the “total”
uncertainty in the relation between [ and I. To assess quantitatively the repartition of
total uncertainty between model and parameters, a parametric analysis on the posterior
statistics of H is conducted for varying combinations of measurement uncertainty in
field test data and observational uncertainty in triaxial test data.

Measurement uncertainty in qt1, (N1)60 and Vs1 is assumed to be normally
distributed and unbiased. Observed values are assumed as mean values. The standard
deviation for each observation is calculated by multiplying mean values by the
coefficient of variation of measurement uncertainty. More than one value was assigned
to such COVs to assess the effect of epistemic parameter uncertainty on total
uncertainty. In addition to the null uncertainty case (i.e., COVm([)=0), sets of values
were assigned subjectively based on suggested values for non-normalized
measurements from the geotechnical literature (Kulhawy & Trautmann 1996, Moss
2008): COVm(qt1)=[0.10,0.15,0.20]; COVm[(N1)60]=[0.15,0.30,0.45];
COVm(Vs1)=[0.05,0.10,0.15]. For each set, the central value is deemed to be the most
plausible based on the geotechnical literature.
The observational uncertainty factors 'I are also assumed to be normally
distributed and unbiased for each measured value of I. Zero-mean Gaussian priors are
used for 'In, with n=1,…,N. The standard deviations of 'In are assigned by
multiplying each observation by the coefficient of variation of measurement
@seismicisolation
@seismicisolation
uncertainty related to triaxial compression testing. As for epistemic uncertainty in field

measurements, the effect of observational uncertainty is assessed parametrically by
defining the set of values COVm(I)=[0.05,0.10,0.15] (e.g., Zhang et al. 2012), with
0.10 being the most plausible value.
For each model and for each combination of measurement uncertainty magnitudes,
the total number of random variables is given by 2N+2, i.e. the sum of: (a) the number
of available in-situ measurements N; (b) the number of observational uncertainty
parameters (which is equal to the number of measurements N); and (c) the number of
parameters of the Normal distribution (i.e., 2).
3.2. Implementation of the hybrid MCMC algorithm
The calculation of the posterior distributions of the random variables defined and
modeled as described in the previous section is pursued via the hybrid Markov Chain
Monte Carlo (MCMC) simulation algorithm illustrated in Zhang et al. (2012). Such
algorithm efficiently combines the features of the Metropolis algorithm (Metropolis et
al. 1953) and Gibbs sampler (Geman et al. 1984). In the hybrid method, samples are
drawn using a proposal distribution similar to that in the Metropolis algorithm.
Differently from the Gibbs sampler, the hybrid chain is applicable even when the
probability distributions of a group of variables conditioned on other variables are not
available in closed form (Zhang et al. 2012).
Based on Bayes’ theorem, the unscaled posterior distribution can be expressed as
q P H , V H , [1 ,..., [ N , 'I1 ,..., 'I N I1c ,..., I Nc f PH f V H

N ° ª I c k [ k2 'I P º ½° (7)

n 1
®) «
°̄ «¬
n 1
VH
n H
» f [ n f 'I n
»¼
¾
°¿
in which f() denotes the prior distribution of an observed variable and ) is the
cumulative standard Normal distribution function.
Markov chains are initialized by sampling initial points randomly from the
respective prior distributions. Subsequently, the updating process, which occurs in Q
user-defined iterations, is structured according to a sequential procedure relying on
Bayes’ theorem. For the i-th iteration (i=2,…,Q):
1) Updating of PH
Draw PH* from a Normal proposal distribution PH* ~ N PHi 1 , V PH
Accept PH* as P Hi with a probability
ª q PH* , V Hi 1 , [1 i 1 ,.., [ Ni 1 , 'I1 i 1 ,.., 'I Ni 1 º
rPH min « i 1 i 1 i 1 i 1 i 1 i 1
,1»
¬« q PH , V H , [1 ,.., [ N , 'I1 ,.., 'I N ¼»
otherwise P Hi = PHi 1
2) Updating of V H
@seismicisolation
@seismicisolation
Draw V H* from a Normal proposal distribution V H* ~ N V Hi 1 , V VH

, with the constraint
VH*
>0
Accept V H* as V Hi with a probability
ª q PHi , V H* , [1 i 1 ,.., [ Ni 1 , 'I1 i 1 ,.., 'I Ni 1 º
rVH min « i 1 i 1 i 1 i 1 i 1
,1»
¬« q PH , V H , [1 ,.., [ N , 'I1 ,.., 'I N
i
»¼
otherwise V Hi = V Hi 1
3) Updating of [1 ,...,[ N
for n=1:N
Draw [ n* from a multivariate Normal proposal distribution [ n* ~ N [ ni 1 , V [n
Accept [ n* as [ ni with a probability
ª q PHi , V Hi , [1 i 1 ,.., [ n* ,.., [ Ni 1 , 'I1 i 1 ,.., 'I Ni 1 º
r[n min « i 1 i 1 i 1 i 1 i 1
,1»
¬« q PH , V H , [1 ,..[ n ,.., [ N , 'I1 ,.., 'I N
i i
¼»
otherwise [ ni = [ ni 1
end
4) Updating of 'I1 ,...,'I N

for n=1:N
Draw 'I n* from a multivariate Normal distribution 'In* ~ N 'Ini 1 , V 'In
Accept 'I n* as 'I ni with a probability
ª q PHi , V Hi , [1 i ,.., [ Ni , 'I1 i 1 ,..'In* ,.., 'I Ni 1 º
r'In min « i 1 i 1 i 1
,1»
¬« q PH , V H , [1 ,.., [ N , 'I1 ,.., 'In ,.., 'I N
i i i i
»¼
otherwise 'I ni = 'I ni 1
end
In this study, four chains were initialized and compiled for each analysis in order to
assess the convergence of the MCMC algorithm. A total of Q=10,000 iterations were
performed for each chain. Following the compilation of the Markov chains, the values
in the initial 20% of each chain (i.e., 2000 iterations) were identified as pertaining to
the burn-in phase, and discarded to reasonably assume that samples of the posterior
distribution comprised post-convergence values only. The mean and standard deviation
of the size-8000 post-convergence part of the samples of the posterior distributions of
PH and VH were recorded. The sample mean and standard deviations of the samples of
Normal distribution parameters PHc and VHc (c=1,..,4, corresponding to the number of
chains) were calculated for each analysis. The mean values / PH and /VH define the
central tendency value of the mean and standard deviation of H, respectively,
considering all chains. Plots illustrating the MCMC algorithm are not shown here.
@seismicisolation
@seismicisolation
The key factor affecting the efficiency of the Markov chain is the spread of the
proposal distribution. In case of normal or multivariate normal proposal distributions,
the standard deviation can be taken as the reference dispersion statistic. An excessively
wide dispersion statistic will result in low acceptance probabilities for proposed
sampled values, while an excessively low dispersion statistic causes a slow movement
of the Markov chain and a high number of iterations to collect a sufficient number of
representative values in the posterior sample. Gelman et al. (1995) suggested that a
suitable tentative value for the standard deviation of the proposal distributions may be
comparable to the posterior standard deviation of the variables to be updated. Here,
standard deviations of proposal distributions were assigned as being equal to the mean
of the prior standard deviation. Table 3 shows the output values of / PH and /VH for
combinations of COVm([) and COVm(I) for the qt1, (N1)60 and Vs1 models, respectively.
The posterior means / PH resulted to be identically zero, attesting for the unbiasedness
of the regressed deterministic models. For all models, the 4 chains showed excellent
convergence to extremely similar posterior distributions. Figure 2 shows the
cumulative distribution functions of an example output set of Gaussian model factors H
for the qt1-I model and for COVm(qt1)=0.15 and COVm(I)=0.10. The parameters of the
distributions are the mean values / PH and /VH of the posterior samples of means and
standard deviations PHc and VHpertaining to c=1,…,4 chains. The distributions are
essentially coincident (with / PH =[0.0004,-0.0016,0.0007,-0.0012] and /VH
=[1.470,1.475,1.467,1.471], attesting for the excellent convergence of the 4 chains.
Similar degrees of convergence among chains were observed for the (N1)60 and Vs1
models. Acceptance rates were approximately 0.37 and 0.28 for PH and VH, respectively,
for all of the analyses performed. These values perfectly fit in the range corresponding
to efficient Markov chains as observed by Gelman et al. (1995).
Inspection of Table 3 suggests several main observations. First, posterior standard
deviations are significant in magnitude. Recalling that these are standard deviations of
zero-mean Gaussian distributions, results attest for the relevance of uncertainty in
design values of effective strength when obtained via models from in-situ tests.
Second, values of /VH are significantly different among models. This disparity is
due to the aggregate effect of: (a) aleatory uncertainty, i.e. the effective dispersion of
observations around the deterministic models; (b) the different magnitudes of
measurement uncertainty in in-situ tests; and (c) the different numerosity of available
0.8
CDF(H)
0.6
0.4
0.2
0
-4 -3 -2 -1 0 1 2 3 4
H[°]

Figure 2. Output Gaussian model factors from 4 Markov chains for the qt1-I model
@seismicisolation
@seismicisolation
Table 3. Means of posterior statistics from the hybrid MCMC algorithm

qt1 (N1)60 Vs1
COVm(I) COVm(qt1) /VH COVm[(N1)60] /VH COVm(Vs1) /VH
[°] [°] [°]
0.05 0.00 1.26 0.00 1.52 0.00 1.83
0.10 0.00 1.42 0.00 1.66 0.00 2.01
0.15 0.00 1.48 0.00 1.74 0.00 2.05
0.05 0.10 1.25 0.15 1.50 0.05 1.83
0.10 0.10 1.39 0.15 1.63 0.05 1.97
0.15 0.10 1.44 0.15 1.70 0.05 2.04
0.05 0.15 1.24 0.30 1.50 0.10 1.84
0.10 0.15 1.37 0.30 1.62 0.10 1.97
0.15 0.15 1.41 0.30 1.67 0.10 2.03
0.05 0.20 1.24 0.45 1.50 0.15 1.86
0.10 0.20 1.36 0.45 1.61 0.15 1.96
0.15 0.20 1.39 0.45 1.66 0.15 2.03
observations among models. It is difficult to assess a priori the relative weight of the
three components in determining the magnitude of total uncertainty. For instance,
comparison of the qt1 and Vs1 models shows that the posterior standard deviations of the
latter exceeds those of the former by a factor of approximately 1.5. This difference may
be explained by the larger standard errors of the preliminary deterministic regressions
(1.27° and 1.98° for qt1 and Vs1, respectively, as shown in Table 3) possibly resulting
from the larger scatter of Vs1 observations round the deterministic model (the
coefficients of determination are 0.92 and 0.67, respectively) as well as by the larger
sample numerosity of qt1 (16 data pairs versus 12). In this case, in-situ testing
uncertainty presumably played a marginal role, since the set of COVms were coincident
for the two models. As another example, if one compares the coefficients of
determination (0.77 and 0.67, respectively) of the (N1)60 and Vs1 models, and observing
that sample numerosities are essentially the same (16 vs. 15) it may be inferred that
aleatory uncertainty plays a more relevant role in determining the greater total
uncertainty in the (N1)60 model with respect to the Vs1 model, despite the significantly
larger magnitude of measurement uncertainty in the former.
As a third main observation of the results of the MCMC analyses, and as expected
from the above discussion, there is a moderately strong inverse relationship between
COVm([) and /VH for constant values of COVm(I)]. Though the estimation procedures
illustrated herein focus on the quantification of transformation uncertainty, the output

factors implicitly reflect information regarding the epistemic measurement uncertainty
in cone penetration testing as well as other sources of epistemic uncertainty as detailed
earlier in the paper. Hence, in the practical implementation of a geotechnical
uncertainty model such as the additive one proposed by Phoon & Kulhawy (1999a,
1999b), the magnitude of the measurement uncertainty parameter (e.g., a COV of
measurement uncertainty) should correspond to the one used in the estimation of the
transformation model factor. Table 3 also highlights the string direct relationship
between COVm(I) and /VH for any given value of COVm([). This result attests for the
fact that if observations are modeled as random variables, additional uncertainty is
input into the [-I relation, thus increasing the magnitude of total uncertainty as well as
model uncertainty.
@seismicisolation
@seismicisolation
4. Utilization of results in geotechnical design
For the sake of practicality in formulating an uncertainty model, it is more convenient

to refer to the null measurement uncertainty case for field data, with plausible values of
observational uncertainty for I. Hence, The values 1.42°, 1.66° and 2.01° can be taken
as reference values of /PH for engineering purposes for the qt1, (N1)60 and Vs1 relations,
respectively, unless specific information regarding the magnitude of in-situ
measurement uncertainty is available.
For geotechnical design purposes, it is convenient to express the outputs of the
Bayesian analysis described in the previous sections in a simple and practical format.
The proposed design format for effective friction angle is
Idc k1 [ k2
MI (8)
in which the factor MI depends from the user-defined level of conservatism through the
probability of non-exceedance pt:
MI )H1 pt (9)
where )H 1 is the inverse of the cumulative Normal distribution with mean / PH and
standard deviation /VH . The lower the value of pt, the higher the target degree of
conservatism in the design value of I. Table 4 contains the model factors MI for the
three models investigated herein, for several notable target probability levels: 0.001,
0.005, 0.01, 0.05 and 0.10 (with COVm(I)=0.10).
Figures 3a, 3b and 3c plot the source data points along with the deterministic
regressions and the design curves for pt=0.005 and 0.050 for the: (a) qt1; (b) (N1)60; and
(c) Vs1 models, respectively.
The model factors calibrated as described herein can be effectively and easily
implemented in design. As a practical calculation example, consider the case in which a
measurement of shear wave velocity Vs=210 m/s is available for a sand deposit at depth
8 m, where the vertical effective stress is calculated to be Vv0=83 kPa. It is of interest
to assign a design value to effective friction angle for a target probability of non-
exceedance of 0.01. From Eq. (4) and the deterministic coefficients in Table 2 for the
Vs-I model, the normalized shear wave velocity is Vs1=220 m/s. The deterministic
value of friction angle is calculated as Idet=3.9·2200.44=41.9°. The model factor
corresponding to pt=0.01 is MI=-4.7°. Hence, the terminal design value is
Mdc Mdetc MM =37.2°. The magnitude of the model factor relative to the deterministic
value (which does not account for uncertainty) is approximately 11%.
Table 4. Model factor MI for notable target probability levels

pt MI qt1 >q@ MI >(N1)60@>q@ MI Vs1 >q@
0.001 -4.4 -5.1 -6.2
0.005 -3.7 -4.3 -5.2
0.01 -3.3 -3.9 -4.7
0.05 -2.3 -2.7 -3.3
0.10 -1.8 -2.1 -2.6
@seismicisolation
@seismicisolation
5. Concluding remarks
The investigation on the uncertainty in effective friction angle of sands illustrated in

this paper highlights the relevance of uncertainty in widely used geotechnical
procedures, such as the derivation of design values from in-situ testing. The significant
magnitudes of the model factors calibrated herein attests for the importance and
convenience of addressing uncertainty in geotechnical analysis in order to attain a
target reliability level expressed in probabilistic terms, thereby pursuing safety and
performance in a way which is as rigorous as possible.
The results of the analyses presented in this paper are directly applicable only to
design problems in which friction angle is the single dominant uncertainty. Nonetheless,
the quantitative information regarding the model factors can be used in cases in which
other design parameters are modeled as random variables. For instance, in a
multivariate design case, distribution parameters of the model factors allow the
probabilistic description of friction angle in a second-moment sense (e.g., for
implementation in a FOSM framework) by considering the deterministic value as the
central tendency parameter and by quantifying the second central moment (i.e.,
standard deviation) based on the distribution parameters of the model factor as obtained
herein. In a higher-level analysis relying on probabilistic simulation, the distribution
parameters of model factors allow the direct generation of simulation input samples of
friction angle, possibly including statistical dependence among random variables using
copula theory or other multivariate distribution modeling techniques.
Another important aspect of reliability-based design which is not explicitly
addressed herein is the selection of the “physically characteristic” value of the field
measurement which servers as input to the uncertainty-based estimation of friction
angle, e.g. the “representative” value of cone resistance over a soil volume in which the
measured parameter is spatially variable. Formally, the acknowledgment of this source
of epistemic uncertainty entails the increase in the epistemic uncertainty associated
with the independent variable. The magnitude of such an increase is extremely variable
and difficult to quantify aprioristically, as it is largely case- and site-specific. As has
been shown in the paper, there exists an inverse relationship between the magnitude of
epistemic uncertainty in field measurement and the standard deviation of the model
factor. By not considering this additional epistemic uncertainty explicitly,
(a) (b) (c)

50 50 50
45 45 45
40 40 40
I[°]
I[°]
I[°]
35 35 35
30 30 30
25 25 25
20 20 20
0 200 400 0 20 40 60 140 160 180 200 220
qt1 (N1)60 Vs1[m/s]

Figure 3. Source data points, deterministic regressions and design curves for: deterministic case (black);
pt=0.005 (dark gray) and pt=0.050 (light gray)
@seismicisolation
@seismicisolation
unconservatism in the characterization of model factors is avoided.

As any output of a statistical/probabilistic analysis, the quantitative results obtained
and presented herein depend from the source datasets, and are likely to vary to some
extent if new data become available to update them. Model factors are thus not static
quantities; rather, the progressive increase in high-quality data would allow updating of
quantitative estimates.
The calibration procedure shown herein demonstrates the possibility of making
available the results from complex probabilistic methods in a simple format for
practical implementation in routine design. The proposed format is compatible with a
“full reliability-based” approach which bypasses the necessity to assign a
“characteristic value”. The progressive compilation of sets of model factors derived
through rigorous procedures would be beneficial in the drafting and refinement of
design codes based on probability and reliability concepts.
Acknowledgments
The first Author is grateful to Dr. Alessandro Bessi for his assistance in performing the
MCMC calculations.
References
American Petroleum Institute API RP2A (2011). ANSI/API Recommended Practise 2 Geo. Geotechnical and
Foundation Design Considerations, Modified version of ISO 19901-4:2003.
Bureau Veritas (2004). Classification of mooring systems for permanent offshore units. Guidance Note NI
493.
Ching, J., Chen, J.-R., Yeh, J.-Y., Phoon, K.-K. (2012). Updating uncertainties in friction angles of clean
sands, Journal of Geotechnical and Geoenvironmental Engineering 138 (2), 217-229.
Christian, J.T., Baecher, G.B. (2011). Keynote lecture: Unresolved problems in geotechnical risk and
reliability, Proceedings of Georisk 2011 – Geotechnical Risk Assessment and Management, GSP 224,
Atlanta, USA, ASCE Press, Reston, Virginia, 50-63.
EC1 (1994). Eurocode 1: Basis of design and actions on structures (Part 1: Basis of design). European
Committee for Standardisation, ENV 1991-1:1994.
EC7 (1995). Eurocode 7: Geotechnical Design – Part 1: General rules, together with the United Kingdom
National Application Document, DD ENV 1997-1:1995. British Standards Institution, London.
Gelman, B.A., Carlin, B.P., Stem, H.S., Rubin, D.B. (1995). Bayesian data analysis, Chapman & Hall,
London.
Geman, S., Geman, D. (1984). Stochastic relaxation, Gibbs distributions, and the Bayesian restoration of
images, IEEE Transactions on Pattern Analysis of Machine Intelligence 6, 721-741.
Ghiassian, H., Jalili, M. (2008). Empirical relationships between internal friction angle and modulus of
elasticity of sandy soils based on in-situ and checking triaxial tests, Proc., 3rd Int. Conf. on Site
Characterization (ISC-3, Taipei), Taylor & Francis, London, (CD-ROM).
Gilbert, R.B., Tang, W.H. (1995). Model uncertainty in offshore geotechnical engineering, Proc. of 27th
Annual Offshore and Technology Conference, Houston, USA, OTC7757, 557-567.
Goble, G. (1999). Geotechnical related development and implementation of load and resistance factor design
(LRFD) methods, NCHRPSynthesis276, Transportation Research Board, Washington, D.C., USA.
Green, R., Becker, D. (2001). National report on limit state design in geotechnical engineering: Canada,
GeotechnicalNews 19 (3), 47-55.
Hatanaka, M., Uchida, A. (1996). Empirical correlation between penetration resistance and internal friction
angle of sandy soils, Soils and Foundations 36 (4), 1–9.
Honjo, Y. (2011). Keynote lecture: Challenges in geotechnical reliability based design, Proceedings of the
International Symposium on Geotechnical Safety and Reliability ISGSR 2011, Vogt, Schuppener,
Straub & Bräu (eds.), Munich, Bundesanstalt für Wasserbau, CD-ROM.
@seismicisolation
@seismicisolation
Kulhawy, F.H., Mayne, P.W. (1990). Manual on estimating soil properties for foundation design, Report EL-
6800, Electric Power Research Institute, Palo Alto, USA, 306p.
Kulhawy, F.H., Birgisson, B., Grigoriu, M.D. (1992). Reliability-based foundation design for transmission
line structures: transformation models for in-situ tests, Report EL5507(4), Electric Power Research
Institute, Palo Alto, USA, 113 p.
Kulhawy, F.H., Trautmann, C.H. (1996). Estimation of in-situ testing uncertainty, Uncertainty In the
Geologic Environment: From Theory to Practice, ASCE Press, Reston, USA, 269-286.
Kulhawy, F.H., Phoon, K.K. (2002). Observations on geotechnical reliability-based design development in
North America, Proc. of the Int. Workshop on Foundation Design Codes and Soil Investigation in view
of International Harmonization and Performance. Tokyo, 31-48.
Mayne, P. W., Christopher, B. R., Berg, R. R., DeJong, J. (2002). Subsurface investigations—Geotechnical
site characterization, FHWA-NHI-01-031, National Highway Institute, Federal Highway Administration,
Washington, D.C., 301p.
Mayne, P.W. (2006). The 2006 James K. Mitchell Lecture: Undisturbed sand strength from seismic cone
tests, Geomechanics and GeoEngineering 1 (4), Taylor & Francis Group, London, 239-257.
Mayne, P.W. (2007). Invited Overview Paper: In-situ test calibrations for evaluating soil parameters,
Characterization & Engineering Properties of Natural Soils, Vol. 3 (Proc. 2nd Singapore Workshop),
Taylor & Francis Group, London, 1602-1652.
Metropolis, N., Rosenblueth, A.W., Rosenblueth, M.N., Teller, A.H., Teller, E. (1953). Equation of state
calculations by fast computing machines, Journal of Chemical Physics 21, 1087-1092.
Meyerhof, G.G. (1994). Evolution of safety factors and geotechnical limit state design. 2nd Spencer J.
Buchanan Lecture. Texas A&M University, College Station, USA.
Moss, R.E.S. (2008). Quantifying measurement uncertainty of thirty-meter shear-wave velocity, Bulletinof
theSeismologicalSocietyofAmerica98ȋ͵Ȍǡͳ͵ͻͻȂͳͶͳͳǤ
Osborne, J.J., Teh, K.L., Houlsby, G.T., Cassidy, M.J., Bienen, B., Leung, C.F. (2010). “InSafeJIP”
Improved guidelines for the prediction of geotechnical performance of spudcan foundations during
installation and removal of jack-up units, RPS Energy Report Number EOG0574-Rev1. Final Guielines
of the InSafe Joint Industry Project. 124p.
Phoon, K.K., Kulhawy, F.H. (1999). Characterisation of geotechnical variability. Canadian Geotechnical
Journal 36(4), 612-624.
Phoon, K.K., Kulhawy, F.H. (1999). Evaluation of geotechnical property variability. Canadian Geotechnical
Journal 36(4), 625-639.
Phoon, K.K., Becker, D.E., Kulhawy, F.H., Honjo, Y., Ovesen, N.K., Lo, S.R. (2003). Why consider
reliability analysis for geotechnical limit state design, Proc. of LSD2003: International Workshop in
Limit Stet Design in Geotechnical Engineering Practice. Phoon, Honjo and Gilbert (eds). World
Scientific Publishing Company, Cambridge, USA.
Phoon, K.-K. (2008). Numerical recipes for reliability analysis – a primer, Reliability-Based Design in
Geotechnical Engineering, Phoon, K.K. (ed.), Taylor and Francis, London, 1-75.
Pillai, V.S., Stewart, R.A. (1994). Evaluation of liquefaction potential of foundation soils at Duncan Dam,
Canadian Geotechnical Journal 31 (6), 951-966.
Plewes, H.D., Pillai, V.S., Morgan, M.R., Kilpatrick, B.L. (1993). In-situ sampling, density measurements,
and testing of foundation soils at Duncan Dam, Proceedings 46th Annual Canadian Geotechnical
Conference, Saskatoon: 223-235. Re-published in Canadian Geotechnical Journal 31 (6), 927-938.
Taylor, B.B., Lewis, J.F., Ingersoll, R.W. (1993). Comparison of interpreted seismic profiles to geotechnical
borehole data at Hibernia, Proc. 4th Canadian Conference on Marine Geotechnical Engineering, Vol. 2,
St. John’s, Newfoundland: 685-708.
Thompson, G.R., Long, L.G. (1989). Hibernia geotechnical investigation and site characterization, Canadian
Geotechnical Journal 26 (4), 653-678.
Whitman, R.V. (1984). Evaluating calculated risk in geotechnical engineering, Journal of Geotechnical and
Geoenvironmental Engineering 110 (2), 145-188.
Zhang, J., Tang, W.H., Zhang, L.M., Huang, H.W. (2012). Characterising geotechnical model uncertainty by
hybrid Markov Chain Monte Carlo simulation, Computers and Geotechnics 43, 26-36.
@seismicisolation
@seismicisolation
IOS Press, 2013
doi:10.3233/978-1-61499-163-2-228
Experiences with Limit State Approach for

Design of Spread Foundations in the Czech
Republic
Martin VANÍEK a,1 and Ivan VANÍEK a
a
Czech Technical University in Prague
Abstract. The first experience with limit state design approach was introduced in
the Czech Republic already 45 years ago for shallow foundations. However just
the experiences with the second version which was implemented in 1987 (25 years
ago) will be discussed here in more detail. The paper describes both limit states
Ultimate and Serviceability limit states.
Mainly the focus will be on:
- definition and distinction of the design into 3 geotechnical categories
- demands on the site investigation
- nominal values of soil properties for different geotechnical categories
- design approaches within limit states
- definition of material partial safety factors
- design calculations for both limit states
- evaluation of up to date experience with respect to probability of failure.
Keywords. Spread foundations, Limit state approach, bearing capacity, settlement
Introduction
Limit state approach was for the first time introduced for the geotechnical design in the
Czech Republic in 1966 when the standard SN 73 1001 (Foundation of structures.
Subsoil under shallow foundations) was published. Only the basic principles will be
defined as new standard with the same number and name was published in 1987.
There were several reasons why the standard was redrafted. The most important
are as follows:
x The classification system for foundation subsoil suggested in 1966 showed
itself as insufficient, as for example for cohesive soils used only 3 classes
based on the plasticity index, while for sandy soils it used 7 classes. However
in the Czech Republic there are not so many non-plastic
sandy soils. Moreover for cohesive soils there were defined parameters for
undrained shear strength required for bearing capacity calculation based on
soil consistency and hence the values were very close.
x Suggested approaches for the total settlement analysis were generally giving
significantly higher values than those measured afterwards in reality.
1
@seismicisolation
@seismicisolation
M. Vaníček and I. Vaníček / Experiences with Limit State Approach 229
Therefore higher attention is devoted to the standard from 1987, which adopted
most of the principles from the original standard from 1966, however it differs from the
original version mainly in those 2 above mentioned points and though in other points
certain corrections were also made. Just before the issue of the new version in 1987 the
drafters were informed by Prof. Krebs Ovesen about the first proposal of the
Eurocode 7.
1. First standard from the 1966
The standard distinguishes 3 fundamental approaches for the design of shallow

foundations with respect to the severity of the superstructure and complexity of the
subsoil conditions:
a) For modest superstructures in simple subsoil conditions it is possible to use
values of derived nominal bearing capacity (which are published in the standard
for each soil class).
b) For demanding superstructures in simple subsoil conditions or for modest
superstructures in complex subsoil conditions limit states calculations are
required. The calculations are performed for nominal mechanical parameters of
subsoil, which are tabulated for each soil class in the standard.
c) For demanding superstructures in complex subsoil conditions limit state
calculations are required. However, the calculations are performed for
statistically evaluated mechanical properties from performed tests for given
subsoil.
Simple subsoil conditions were defined as: smooth terrain, subsoil within the site is
not changing significantly and the strata are more or less of the same thickness and
roughly horizontal.
Modest superstructures from foundation engineering point of view are defined as
follows: houses up to 5 storeys and other buildings up to 3 storeys, which are not
affected by differential settlement of foundation elements up to the range of 20 to
30mm.
Ultimate state (limit state of bearing capacity) for homogeneous subsoil and centric
loading (Fig. 1) was calculated from the equation:
qf = n1(r) B/2 Nr + n2(r) D Nqr + c(r) Ncr (1)
where values of the design bearing coefficients Nr, Nqr, Ncr are defined for undrained
u(r), resp. drained ’(r) values of the angle of internal friction. Note: for u = 0: Nr = 0,
Nqr = 1 and Ncr = 5.1.
The limit value of bearing capacity was finally compared with the extreme value of
contact pressure (for extreme loading).
D
B
Figure 1. Sketch of the situation for the basic equation of the bearing capacity
@seismicisolation
@seismicisolation
230 M. Vaníček and I. Vaníček / Experiences with Limit State Approach
For design values of shear parameters applies:
(r) = (n) · k (2)
c(r) = c(n) · k (3)
where k is the standard value of the coefficient of uniformity and (n) and c(n) are
recommended table nominal values for the design approach ad b) or evaluated values
after statistical analysis for the design approach ad c).
For design values of weight density applies:
n(r) = n1(n) · n (4)
where n = 0.9 is a coefficient of loading

The last coefficient - coefficient of calculation reliability m - was applied on
bearing capacity factors, when m = 0.5 was applied for (r) > 30° and was linearly
increasing with lower value of (r) up to m = 1 for (r) = 0.
A short note can be made to the selection of coefficient of uniformity k for the
angle of internal friction for the design approach ad c). The minimum of tested samples
is N = 5. During the statistical analysis of the measured values it is assumed that the
basic set of examined values is close to the normal distribution – Gauss-Laplace. Hence
the corresponding coefficient of uniformity k is determined from the equation (5):
k=1 - (5)
where is determined from the Fig. 2 for the probability of 0.999, namely with respect
to the amount of samples N and variability of the measured values v. Value of = 0.3
and the corresponding value of coefficient of uniformity k = 0.7 are limiting. If for the
given number of measured values N and its variability v the value of > 0.3 it is
necessary to either increase the number of samples N or to divide the measured values
more concisely to statistically more homogeneous sets.
@seismicisolation
@seismicisolation
Number of measured values N M. Vaníček and I. Vaníček / Experiences with Limit State Approach 231
Coefficient of variability v
Figure 2. Determination of the coefficient of uniformity k = 1 – as function of the number of measured
values N and coefficient of variability v for probability of 0.999 (according to SN 731001:1966)
For the calculation of the total settlement two equations were recommended, one
based on the theory of elasticity and a second one based on the stress-strain method.
For both methods the coefficient of calculation reliability m = 1.25 was applied and for
the first one as well correction factor m1 for different types of soils (m1 is in the range
of 0.5 – 0.8) and for the second method correction factor m2 which is different for
normally consolidated soils (m2 = 0.8) and for over-consolidated soils (m2 = 0.5).
2. Second standard from the 1987

Firstly new soil classification system was applied for this standard. This new system is
in principle based on US Bureau of Reclamation System with some modifications, as
we have had very good experience with this system from the view of small earth fill
dams design. In more details this new system is described in Šimek et al. (1990) and in
shorter version in Vaní ek I. and Vaní ek M. (2008).
Also a new equation for the total settlement calculation was recommended and
which is based on large in situ experiments, e.g. Havlí ek (1978), or Sey ek (1995).
Numerous application problems show that in most cases the calculated settlements and
deformations are greater than the real measured values (Vaní ek, 1982). This
difference is most often attributed to some type of sample failure during the process of
sample extraction and/or sample handling before testing. In order to explain the above-
mentioned difference, extensive laboratory and on-site tests were carried out. High-
precision sensors registering vertical deformations were mounted under modelled real-
life spread foundations. The results manifested that only negligible vertical movements
were measured in subsoil despite the fact that according to the theory of elastic half-
space there should be stress increase and hence also deformation. The differences in
results were also determined for different types of soils. The differences found were
@seismicisolation
@seismicisolation
attributed to the structural strength of soil, i.e. the bond between individual grains,
which develops in time for long-term acting initial stress. It is becoming evident that
additional bonds arise in soils exposed to long-term action of geostatic stress, which are
manifested by the appearance of over-consolidation stress even in normally
consolidated soils or by its increase in over-consolidated soils. This fact is attributed to
secondary consolidation, or the appearance of additional diagenetic bonds (such as
cementation). These bonds, sometimes also referred to as "cold welding“, are of
relatively fragile nature, and they may easily be broken even by mere sample unloading
to zero external load. Therefore the recommendation regarding settlement calculation
will be shown hereinafter.
2.1. Principles for the spread foundation design
The approach to the design of shallow foundations is in principle the same as the
approach suggested in 1966. There are also 3 design approaches, even so the
specification on the complexity of structures is not as detailed, it only specifies that a
modest superstructure is not sensitive to differential settlement and has sufficient safety
margin within plastic deformation area. Here for the first time the term – design
according to the 1st, 2nd or 3rd Geotechnical Category – is used, when:
x For the 1st Geotechnical Category the effects of the supposed service design
loading for basic combinations d,s are compared with tabled nominal values
of the bearing capacity of ground Rd,t:
d,s Rd,t (6)
x nd
For the 2 Geotechnical Category the effects of the supposed extreme loading
d,e for the most dangerous combination are compared with the design bearing
capacity of the ground Rd, which is calculated from tabulated nominal
mechanical properties recommended for different types of soils or are based
on mechanical properties, which are very well known for a certain region
(local standard characteristics).
x For the 3rd Geotechnical Category the effects of the supposed extreme loading
d,e for the most dangerous combination are compared with the design bearing
capacity of the ground Rd, which is calculated from standard mechanical
properties determined via tests.
In principle the procedure of calculation for 2nd and 3rd Geotechnical Category is
the same.
d,e Rd (7)
2.2. Soil classification system and indicative nominal soil characteristics
Czech standard SN 73 1001 (1987) distinguishes 5 types of gravel soils (G1 – G5), 5
types of sandy soils (S1 – S5) and 8 types of cohesive soils (F1 – F8), according to the
Figures 3 and 4. Selected indicative nominal soil characteristics are presented in Tables
1 and 2, where is Poisson’s ratio, unit weight and is the coefficient expressing the
ratio between Eoed and Edef , when Eoed = 1/ · Edef.
@seismicisolation
@seismicisolation
Figure 3. Classification of soils with grains smaller than 60 mm (according to SN 731001:1987)
Figure 4. Plasticity chart (according to SN 731001:1987)
@seismicisolation
@seismicisolation
Table 1. Indicative nominal soil characteristics for cohesive soils F1 – F4 (according to SN 731001:1987)
Soil type Property Consistency

Symbol Soft Firm Stiff Hard
– – Sr > 0.8 Sr < 0.8 Sr > 0.8 Sr < 0.8
, , [kNm-3] = 0.35; = 0.62; = 19.0
Edef [MPa] 5–10 10–20 12–24 15–30
determined by tests
cu [kPa] 40 70 70 70–80
MG
u [°] 0 0 10 12–15
cef [kPa] 4–12 8–16 16–32 16–24 determined by tests
ef [°] 26–32
, , [kNm-3] = 0.35; = 0.62; = 19.5
Edef [MPa] 4–8 7–15 10–12 18–25
determined by tests
cu [kPa] 30 60 60 60–70
CG
u [°] 0 0 10 12–15
ef [°] 24–30
, , [kNm-3] = 0.35; = 0.62; = 18.0
Edef [MPa] 3–6 5–8 8–12 15–15
determined by tests
cu [kPa] 30 60 60 60–70
MS
u [°] 0 0 10 12–15
ef [°] 24–29
, , [kNm-3] = 0.35; = 0.62; = 18.5
Edef [MPa] 2.5–4 4–6 5–8 8–12
determined by tests
cu [kPa] 30 50 70 70–80
CS
u [°] 0 0 5 8–14
ef [°] 23–27
Table 2. Indicative nominal soil characteristics for sandy soils S1 – S5 (according to SN 731001:1987)
Edef ef Factors to be
Soil [MPa] [°]
cef considered when
type
[kNm-3] ID = ID = ID = ID = [kPa] determining the
Symbol values in the range
0.33–0.67 0.67–1.0 0.33–0.67 0.67–1.0
SW 0.28 0.78 20 30–60 50–100 34–39 37–42 0 ID, w, % of gravel,
SP 0.28 0.78 18.5 15–35 30–50 32–35 34–37 0 particle shape,
S–F 0.30 0.74 17.5 12–19 17–25 28–31 30–33 0 angularity
SM 0.30 0.74 18 5–15 28–30 0–10 Amount of fine

particles and soil
SC 0.35 0.62 18.5 4–12 26–28 4–12 consistency
Table 3. Values of the bearing capacity Rd,t for F soils – valid for foundation depth D = 0.8 – 1.5 m and width
B 3m (according to SN 731001:1987)
Tabulated design bearing capacity Rd,t [kPa]
Soil type
Consistency
Symbol
Soft Firm Stiff Hard
MG 110 200 300 500
CG 100 175 275 450
MS 100 175 275 450
CS 80 150 250 400
ML, MI 70 150 250 400
CL, CI 50 100 200 350
MH, MV, ME 50 100 200 350
CH, CV, CE 40 80 160 300
@seismicisolation
@seismicisolation
2.3. Tabulated values of the bearing capacity Rd,t for different types of soils
Tabulated design values of the bearing capacity Rd,t, which are used for the design of
spread footings falling under the 1st Geotechnical Category, are shown in Tables 3 and
4.
Table 4. Values of the bearing capacity Rd,t for S soils – valid for foundation depth D = 1.0 m and for dense
sands (S1 – S3) or for consistency stiff to firm for S4 and S5 (according to SN 731001:1987).
Tabulated design bearing capacity Rd,t [kPa]
Soil type
Width of foundation B [m]
Symbol
0.5 1 3 6
SW 300 500 800 600
SP 250 350 600 500
S-F 225 275 400 325
SM 175 225 300 250
SC 125 175 225 175
2.4. Determination of the design bearing capacity Rd
The following general equation is used for the design bearing capacity Rd determination
for foundations with horizontal footing bottom, for both undrained and drained
conditions:
Rd = cd Nc sc dc ic + 1 d Nd sd dd id + 2 b/2 Nb sb db ib (8)
where:
1 and 2 – effective unit weight above or below footing bottom [kNm-3]. Note: For
drained conditions the ground water level should be taken into consideration
b – effective width or diameter of foundation [m]
Nc, Nd, Nb – bearing capacity coefficients [-]
d – foundation depth [m]
cd – design value of cohesion [kPa],
sc, sb, sd – shape factor [-]
dc, db, dd – depth factor [-]

ic, ib, id – load inclination factor [-]
The bearing capacity coefficients are determined from following equations:
Nc = (Nd – 1) / tan d valid for d > 0

Nc = 2 + valid for d = 0
Nd = tan2 (45° + d / 2). exp ( tan d) (9)
Nb = 1.5 (Nd – 1) tan d
where d is the design value of the angle of the internal friction.

The shape factors are calculated from the following equations:
sc = 1 + 0.2 b/l
sd = 1 + b/l sin d (10)
sb = 1 - 0.3 b/l
@seismicisolation
@seismicisolation
where b and l are the dimensions of a rectangular footing. For square or circular footing
b = l.
Depth factors are calculated from the following equations:
dc = 1+0.1 (d/b)
dd = 1+0.1 (d/b · sin 2d) (11)
db = 1
where b is the width or diameter of the footing.

Inclination factors are calculated from the following equations:
ic = id = ib = (1 – tan )2 (12)
where is the angle of inclination of the forces resultant from vertical. Note: For this
case also resistance in footing bottom should be taken into account. For > 30° an
individual approach is needed.
Note: For eccentric loading an effective footing area Aef = bef · lef is applied for the
design bearing capacity as well as for the shape and depth factors.
Design values of the angle of internal friction d and cohesion cd are determined
from the nominal values or local values (for 2nd GC) or from recommended values after
measured statistical data evaluation (for 3rd GC) divided by the material partial factor
for subsoil m .
d = / m, cd = c / m,c (13)
Material partial factor for is m, = 1.5 valid for 12° or

m, = / ( – 4) valid for > 12°
Material partial factor for c is m,c = 2
2.5. Determination of the serviceability limit state – limit state of deformation

The serviceability limit state calculations of deformation are used to prove that service
design loading of subsoil would not exceed allowable deformation of subsoil. Hence,
the superstructure settlement, either uniform or differential, would not cause
inadmissible superstructure deformation or such change of superstructure location that
would cause significant difficulty of the superstructure ordinary usage.
For the 1st geotechnical category the serviceability limit state is not assessed.
For the 2nd geotechnical category the settlement analysis is performed using the
tabled nominal values of subsoil properties (see e.g. Tables 1 and 2). If available, local
nominal values should be used instead.
For the 3rd geotechnical category the settlement analysis is performed using
nominal values of deformation characteristics determined during ground investigation.
The equation for the total settlement analysis fall in the “stress-strain” method
category, it assumes 1D deformation along with structural strength:
@seismicisolation
@seismicisolation
n V z ,i mi V or ,i
s ¦
i 1 Eoed ,i
hi (14)
where:
s – total settlement
n – total number of layers
z,i – vertical component of stress (surcharge) in the middle of layer i caused by the
bearing pressure overload at foundation level ol
mi – corrective coefficient of surcharge, which is for layer i determined based on
soil type from Tab. 5. and represents the influence of structural strength
or,i – original vertical geostatic stress in the middle of layer i
hi – thickness of layer i
Eoed,i – design oedometric modulus of subsoil layer i
Table 5. Values of corrective coefficient of surcharge m (according to SN 731001:1987)

Type of subsoil m
Highly compressible fine-grained soils
with deformation modulus Edef < 4 MPa
normally consolidated
soft or firm consistency 0.1
(all 3 conditions must be fulfilled)
Embankments and other bulk soils, subsoils subsequently surcharged and up to now unconsolidated
Rocks within classes R1, R2; sound Mesozoic and Cainozoic sedimental rocks within classes R4, R5
Fine-grained soils within F classes, whereby does not belong other coefficient m
Sands and gravels within types SW, SP, GW, GP under the groundwater level 0.2
Rocks within class R3
Sands and gravels within types SW, SP, GW, GP above the groundwater level
Sands and gravels within types S-F, SM, SC, G-F, GM, GC 0.3
Rocks within classes R4, R5 – except sound Mesozoic and Cainozoic sedimental rocks
Rocks within class R6 (eluvia) 0.4
Loess and loess loams above groundwater level, if it is possible to avoid their water saturation 0.5
In the above mentioned calculation model the vertical surcharge z is reduced for
each layer just to its effective component (z – m·or), which contributes to the
deformation. Note: SN 731001:1987 takes great care of the determination of the
vertical surcharge component z. Therefore it allows determining of this component
under selected point of flexible or rigid foundation.
Reducing the vertical surcharge just to its effective component the settlement
analysis is limited only to the real deformation zone. Deformation zone under the
foundation is limited area within which indispensable deformations are generated from
superstructure surcharge (see Fig 5). The Czech standard SN 731001:1987 also
defines limiting values for settlement, that is, both for final total average settlement (60
– 200 mm) and relative differential settlement (0.0015 – 0.006).
@seismicisolation
@seismicisolation
ex isting
Gr ound l ev el
pr oposed
F oundation
l ev el Vol
h1
h2
Ground e.g.
water
lev el m =0,3
m =0,1
Vor
Vz
m·Vor
Deformation
zone depth zz
Figure 5. Calculation model for the determination of total settlement (effective surcharge is hatched).
According to SN 731001:1987
3. Conclusion
The article summarises the approaches used in the Czech Republic for the design of
shallow foundations according to limit states in the period of last 45 years. It is obvious
that the focus is on partial safety factors for material (m), because from the limit states
point of view the probability approach to the subsoil properties is preferred. This partial
@seismicisolation
@seismicisolation
safety factor is however applied directly on the angle of internal friction and not on
the tan as it is now suggested in Eurocode 7. Thus, this approach allows direct use for
shape and depth coefficients calculations for the limit bearing capacity.
The described approach is therefore most similar to the Design Approach 3 (DA3)
in current Eurocode 7. The most meaningful way forward would hence be to propose to
used further the DA3, however in the newest proposal of National annex (yet not
approved) it is proposed to follow Design Approach 1. DA1 allows the use of not only
the already used principle, but also allows to design the foundation structure itself in
compatible way with structural Eurocodes.
The calculation model suggested for the bearing capacity of shallow foundations is
closest to the model proposed by B. Hansen (1970) with the modification to simplify
the calculation namely for inclination factors. The calculation model proposed in SN
731001:1987 for the settlement analyses is based on actual in-situ measurements and
correspondence between analysis and follow up monitoring and is much better
compared to those valid in between 1966 and 1987. The current version of described
standard SN 731001:1987 was officially cancelled in 2010 when Eurocode 7 was
fully implemented, however the calculation models described here are still applicable
in accordance to EN SN 1997-1 (Eurocode 7).
Even thought the original idea was to guarantee the design safety with probability
of about 10-4 up to now experience shows that this safety (probability of failure) is
rather in the range of 10-6. Since 1987 only 2 foundation failures are known to the
authors even if during this period about 1-2 million of spread foundations were
constructed. Discussion to the level of this indemnity is therefore possible, especially if
we know that probability of failure of other geotechnical structures is significantly
higher (e.g. for earth fill dams the probability of failure is 1:100).
References
SN 73 1001 (1966). Foundation of structures. Subsoil under shallow foundations, (In Czech – Základová
pda pod plošnými základy), ÚNM Praha, 64 p.
SN 73 1001 (1987). Foundation of structures. Subsoil under shallow foundations. (In Czech). ÚNM Praha,
75 p.
SN EN 1997-1 (2006). Eurocode 7: Geotechnical design – Part 1: General rules. (In Czech). NI Praha,
138 p.
Hansen, B. (1970). Bearing capacity, Danish Geot. Inst., Bull. No 28.
Havlí ek, J. (1978). Foundation settlement. (In Czech). Final Report of the Research Project C 52-347-018,
SG Praha.
Sey ek, J. (1995). Calculation of 2nd Limit State of Shallow Foundations. (In Czech), PhD thesis, Czech
Technical University in Prague.
Šimek, J., Jesenák, J., Eichler, J. and Vaní ek, I. (1990). Soil Mechanics. (In Czech). SNTL Praha, 388 p.
Vaní ek, I. (1982). Soil Mechanics. (In Czech). Publishing company of the Czech Technical University in
Prague, 331 p.
Vaní ek, I. and Vaní ek, M. (2008). Earth Structures in Transport, Water and Environmental Engineering.
Springer. 637 p.
@seismicisolation
@seismicisolation
@seismicisolation
@seismicisolation
Code Development
@seismicisolation
@seismicisolation
@seismicisolation
@seismicisolation
IOS Press, 2013
doi:10.3233/978-1-61499-163-2-243
AASHTO Geotechnical Design

Specification Development in the USA
Tony M. ALLEN,1
Washington State Department of Transportation, USA
Abstract. A complete, formal geotechnical design code of practice addressing all

aspects of geotechnical engineering really does not exist in the USA at the national
level. However, a geotechnical design code has been developed at the national
level in response to the needs of the structural engineers as part of the American
Association of Highway and Transportation Officials (AASHTO) specifications,
primarily focusing on structure foundations, buried structures, and retaining walls.
Since 1994, AASHTO began migrating from Allowable Stress Design (ASD) to
Load and Resistance Factor Design (LRFD), the USA equivalent to Limit States
Design (LSD). AASHTO fully adopted LRFD, at least in the transportation Sector,
in 2007, and ceased updating their Allowable Stress Design and Load Factor
Design specifications in 2002.
A key issue for the USA geotechnical community regarding geotechnical
design code has been to strike a balance between prescriptive minimum design
requirements and levels of safety (i.e., load and resistance factors) and the
flexibility needed by geotechnical engineers to apply engineering judgment to
address site specific issues and local experience.
Summarized is the historical development of geotechnical foundation design
code in the USA for transportation applications. With regard to the geotechnical
portions of the current AASHTO LRFD Bridge Design Specifications, the
development and selection of load and resistance factors is discussed. Key
considerations for geotechnical design specification development are identified.
Finally, gaps and future development needs for USA geotechnical design codes of
practice are presented.
Keywords. foundation design, limit states, LRFD, resistance factors, calibration.
Introduction
A complete, formal geotechnical design code of practice addressing all aspects of

geotechnical engineering really does not exist in the USA at the national level.
However, a geotechnical design code has been developed at the national level in
response to the needs of the structural engineering community as part of the American
Association of Highway and Transportation Officials (AASHTO) specifications,
primarily focusing on structure foundations, buried structures, and retaining walls in
transportation applications. While other national design codes contain limited
geotechnical specification guidance (e.g., the International Building Code – IBC), the
AASHTO LRFD design specifications, which are focused on the transportation sector,
contain the most complete geotechnical design specifications and are the most widely
1
State Geotechnical Engineer, Washington State Department of Transportation, P.O. Box 47365,
Olympia, WA, 98504-7365, USA; E-mail: allent@wsdot.wa.gov
@seismicisolation
@seismicisolation
244 T.M. Allen / AASHTO Geotechnical Design Specification Development in the USA
used. AASHTO began migrating in 1994 from Allowable Stress Design (ASD) to Load
and Resistance Factor Design (LRFD), using the Ontario bridge Limit States Design
(LSD) code, which was first developed in 1979, as a starting point. The LRFD
approach fits within the broader framework of LSD. The foundations portion of the
AASHTO LRFD Bridge Design Specifications was completely rewritten and published
in 2005 through a task force of Federal Highway Administration, State Transportation
Department, and consultant experts. AASHTO fully adopted LRFD in 2007, and
ceased updating their Allowable Stress Design and Load Factor Design specifications
in 2002.
State transportation agency specific geotechnical design codes are also becoming
available. These agency specific design codes are developed to provide state agency
specific implementation of the AASHTO LRFD specifications, to address areas of
geotechnical practice not covered in the AASHTO specifications, and to define
geotechnical practice for use in contracting methods such as design-build. Examples of
such state transportation geotechnical manuals that augment the AASHTO
specifications include the Washington State Department of Transportation (DOT)
Geotechnical Design Manual (WSDOT 2011), the Florida DOT Soils and Foundations
Handbook (FDOT 2012), and the KDOT Geotechnical Manual (KDOT 2007). A
complete listing of state DOT geotechnical design manuals is provided by FHWA
(2007).
A key issue for the USA geotechnical community regarding geotechnical design
code has been to strike a balance between prescriptive minimum design requirements
and levels of safety (i.e., load and resistance factors) and the flexibility needed by
geotechnical engineers to apply engineering judgment to address site specific issues
(Goble 1999). Another key issue is whether or not to maintain levels of safety used in
past successful practice versus the level of safety that theoretically should be used
based on reliability theory analysis. Finally, past geotechnical design practice has not
clearly separated the geotechnical limit states (e.g., service and strength) in a way that
is compatible with the categories of limit states used by the structural engineering
community. Moving from ASD to LRFD requires these limit states to be separated so
that the level of safety used in the design is properly matched with the limit state under
consideration.
This paper summarizes the development of geotechnical foundation design code in

the USA, focusing on transportation applications. With regard to the geotechnical
portions of the current AASHTO LRFD Bridge Design Specifications, the development
and selection of load and resistance factors is discussed, as well as how past ASD
geotechnical design procedures were adapted to LSD. Finally, gaps and future
development needs for USA geotechnical foundation design codes of practice are
presented.
1. Overview of AASHTO Design Code Development in the USA
The AASHTO specifications are intended to be a comprehensive design code for the
design of bridges, walls, underground structures, and other transportation sector
structures. The primary focus of the committee and the design code it produces is
structural design, and most of the committee members are structural engineers. The
design specifications are organized to be most efficient for the structural design of
bridges. However, geotechnical design specifications are included for foundations,
@seismicisolation
@seismicisolation
T.M. Allen / AASHTO Geotechnical Design Specification Development in the USA 245
walls, and underground structures, primarily in chapters 10 (foundations) and 11

(abutments and walls). Portions of Chapter 3 (loads and load factors) and Chapter 12
(buried structures and tunnel liners) also include geotechnical design provisions, both
for static and seismic loads.
Beginning in 1997, geotechnical engineers from the state departments of
transportation were added to the committee as members, primarily on the technical
committee that specifically focuses on development of the foundation and wall design
chapters (i.e., chapters 10 and 11, and part of Chapter 3) of the AASHTO LRFD Bridge
Design Specifications. For those state transportation departments that do not have
geotechnical engineers specifically appointed to the committee, the state’s structural
engineering representative on the committee is expected to get input from their
respective geotechnical departments to help review the specifications and recommend
voting decisions when changes to the geotechnical portions of the specifications are
proposed. New design specifications, and any changes made to existing specifications,
are developed by or through the technical committee responsible for the subject area (in
the case of foundations and walls, the T-15 Technical Committee is the one responsible
for this subject area) and then submitted by that technical committee to the full Bridge
and Structures Subcommittee. Note that a two-thirds majority of the voting members
(one vote per state) of the full Subcommittee is required to approve changes to the
design specifications. For additional information regarding the subcommittee, its
structure, and its activities, see the website for the AASHTO Bridge and Structures
Subcommittee (AASHTO 2012b).
The AASHTO LRFD specifications (AASHTO 2012a) include mandatory design
code in the left column of each page and commentary in the right column of each page.
The AASHTO design specification provisions (referred to as “articles”) and associated
commentary are typically developed from what is considered widely accepted design
practice. For example, geotechnical design methodologies found in national design
manuals (e.g., FHWA manuals), textbooks, and peer reviewed journal publications and
research reports are typically used as the basis of the provisions included in the
AASHTO specifications. In some cases, research specifically targeted at development
of the AASHTO LRFD specifications is used to develop the design specifications,
subject to review and revision conducted by the responsible AASHTO Bridge and
Structures technical committee, in this case T-15, and the subcommittee as a whole.
The technical committee will also typically obtain input from a wide range of private
sector organizations that would have an interest in the design specifications being
considered, as well as academia, before finalizing the specification changes to be sent
on to the full subcommittee for final review and voting.
The AASHTO design specifications are intended to represent what the state
departments of transportation engineering leadership consider to be the standard of
practice. However, due to the political structure of the USA, being made up of
individual states, each state can decide how it will implement the AASHTO design
specifications in their state. That adoption is usually addressed through state bridge
and/or geotechnical design manuals or design policy memorandums that explain and
augment how the AASHTO design specifications will be applied for transportation
projects in their respective states. State geotechnical design manuals and other
documents may also include design requirements for issues not specifically addressed
in the AASHTO manual. Therefore, it is the combination of the AASHTO LRFD
Bridge design specifications with the bridge and geotechnical design manuals or policy
memorandums in each state transportation department that define the standard of
@seismicisolation
@seismicisolation
practice for the transportation sector. Note that this description is not intended to be a
legal definition of “standard of practice,” but is simply a general description of what is
typical of state transportation department engineering practice. State transportation
departments may also directly use national manuals produced, for example, by the
Federal Highway Administration (FHWA) to augment the AASHTO LRFD Bridge
Design Specifications.
Detailed geotechnical design specification development has been slower than the
development of the structural design specifications. Geotechnical engineers in the
USA have been reluctant to develop codes to govern their design practice due to fear of
being too prescriptive and inflexible to apply engineering judgment to adapt the design
practices available to site-specific conditions (Goble 1999; DiMaggio, et al., 1999).
This reluctance is especially strong with regard to the selection of the level of safety to
use (i.e., the factor of safety, or load and resistance factors), as well as how to apply
ASD geotechnical design procedures to LSD based design (Goble 1999; Christian
2003). Furthermore, national design codes may not be able to accommodate the design
practices successfully used locally.
Practicing geotechnical professionals typically rely heavily upon past successful
practices. These past practices have defined the levels of safety needed to have a
successful design. Furthermore, these past ASD practices tended to combine multiple
limit states into one calculation. LSD based design codes require a clear separation
between the various limit states so that the level of safety used can be more accurately
targeted to the consequences of failure and the likelihood a given load combination will
occur. Because of these issues, geotechnical engineers have been slow to accept and
use LSD based design codes such as the AASHTO LRFD Bridge Design Specifications,
though at this point, LSD based design has become more widely accepted and used.
2. Design Approach Used to Achieve Desired Level of Safety in the AASHTO

LRFD Bridge Design Specifications
The AASHTO specifications use load and resistance factors (i.e., LRFD) to account
and design for the various sources of variability in the applied loads and the resistance
to those loads available when designing structures such as bridges and walls. The
general approach used to address the level of safety needed and how the North
American (e.g., USA) approach differs from the European approach is illustrated in
Figure 1. In the LRFD approach, individual loads within a given load group are
increased using a load factor to account for the uncertainty in each load. Then the total
resistance available to resist the applied loads is reduced using a resistance factor to
account for uncertainty in the resistance available. Figure 1 also illustrates the
difference between the LRFD and ASD approaches.
Eq. (1) illustrates the North American LSD approach.
¦ Q i ni d I Rn (1)
where, Ji is a load factor applicable to a specific load type Qni; the summation of JiQni
terms is the total factored load for the load group applicable to the limit state being
considered; I is the resistance factor; and Rn is the nominal unfactored resistance
available (either ultimate or the resistance available at a given deformation).
@seismicisolation
@seismicisolation
Design Rd Qd Q
Model
C, Is C/fc, Isfc Q x Jf
(a) European LSD approach (Design Models 1 and 3)
(b) North American LSD approach

Rn
Design Rn/FS
Model Q
C, Is Safety
Factor, FS >
(c) North American ASD approach

Figure 1. Approaches to address level of safety needs in design.
For geotechnical design, three categories of limit states are typically considered.
These limit state categories include service, strength, and extreme event. Of these,
there are several subcategories of limit states that affect which specific set of loads (i.e.,
load groups) are considered. For geotechnical design, the primary subcategories within
the AASHTO design specifications include Service I, Strength I (most common for
geotechnical design) and IV (used for long span bridges where dead load dominates),
and Extreme Event I (seismic) and II (vessel or vehicle collision and scour). The
details of these load groups for these limit states are provided in AASHTO (2012a),
specifically Article 3.4.1, and especially tables 1 and 2 in that article.
Properly combining these groups of loads for design purposes can be complicated,
as some loads require the use of either a maximum or minimum value. The selection of
the maximum or minimum value depends on whether the load is contributing to the
instability of the element or feature being designed (in this case the maximum load
factor is used) or contributing to the stability of the element or feature being designed
(in this case the minimum load factor is used). The principle is to select combinations
of maximum and minimum load factors to produce the most extreme factored force
effect. Many of the geotechnical load factors provided in the AASHTO specifications
have both maximum and minimum values. For examples of how maximum and
minimum load factors have been combined for specific geotechnical design situations,
see WSDOT (2011) and Berg, et al. (2009).
As illustrated in Figure 1 and Eq. 1, each load typically has its own load factor
associated with it to address the uncertainty in each load. For application of load
factors, forces are not broken down into vertical and horizontal components to apply
maximum and minimum load factors to the components of the single load that
contribute to instability or stability of the element, respectively – only a single load
factor is applied to the single resultant force for each source of load. For resistance, a
@seismicisolation
@seismicisolation
single resistance factor is typically applied to the available resistance to address

uncertainty, though there are cases where different resistance factors are applied to
different components of the resistance. An example of this is drilled shaft design, in
which the side resistance has a different resistance factor from the base resistance.
Phoon, et al. (2003) discuss the various approaches to applying resistance factors and
the advantages and disadvantages of each approach.
It is the combination of load and resistance factors that determines the level of
safety in the design. For LRFD, the level of safety is captured through a reliability
index, E, which can also be represented by a probability of failure, Pf. For a given
category of limit states (e.g., the strength limit state), the target reliability index, EW,
used to establish the load and resistance factor combination is consistent for all limit
states within that category. The various categories of limit states (e.g., service, strength,
extreme event) represent differences in both the consequences of failure and
differences in the probability that a given loading will occur within the specified design
life of the structure. For example, the strength limit states consider loads that have a
high probability of occurring (e.g., static loads due to structure weight) and a failure
consequence that is high due to the potential for collapse, since strength limit state
design is focused on the ultimate strength of the component or supporting soil. The
consequences of a service limit state failure (e.g., excessive deformation, but no
collapse and subsequent loss of life) may be less than the consequences of a strength or
extreme event limit state occurrence. However, for extreme events, the consequences
of failure are high and similar to those associated with the strength limit state, yet the
load combination (e.g., combinations that include earthquakes) may have a very low
probability of occurrence in comparison to the strength limit state load combination.
Because of the low probability of occurrence of the load combination, more severe
consequences of failure and a less stringent failure criterion can be acceptable for
extreme event limit states.
The AASHTO Bridge Subcommittee has determined that the strength limit states
should target a E value of 3.5 (approximately 1 in 5,000 probability of failure), not
considering redundancy. The level of safety in the foundation units supporting the
bridge piers or other structures must be at that level. However, foundation units may
contain multiple foundation elements that provide some redundancy in the foundation.
The foundation elements (e.g., an individual pile or shaft in multiple pile or shaft
foundation units) do not necessarily need to be designed to the 1 in 5,000 probability of
failure, since it is the level of safety in the foundation unit as a whole that is most
important to be consistent with the level of safety in the structure the foundation unit
supports, assuming that the load in the foundation elements has the ability to
redistribute should one of the elements not perform as expected. The AASHTO LRFD
Bridge Design Specifications (specifically AASHTO 2012a – Article C10.5.5.2.1)
allow individual elements within a foundation unit to be designed for a higher
probability of failure than the 1 in 5,000 the structure must meet for the strength limit
state.
For service and extreme event limit states, the target level of safety to use is under
development. Currently, load and resistance factors for service and extreme event limit
states are near 1.0 to reflect differences in failure consequences or probability of
occurrence of the load combination relative to the strength limit state.
Additional information on the selection of a EW for various limit states is provided
in Allen, et al. (2005). Additional perspectives on this issue are also provided by
Phoon, et al. (2003). Recommendations on the effect of failure consequences on the
@seismicisolation
@seismicisolation
selection of the target reliability level are also provided by ISO (1998) and Santamarina,
et al. (1992). Reliability levels needed for various types of limit states are also
discussed by Meyerhof (1994). In general, there appears to be agreement that EW for
service and extreme event limit states should be lower (i.e., higher Pf) that what is used
for the strength limit state, but how much lower is yet to be determined. Allowance of
a higher Pf (lower EW) for service and extreme event limit states will result in load and
resistance factors that are closer to 1.0 than is the case for the strength limit state.
3. Historical Development of Geotechnical LRFD Foundation Design
With regard to the AASHTO specifications, a fairly complete history of the

development of geotechnical load and resistance factors for foundation design up
through 2005 is provided in Allen (2005). Key points from that historical development
summary are provided herein. Geotechnical LRFD, or Limit States Design, was
developed in the USA with consideration to three main issues: 1) the separation of
geotechnical design procedures into clearly defined limit states, 2) definition of what is
a load and what is a resistance, especially when dealing with geotechnical elements
contained within a continuum and soil-structure interaction problems, and 3) the
establishment of load and resistance factors to meet a level of safety that is consistent
with the rest of the AASHTO LRFD Bridge Design Specifications yet consistent with
past design practice. Not all of these issues have been completely resolved, but
sufficient progress regarding each of these has been made to develop a useable design
code.
3.1. Definition of Geotechnical Limit States
The combination of more than one limit state into a single design procedure is fairly
common in US geotechnical engineering practice. For example, Peck, et al. (1974)
provide spread footing design charts that are partially based on the limit to prevent
shear failure of the soil in bearing and partially on excessive settlement. Similarly,
presumptive footing bearing capacities for various common soil or rock conditions are
typically based on settlement limited maximum values. Because of this past practice,
the typical geotechnical engineer had little experience with bearing resistance values
greater than these settlement limited values. Though well accepted bearing capacity
theory such as described in Munfakh, et al. (2001) could be used to calculate much
higher bearing resistance values that are representative of ultimate conditions,
geotechnical engineers rarely took advantage of higher bearing values.
It is important to keep in view the definition of each limit state when attempting to
address this issue. For example, for the strength and extreme events limit states,
definition of the limit state is typically exceedance of the maximum resistance available
by the applied loads (ISO 1998). Exceedance of this limit state usually results in
collapse of the considered structure or structure components, or at least large
uncontrolled and unpredictable deformation which contributes to the collapse of the
structure. For the service limit state, definition of the limit state is exceedance of a
defined acceptable deformation to insure functionality of the considered structure or
structure component. Exceedance of this limit state usually does not result in collapse
@seismicisolation
@seismicisolation
of the structure but may require eventual repair or replacement of the considered
structure due to reduced life or functionality (ISO 1998).
Considering these limit state definitions, past design practices must be refitted to
accommodate the separation of the limit states contained within LSD based design
procedures. For example, in past ASD practice, for very dense foundation soil, the
allowable bearing resistance would be typically limited to 500 to 600 kPa. However,
bearing capacity theory, which is used to estimate the bearing resistance at shear failure
of the foundation soil, a calculation applicable to the strength limit state, could result in
a maximum bearing resistance value of 2,500 to 5,000 kPa. Even using typical LRFD
resistance factors for bearing resistance at the strength limit state (i.e., 0.50 as specified
in AASHTO 2012a, Article 10.5.5.2.2), the strength limit state bearing resistance
allowed for the strength limit state is much higher than previous geotechnical
experience would allow. To provide the needed separation, settlement limited bearing
resistance design methods have been included in the service limit state articles of the
specifications, and ultimate bearing capacity methods such as described in Munfakh, et
al. (2001) have been included in the strength limit state articles, as the load and
resistance factor combination for the strength limit state would result in excessively
conservative designs if applied to settlement limited bearing resistance design methods.
A related issue with regard to the blurring of the limit states in geotechnical design
is how ultimate values are defined. For the strength limit state, failure is defined by the
ultimate resistance, which typically results in shearing of the soil whereas for the
service limit state, failure is defined by a maximum allowed deformation (ISO 1998).
However, in a number of cases, the ultimate geotechnical resistance of a foundation
element is defined by a deformation criterion. In such cases, is the ultimate
geotechnical resistance really a deformation limited resistance and therefore
appropriate for a service limit state design and not a strength limit state design?
For example, for pile foundations, ultimate bearing resistance of a pile is often
defined using a criterion such as that proposed by Davisson, as shown in Figure 2. In
the figure, Qf is the failure load, F is the total pile settlement at failure, Lp is the pile
length, A is the pile cross-sectional area, and E is the pile modulus. The current
AASHTO LRFD specifications specifically refer to this method (AASHTO 2012a).
For a 300 mm diameter pile, this criterion would define ultimate bearing resistance as
the resistance at approximately 33 mm of vertical displacement for a typical bearing

resistance for a pile of this diameter. Yet for service limit state design, the resistance
allowed would typically be defined at a vertical displacement of 25 mm. Therefore, in
this example, the strength limit state ultimate bearing resistance would occur at almost
the same displacement as the service limit state resistance.
It can be observed from Figure 2 that the deformation based failure criterion does
not capture the true ultimate bearing resistance of the pile, in that there is significant
reserve capacity beyond the bearing value defined as failure (i.e., Qf). However,
current design practice assumes that this value is an ultimate resistance, and any
additional resistance available beyond this point is assumed to not be reliable or useful.
A similar problem occurs with the design of drilled shafts for bearing resistance.
Ultimate end bearing resistance for drilled shafts is typically defined as the resistance at
a deformation of 5 percent of the shaft diameter (Brown, et al., 2010). However,
drilled shaft bearing resistance is complicated by the fact that side resistance reaches its
ultimate value at a much smaller deformation (e.g., approximately 0.5 percent of the
shaft diameter). These deformation based ultimate bearing resistance definitions
@seismicisolation
@seismicisolation
0.0
ELASTIC COMPRESSION
OF PILE QL p
5.0
AE
PILE SETTLEMENT, MM
10.0 Qf
X
F
15.0 FAILURE
CRITERION
20.0 (X = 3.8 + B/36,600
B = Pile Dia., MM)
25.0
30.0
35.0
0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300
APPLIED LOAD, Q, KN
Figure 2. Davisson criteria for pile foundation load test interpretation (adapted from Davisson 1972).
complicate the determination of ultimate shaft bearing resistance for the strength and
extreme event limit states.
For deep foundations, full scale load tests are the most common way of obtaining
measured ultimate bearing resistance values. While a plunging failure, which comes
closest to a true ultimate bearing resistance, can occur, more commonly the load test is
considered to have reached its ultimate value once a predefined deformation limit has
been exceeded, such as shown in Figure 2, as the foundation resistance will continue to
increase as the deformation increases, though at a decreasing rate.
It has been decided by the AASHTO specification writers to continue with these
deformation based definitions of ultimate foundation bearing resistance to be consistent
with past design practice and to keep the ultimate resistance practical to achieve and
determine. Furthermore, the foundation load test results used in the calibration of
resistance factors have used these deformation based definitions of ultimate resistance.
Therefore, the strength limit state resistance factors have been derived using
deformation based definitions of ultimate resistance.
3.2. Separation of Loads from Resistance in Geotechnical LSD
For some aspects of geotechnical design, it is not always clear how to separate and
separately factor the applied loads and available resistance (Goble 1999), or to separate
destabilizing forces from restoring forces when considering when to use maximum
versus minimum load factors. DiMaggio, et al. (1999) found that this issue has
@seismicisolation
@seismicisolation
affected the development of LSD in Europe as well as in the USA, and that the solution
to this problem is neither simple nor forthcoming. Handling slope stability analysis in
LSD is one of the most problematic examples of this. Furthermore, most slope stability
programs in use in the USA are designed to conduct the analysis in a way that produces
a slope stability safety factor.
The AASHTO specifications have accommodated this problem by having slope
stability analyses conducted within the service limit state, even though in concept,
slope stability analysis and design are really strength limit or extreme event limit state
activities. The reason this has been done is that for the service limit state, at least
currently, all load factors applicable in the slope stability analysis situation are 1.0.
This eliminates the need for maximum and minimum load factors and the need to
differentiate destabilizing forces from restoring forces. Furthermore, for slope stability
design, a resistance factor significantly less than 1.0 is used to provide the level of
safety needed, and the inverse of the safety factor produced by the slope stability
analysis program is equated to the target maximum resistance factor needed.
Essentially, this is a “work-around” until a better approach to address this issue in the
context of LSD is developed.
Monte Carlo simulation can be applied to slope stability analysis, which can
account for variability in the input parameters. The Monte Carlo technique does not
specifically account for any model uncertainties, though it is likely that model
uncertainties will be fairly small for typical slope stability situations. The exception to
this is slopes (or walls) with reinforcement elements, as limit equilibrium based
methods may poorly model such situations (Rowe and Ho 1993; Allen and Bathurst
2002). The Monte Carlo approach could be used to more directly evaluate the
reliability of a slope stability design (WSDOT 2011), at least without reinforcement
elements, though not in the format currently used in AASHTO’s version of LSD.
Soil-structure interaction problems can also be problematic with regard to their
application to LSD. In such cases, the soil loads and stiffness affect the structure loads,
making it difficult to decide how to factor the applied loads and available resistance to
produce a safe result. Lateral loading and analysis of deep foundations is an example
of this. The geotechnical portion of lateral load analysis includes the determination of
the soil springs used in the structural modeling. Soft soil springs allow more
deformation in the structure, whereas stiff soil springs attract more load into the
structure. Either case can produce the critical design case, depending on the limit state
being considered and the structure details. This has hindered the development of a soil
strength limit state for lateral loading of foundation elements, an issue that is currently
under discussion. For structural design of the foundations and the structure the
foundations support, the distribution of loads between the foundations and above
ground structure is assessed by not factoring the loads applied to the foundations
(AASHTO 2012a; WSDOT 2011). Once the load distribution is determined, then the
loads applied to each structural element are factored and equated to the factored
resistance in each structural element (see WSDOT 2011, specifically Section 8.6.1).
3.3. Establishment of Load and Resistance Factors for Geotechnical Design
Early development work to determine load the resistance factors for geotechnical
design heavily relied upon a technique known as calibration by fitting to ASD.
Essentially, this technique provides a way to determine load and resistance factors that
will produce the same level of safety as has been used in ASD past practice.
@seismicisolation
@seismicisolation
Calibration by fitting to ASD conducted for the case where two load sources are
considered uses the following equation:
J DL DL LL J LL
I (2)
DL 1 FS
LL
where, I the resistance factor, JDL is the load factor for the dead load, JLL is the load
factor for the live load, DL/LL is the dead load to live load ratio, and FS is the
allowable stress design factor of safety. As can be seen from this equation, there is no
consideration of the statistical parameters associated with the loads and the resistance,
and therefore no consideration of the margin of safety inherent in the FS.
Note that this equation simply provides a way to obtain an average value for the
load factor considering the use of two different load factors from two different sources,
and considering the relative magnitudes of each load. In the most basic terms, the ASD
FS is simply the average load factor divided by the resistance factor. There is no
consideration of the actual bias or variability of the load or resistance prediction
methods when “calibration” by fitting to ASD is used, nor is there any consideration to
the probability of failure, Pf. All that is being done here is to calculate the magnitude
of a resistance factor, for a given set of load factors, that when combined with the load
factors, provides the same magnitude of FS as is currently used for ASD.
The key benefit of this technique is that it is useful for providing benchmark load
and resistance factors that are representative of past design practice. Problems with this
technique include the following:
x Past practice may vary widely across the USA (i.e., when establishing national
specifications, which past practice should be used to establish load and
resistance factors?),
x The actual level of safety remains unknown and is not likely to be consistent
for the various aspects of design, and
x Perception may have a role in defining the level of safety implied by
successful past practice. Factors affecting this perception include the potential
for built-in conservatisms in how the design procedures are applied that are
not captured by the ASD safety factor used (e.g., some practices may base
“ultimate” values using severely limited deformations), and use of design
procedures that have a bias toward producing excessively conservative
predictions.
The preferred approach is to use reliability theory to establish load and resistance
factors if adequate data are available to establish statistical input parameters. Through
reliability theory calibrations, load and resistance factors can be established that
provide a consistent level of safety. The general process used to perform calibration
using reliability theory is to:
x Gather the data needed to statistically characterize the key load and resistance
random variables, developing parameters such as the mean, COV, and
distribution type (e.g., normal, lognormal),
@seismicisolation
@seismicisolation
x Estimate the reliability inherent in current design methods using statistical

data gathered that are characteristic of the variability and bias in the design
methods and input parameters used,
x Select a target reliability based on the margin of safety implied in designs used
in previous practice, considering the need for consistency with the reliability
used in the development of the rest of the AASHTO LRFD specifications, and
considering levels of reliability for geotechnical design as reported in the
literature, and
x Determine load and resistance factors consistent with the selected target
reliability.
Two basic calibration approaches have been used to develop the load and
resistance factors for the geotechnical portions of the AASHTO specifications. The
calibrations conducted by Barker, et al. (1991) are an example of one approach, in
which they developed statistics for key input parameters such soil friction angle, soil
unit weight, and other parameters, attempting to characterize inherent spatial variability,
systematic error, and design model error. Design model error is usually obtained from
measurement of the performance of full scale structures, or possibly bench scale
laboratory model data, and represents the ability of the design model to make accurate
predictions of load or resistance (i.e., how well does theory match reality?). However,
it is not usually straight-forward to separate the other sources of error from the model
error, as the model error may also contain error from spatial and systematic sources
(Phoon 2005). The calibrations conducted by Paikowsky, et al. (2004) are an example
of the other approach, in which these three sources of error are considered to be
contained with the statistics used. Allen, et al (2005), Allen (2005), and Phoon (2005)
provide additional details regarding these two approaches and how they are carried out,
and examples that illustrate these two approaches and the pitfalls that can occur, as well
as the application of the calibration results to the determination of the load and
resistance factors provided in the AASHTO LRFD specifications.
Load factors were established through the original calibration work done for the
overall LRFD bridge design specifications (Nowak 1999; Kulicki, et al. 2007). They
were in general not modified for adoption to geotechnical design; instead, the
calibration effort focused on modifying the resistance factors needed to meet overall
level of safety requirements (Allen 2005).
Barker, et al. (1991) also investigated the probability of failure, Pf, implied by the
safety factors used in past design practice, with consideration to the statistical data
gathered. They discovered that the Pf was higher for foundation systems with many
elements (e.g., pile foundations) than for systems with fewer elements, such as shaft
and footing foundations. This was considered when establishing the target reliability
index, EW, to use for the various foundation types to determine the resistance factors
needed (Isenhower and Long 1997; Allen 2005; Allen, et al. 2005; AASHTO 2012a).
Allen (2005) summarized the historical progression of resistance factor
development for footings, drilled shafts, and driven piles, considering both calibration
by fitting to ASD and reliability theory calibrations. Initially, the available data
suitable for conducting reliability theory calibrations were rather limited. Since those
initial calibrations, additional database development and calibrations were conducted
for some foundation types and design methods. Also note that initial LRFD
calibrations conducted by Barker, et al. (1991) used load factors from an older
@seismicisolation
@seismicisolation
AASHTO design code. Allen (2005) updated those calibrations using the newer LRFD
load factors (described in the table as “current practice”).
A portion of the resistance factor development for drilled shafts as reported by
Allen (2005) is summarized in Table 1. The load factors used in the initial calibrations
as well as the calibrations used to determine the resistance factors currently provided in
the AASHTO specifications are summarized in Table 2. See Allen (2005) for
additional calibration result summaries for various foundation types.
While reliability theory calibrations have the potential to produce more accurate
and consistent results, the final resistance factors selected in the AASHTO LRFD
specifications for the various foundation types and limit states did not necessarily fully
rely on the reliability theory calibration results. In general, estimation of geotechnical
resistance, and loads, can involve considerable engineering judgment, and this
judgment is difficult to quantify statistically. Furthermore, the available data upon
which the needed input statistics were based were limited in terms of quantity and
quality in some cases. In other cases, considerable high quality data were available to
develop the needed statistics for reliability theory input. Therefore, the final selection
of resistance factors considered both the statistical reliability of the method and the
level of safety implied by past successful design practice. The degree of reliance on
past successful design practice depended upon the quality of the database available, and
how well that data modeled reality. In addition, the calibration studies summarized by
Allen (2005) were conducted over a considerable period of time, and changes and
upgrades in design procedures have occurred since those studies were conducted.
Judgments were made regarding how the available calibration results apply to the
newer design methods.
There is some inherent reliability in past successful design practice and the safety
factors used in those design practices. Therefore, if it is decided to deviate from those
past successful design practices, there must be a strong reason for doing it. A large and
reliable database coupled with reliability theory could be an adequate reason for
deviating from past practice, especially if it is recognized that past practice has been
excessively conservative, or if past practice has resulted in a higher than acceptable
failure rate. This philosophy was used in the establishment of the resistance factors for
design in the AASHTO LRFD Bridge Design Specifications.
In summary, the AASHTO LRFD design specifications load and resistance factors
were determined using calibration by fitting to ASD, reliability theory in cases where
adequate data to develop statistics were available, or a combination of the two (Allen
2005; AASHTO 2012a), primarily based on the original work by Barker, et al. (1991)
and Paikowsky, et al. (2004). For those cases where the selected resistance factor was
developed through strong reliance on the reliability theory calibration results, average
measured soil properties were used to determine predicted values in those calibrations.
This means that theoretically, average soil/rock property values could be used in
combination with those reliability theory derived resistance factors to achieve the
desired level of safety. However, if the amount of property data is not adequate to
reliably establish average design parameters, more conservative property values should
be considered.
@seismicisolation
@seismicisolation
256
Table 1. Summary of calibration results for drilled shaft bearing resistance (after Allen 2005).
I from
I from I I from
I from Reliability
Calibration Recom- Calibration I from Reliability New
Strength Condition Reliability ASD FS Theory
Design ASD FS by fitting to mended by fitting to Theory, Recom-
T.M. Allen / AASHTO Geotechnical Design Specification Development in the USA

Limit and Theory Current (NCHRP
Method Used ASD in ASD Recommended in mended
State Location (NCHRP Practice 343 Stats.),
(NCHRP NCHRP (Current NCHRP 507 I
343) Updated to
343) 343 Practice)
Current J’s
0.24 to 0.28,
Side D-method
2 depending on constr.
Bearing Resistance (Reese and 2.5 0.61 0.65 2.5 0.55 0.455
method for clay (0.30
in clay O’Neill 1988)
0.72 recommended)
(0.674) 0.604 0.24 to 0.28,
Base Total Stress
depending on constr.
Bearing Resistance (Reese and 2.75* 0.55 0.55 2.75* 0.50 0.405
method (0.30
in clay O’Neill 1988)
recommended)
0.25 to 0.73,
Side E-method
depending on constr.
Bearing Resistance (Reese and 2.5 0.612 - 2.5 0.55 0.555
method (0.40
in sand O’Neill 1988)
recommended)
- - 0.25 to 0.73,
Base
Reese and depending on constr.
Bearing Resistance 2.75* 0.55 - 2.75* 0.50 0.505
O’Neill 1988 method (0.40
in sand
recommended)
Side and
Base 0.52 to 0.69, 0.55 for
Resistance Reese and depending on constr. side,
Bearing - - - - 2.5 0.55 -
(sand/clay O’Neill 1988 method (0.50 to 0.70 0.50 for
mixed recommended) base5
profile)
*Implied in NCHRP 343, due the logic that there is greater uncertainty in the base resistance due to the need for greater deformation to mobilize base resistance.
2
The value shown in the table is slightly lower than the value shown in Appendix A of Barker, et al. (1991). The difference appears to be the result of using a DL/LL ratio of 2.0
rather than 3.0 when performing the calibration by fitting to ASD.
4
Comparitive or other calibrations conducted by the writer using reliability theory (Monte Carlo Method).
5
The recommended resistance factor is applicable to both the Reese and O’Neill (1988) and the O’Neill and Reese (1999) methods.
@seismicisolation
@seismicisolation
Table 2. Load Factor Design (LFD) and LRFD load factors used in resistance factor calibration (after Allen
2005).
LFD Load Factors
2004 LRFD
(Used in the NCHRP
Load Factors
343 Calibrations)
Dead load, JDL 1.30 1.25
Live load, JLL 2.17 1.75
Since 2005, additional resistance factor evaluations were conducted for driven
piles. One reason for this was that the resistance factors for pile design, which were
based in part on the work by Paikowsky, et al. (2004), were more conservative than had
been used in many years of past practice. Pile foundation design using that past
practice have performed very well with virtually no failures or even poor performance.
Therefore, some of the resistance factors for pile foundation design were increased
somewhat to be more consistent with that past successful practice.
4. Future Geotechnical Development Activities and Needs for the AASHTO LRFD
Specifications
Since the AASHTO LRFD Bridge Design Specifications became mandatory in 2007, a
number of research projects to develop load and resistance factors have been funded
and several of them completed. Several of these have been funded by state
transportation departments in an effort to determine how to adopt the LRFD
specifications to local practices. Others have been funded at the national level to fill
gaps in the AASHTO specifications with regard to geotechnical load and resistance
factors. Examples include Allen (2007), Smith, et al. (2011), Basu and Salgado (in
press), Paikowsky, et al. (2010), Paikowsky, et al. (2005), Schneider (2009), Yang and
Liang (2009), TRB E-Circular E-C136 (2009), and Long, et al. (2009).
A key hindrance to continued development of load and resistance factors as new
data become available or as new methods are developed is that the information
contained in the databases developed is not complete, making the data less useful for
conducting future calibrations. In most cases, foundation design calibration work and
the databases generated and published have not included enough detail for each case
history to allow the calibrations to be updated once new or improved design methods
are developed. This creates the need to re-gather the details of the case histories used
in previous calibrations to accomplish calibrations of new or updated design procedures,
as model uncertainty and bias for each design methods is unique.
The AASHTO Bridge Subcommittee, at least informally, adopted a recommended
practice for conducting calibrations, the development and documentation of the
databases generated used as the basis for these calibrations, and the selection of load
and resistance factors based on such work. It should be anticipated that most future
calibration efforts will follow these recommended practices, especially for research
funded through the AASHTO Bridge Subcommittee, which will help to improve the
database completeness problem. Those recommended practices are contained in
NCHRP Report 20-07/186 (Kulicki, et al., 2007). This report heavily references TRB
Circular E-C079 (Allen, et al. 2005) for the details of calibration procedures and
database content. Since those documents were produced, Bathurst et al. (2008) and
@seismicisolation
@seismicisolation
Bathurst et al. (2011) provided some refinements to the reliability theory calibration
approach that should be used, and they should be considered as an update to the TRB
Circular. Another important issue that is yet to be addressed is how to give credit to
the amount and quality of the site specific data available when selecting load and
resistance factors. This issue is partially addressed for pile foundation design, in that
higher resistance factors can be used if dynamic pile tests or pile load tests are
conducted (AASHTO 2012a, Zhang 2004, Allen 2005). This issue is not adequately
addressed for other foundation systems or situations. Most of the calibration effort has
focused on the strength limit state. Efforts for other limit states (e.g., service, extreme
event – seismic) are on-going and needed.
5. Concluding Remarks
The development of LSD in USA geotechnical practice has come far in the last 15
years since it was first implemented in 1994. Even 10 years after LRFD was first
adopted, at least in the transportation sector, geotechnical engineers were reluctant to
move forward with LSD (Goble 1999; Christian 2003). However, since that time, most
state transportation departments as well as the Federal Highway Administration have
adopted LSD, termed LRFD in the USA, for their geotechnical design practices. While
initially, load and resistance factors were developed to provide a level of safety that
was consistent with past practice, reliability based approaches have been used to start
developing load and resistance factors that better reflect a level of safety that is
consistent with the structures the foundations support. Geotechnical engineers are now
gaining the confidence they need to start using load and resistance factors that may be
more or less conservative than the equivalent level of safety they have used in past
practice, educating themselves using what is available in the literature or in university
level classes on the subject that are now becoming available, or having research
conducted that demonstrates how LRFD should be applied to their local geotechnical
practices. As indicated herein, a significant number of studies to develop load and
resistance factors for use in LRFD have already been published, and the amount of
research being conducted on this subject is growing, both for public and private sector
application.
What is most important now is to make sure that the databases created to
accomplish this work are consistent in the level of detail included, so that future
calibration efforts can build on the previous efforts. Not only can these databases be
used to improve the effectiveness and consistency of the load and resistance factors
developed, but they can also be used to improve the accuracy of the design methods
themselves (Bathurst, et al. 2011; Huang, et al. 2012 – in press). Development and
maintenance of these databases is expensive and requires a long-term commitment.
Furthermore, they have to be updated to keep up with the rapid development of
computer software. The recommended practice described by Kulicki, et al. (2007) and
Allen et al. (2005) provides an excellent starting point to accomplish this.
Finally, the use of statistics and reliability theory techniques is new to most of the
geotechnical engineering community (Christian 2003). It is important for those who
develop design specifications as well as those who do the reliability based research to
be able to communicate to the average geotechnical engineer what this all means to
them and how it relates to their current design practice. Once this is accomplished,
only then can LSD move forward in geotechnical practice unhindered.
@seismicisolation
@seismicisolation
References
AASHTO (2012a). AASHTO LRFD Bridge Design Specifications, American Association of State Highway
and Transportation Officials, 6th Edition, Washington, D.C., USA.
AASHTO (American Association of State Highway and Transportation Officials) (2012b). Subcommittee on
Bridges and Structures, http://bridges.transportation.org/Pages/default.aspx.
Allen, T.M., and Bathurst, R.J. (2002). “Soil Reinforcement Loads in Geosynthetic Walls at Working Stress
Conditions,” Geosynthetics International, Vol. 9, Nos. 5-6, pp. 525-566.
Allen, T. M. (2005). Development of Geotechnical Resistance Factors and Downdrag Load Factors for
LRFD Foundation Strength Limit State Design, Publication No. FHWA-NHI-05-052, Federal Highway
Administration, Washington, DC, 41 pp.
Allen, T.M. (2007). “Development of a New Pile Driving Formula and Its Calibration for Load and
Resistance Factor Design,” Transportation Research Record 2004, Washington, DC, pp. 20-27.
Allen, T.M., Nowak, A.S. and Bathurst, R.J. (2005). Calibration to Determine Load and Resistance Factors
for Geotechnical and Structural Design, Transportation Research Board Circular E-C079, Washington,
DC, 93 p.
Bathurst, R.J., Allen, T.M., and Nowak, A.S. (2008). “Calibration Concepts for Load and Resistance Factor
Design (LRFD) of Reinforced Soil Walls,” Canadian Geotechnical Journal, Vol. 45, pp. 1377-1392.
Bathurst, R.J., Huang, B., and Allen, T.M. (2011). “Load and Resistance Factor Design (LRFD) Calibration
for Steel Grid Reinforced Soil Walls,” Georisk, Vol. 5, Nos. 3 and 4, pp. 218-228.
Barker, R. M., Duncan, J. M., Rojiani, K. B., Ooi, P. S. K., Tan, C. K. and Kim. S. G. (1991). Manuals for
the Design of Bridge Foundations. NCHRP Report 343, TRB, National Research Council, Washington,
DC.
Basu, D., and Salgado, R. (2012 - in press). “Load and Resistance Factor Design of Drilled Shafts in Sand,”
ASCE Journal of Geotechnical and Geoenvironmental Engineering.
Brown, D.A., Turner, J.P., and Castelli, R.J. (2010). Drilled Shafts: Construction Procedures and LRFD
Design Methods – Geotechnical Engineering Circular No. 10, FHWA NHI-10-016. National Highway
Institute, Federal Highway Administration, U.S. Department of Transportation, Washington, DC.
Berg, R. R., Christopher, B. R., and Samtani, N. C. (2009). Design of Mechanically Stabilized Earth Walls
and Reinforced Slopes, No. FHWA-NHI-10-024 Vol I and NHI-10-025 Vol II, Federal Highway
Administration, 306 pp (Vol I) and 378 pp (Vol II).
Christian, J. T. (2003). “Geotechnical Acceptance of Limit State Design Methods,” LSD2003: International
Workshop on Limit State Design in Geotechnical Engineering Practice, Phoon, Honjo, & Gilbert (eds),
World Scientific Publishing Company, pp. 1-7.
Davisson, M. T. (1972). “High Capacity Piles.” Proceedings Lecture Series, Innovations in Foundation
Construction, Illinois Section, American Society of Civil Engineers, Reston, VA.
DiMaggio, J., Saad, T., Allen, T., Christopher, B., DiMillio, A., Goble, G., Passe, P., Shike, T., and Person,
G. (1999). Geotechnical Engineering Practices in Canada and Europe, Report No. FHWA-PL-99-013,
Washington, DC, 74 pp.
FHWA (Federal Highway Administration) (2007). Geotechnical Technical Guidance Manual,

http://flh.fhwa.dot.gov/resources/manuals/pddm/Geotechnical_TGM.pdf#2.6.
Florida State Department of Transportation (2012). Soils and Foundations Handbook, Gainesville, FL, 188
pp.
Goble, G. (1999). Geotechnical Related Development and Implementation of Load and Resistance Factor
Design (LRFD) Methods, NCHRP Synthesis 276, Transportation Research Board, Washington, D.C.,
69 pp.
Huang, B., Bathurst, R.J., and Allen, T.M. (2012 - in press). “Load and Resistance Factor Design (LRFD)
Calibration for Steel Strip Reinforced Walls,” ASCE Journal of Geotechnical and Geoenvironmental
Engineering.
International Standards Organization (ISO) (1998). ISO 2394 – General Principles on Reliability of
Structures, Annex E.
Isenhower, W. M., and Long, J. H. (1997). “Reliability Evaluation of AASHTO Design Equations for Drilled
Shafts,” Transportation Research Record 1582, Transportation Research Board, Washington D.C., pp.
60-67.
Kansas State Department of Transportation, 2007, KDOT Geotechnical Manual.
Meyerhof, G. G. (1994). Evolution of Safety Factors and Geotechnical Limit State Design, the Second
Spencer J. Buchanan Lecture, Texas A&M University, College Station, TX, 32 pp.
Kulicki, J. M., Prucz, Z., Clancy, C.M., Mertz, D.R., and Nowak, A.S. (2007). Updating the Calibration
Report for AASHTO LRFD Code, NCHRP 20-07/186, Transportation Research Board, Washington
D.C., 156 pp.
@seismicisolation
@seismicisolation
Long, J.H., Hendrix, J., and Baratta, A. (2009). Evaluation/Modification of IDOT Foundation Piling Design
and Construction Policy, Research Report FHWA-ICT-09-037, Illinois Center for Transportation, 58
pp.
Munfakh, G., Arman, A., Collin, J. G., Hung, J. C.-J., and Brouillette, R. P. (2001). Shallow Foundations
Reference Manual. Publication No. FHWA-NHI-01-023, Federal Highway Administration,
Washington, D.C.
Nowak, A. S. (1999). Calibration of LRFD Bridge Design Code, NCHRP Report 368, Transportation
Research Board, Washington, DC.
O’Neill, M. W. and Reese, L. C. (1999). Drilled Shafts: Construction Procedures and Design Methods.
Report No. FHWA-IF-99-025, Federal Highway Administration, Washington, D.C.
Paikowsky, S. G., C. Kuo, G. Baecher, B. Ayyub, K, Stenersen, K. O’Malley, L. Chernauskas, and M.
O’Neill (2004). Load and Resistance Factor Design (LRFD) for Deep Foundations, NCHRP Report
507, Transportation Research Board, Washington, DC.
Paikowsky, S.G., Fu, Y., and Lu, Y. (2005). LRFD Design Implementation and Specification Development,
NCHRP 20-07/183, Transportation Research Board, Washington, D.C., 30 pp, plus Microsoft Access
Database.
Paikowsky, S.G., Canniff, M. C., Lesny, K., Kisse, A., Amatya, S., and Muganga, R. (2010). LRFD Design
and Construction of Shallow Foundations for Highway Bridge Structures, NCHRP Report 651,
Transportation Research Board, Washington D.C., 140 pp.
Peck, R. B., Hansen, W. E., and Thornburn, T. H. (1974). Foundation Engineering. Second Edition, John
Wiley and Son, Inc., New York, 514 pp.
Phoon, K.K., Kulhawy, F. H., and Grigoriu, M.D. (2003). “Development of a Reliability-Based Design
Framework for Transmission Line Structure Foundations,” ASCE Journal of Geotechnical and
Geoenvironmental Engineering, Vol. 129, No. 9, pp. 798-806.
Phoon, K.K., (2005). “Reliability-Based Design Incorporating Model Uncertainties,” 3rd International
Conference on Geotechnical Engineering Combined with 9th Yearly Meeting of the Indonesian Society
for Geotechnical Engineering, Samarang, Indonesia, pp 191-203.
Rowe, R. K. and Ho, S. K. (1993). “Keynote Lecture: A Review of the behavior of reinforced soil walls.”
Earth Reinf. Practice, Balkema, Rotterdam, pp. 801-830.
Reese, L. C., and O’Neill, M. W. (1988). Drilled Shafts: Construction Procedures and Design Methods.
FHWA Publication No. FHWA-HI-88-042, 564 pp.
Santamarina, J. C., Altschaeffl, A. G., and Chameau, J. L. (1992). “Reliability of Slopes: Incorporating
Qualitative Information,” Transportation Research Board, TRR 1343, Washington, D.C., pp. 1-5.
Schneider, J.A. (2009). “Uncertainty and Bias in Evaluation of LRFD Ultimate Limit State for Axially
Loaded Driven Piles,” Deep Foundations Institute Journal, Vol. 3, No. 2, pp. 25-36.
Smith, T., Babas, A., Gummer, M., and Jin, J. (2011). Recalibration of the GRLWEAP LRFD Resistance
Factor for Oregon DOT, Final Report – SPR 683, FHWA-OR-RD-11-08, 92 pp.
Transportation Research Board (2009), Implementation Status of Geotechnical Load and Resistance factor
Design in State Departments of Transportation, E-C136, Transportation Research Board, Washington,
D.C., 42 pp.
Washington State Department of Transportation (2011). Geotechnical Design Manual M46-03.06, Olympia,
WA, USA.
Yang, L., and Liang, R. (2009). “Incorporating Setup into Load and Resistance Factor Design of Driven Piles
in Sand,” Canadian Geotechnical Journal, Vol. 46, pp. 296-305.
Zhang, L. (2004). “Reliability Verification Using Proof Pile Load Tests,” ASCE Journal of Geotechnical and
Geoenvironmental Engineering, Vol. 130, No. 11, pp. 1203-1213.
@seismicisolation
@seismicisolation
IOS Press, 2013
doi:10.3233/978-1-61499-163-2-261
Lessons learned from LRFD calibration of

reinforced soil wall structures
Richard J. BATHURST1, Tony M. ALLEN2, Yoshihisa MIYATA3
and Bingquan HUANG4
Abstract. This paper reports lessons learned from reliability theory-based LRFD
calibration for internal limit states for reinforced soil walls subjected to soil self-
weight under operational conditions. The example of the ultimate pullout limit
state for steel strip reinforced soil walls is used to demonstrate key issues. A
unique feature of the general approach is the use of bias values that include the
influence of model error and other sources of variability in input parameters on
nominal load and resistance values. The paper shows how bias values can be used
to: a) develop analytical load and resistance models that are statistically more
accurate and avoid unwanted dependencies; b) select load factors satisfying a
target exceedance value, and; c) compute resistance factors meeting a target
probability of failure (reliability index value). The general approach has
application to rigorous LRFD calibration of a wide range of geotechnical soil-
structure design problems.
Keywords. LRFD calibration, reliability theory, bias, reinforced soil walls, steel
strip, pullout
1. Introduction
Design practice for reinforced soil walls in the USA, Canada and Japan is moving to
load and resistance factor design (LRFD). This has generated renewed interest in
procedures to calibrate limit state equations based on reliability theory.

The objective of LRFD design is to select or size a component (element) so that
the probability of failure does not exceed a target value that is based on experience,
project requirements, system redundancy and/or dictated in codes. The objective of
LRFD calibration is to compute load and resistance factors that are consistent with an
acceptable target probability of failure. Issues related to LRFD calibration are the focus
of this paper.
The authors have carried out LRFD calibration for simple limit states for a number
of different wall systems including steel strip and steel grid walls (Huang et al. 2012;
Bathurst et al. 2011a), multi-anchor steel walls (Bathurst et al. 2011c) and geosynthetic
1
Corresponding Author: Professor and Research Director, GeoEngineering Centre at Queen’s-RMC,
Department of Civil Engineering, Royal Military College of Canada, Kingston, Ontario, K7K 7B4
CANADA; Email: bathurst-r@rmc.ca
2
State Geotechnical Engineer, Washington State Department of Transportation, State Materials
Laboratory, Olympia, Washington, 98504-7365, USA; E-mail: allent@wsdot.wa.gov
3
Associate Professor, Department of Civil and Environmental Engineering, National Defense Academy, 1-
10-20 Hashirimizu, Yokosuka 239-8686, JAPAN: E-mail: miyamiya@nda.ac.jp
4
AMEC Earth & Environmental, 5681-70th St, Edmonton, Alberta, T6B 3P6 CANADA; E-mail:
bing.huang@amec.com
@seismicisolation
@seismicisolation
262 R.J. Bathurst et al. / Lessons Learned from LRFD Calibration of Reinforced Soil Wall Structures
reinforced wall systems (Bathurst et al. 2011b). A common feature of these calibration
exercises is the use of bias values and their associated statistics to perform calibration
of simple limit state functions. The limit state for ultimate steel strip pullout in
conventional reinforced soil walls is used in this paper to demonstrate how bias values
can be employed to:
a) Develop analytical load and resistance models that are statistically more accurate;
b) Avoid and test for unwanted dependencies;
c) Select load factors satisfying a target exceedance value, and;
d) Compute resistance factors meeting a target probability of failure (reliability
index value).
This paper summarizes the lessons learned which will benefit other researchers,
engineers and code developers engaged in LRFD calibration for a wide range of other
simple geotechnical soil–structure interaction problems.
2. General approach
This paper is restricted to linear limit state functions having only one load term which
can be expressed as:
g φR γ Q (1)
Here, Qn = nominal (calculated) load; Rn = nominal (calculated) resistance; γQ = load

factor; and ϕ = resistance factor. The nominal values are assumed to be computed using
closed-form equations which are deterministic. The expectation is ϕ ≤ 1 and γQ ≥ 1, and
these limits are imposed on calibration outcomes.
True probability of failure for a given limit state (i.e. probability that g < 1) should
be based on the distribution of measured load (Qm) and measured resistance (Rm)
values for the simple (single load) limit state introduced earlier rather than the nominal
(calculated) values. The reason is that it is rarely the case that calculated values and
measured values are the same in geotechnical practice. The ratio of measured value to
nominal (calculated) value is called bias. Bias values are used to adjust calculated
values from deterministic theory-based, semi-empirical or empirical equations to better
match observed (measured) values. Bias values can be understood to capture model
bias (i.e. the intrinsic accuracy of the deterministic theoretical, semi-empirical or
empirical model representing the mechanics of the limit state under investigation),
random variation in input parameter values, spatial variation in input values, quality of
data and, consistency in interpretation of data when data are gathered from multiple
sources, which is the typical case (Allen et al. 2005). The transformation from nominal
load and resistance values to measured values is made by multiplication of each
nominal load and resistance value by the corresponding bias value (i.e. Qm = QnXQ and
Rm = RnXR). Algebraic manipulation leads to the limit state function Eq. (1) expressed
in terms of load and resistance bias values as (e.g. Huang et al. 2012):
ೂ
g X X (2)

@seismicisolation
@seismicisolation
R.J. Bathurst et al. / Lessons Learned from LRFD Calibration of Reinforced Soil Wall Structures 263
This transformation is only strictly valid if there are no dependencies between load and
resistance bias values and their corresponding nominal load and resistance values (i.e.
they are uncorrelated). Simple statistical tests introduced later in the paper are used to
check that this criterion is satisfied.
Clearly, the use of bias values to perform LRFD calibration using Eq. (2) requires
a sufficient database of measured loads and a related database of measured resistance
values. The accumulation and interpretation of quality measurement data from
monitored reinforced soil structures and laboratory testing of reinforcement
components has been a major effort of the authors.
3. Selection of probability of failure
The objective of LRFD calibration for a limit state function is to select values of
resistance factor and load factor such that ϕ ≤ 1, γQ ≥ 1 and a target probability of
failure (Pf (g < 1)) is satisfied. The choice of probability of failure (or equivalently,
reliability index β) is prescribed by codes, experience and system redundancy. Past
geotechnical design practice has led to a probability of failure for foundations, in
general, of approximately 1 in 1000. However, reinforced soil walls are highly
redundant systems with multiple layers of reinforcement. Hence, the tensile failure (or
pullout) of a single reinforcement layer will not result in failure of the wall. This
redundancy concept is similar to pile groups that are designed to a target reliability
index value of β = 2.0 to 2.5 because of potential load shedding from a failed pile to
other piles in the group (Barker et al. 1991; Paikowsky et al. 2004; Allen 2005). A
value of β = 2.33 corresponds to a probability of failure of 1 in 100 (Pf = 0.01). This is
the reference value that has been used for the internal limit states calibration for
reinforced soil wall systems by the authors in previous related work.
4. Load and resistance models

The LRFD calibration procedure described in this paper is based on load and resistance
bias statistics. As demonstrated later in this paper, the magnitude and frequency
distribution of these values will be sensitive to the accuracy of the underlying
deterministic models adopted for the limit state. Many load and resistance equations in
reinforced soil wall design codes include empirical coefficients that were originally
selected for allowable stress design (ASD) practice (classical factor of safety
approach). These equations, when first developed, typically over-predicted load values
when compared to measured loads (where available) and under-predicted resistance
values (tensile or pullout capacity) from laboratory tests on reinforcement elements.
This may be expected since initial model development with limited data was purposely
conservative to encourage safe ASD outcomes. However, the amount of conservatism
was often subjective, and excessive conservatism often led to design curves that did not
capture the measured trends in the data or qualitative expectations based on the
mechanics of the soil-structure interaction mechanism. Furthermore, in some cases, the
original model development and calibration for ASD practice was carried out two or
three decades ago, and re-examination of the accuracy of the underlying load and
resistance models has not kept pace.
@seismicisolation
@seismicisolation
5. Example
In this paper, LRFD calibration issues are referenced to the example of the ultimate
pullout limit state function for a steel reinforcement strip found in a conventional
mechanically stabilized earth wall. This example is used because there is a large
database of quality load data available from full-scale instrumented wall structures
(Miyata and Bathurst 2012a). A large database of pullout tests is also available for steel
strip reinforcement in combination with a wide range of frictional soils (Miyata and
Bathurst 2012b). Some features of the details of the implementation of the underlying
deterministic equations used here are simplified and the range of measurement data is
purposely broad to maximize the number of data points in each data set. For example,
the pullout data are exclusively from Japanese sources while the load data are from
monitored walls in the USA, Japan and Europe. Some soils used in Japan are not
recommended in USA specifications for the same type of wall structure. As another
example, no distinction is made between pullout test results using different pullout test
methodologies.
5.1. Load models
Figures 1 and 2 compare calculated load data using two different load models and
measured load data from a database of steel strip reinforced soil walls. The database is
comprised of 93 measurements taken from 16 different structures constructed with
granular backfill soil having a friction angle in the range of 35 ≤ φ ≤ 45o (Miyata and
Bathurst 2012a). The maximum tensile load (Tmax) in a reinforcement layer is
expressed as:
T KS σ (3)
where: Sv = reinforcement spacing; K = coefficient of earth pressure; σv = vertical

pressure at the elevation of the reinforcement strip; and, z = depth of layer below the
crest of the wall.
In current codes (e.g. PWRC 2003; AASHTO 2012; BSI 2010) a bilinear
distribution for coefficient K is recommended which can be expressed as:
K K 1 K for z ≤ zo = 6m (4a)
౥ ౥
and
K = Ka for z > z o = 6 m (4b)
Here Ka = (1 − sinφ)/(1 + sinφ) and Ko = 1 − sinφ. The earth pressure coefficient K is

normalized with Ka in Figure 1a and it can be seen to vary slightly as a function of
friction angle. Superimposed on the figure are back-calculated values of K/K a using Eq.
(3) and measured reinforcement loads. The bilinear model results in larger K/K a values
with decreasing z < zo which is also the visual trend in the back-calculated values.
There are four data points that may be treated as outliers. Using the φ = 40o design
curve as a reference, the exceedance level (i.e. fraction of measured load values greater
than calculated values) is about 58%. Applying a multiplier to the bilinear reference
@seismicisolation
@seismicisolation
a) a)
b) b)
c) c)
d) d)
Figure 1. Bilinear load model Figure 2. Exponential load model
@seismicisolation
@seismicisolation
curve reduces the exceedance value as illustrated in the inset table. Miyata and Bathurst
(2012a) concluded that, when bilinear load models of this type were originally
proposed in the late 70’s, the exceedance rate for ASD design practice was about 37%.
Since that time there is now more data (n = 93 viz n = 34 based on original documents).
The larger data set with the same original model now gives an exceedance value of
58%. This observation illustrates that load factors based on exceedance values should
be re-examined periodically as the amount of measured data increases over time. To the
best knowledge of the authors, this is seldom if ever done. Also shown in Figure 1a is
the design curve with load factor γQ = 1.75 which corresponds to an exceedance rate of
about 4%. This value is close to the 3% exceedance rate that was adopted by bridge
engineers during the development of modern LRFD for bridge superstructure elements
in North American codes (Nowak 1999; Nowak and Collins 2000; Allen et al. 2005).
The applicability of this low exceedance rate to reinforced soil structures is discussed at
the end of the paper.
Figure 1b shows that that measured load values increase in a generally linear
manner with calculated values. This indicates that, at least qualitatively, the bilinear
load model is accurate. However, the mean load bias value μQ = 1.12, which means that
on average measured load values are 12% higher than calculated values based on the
currently available measured load data.
Load bias values are plotted against depth z and nominal (calculated) load values
in Figure 1c and 1d, respectively. If the outlier points are removed (i.e. the data is
filtered), the visual impression is that the bias values are independent of z and T max
(calculated). In fact, a regressed linear line fitted to all data points in both plots includes
a zero slope within 95% confidence limits on the estimate of slope. The Spearman’s
rank coefficient test, which is a measure of the strength of a monotonic relationship
between data pairs, showed that there was no relationship between independent and
dependent parameters at a level of significance of 5%. Hence, at this point the current
bilinear load model is judged to do reasonably well regardless of whether or not the
visual outliers are removed from analyses. Furthermore, there is no evidence in the
source documents why these data are anomalous. The quantitative effect of removing
the four outlier points is a small improvement in bias values (i.e. μQ decreases from
1.12 to 1.07, and COVQ decreases from 0.332 to 0.271).
An alternative model for coefficient K is an exponential function of the form:

K K a (5)

This function has the following advantages: a) It is smoothly continuous; b) The

independent parameters remain depth z and soil friction angle φ; c) There are three
constant coefficients (a, b and c) – this is the same number of coefficients found in the
current bilinear expressions (i.e. Eq. (4a)), and; c) The maximum value of K occurs at z
= 0 and the minimum value is strongly asymptotic with increasing z. Hence, maximum
and minimum values for K are a function of soil friction angle as is the case with
bilinear expressions found in current codes.
The accuracy of this model can be referenced to the plots in Figure 2. The constant
coefficients in this model were computed using the non-linear optimization utility
(Solver) in Excel. The objective function was μQ = 1 and solutions were constrained to
a > 1, b > 1 and c = 1. The coefficients were adjusted slightly to give convenient values
of a = 1.14 and b = 3.36. Hence, z = 0 gives K = 4.5×Ka and z →∞ gives K = 1.14×Ka.
@seismicisolation
@seismicisolation
The corresponding load bias statistics are μQ = 1.00 and COVQ = 0.316. The need to
remove the outlier points identified earlier is visually less compelling using this model.
However, if these data points are removed, the bias statistics improve slightly to μQ =
1.00 and COVQ = 0.283. The independence of load bias values with depth z and Tmax
(calculated) is visually apparent in Figures 2c and 2d and confirmed using the
quantitative tests described earlier.
5.2. Resistance (pullout) models
The ultimate pullout capacity of a reinforcement strip in a soil retaining wall is

computed as:
P 2f σ bL (6)
where: f is a dimensionless empirical interface shear coefficient; σv = vertical pressure

at elevation of the reinforcement strip; b = strip width; and, L e = anchorage (pullout)
length. The value of f in this example is computed using the default model for
frictional soils in the range of 35 ≤ φ ≤ 45o. Japanese, USA and UK design codes use
bilinear equations to compute coefficient f . The Japanese equations for different soil
and reinforcement types have the following format:
f f 1 tan ψ for z ≤ zo (7a)

౥ ౥
and
f tan ψ for z > z o (7b)
In this example, the default values for the three constant coefficient terms are taken
from the Japanese (PWRC 2003) design code for ribbed steel strips in granular soils,
i.e. f = 1.5, zo = 6 m and ψ = 36o.
Figure 3a shows back-calculated values of coefficient f using Eq. (6) and
measured Pmax values from pullout test results. Superimposed on the plot is Eq. (7) with
different multipliers. In LRFD terminology, these multipliers are equivalent to
resistance factors (ϕ). A review of Japanese past practice shows that the original design
curve (multiplier = 1) adopted in the 1980’s corresponded to a minimum exceedance
rate of 78% (Miyata and Bathurst 2012b). The exceedance rate using available data
today is about 57%. Figure 3b shows measured versus calculated Pmax values. The
general trend is that measured values increase with calculated values. However,
resistance (pullout) bias statistics show that the measured values are on average 39%
greater than calculated values (μR = 1.39) and the spread in bias values is high, as
indicated by COVR = 0.559. Pullout bias values are visually correlated with depth z and
calculated Pmax values as shown in Figures 3c and 3d, respectively. The 95%
confidence limits on the slope of the linear regressed lines shown in the figures confirm
these dependencies, as does the Spearman’s rank correlation test at a level of
significance of 5%.
The accuracy of load predictions in terms of mean bias and coefficient of variation
of bias values can be improved using an exponential function of the same general form
as Eq. (5) but adapted to the pullout case as (Miyata and Bathurst 2012b):
@seismicisolation
@seismicisolation
a)
a)
b) b)
c) c)
d) d)
Figure 3. Bilinear pullout model Figure 4. Exponential pullout model
@seismicisolation
@seismicisolation
‫כ‬౥ ౟
f tan ψ (8)

This equation has the following advantages:
a) The equation has three constant coefficients which is the same number of
coefficients as the bilinear models currently recommended by PWRC (2003);
b) The minimum value of interface shear coefficient remains f tan ψ , and;
c) The function is smoothly continuous with depth z.
The constant coefficient terms were selected after minor adjustment from the results
of optimization (i.e. same method used to fit the exponential function for maximum
tensile load). The final values are, f = 7.0, Ψi = 33o and d = 0.420. The value for
coefficient Ψi is within a few degrees of the default value (36o) currently recommended
in the Japanese code for the case of granular soil in combination with ribbed steel
strips.
Back-calculated f values using Eq. (6) and measured pullout capacity test data are
shown in Figure 4a, and measured versus calculated Pmax values are plotted in Figure
4b. The visual impression is that the exponential model is more accurate than the
bilinear model. This is confirmed quantitatively by the matching bias statistics which
have a mean μR = 1.00 and COVR = 0.475. Figures 4c and 4d show no visual indication
of bias dependency with depth and calculated pullout capacity values, respectively.
This was confirmed quantitatively using the zero slope test as before and Spearman’s
rank correlation test at a level of significance of 5%.
6. Calculation of resistance factors
The probability of failure using bias values is notionally related to the area of overlap
between the load bias (XQ) and resistance bias (XR) frequency distribution curves
illustrated in Figure 5. Load and resistance factors (γQ, ϕ) are selected to shift the two
curves apart along the horizontal axis so that the probability of failure (overlap) does
not exceed an acceptable value. As noted earlier, a value of Pf = 0.01 has been
recommended for highly strength redundant systems such as reinforced soil walls with
multiple reinforcement layers.
If the load and resistance bias frequency distributions can be closely fitted to log-
normal distributions over many standard deviations from the mean, then the bias
statistics for the entire data sets can be used to compute Pf. However, this is not always
the case. In fact, it is very common that deviations from idealized log-normal
distributions will occur in the tails of these distributions for soil-structure problems
such as reinforced soil walls. Because it is the overlap between the upper tail of the
load frequency distribution and the lower tail of the resistance frequency distribution
that will largely influence the true probability of failure, it may be necessary to select
other log-normal distributions to capture the data in the tails (Allen et al. 2005).
Figure 6a shows bias values for reinforcement load using the bilinear load model
plotted as a cumulative distribution function (CDF) plot. Log-normal approximations
have been fitted to the entire data set (n = 93) and to the upper tail comprised of the
four data points previously identified as possible outliers. In Figure 6b the four outliers
@seismicisolation
@seismicisolation
Figure 5. Idealized frequency distribution curves for unfactored and factored load and resistance bias values
Table 1. Load and resistance bas statistics
a) Bilinear models
Load bias Resistance bias*
Figure 6a Figure 6b Figure 8
Column → 1 2 3 4 5 6
Fit to all Fit to all data Fit to all Fit to all data Fit to all Fit to all data
Parameter data upper tail data upper tail data lower tail
n 93 93 89 89 70 70
Mean 1.12 0.900 1.07 1.49 1.45 0.750
COV 0.332 0.556 0.271 0.027 0.577 0.267
*Resistance bias values are correlated with nominal (calculated) pullout (resistance) values
b) Exponential models
Load bias Resistance bias

Figure 7a Figure 7b Figure 9
Column → 1 2 3 4 5 6
Fit to all
Parameter Fit to all Fit to all data Fit to all Fit to all data Fit to all
data
data upper tail data upper tail data
lower tail
n 93 93 89 89 70 70
Mean 1.00 1.00 1.00 1.20 1.00 1.00
COV 0.316 0.316 0.283 0.108 0.475 0.50
at the upper tail are removed (n = 89) and again a log-normal approximation fitted to
the entire distribution and through the upper tail.
Figure 7a shows similar data using the exponential load model to compute the load
bias values. In this case, the log-normal approximation to the entire data set (n = 93) is
judged to be an acceptable fit to the upper tail of the load bias distribution. The four
possible outliers in the original data set are removed in Figure 7b and log-normal
approximations fitted to the entre data set (n = 89) and the upper tail.
Figures 8 and 9 show a similar log-normal fitting exercise applied to the pullout
bias data sets. However, the fit-to-tail is now at the bottom of the CDF plots. No
@seismicisolation
@seismicisolation
a) a)
b) b)
Figure 6. Bilinear load model Figure 7. Exponential load model
outliers have been removed in these two plots. The possible outlier identified in Figure
9 has been ignored in the fit-to-lower tail. The difference in bias statistics for the two
approximations is negligible in the figure.
It is clear from the fit-to-tail approximations shown in Figures 6 through 9 that
judgment plays a role in the choice of approximation to the distribution tails.
@seismicisolation
@seismicisolation
Load and resistance bias statistics are summarized in Table 1.

Resistance factors can be computed for a target probability of failure by trial and
error using Monte Carlo simulation and combinations of the bias mean and COV in
Table 1. For demonstration purposes, it is more convenient to compute resistance
factors to sufficient accuracy using the following closed-form solution:
μR
γQ
μQ
(1+COV ) (1+COV )
2
Q
2
R
ϕ = (9)
{
exp β
⎣(
LN ⎡ 1+COVQ
2
)(1+COV )⎤⎦ }
2
R
This equation is applicable to load and resistance bias values having log-normal
distributions, which is the case in this example. The derivation of this equation can be
found in the paper by Bathurst et al. (2011c).
The results of calculations for β = 2.33 (Pf = 0.01) and using all permutations of
bias statistics in Table 1 for bilinear and exponential load and resistance models are
plotted in Figure 10a and 10b, respectively. The figure inset legends identify the
combinations of bias statistics from Table 1 used to generate the ϕ-γQ lines. The two
figures show that the bilinear models, which are generally less accurate than the
exponential models, give a wider range of possible resistance factors (i.e. the vertical
spread in ϕ-γQ lines is greater). Two fit-to-tail curves are identified by the heavy lines
in each figure. Since the choice of fit-to-tail is subjective, the choice of which ϕ-γQ line
should be used for calibration requires judgment.
For each calibration line, the resistance factor can be computed using an
appropriate load factor. For example, Japanese past practice has shown that an
exceedance value of 37% of the calculated load is acceptable. This corresponds (by
interpolation) to a load factor (multiplier) γQ = 1.20 using the bilinear load model
(Figures 1a and 1b). The same exceedance value using the exponential load model
gives a lower value, i.e. γQ = 1.10. The resulting resistance factor using the bilinear
load and pullout models is ϕ = 0.28 to 0.32 and for the exponential load and resistance
models is ϕ = 0.27. Other load factors could be used. For example, the current
AASHTO (2012) code recommends a load factor of γQ = 1.35 for the calculation of
reinforcement load due to soil self-weight plus any permanent surcharge loading. This
value leads to slightly higher resistance factors using the same ϕ-γQ lines in Figure 10.
However, the corresponding exceedance values are much lower than the 37% value
identified earlier (i.e. 24% and 11% in Figures 1a and 2a using currently available load
data).
7. Discussion and conclusions
This paper has demonstrated a number of key steps and issues related to LRFD
calibration of a simple linear limit state function with a single load term. The example
used to develop these points is the ultimate pullout limit state for steel strips used in
reinforced soil walls.
Bias values have been used to quantify the accuracy of current steel strip
reinforcement load and resistance (pullout) models found in design codes. Bias values
@seismicisolation
@seismicisolation
Figure 8. Bilinear pullout model Figure 9. Exponential pullout model
computed for the current bilinear resistance (pullout) model used in practice today
show that this model has variable accuracy with depth. Furthermore, it violates the
criterion that bias values should be independent of nominal (calculated) pullout values.
This criterion must be satisfied in order to transform the original limit state function
expressed in terms of nominal load and resistance values to a linear function in terms of
load and resistance bias values.
The paper shows how bias statistics can guide the development of load and
resistance models that are more accurate on average and ensure that the model
predictions are uncorrelated with depth. In this paper exponential models are used, but
the authors have demonstrated in other related papers that multi-linear formulations
may be adequate (Bathurst et al. 2011a; Huang et al. 2012).
The example problem demonstrates that LRFD calibration requires special
attention be paid to the upper tail of the load bias CDF plot and the lower tail of the
resistance CDF plot since it is the region of tail overlap that strongly influences the true
probability of failure. The sensitivity of LRFD calibration outcomes on bias statistics
(mean and COV values) extracted from approximations to unfiltered and filtered CDF
data plots, and best-fit-to-tail in CDF plots has been demonstrated quantitatively.
The selection of the load factor(s) used in limit state equations should reflect
current accepted levels of load exceedance as a practical strategy to link current
practice to LRFD calibration outcomes. In the development of current LRFD equations
for bridge superstructure design, the load factor for dead load was based on a load
exceedance rate of 3%, which corresponds to two standard deviations below the mean
of measured loads. An important difference between measured loads in steel structures
and measured reinforcement loads in soil-structures (such as soil retaining walls) is that
the spread in measured loads for the latter is much larger (i.e. higher COV values).
@seismicisolation
@seismicisolation
a) Bilinear load and resistance (pullout) models

b) Exponential load and resistance (pullout) models
Figure 10. Computed resistance factors for probability of failure Pf = 0.01 (β = 2.33)
Hence, load factors based on small exceedance values common in structural

engineering applications are not practical for some limit states for reinforced soil
retaining walls. Whether consciously or not, larger exceedance values are acceptable in
@seismicisolation
@seismicisolation
geotechnical engineering as demonstrated by comparison of measured loads with

design curves originally developed for pullout ASD practice. Nevertheless, as
demonstrated in this paper, the same acceptable target probability of failure (reliability
index β) can be achieved using different combinations of load and resistance factor.
Hence, there is no compelling reason to impose the load factor based on 3%
exceedance in structural engineering practice on soil-structure limit states. The most
important outcome from LRFD calibration is that an acceptable target probability of
failure is satisfied and the computed resistance factor and load factor satisfy ϕ ≤ 1 and
γQ ≥1, respectively.
It is clear that large databases of quality measured load and resistance data are
required for LRFD calibration of simple linear limit state functions. Sufficiently large
and high quality databases of the type compiled by the authors for steel strip reinforced
soil walls may not be available for similar limit state function calibration in other
systems.
Judgment in the selection of data from multiple sources and fitting to frequency
distribution plots is an unavoidable feature of LRFD calibration. Additional discussion
on these issues can be found in the TRB Circular E-C079 (Allen et al. 2005).
Finally, it is prudent to remind design engineers that limit state functions that are
empirically adjusted and calibrated against physical data should only be used for design
when project-specific conditions fall within the physical data envelope that was used to
calibrate the underlying deterministic models and to compute the load and resistance
factors.
Acknowledgments
The authors are grateful for the financial support from the following departments and
agencies: Natural Sciences and Engineering Research Council (NSERC) of Canada;
Department of National Defence (Canada); Japan Ministry of Education, Culture,
Sports, Science and Technology; Japan Ministry of Defense; JSPS Invitation
Fellowship Program for Research in Japan; Ministry of Transportation of Ontario and
the following US State Departments of Transportation - Alaska, Arizona, California,
Colorado, Idaho, Minnesota, New York, North Dakota, Oregon, Utah, Washington and
Wyoming.
References
Allen, T.M. (2005). Development of Geotechnical Resistance Factors and Downdrag Load Factors for
LRFD Foundation Strength Limit State Design, Publication No. FHWA-NHI-05-052, Federal Highway
Administration, Washington, DC, 41 pp.
Allen, T.M., Nowak, A.S. and Bathurst, R.J. (2005). Calibration to Determine Load and Resistance Factors
for Geotechnical and Structural Design. Circular E-C079, Washington, DC: Transportation Research
Board, National Research Council.
American Association of State Highway and Transportation Officials (AASHTO) (2012). LRFD Bridge
Design Specifications. 6th edition. AASHTO: Washington, DC, USA.
Barker, R.M., Duncan, J.M., Rojiani, K.B., Ooi, P.S.K., Tan, C.K. and Kim. S.G. (1991). Manuals for the
Design of Bridge Foundations. NCHRP Report 343, TRB, National Research Council, Washington,
DC.
@seismicisolation
@seismicisolation
Bathurst , R.J., Huang, B. and Allen, T.M. (2011a). Load and resistance factor design (LRFD) calibration for
steel grid reinforced soil walls, Georisk, 5(3-4): 218-228.
Bathurst, R.J., Huang, B. and Allen, T.M. (2011b). Interpretation of installation damage testing for
reliability-based analysis and LRFD calibration, Geotextiles and Geomembranes, 29(3): 323-334.
Bathurst, R.J., Miyata, Y. and Konami, T. (2011c). Limit states design calibration for internal stability of
multi-anchor walls, Soils and Foundations, 51(6): 1051-1064.
British Standards Institution (BSI) (2010). Code of Practice for Strengthened/Reinforced Soil and Other
Fills. BSI, Milton Keynes, BS 8006.
Huang, B., Bathurst, R.J. and Allen, T.M. (2012). Load and resistance factor design (LRFD) calibration for
steel strip reinforced soil walls, ASCE Journal of Geotechnical and Geoenvironmental Engineering,
138(8): 922-933.
Miyata, Y. and Bathurst, R.J. (2012a). Measured and predicted loads in steel strip reinforced c-φ soil walls in
Japan, Soils and Foundations, 52(1): 1-17.
Miyata, Y. and Bathurst, R.J. (2012b). Analysis and calibration of default steel strip pullout models used in
Japan, Soils and Foundations, 52(3): 481-497.
Nowak, A.S. (1999). Calibration of LRFD Bridge Design Code. NCHRP Report 368, National Cooperative
Highway Research Program, Transportation Research Board, Washington, D.C.
Nowak, A.S. and Collins, K.R. (2000). Reliability of Structures. McGraw-Hill, New York.
Paikowsky, S.G. (2004). Load and Resistance Factor Design (LRFD) for Deep Foundations. NCHRP Report
507, National Cooperative Highway Research Program, Transportation Research Board of the National
Academies, Washington, D.C. 126 pp.
Public Works Research Center (PWRC) (2003). Design Method, Construction Manual and Specifications for
Steel Strip Reinforced Retaining Walls. 3rd revised Edition. Public Works Research Center, Tsukuba,
Ibaraki, Japan, 302 pp. (in Japanese).
@seismicisolation
@seismicisolation
IOS Press, 2013
doi:10.3233/978-1-61499-163-2-277
Geotechnical Design Code Development

in Canada
Gordon A. FENTON
Visiting Professor, Faculty of Civil Engineering and Geosciences, Delft University of
Technology, Delft, The Netherlands, and
Professor, Department of Engineering Mathematics, Dalhousie University, Halifax,
Nova Scotia, Canada; email: Gordon.Fenton@dal.ca
Abstract. Canada has two national codes of practice which include geotechnical
design provisions: the National Building Code of Canada and the Canadian High-
way Bridge Design Code. Both of these codes have been using a load and resistance
factor format for about two decades now, but are still in the process of adopting a
reliability-based design framework. This paper describes the advances planned for
these codes.
Keywords. geotechnical code development, reliability-based geotechnical design,
load and resistance factor design, Canadian codes, code comparison
1. Background
Geotechnical design codes have been migrating towards reliability-based design con-
cepts for several decades now. Prior to 1979, geotechnical design in Canada was based
on working stress design (WSD) which involved satisfying an equation of the form
R̂ ≥ Fs ∑ F̂i (1)
i
where R̂ is the characteristic (nominal or design) resistance, Fs is a factor of safety, and

F̂i is the ith characteristic (nominal or design) load effect. The factor of safety was tra-
ditionally used to account for all sources of uncertainty. However, in recent decades, it
has been recognized that not all sources of uncertainty are equal. For example, usually
live loads are less certain than dead loads, concrete strengths are less certain than steel
strengths, and soil strengths are less certain than most other engineering properties. It
has thus made sense to break up the global factor of safety, Fs , into a sequence of partial
factors, one for each source of uncertainty in the design.
In most modern code implementations, the resulting set of partial factors has been
separated into two distinct groups (which are nevertheless inversely related). These are
the load and resistance factors which lead to a design methodology referred to as Load
and Resistance Factor Design (LRFD). The partial factors are individually related to the
variability of the quantity that they are factoring and are used to scale the characteristic
design values to more conservative values such that the overall probability of design fail-
@seismicisolation
@seismicisolation
278 G.A. Fenton / Geotechnical Design Code Development in Canada
ure is acceptably small. In general, this means that loads are scaled up (so long as they
are acting in a way that reduces overall system safety) and resistances are scaled down
so that the final factored design values are acceptably conservative. Under the LRFD ap-
proach in most codes, designs must satisfy an equation of the following generalized form
(although the right hand side is often expressed more precisely as a series of possible
load combinations),
ϕg R̂ ≥ ∑ Ii ηi αi F̂i (2)
i
where ϕg is a geotechnical resistance factor, R̂ is the characteristic geotechnical resis-

tance (based on characteristic ground parameters), and, for the ith load, Ii is a structure
importance factor, ηi is a load combination factor, αi is the load factor, and F̂i is the
characteristic load effect.
In this paper the word characteristic is used because it suggests a value which char-
acterizes (in some sense) a design parameter which is uncertain, e.g. a load or resistance,
having a random distribution. The words nominal or design have also been commonly
used to equivalently describe such design parameters, but these words do not carry the
suggestion of the underlying random distribution of the design parameter, and so will
not be used here. Some design codes (e.g. the Eurocode) provide a specific statistical
definition of the word characteristic as the 5th or 95th percentile, whichever is worst.
Most other geotechnical design codes provide only vague definitions for the characteris-
tic value. For example, probably the most popular definition is “a conservative estimate
of the mean.”
In Canada, the LRFD approach is embedded within a Limit State Design (LSD)
framework, where the LRFD formulation is satisfied for each of a sequence of possible
failure modes, or limit states. Typically the load and resistance factors are specifically
selected for the limit state under consideration. For example, designing against the limit
state of bearing capacity failure would usually involve different factors than designing
against the limit state of excessive settlement.
The load and resistance factor method typically appears in one of two forms in
geotechnical design codes around the world;

1) the partial resistance factor approach, in which the individual components of ground
strength, e.g. cohesion and friction, are factored separately. The rational behind this
approach is that the components of strength have different levels of uncertainty –
for example, cohesion is generally deemed to be more uncertain than friction angle.
This is analogous to how live and dead loads are factored separately.
2) the total resistance factor approach, in which the geotechnical resistance is computed
in the traditional way using best estimates of the ground parameters (i.e. character-
istic values) and then the final result is factored. This approach is more analogous to
how resistances are factored in structural engineering where each engineering mate-
rial (e.g. concrete, steel, and wood) has its own resistance factor. The ground is then
viewed as just another engineering material.
In 1979 and then again in 1983 the Ontario Highway Bridge Design Code (OHBDC)
adopted the partial resistance factor approach from Danish practice, in which compo-
nents of ground strength (e.g., cohesion and friction angle) were individually factored. In
1983, the LSD approach became mandatory in the bridge code. Unfortunately, the partial
factor format did not lead to design consistency with the working stress design approach
@seismicisolation
@seismicisolation
G.A. Fenton / Geotechnical Design Code Development in Canada 279
and so was not readily accepted by geotechnical engineers. Another complaint levied
against the partial resistance factor approach is that by modifying the ground proper-
ties away from their characteristic values, the resulting predicted failure mechanism was
sometimes significantly different than the actual failure mechanism in the ground. Many
geotechnical engineers found that the myriad of resistance factors that the approach in-
volved made it difficult to retain a clear understanding of the geotechnical problem be-
ing considered. In addition, the 1983 edition of OHBDC applied both a partial factor to
soil properties along with a load factor to active and passive earth pressures. This double
factoring of the ground led to increases of approximately 30% in footing widths for can-
tilever retaining walls (Green and Becker, 2000) beyond what traditional designs called
for.
In 1991, the OHBDC switched to the total resistance factor approach, where the
characteristic (nominal) geotechnical resistance was computed using traditional (work-
ing stress design) methods and then factored at the end. In general, this approach was
preferred by the geotechnical community because it was simpler and more familiar (sim-
ilar to the traditional factor-of-safety approach, Eq. 1), because it more closely preserved
the best estimate of the failure mechanism, and because it seemed to be in better harmony
with the approach taken by structural engineers of factoring each engineering material.
The 1991 edition of the Ontario Highway Bridge Design Code (OHBDC) was the
third and last edition of the OHBDC. In 2000, the 9th edition of the CAN/CSA-S6 code,
renamed the Canadian Highway Bridge Design Code (CHBDC) became a national stan-
dard and was largely modeled on OHBDC 1991. The most recent edition of CHBDC was
published in 2006 (10th Edition). The geotechnical design code provisions in the current
CHBDC are little changed from the 1991 OHBDC.
The National Building Code of Canada (NBCC) permitted both working stress and
limit states design in their 1995 edition. In the next edition, 2005, limit state design be-
came mandatory for geotechnical designs, although the geotechnical resistance factors
themselves are still not part of the Building Code, even in the most recent edition (Na-
tional Research Council, 2010). In other words, the NBCC is still lagging well behind
the CHBDC with respect to geotechnical design.
2. Present Status
At this time, geotechnical design in Canada follows the total resistance factor approach
within a Limit State Design framework, as do most other geotechnical design codes in
North America, e.g., American Association of State Highway and Transportation Of-
ficials (AASHTO, 2007). The resistance factors to be used in Canada appear in the
CHBDC (Canadian Standards Association, 2006) and in the User’s Guide to the NBCC
(National Research Council, 2010), with the latter being nearly identical to those speci-
fied in the CHBDC.
It is of interest to compare the resistance factors specified in the CHBDC to those
specified in other codes from around the world. To this end, a very simple example in
which the required area of a spread footing designed against bearing failure is considered.
Characteristic dead and live loads of F̂D = 3700 kN and F̂L = 1000 kN, respectively, are
to be supported by a weightless soil having characteristic soil properties ĉ = 100 kPa
and φ̂ = 30 ◦ (note that the distinction between drained and undrained parameters is not
@seismicisolation
@seismicisolation
made here since this is not important to the illustration being made – either condition can
be assumed). The resulting required areas are shown in Table 1. The design must satisfy
the following equation (which also corresponds to Design Approach 2 in the Eurocode);
ϕgu R̂u ≥ αL F̂L + αD F̂D (3)
where the importance factor and load combinations factors appearing on the right hand
side of Eq. (2) are both 1.0 for this simple load combination, and the subscript u on
the left hand side (resistance side) denotes that this is an ultimate limit state (ULS). For
a weightless soil, the characteristic ultimate geotechnical resistance, R̂u , is equal to the
footing area, A, times the characteristic ultimate soil bearing capacity, ĉN̂c , i.e.,
R̂u = AĉN̂c (4)
The characteristic bearing capacity factor, N̂c , is given by (e.g. Prandtl, 1921, and Mey-
erhof, 1951, 1963), as

exp{π tan φ̂ } tan2 π4 + φ̂2 − 1
N̂c = (5)
tan φ̂
so that the minimum required footing area is computed from Eq. (3) as
αL F̂L + αD F̂D
A= (6)
ϕgu ĉN̂c
For the given problem, N̂c = 30.14, so that ĉN̂c = 100(30.14) = 3014 kPa and
1000αL + 3700αD
A= (7)
3014ϕgu
In the case of the partial factor approach, where the components of the ground shear
strength are factored separately, it can be seen that the N̂c term involves both tan(φ̂ ) and
φ̂ . Thus, applying partial factors yields a ‘factored’ N̂c value which will be referred to
here as N̂ f and which is computed as

π φ̂ f
exp{πϕφ tan φ̂ } tan2 4 + 2 −1
N̂ f = (8)
ϕφ tan φ̂
where ϕφ is the partial factor applied to tan(φ ) and φ f is the ‘factored’ friction angle
defined as

φ f = tan−1 ϕφ tan φ̂ (9)
so that the minimum footing area required for the partial factor approach becomes
αL F̂L + αD F̂D
A= (10)
ϕc ĉN̂ f
@seismicisolation
@seismicisolation
where ϕc is the partial factor associated with the cohesion component of shear strength.
In Table 1, where a range in factors is given, the midpoint of the range is used.
Table 1 Values of load and resistance factors suggested by various sources along
with the footing area each would require in a bearing capacity design ex-
ample assuming similarly defined characteristic loads. All factors are ap-
plied in a multiplicative fashion.
Source Dead Load Live Load tan(φ ) c Bearing Area

Factor Factor Factor Factor Factor (m2 )
CFEM (1992) 1.25 1.5 0.8 0.5-0.65 5.22
NCHRP 343 (1991) 1.3 2.17 0.35-0.6 4.88
NCHRP 12-55 (2004) 1.25 1.75 0.45 4.70
Denmark (1985) 1.0 1.3 0.83 0.56 4.13
AASHTO (2007) 1.25 1.75 0.45-0.55 4.23
AS 5100 (2004b) 1.2 1.8 0.35-0.65 4.14
CHBDC (2006) 1.2 1.7 0.5 4.07
AS 4678 (2002b) 1.25 1.5 0.75-0.95 0.5-0.9 3.89
EC 7 DA 1 (2004) 1.0 1.3 0.8 0.8 3.06
EC 7 DA 2 (2004) 1.0 1.3 0.71 3.04
Table 1 illustrates that a range in conservatism apparently exists across this selection of
codes under the above assumptions. The 1992 Canadian Foundation Engineering Manual
(CFEM, Canadian Geotechnical Society, 1992) is perhaps the most conservative, with a
required bearing area of 5.22 m2 . The least conservative (apparently) are the two Design
Approaches (DA 1 and 2) of Eurocode 7 (2004) with required bearing areas of about
3.05 m2 .
However, Table 1 also assumes that the characteristic design parameters are the same
for all codes. A more complete comparison of the levels of safety inherent in each design
code involves a more careful consideration of how all of the parameters entering the
design process are defined and factored, particularly with respect to characteristic values.
Such a comparison is considered next.
2.1. Characteristic Loads and Bias Factors
Some codes specify that the characteristic load is equal to the mean, others suggest using
a ‘cautious estimate of the mean’, while others specify the use of an upper (or lower)
quantile. Similarly, the characteristic resistance may be computed using mean strength
parameters, or using quantiles of the strength parameters. In general, the difference be-
tween the characteristic design value and its mean is usually captured by a bias factor
usually defined as the ratio of the mean to characteristic value, i.e.,
μR μL μD
kR = , kL = , kD = (11)
R̂u F̂L F̂D
where k is the bias factor and μ is the mean of the subscripted variable. Introducing the
dead to live load ratio, RD/L = μD / μL , allows Eq. (3) to be re-expressed as
μR ≥ Fs (μL + μD ) (12)
@seismicisolation
@seismicisolation
where Fs is a global factor of safety, defined as

kR αL αD RD/L 1
Fs = + (13)
ϕgu kL kD 1 + RD/L
Note that Eq. (12) is seen to take on a similar role (and definition as ratio of mean
resistance to mean load) as does the traditional factor of safety used in working stress
design approaches. If the coefficients of variation of the loads and resistances are approx-
imately the same worldwide, then the global factor of safety provides a simple measure
of the relative safety of a code design which then allows the safety level of various codes
to be compared. Ellingwood (1999) notes that probability models for loads collected in
research programs in North America and Europe agree reasonably well, and so the as-
sumption that coefficients of variation are similar, at least between North America and
Europe, is deemed to be reasonable. In this paper the global factor of safety provided by
the following design codes are compared for shallow foundations at the bearing capacity
ultimate limit state;
1) The National Building Code of Canada (NBCC) published by the National Research
Council of Canada (2010),
2) The Canadian Highway Bridge Design Code (CHBDC) published by the Canadian
Standards Association (2006),
3) AASHTO LRFD Bridge Design Specifications (AASHTO), published by the Amer-
ican Association of State Highway and Transportation Officials (2007),
4) The Eurocode, in particular Eurocode 0 (Basis of Structural Design, British Stan-
dard, 2002a), Eurocode 1 (Actions on Structures – Part 1-1: General Actions, British
Standard, 2002b) and Eurocode 7-1 (Geotechnical Design – Part 1: General Rules,
British Standard, 2004),
5) Australian Standard AS5100 (Bridge Design, Standards Australia, 2004)
To compare the level of safety between each of these codes, a hypothetical geotechnical
system will be considered which has dead to live load ratio RD/L = 3.0.
The Eurocode is reasonably specific as to how characteristic loads are defined. With
respect to dead loads, the Eurocode 0 (British Standards, 2002a) states that the variability
of permanent actions (i.e. dead loads) may be neglected if they do not vary significantly
over the design working life. In other words, if the coefficient of variation of dead loads,
vD , is less than about 10%, then the dead loads can be considered to be non-random and
F̂D = μD so that kD = 1.0. The other codes considered are less specific about the definition
of characteristic dead loads, but generally indicate that F̂D is to be estimated using mean
structural component weights. Bartlett et al. (2003) suggest that some dead load com-
ponents are often forgotten or missed in the estimation process, so that in practice the
characteristic (design) dead load is generally somewhat less than the true mean dead load
and the dead load bias factor is more like 1.05 (see also Ellingwood et al., 1980). Since
this error is probably common to all localities, it will be assumed here that kD = 1.05 for
all codes considered.
With respect to live loads, the North American codes define the characteristic live
load as the mean maximum live load exerted on the structure over its design lifetime –
for example, Clause 4.3.1 of ASCE-7 (2010) states that uniformly distributed live loads
are the mean of the maximum load over the design lifetime. Although the NBCC does
not specifically define the characteristic live load, Bartlett et al. (2003) implies that it has
@seismicisolation
@seismicisolation
the same definition as ASCE-7. Both codes specify acceptable characteristic live load
values which are typically somewhat higher than the actual mean maximum live load. For
example, both the Canadian and US codes specify a uniform live load for office space of
2.4 kPa. Bartlett et al. (2003) suggest that, after reductions for influence or tributary area,
the code specified characteristic live load is typically about 10% higher than the actual
mean value, so that kL = 0.9 was adopted by Bartlett et al. in their calibration efforts for
the 2005 edition of the NBCC. As also reported by Bartlett et al., this bias value is in
reasonable agreement with ASCE-7.
The Eurocode 0 (British Standard, 2002a) states in Clause 4.1.2(7) that, for variable
actions, the characteristic value shall correspond to one of; an upper value with an in-
tended probability of not being exceeded or a lower value with an intended probability
of being achieved, during some specific reference period; or a nominal value, which may
be specified in cases where a statistical distribution is not known. This is a fairly vague
definition, but Clause 4.1.2(4) suggests that an “upper value” (which would be of inter-
est for loads) corresponds to a 5% probability of being exceeded (95% fractile). Clause
4.1.2(4) further states that the action may be assumed to be Gaussian. If this is assumed,
then the 95% fractile is given by
F̂L = μL (1 + 1.645vL) → kL = 1/ (1 + 1.645vL) (14)
where vL is the coefficient of variation of the maximum lifetime live load. Both Allen
(1975) and Bartlett et al. (2003) use vL = 0.27. The author is not sure what value of
vL was assumed in the Eurocode, but Ellingwood (1999) suggests that Europe uses a
similar value to that used in North America. If this is the case, then the Eurocode is using
kL = 0.69, which is very close to Allen’s (1975) suggested bias of 0.7.
Another approach to estimating the live load bias factor employed in Europe is to
consider the characteristic office occupancy uniform live load specified in the European
and North American codes, which are 3.0 and 2.4 kPa, respectively. If the live load bias
factor of kL = 0.9, adopted by Bartlett et al. (2003), is assumed true for North America,
then μL = 0.9(2.4) = 2.16 kPa. If it is further assumed that this mean live load is at least
approximately true in Europe, then the European live load bias factor is kL = 2.16/3.0 =
0.72. On the basis of both of the above approximate calculations, it appears likely, then,
that the Eurocode uses a live load bias factor of approximately kL = 0.70.
The Australian Standard AS5100.1 (Standards Australia, 2004a) specifically defines
load actions for ultimate limit state as “an action having a 5% probability of exceedance
in the design life” in Clause 6.5. This is the same as used in the Eurocode (albeit more
clearly specified). In addition, since the Australian- New Zealand “Structural Design Ac-
tions” Standard AS/NZS 1170 (Standards Australia, 2002a) specifies that the character-
istic uniform live load for office buildings is 3.0 kPa, which is the same as the Eurocode,
it appears that the live bias factor for Australia is also kL = 0.70.
2.2. Characteristic Resistance and Bias Factors
The estimation of the resistance of the ground to imposed loads is generally a multi-step
process: 1) take measurements of the ground properties, 2) correlate the measurements
with characteristic engineering parameters (e.g. cohesion and friction angle), and 3) use
the characteristic parameters in a prediction model. Each step introduces errors, and so
@seismicisolation
@seismicisolation
the characteristic resistance and associated resistance factor (discussed later), along with
the loads and load factors, must be found in such a way to ensure a safe design. Eurocode
7-1, Clause 2.4.5.2 (British Standard, 2004) provides a number of requirements for the
selection of characteristic properties, such as “The characteristic value of a geotechnical
parameter shall be selected as a cautious estimate of the value affecting the occurrence
of the limit state” and “If statistical methods are used, the characteristic value should be
derived such that the calculated probability of a worse value governing the occurrence
of the limit state under consideration is not greater than 5%. NOTE: In this respect, a
cautious estimate of the mean value is a selection of the mean value of the limited set
of geotechnical parameter values, with a confidence level of 95%; where local failure is
concerned, a cautious estimate of the low value is a 5% fractile.” The Eurocode 0 (British
Standard, 2002) states that “where a low value of material or product property is un-
favourable, the characteristic value should be defined as the 5% fractile value.” Accord-
ing to Schneider (2012), the characteristic ground parameters should be selected as a 5%
fractile value of the sample mean, using the distribution of the sample mean, rather than
√
that of the samples directly (the sample mean having standard deviation s/ n, where s
is the sample standard deviation, and n is the number of samples used to estimate s). The
author notes that a 5% fractile value based on the sample mean will generally be quite a
bit less conservative than a 5% fractile based on the samples themselves.
Hicks (2012) interprets Clause 2.4.5.2 of Eurocode 7-1 as meaning that the charac-
teristic soil parameters are to be selected so as to ensure a 95% confidence in the geotech-
nical system being designed. While this is a reasonable interpretation, it will involve both
the distribution of the applied maximum lifetime load and an appropriate spatial averag-
ing of geotechnical parameters over the actual failure surface (or failure domain). The
author feels that it is probably easier to develop a design code using characteristic soil
parameters based on fractiles of the soil parameter distribution at this point in time.
In any case, the above discussion about characteristic values used in the Eurocode
refers to the selection of characteristic strength parameters (e.g. cu or φ ) rather than to
the characteristic resistance appearing in Eq. (3). The characteristic geotechnical resis-
tance, R̂u , would then be computed employing a (probably non-linear) model which
uses these characteristic ground parameters. Thus, the final bias of the characteristic re-
sistance depends not only on the distribution of the ground properties, but also on the
model used to predict R̂u . It will be assumed here that the coefficient of variation, vR , of
R̂u is approximately equal to the coefficient of variation of the ground parameters used
in the model, which are typically in the range of 0.1 to 0.3 (e.g., Meyerhof, 1995 and
Phoon and Kulhawy, 1999). Note that geotechnical resistance often involves an average
of ground properties, e.g. along a failure surface, which will have a smaller variability
than the point variability suggested in the literature. Thus, a reasonable value for the re-
sistance variability is deemed to be about vR = 0.15, which will be assumed here. Similar
to Eq. (14), the resistance bias factor assumed in the Eurocode can be computed from
R̂u = μR (1 − 1.645vR) → kR = 1/ (1 − 1.645vR) (15)
which for vR = 0.15 gives kR = 1.33.

The Australian Standard AS5100.3 (Standards Australia, 2004b) states that “the
characteristic value of a geotechnical parameter should be a conservatively assessed
value of the parameter.” Although the author was unable to find a more precise defini-
@seismicisolation
@seismicisolation
tion, the wording here suggests that the Australians are following the Eurocode approach.
Thus, a bias factor of kR = 1.33 will be assumed for Australia as well.
In North America, Commentary Clause C10.4.6.1 of AASHTO (2007) says that
“For strength limit states, average measured values were used to calibrate the resistance
factors”, which suggests that kR = 1.0. However, the commentary goes on to say that “it
may not be possible to reliably estimate the average value of the properties needed for
design. In such cases, the Engineer may have no choice but to use a more conservative
selection of design properties” which suggests that in practice, kR > 1.0.
Clause 8.5 of the Canadian Foundation Engineering Manual (Canadian Geotechni-
cal Society, 2006) states that “Frequently, the mean value, or a value slightly less than
the mean is selected by geotechnical engineers as the characteristic value.” Commen-
tary K of the NBCC User’s Guide (National Research Council of Canada, 2011) says
that “the [characteristic] resistance is the engineer’s best estimate of the ultimate resis-
tance.” Becker (1996a) claims “The design values do not necessarily need to be taken
as the mean values, although this is common geotechnical design practice.” All of these
statements suggest that kR = 1.0, or perhaps slightly greater than 1.0. However, Becker
(1996a) later argues that the characteristic resistance is typically selected to be somewhat
below the mean, due to sampling uncertainties, and he subsequently uses kR = 1.1 in his
NBCC development paper (Becker, 1996b). Based on Becker’s reasoning, the value of
kR = 1.1 will be assumed to apply to all of the North American design codes considered
here.
2.3. Load Factors
Load factors are designed to reflect uncertainty in the lifetime loads experienced by a
structure or foundation. The basic idea is to set the factored loads, αL F̂L and αD F̂D , to
values having sufficiently low probability of being exceeded by the true (random) life-
time loads. Considering, for example, live loads (with dead loads following the same
reasoning), the factored live load which has probability ε of being exceeded by the true
live load over the design lifetime can be approximated as
αL F̂L = μL (1 + zε vL ) (16)
in which zε is the standard normal point with exceedance probability ε , i.e. the point such
that Φ(−zε ) = ε , where Φ is the standard normal cumulative distribution function. Note
that Eq. (16) assumes that the live load is (at least approximately) normally distributed.
Rearranging Eq. (16) leads to an expression for the load factor, which is

μL
αL = (1 + zε vL ) = kL (1 + zε vL ) (17)
F̂L
ASCE-7 (American Society of Civil Engineers, 2010) found that their load factors
are well approximated by Eq. (17) when they set zε = ωL β , where β is the target reli-
ability index and ωL = 0.8 when L is a principle action or ωL = 0.4 when L is a com-
panion action. Equation (17) can be used for other load types simply by changing the
subscript. Note that Eq. (17) suggests that load factors are independent of the resistance
distribution. It also states that the load factors are very dependent on how the charac-
teristic load is defined, i.e. on the load bias factor, k. If designs have a common target
@seismicisolation
@seismicisolation
reliability index, β , and kL = 0.9 in North America and kL = 0.7 in Europe and Australia,
as suggested above, then one would expect the load factors in Europe and Australia to be
lower than those used in North America if Eq. (17) is accurate. As will be seen, the Euro-
pean and Australian load factors are generally higher than those used in North America
– the European and Australian codes compensate for their higher load factors through
higher resistance factors. In other words, Eq. (17) cannot be used as a general formula
for load factors. The magnitude of the resistance factors (and bias factors) must still be
considered.
Table 2 gives the load factors as specified by the various design codes considered
here. The last column of the table gives the total load factor, αT , for a given mean dead
to live load ratio, which scales the total mean load, μL + μD , to be equal to the sum of
factored live and dead loads. The total load factor can be seen in Eq. (13) and is defined
by

αL αD RD/L 1
αT = + (18)
kL kD 1 + RD/L
Table 2 Load and bias factors for various design codes.
Source kL kD αL αD αT
NBCC 2012 0.9 1.05 1.50 1.25 1.31
CHBDC 2006 0.9 1.05 1.70 1.20 1.33
AASHTO 2007 0.9 1.05 1.75 1.25 1.38
Eurocode 7 0.7 1.05 1.50 1.35 1.50
AS5100.3 0.7 1.05 1.80 1.20 1.50
The dead load factor for the Eurocode (1.35) is larger than the dead load factors used in
North America (1.2 to 1.25) which, when combined with the smaller value of kL , yields a
final αT value which is significantly larger than that appearing in the Canadian codes and
in AASHTO. The Australian Standard AS5100 has the second highest αT value because
of their relatively high live load factor, αL , and low live load bias factor, kL .
Table 3 shows the total effective load factor, the resistance bias, the resistance factor,
and the global factor of safety for the five design codes considered with respect to shallow
foundation bearing capacity (assuming Design Approach 2 and the GEO limit state for
the Eurocode).
Table 3 Global factor of safety for various design codes.
Source αT kR ϕgu Fs
NBCC 20121 1.31 1.1 0.50 2.88
CHBDC 2006 1.33 1.1 0.50 2.93
AASHTO 2007 1.38 1.1 0.45 − 0.5 3.04 − 3.37
Eurocode 72 1.50 1.33 0.71 2.81
AS5100.3 1.50 1.33 0.35 − 0.65 3.07 − 5.70
1 the NBCC itself does not specify resistance factors.The resistance factors shown above
appear in Appendix K of the NBCC User’s Guide (National Research Council, 2011).
2 based on Eurocode 7 Design Approach 2 for the GEO limit state.
@seismicisolation
@seismicisolation
Perhaps unsurprisingly, and despite the considerable variation in implementation details,

the five codes considered here all arrive at quite similar global factors of safety, Fs , as seen
in the last column of Table 3. Many assumptions were made in arriving at Table 3 about
how characteristic values are actually defined in the various codes, and so there may
actually be more discrepancy between the codes for this particular limit state. However,
it appears likely that codes are calibrated for much the same target failure probability
regardless of the implementation details. The author notes that, if this is the case, there
seems to be little justification in codes being different – we might as well all adopt the
same model and work in common towards a safer and more economical design code.
The model adopted worldwide should be the simplest and easiest to define.
3. Future Directions
Geotechnical engineers are, of course, well aware of the fact that their designs depend
on one of the most uncertain of all engineering materials. Unlike wood, concrete, steel,
and other quality controlled engineering materials, it is not even known how the natural
variability of soil properties should properly be characterized. In addition, geotechnical
engineers are also aware that their uncertainty about the resistance of a geotechnical
system decreases with increased site understanding and site modeling effort. Thus, there
is a real desire amongst the geotechnical community that their designs reflect the degree
of their site and modeling understanding. In other words, geotechnical designs should
become more economical as site and model understanding increases.
To accomplish this, it makes sense to have a resistance factor which is adjusted as
a function of site and model understanding. There are at least two advantages to such
an approach: 1) overall safety can be maintained at a common target maximum failure
probability, and 2) the direct economic advantage to having site and model understanding
can be demonstrated. For example, the current Canadian design codes specify a single
resistance factor for bearing capacity design (0.5). It doesn’t matter how confident one
is in one’s prediction of the bearing capacity of a foundation, the same resistance factor
must be used. Thus, there is no direct advantage to improving the geotechnical response
prediction if only a single resistance factor can be used – one might as well spend the
least one can on the site investigation and modeling since the resistance factor cannot
currently be increased.
The resulting desire for a resistance factor which depends on site and model under-
standing is not new. In 1991, the Institution of Civil Engineers (1991) made the classic
observation that “You pay for a site investigation whether you have one or not,” which
is, as we all know, entirely true. Recognizing this fact, it is of real economic value to
have a sliding resistance factor to reflect the true lifetime cost of the lack or presence
of a site investigation. The Australian Standard for Bridge Design, Part 3: Foundations
and Soil-Supporting Structures (AS 5100.3, Standards Australia, 2004b) has provided a
range in “geotechnical strength reduction factors” since 2004 with worded guidance as
to which end of the scale should be used. For example, AS 5100.3 suggests that the lower
end of the resistance factor range (more conservative) should be used for limited site in-
vestigations, simple methods of calculation, severe failure consequences, and so on. It is
of interest to note that the Australian Standard recommendations for the resistance fac-
tor considers both site and model understanding and failure consequence in their single
factor.
@seismicisolation
@seismicisolation
As is well known, the overall safety level of any design should depend on at least
three things: 1) the uncertainty in the loads, 2) the uncertainty in the resistance, and 3)
the severity of the failure consequences. These three items are all basically independent
of one another and in most modern codes are thus treated separately, as is reasonable. In
most codes, uncertainties in the loads are handled by load and load combination factors,
failure consequences are handled by applying a multiplicative importance factor to the
more uncertain loads (e.g. earthquake, snow, and wind), and uncertainties in resistance
are handled by material specific resistance factors (e.g. ϕc for concrete, ϕs for steel, etc).
Because the ground is so highly uncertain, similarly to earthquake, snow, and wind
loads, it makes sense to apply a partial safety factor to the ground that depends on both
the resistance uncertainty and consequence of failure. This would be analogous to how
wind load, for example, in the NBCC (NRC, 2010) has both a load factor associated
with wind speed uncertainty as well as an importance factor associated with failure con-
sequences. Figure 1 illustrates the basic idea, where the overall partial factor applied to
the geotechnical resistance varies with both uncertainty and failure consequence. The
numbers in the figure are relative to the default central partial factor (i.e. relative to 1.0)
and it is assumed that current geotechnical design approaches in Canada lead to typical
or default levels of site and model understanding so that, for typical failure consequence
geotechnical systems, the central value is what is currently used in design. From this
value, increased site investigation and/or modeling effort leads to lower uncertainty and a
higher resistance factor (and a more economical design). Similarly, for geotechnical sys-
tems with high failure consequences, e.g. failure of the foundation of a major multi-lane
highway bridge in a capital city, the resistance factor is decreased to ensure a decreased
maximum acceptable failure probability. Of particular note in Figure 1 is the fact that
if a geotechnical system with high failure consequences is designed with high site and
model uncertainty, the designer is penalized by a low partial factor.
HIGH
Consequence
1.0 0.8 0.6
TYPICAL 1.2 1.0 0.8

(default)
LOW 1.4 1.2 1.0

Consequence
LOW TYPICAL HIGH
Uncertainty Uncertainty
Figure 1. Floating partial safety factor, relative to the default, applied to geotechnical
resistance (numbers are for illustration only).
Figure 1 suggests that for each limit state (e.g. bearing, sliding, overturning, etc.)
a 3 x 3 matrix of resistance factors would have to be provided. Rather than introduc-
ing the resulting myriad tables, the multiplicative approach taken in structural engineer-
@seismicisolation
@seismicisolation
ing (where the load is multiplied by both a load factor and an importance factor) will
be adopted for geotechnical resistance as well. In other words, the overall safety factor
applied to geotechnical resistance is broken into two parts;
1) a resistance factor, ϕgu or ϕgs , which accounts for resistance uncertainty. This
factor basically aims to achieve a target maximum acceptable failure probabil-
ity equal to that used currently for geotechnical designs for typical failure conse-
quences (e.g. lifetime failure probability of 1/5,000 or less). The subscript g refers
to ‘geotechnical’ (or ‘ground’), while the subscripts u and s refer to ultimate and
serviceability limit states, respectively.
2) a consequence factor, Ψ, which accounts for failure consequences. Essentially, Ψ >
1 if failure consequences are low and Ψ < 1 if failure consequence exceed those of
typical geotechnical systems. For typical systems, or where system importance is
already accounted for adequately by load importance factors, Ψ = 1. The basic idea
of the consequence factor is to adjust the maximum acceptable failure probability of
the design down (e.g. 1/10,000) for high failure consequences, or up (e.g. 1/1,000)
for low failure consequences.
The geotechnical design would then proceed by ensuring that the factored geotechnical
resistance at least equals the effect of factored loads. For example, for ultimate limit
states, this means that in Canada the geotechnical design will soon need to satisfy an
equation of the form
Ψϕgu R̂ ≥ ∑ Ii ηi αui F̂ui (19)

i
which is almost identical to Eq. (2), with the exception that the overall geotechnical re-
sistance factor is expressed as the product of the consequence factor, Ψ, and the ultimate
geotechnical resistance factor, ϕgu , and the loads and load factors appearing on the right-
hand-side are also those specific for the ultimate limit state under consideration (and,
hence, the subscript u). An entirely similar equation must be satisfied for serviceability
limit states, with the subscript u replaced by s. The serviceability geotechnical resistance
factors, ϕgs , will be closer to 1.0 than ϕgu , since serviceability limit states can have larger
maximum acceptable probabilities of occurrence.
In Eq. (19), the value of Ψ is shown as being independent of the ultimate or service-
ability limit states (not subscripted by u or s). It is currently not known if this is true.
It is known that the target maximum failure probabilities for serviceability limit states
are greater than those for ultimate limit states. For example, a typical geotechnical sys-
tem might have a target maximum lifetime failure probability of 1/5, 000 for ultimate
limit states, but only 1/500 for serviceability limit states. If the geotechnical system has
high failure consequences, the lifetime maximum acceptable failure probability might
decrease by the same fraction, i.e. to 1/10, 000 for ULS and to 1/1, 000 for SLS. Thus,
it is suspected that the same (or similar) consequence factor can be used to scale the
target maximum acceptable failure probability for both ULS and SLS designs, since the
probabilities scale by the same amount. This is a topic of further investigation.
The geotechnical resistance factor, ϕgu or ϕgs , depends on the degree of site and
prediction model understanding. Three levels will be considered in the future editions of
the building and highway codes in Canada;
@seismicisolation
@seismicisolation
• High understanding: Extensive project-specific investigation procedures and/or

knowledge are combined with prediction models of demonstrated (or proven) qual-
ity to achieve a high level of confidence in performance predictions,
• Typical understanding: Usual project-specific investigation procedures and/or
knowledge are combined with conventional prediction models to achieve a typical
level of confidence in performance predictions,
• Low understanding: Understanding of the ground properties and behaviour are
based on limited representative information (e.g. previous experience, extrapolation
from nearby and/or similar sites, etc.) combined with conventional prediction mod-
els to achieve a lower level of confidence with the performance predictions.
The resulting table for ULS geotechnical resistance factors to appear in future Canadian
codes will look something like Table 4. NOTE: the numerical values appearing in Table
4 are purely illustrative and have yet to be determined.
The SLS geotechnical resistance factors will appear in a very similar table, an example
of which appears in Table 5. Again, all numbers are for illustration only.
The consequence factor, Ψ, adjusts the maximum acceptable failure probability of the
geotechnical system being designed to a value which is appropriate for the consequences.
Three failure consequence levels will be considered in future Canadian geotechnical de-
sign codes;
• High consequence: the geotechnical system is designed to be essential to post-
disaster recovery (e.g. hospital or lifeline bridge), and/or has large societal and/or
economic impacts.
• Typical consequence: the geotechnical system is designed for typical failure conse-
quences, e.g. the usual office building, bridge, etc. This will be the default failure
consequence level.
• Low consequence: failure of the geotechnical system poses little threat to human
or environmental safety, e.g. storage facilities, temporary structures, very low traffic
volume bridges, etc.
Table 4 Example table of ultimate limit state geotechnical resistance factors (static
loading), ϕgu . All numbers are for illustration only – factors have not yet
been determined.
Limit State Degree of Understanding

Low Typical High
Shallow Foundations
Bearing resistance 0.45 0.50 0.60
Passive resistance 0.40 0.50 0.60
Horizontal resistance (sliding) 0.75 0.80 0.85
Ground Anchors
Static Analysis – tension 0.30 0.40 0.50
Static Test – tension 0.55 0.60 0.65
Deep Foundations – Piles
Static Analysis
Compression 0.35 0.40 0.50
Tension 0.35 0.40 0.45
etc.
@seismicisolation
@seismicisolation
Table 5 Example table of serviceability limit state geotechnical resistance factors

(static loading), ϕgs . All numbers are for illustration only – factors have
not yet been determined.
Limit State Degree of Understanding

Low Typical High
Shallow Foundations
Settlement 0.70 0.90 1.00
Embankments
Lateral displacement 0.60 0.70 0.80
Deep Foundations – Piles
Lateral displacements 0.70 0.80 0.90
etc.
Table 6 illustrates how the consequence factor table will appear in future Canadian
geotechnical design codes. As with Tables 4 and 5, the factors are not finalized and are
for illustration only.
Table 6 Example consequence factor table for the Canadian Highway Bridge De-
sign Code. Numbers are for illustration only.
Consequence Example Consequence

Level Factor, Ψ
High Lifelines, Emergency 0.9
Typical Highway Bridges 1.0
Low Secondary Bridges 1.1
4. Conclusions
The evolution of geotechnical design codes, from traditional working stress design (fac-
tor of safety) to reliability-based design approaches, has been lagging well behind struc-
tural design codes. There is no question that this lag is due to the much larger uncertainty
about the ground than exists with most other engineering materials. A batch of 30 MPa
concrete will have pretty much the same distribution in strength properties whether or-
dered in Calgary or in Tokyo. On the other hand, the ground properties at sites in these
two cities will almost certainly be significantly different – in fact, ground properties will
usually differ from point to point within the same site.
In general, all sources of uncertainty entering into the LRFD equation (e.g. Eq. 2)
are factored to arrive at an acceptably safe design solution. The factors applied are related
to the magnitude of the uncertainty in the parameter being factored. For example, the
uncertainty associated with steel reinforcing is less than that with concrete and so the
steel resistance factor is closer to 1.0 than is the concrete resistance factor. Reduced
uncertainty about an engineering material results in an increased resistance factor. The
ground is simply another engineering material and the resistance factor associated with
the ground should be related to its site specific uncertainty. Since the distribution of the
@seismicisolation
@seismicisolation
ground strength varies from site to site and even within a site, it makes sense to relate
the geotechnical resistance factor to the magnitude of the residual uncertainty, i.e. the
uncertainty remaining after site investigation and modeling efforts have been accounted
for.
The structural design codes in Canada also recognize the fact that some of the pa-
rameters in the LRFD equation are highly variable and so need to be treated with special
consideration in the event that the system being designed is of higher or lesser impor-
tance, i.e. if the failure consequences are higher or lower than usual. Because earthquake,
wind, and snow load are highly variable and have site specific distributions, both the
NBCC and the CHBDC apply importance factors to these loads. The importance fac-
tors increase with increasing system importance. Similarly, ground properties are both
highly variable and site specific, and so the application of a factor to account for system
importance is appropriate on the resistance side.
With the above thoughts in mind, the next editions of the Canadian Highway Bridge
Design Code and the National Building Code of Canada will include several philosoph-
ical changes to their geotechnical design provisions. These include;
• the introduction of three levels of site and model understanding – high, typical, and
low – through the ULS and SLS resistance factors. These factors are intended to
account for site and modeling uncertainties and are aimed at producing a design
with a target maximum acceptable failure probability for typical geotechnical sys-
tems (i.e. systems having typical failure consequence levels). For example, ULS and
SLS maximum acceptable lifetime failure probabilities might be 1/5,000 and 1/500,
respectively, and so these resistance factors would be targeted at these values.
• the introduction of three levels of failure consequence – high, typical, and low –
through a consequence factor which multiplies the factored resistance. The basic
idea of the consequence factor is to allow the target maximum acceptable lifetime
failure probability provided by the resistance factor to be adjusted up or down de-
pending on whether the failure consequences are lower or higher than typical.
Research into the determination of the required resistance and consequence factors for
the Canadian codes is ongoing. The consequence factor is a new idea and work is still
needed to determine when it should and should not be applied. For example, whether
both the consequence factor and importance factors should be applied simultaneously is
unknown, but initially, they will not be.
References
Allen, D.E. (1975). Limit States Design – A probabilistic study, Canadian Journal of
Civil Engineering 2(1), 36–49.
American Association of State Highway and Transportation Officials (2007). LRFD
Bridge Design Specifications, Washington, DC.
American Society of Civil Engineers (2010). Minimum Design Loads for Buildings and
Other Structures, ASCE Standard ASCE/SEI 7-10, Reston, Virginia.
Bartlett, F.M., Hong, H.P. and Zhou, W. (2003). Load factor calibration for the proposed
2005 edition of the National Building Code of Canada: Statistics of loads and load
effects, Canadian Journal of Civil Engineering 30(2), 429–439.
Becker, D.E. (1996a). Eighteenth Canadian Geotechnical Colloquium: Limit states de-
sign for foundations. Part 1. An overview of the foundation design process, Cana-
dian Geotechnical Journal 33(6), 956–983.
@seismicisolation
@seismicisolation
Becker, D.E. (1996b). Eighteenth Canadian Geotechnical Colloquium: Limit states

design for foundations. Part II. Development for the National Building Code of
Canada, Canadian Geotechnical Journal 33(6), 984–1007.
British Standard BS EN 1990 (2002a). Eurocode – Basis of Structural Design, CEN
(European Committee for Standardization), Brussels.
British Standard BS EN 1991-1-1 (2002b). Eurocode 1: Actions on Structures – Part
1-1: General Actions – Densities, Self-weight, Imposed Load for Buildings, CEN
(European Committee for Standardization), Brussels.
British Standard BS EN 1997-1:2004 (2004). Eurocode 7: Geotechnical design – Part 1:
General rules, CEN (European Committee for Standardization), Brussels.
Canadian Geotechnical Society (1992). Canadian Foundation Engineering Manual, 3rd
Ed., Montreal, Quebec.
Canadian Geotechnical Society (2006). Canadian Foundation Engineering Manual, 4th
Ed., Montreal, Quebec.
Canadian Standards Association (2006). Canadian Highway Bridge Design Code,
CAN/CSA-S6-06, Mississauga, Ontario.
Danish Geotechnical Institute (1985). Code of Practice for Foundation Engineering,
DGI-Bulletin No. 36, Copenhagen.
Ellingwood, B.R. (1999). A Comparison of General Design and Load Requirements in
Building Codes in Canada, Mexico, and the United States, Engineering Journal 2,
67–81.
Ellingwood, B.R., Galambos, T.V., MacGregor, J.G. and Cornell, C.A. (1980). Develop-
ment of a Probability Based Load Criterion for American National Standard A58:
Building Code Requirements for Minimum Design Loads in Buildings and Other
Structures, National Bureau of Standards, U.S. Dept. of Commerce, NSC Special
Publication 577, Washington, D.C..
Green, R. and Becker, D. (2000). National report on limit state design in geotechnical
engineering: Canada, in LSD2000: International Workshop on Limit State Design in
Geotechnical Engineering, ISSMGE, TC23, Melbourne, Australia.
Hicks, M.A. (2012). An explanation of characteristic values of soil properties in Eu-
rocode 7, in Modern Geotechnical Design Codes of Practice - Development, Cali-
bration & Experiences, edited by Arnold, P., Fenton, G.A., Hicks, M.A., Schweck-
endiek, T. and Simpson, B., IOS Press, Amsterdam, The Netherlands.
Institution of Civil Engineers (1991). Inadequate Site Investigation, Thomas Telford,
London.
Meyerhof, G.G. (1951). The ultimate bearing capacity of foundations. Géotechnique
2(4), 301–332.
Meyerhof, G.G. (1963). Some recent research on the bearing capacity of foundations,
Canadian Geotechnical Journal 1(1), 16–26.
Meyerhof, G.G. (1995). Development of geotechnical limit state design. Canadian
Geotechnical Journal 32(1), 128–136.
NCHRP (1991). Manuals for the Design of Bridge Foundations, Report 343, National
Cooperative Highway Research Program, Transportation Research Board, NRC,
Washington, DC.
NCHRP (2004). Load and Resistance Factors for Earth Pressures on Bridge Substructurs
and Retaining Walls, Report 12-55, D’Appolinia and the University of Michigan,
National Cooperative Highway Research Program, Transportation Research Board,
NRC, Washington, DC.
National Research Council (2011). User’s Guide – NBC 2010 Structural Commentaries
(Part 4 of Division B), 3rd Ed., National Research Council of Canada, Ottawa.
National Research Council (2010). National Building Code of Canada, 13th Ed., Na-
tional Research Council of Canada, Ottawa.
@seismicisolation
@seismicisolation
Phoon, K-K., and Kulhawy, F.H. (1999). Characterization of geotechnical variability,

Canadian Geotechnical Journal 36(4), 612–624.
Prandtl, L. 1921. Uber die Eindringungsfestigkeit (Harte) plastischer Baustoffe und
die Festigkeit von Schneiden, Zeitschrift fur angewandte Mathematik und Mechanik
1(1), 15–20.
Schneider, H.R. 2012. Dealing with uncertainties in EC7 with emphasis on determi-
nation of characteristic soil properties, in Modern Geotechnical Design Codes of
Practice - Development, Calibration & Experiences, edited by Arnold, P., Fenton,
G.A., Hicks, M.A., Schweckendiek, T. and Simpson, B., IOS Press, Amsterdam,
The Netherlands.
Standards Australia (2002a). Structural Design Actions, Part 1: Permanent, imposed and
other actions, Australian/New Zealand Standard, AS 1170.1:2002, Sydney, Aus-
tralia.
Standards Australia (2002b). Earth-Retaining Structures, Australian Standard AS 4678–
2002, Sydney, Australia.
Standards Australia (2004a). Bridge Design, Part 1: Scope and General Principles,
Australian Standard AS 5100.1–2004, Sydney, Australia.
Standards Australia (2004b). Bridge Design, Part 3: Foundations and Soil-Supporting
Structures, Australian Standard AS 5100.3–2004, Sydney, Australia.
@seismicisolation
@seismicisolation
IOS Press, 2013
doi:10.3233/978-1-61499-163-2-295
Can We Do Better Than the Constant

Partial Factor Design Format?
Kok-Kwang PHOONa,1 and Jianye CHING b
a
Department of Civil and Environmental Engineering, National University of
Singapore, Singapore
b
Department of Civil Engineering, National Taiwan University, Taiwan
Abstract. A partial factor format involves multiplying a nominal resistance such

as the mean bearing capacity by a resistance factor or dividing a characteristic
strength such as the lower 5% undrained shear strength by a partial factor. The
purpose of a partial factor format is to implement reliability-based design without
requiring the engineer to perform reliability analysis. Hence, the performance of a
partial factor format must be measured by its ability to produce designs achieving
a desired target reliability index within an acceptable error margin. It is an article
of faith that it is possible to achieve an approximately consistent reliability index
by recommending a single numerical value for a single resistance factor. This
practice is widely adopted in the form of the Load and Resistance Factor Design
(LRFD) approach. Based on the examples studied in this paper (pile installed in
homogeneous clay and pile installed in clay and sand), it is demonstrated
numerically that the deviation from the target reliability index can be so large,
particularly on the unconservative side, that the purpose of reliability calibration
becomes nearly meaningless. In other words, the partial factor format may not
perform significantly better than an uncalibrated format in achieving the
prescribed target reliability index. In our opinion, it is difficult to provide a
compelling reason to engineers to go through the hassle of changing a code format
in this case. If a design guide is insistent on recommending a single numerical
value in association with a partial factor format, it is more sensible to recommend
a constant quantile value, rather than a constant resistance factor.
Keywords. Partial factor format, reliability calibration, LRFD, MRFD, piles.
Introduction
Reliability-based design is presently implemented in the form of familiar “look and

feel” code formats such as the partial factor format, Load and Resistance Factor Design
(LRFD) format, and Multiple Resistance Factor Design (MRFD) format (Phoon et al.
2003b), among others. The common feature of these formats involves multiplying a
nominal resistance such as the mean bearing capacity by a resistance factor or dividing
a characteristic strength such as the lower 5% undrained shear strength by a partial
factor. For brevity, we broadly term these code formats as partial factor formats.
From the perspective of an engineer, there is no difference between applying
partial factor formats and the prevailing factor of safety format, other than adopting a
different numerical value or more commonly, a set of numerical values for different
1
Department of Civil and Environmental Engineering, National University of Singapore, Block E1A,
#07-03, 1 Engineering Drive 2, Singapore 117576, Singapore; Email: kkphoon@nus.edu.sg.
@seismicisolation
@seismicisolation
296 K.-K. Phoon and J. Ching / Can We Do Better than the Constant Partial Factor Design Format?
resistance and load components mandated in such codes. The key difference lies with
the design goal. In partial factor formats, the numerical values of these factors are not
based purely on experience or precedents, but calibrated by the code developer using
reliability analysis to achieve a desired target reliability index. It is clear that partial
factor formats are meant to implement reliability-based design without requiring the
engineer to perform reliability analysis. The obvious limitation associated with
replacing reliability analysis with a simple multi-factor algebraic design check is that
the target reliability index cannot be achieved exactly.
The performance of a partial factor format must be measured by its ability to
produce designs achieving a desired target reliability index within an acceptable error
margin. When the partial factor format is first introduced into a design code, it should
preferably produce designs comparable to those produced by the factor of safety
method for continuity with past practice and experience. In fact, the target reliability
index is commonly prescribed to comply with this judicious continuity principle.
However, the primary goal must be to maintain a uniform level of reliability – this is
the key basis for switching to reliability-based design in the first place. For a partial
factor format, the ability to maintain a uniform level of reliability is primarily related to
the range of design scenarios covered by the code and the number of available partial
factors that can be “tuned” during the reliability calibration process.
A partial factor format should reveal the maximum deviation from the target
reliability index among the range of design scenarios appearing in the calibration
domain. In principle, application of a partial factor format to a design scenario lying
outside the calibration domain can produce a reliability index far from the target value.
Hence, it is important to state the salient features of the underlying calibration domain
explicitly in association with any partial factor format to avoid conveying the
impression that it can be applied to any design scenario, which is unlikely to be true.
One noteworthy feature of this calibration domain that is distinctive to geotechnical
engineering is that coefficients of variation of geotechnical parameters can vary over a
wide range, because of diverse evaluation methodologies to cater for diverse practice
and site conditions. It is easy to envisage that a single partial factor is unable to
achieve a uniform reliability index if the range of coefficients of variation is
sufficiently large, say between 10% and 70% for undrained shear strength estimated
using different methods as shown in Table 1.

There are three ways to respond to this issue. One, the code developer accepts a
large error band in the actual reliability indices produced by the partial factor format.
Nonetheless, this paper shows that the error can be so large, particularly on the
unconservative side, that the purpose of reliability calibration becomes nearly
meaningless. In other words, the partial factor format may not perform significantly
better than an uncalibrated format in achieving the prescribed target reliability index.
In our opinion, it is difficult to provide a compelling reason to engineers to go through
the hassle of changing a code format in this case. Second, the range of coefficients of
variation is fortuitously narrow for the design scenarios commonly found in a particular
locale. This is an unlikely situation for large countries with diverse geologic conditions
and diverse local practice. Third, the range of coefficients of variation is divided into
sufficiently small segments during reliability calibration such that a constant partial
factor can assure a reasonably uniform reliability level over each segment.
The third method is probably the most realistic approach currently. Phoon et al.
(1995) proposed a simple three-tier scheme (low, medium, high) for coefficients of
variation (Table 1) and calibrated LRFD and MRFD factors for each variability tier. In
@seismicisolation
@seismicisolation
K.-K. Phoon and J. Ching / Can We Do Better than the Constant Partial Factor Design Format? 297
other words, three numerical values are associated with each LRFD/MRFD factor,
rather than recommending a single numerical value for each LRFD factor (a practice
commonly adopted in most structural design codes). The main drawback with this
approach is that the reliability index may be affected by other deterministic/statistical
parameters such as the mean value of a strength parameter. If the mean value of a
strength parameter also varies over a wide range (not unusual in geotechnical
engineering), then it must be considered in the calibration domain. Hence, the
calibration domain involves two dimensions (coefficient of variation and mean of
strength parameter). It is clear that there are nine calibration subdomains if each
dimension is segmented into three parts as shown in Figure 1. There is a potential
practical problem here associated with unwieldy proliferation of reliability-calibrated
factors as the dimension of the calibration domain increases. It is possible to mitigate
this problem slightly by deleting subdomains deemed unrealistic in practice, but the
fundamental proliferation problem remains.
Table 1. Ranges of soil property variability for reliability calibration (Phoon et al. 1995, updated Phoon &
Kulhawy 2008).
Geotechnical parameter Property variability Coefficient of variation (%)
Undrained shear strength Lowa 10 - 30
Mediumb 30 - 50
c
High 50 - 70
Effective stress friction angle Lowa 5 - 10
Mediumb 10 - 15
c
High 15 - 20
Horizontal stress coefficient Lowa 30 - 50
Mediumb 50 - 70
Highc 70 - 90
a - typical of good quality direct lab or field measurements
b - typical of indirect correlations with good field data, except for the standard penetration test (SPT)
c - typical of indirect correlations with SPT field data and with strictly empirical correlations
Fig. 1 Segmentation of 2D parameter space into 9 subdomains for reliability calibration (Phoon et al. 1995).
@seismicisolation
@seismicisolation
There are two objectives in this paper. First, the degrees of deviation from the
target reliability index produced by existing partial factor formats are demonstrated for
piles in homogeneous and layered soils. It will be demonstrated that the performance
of existing partial factor formats in the context of achieving the target reliability index
can be poor under some fairly common circumstances. Second, the quantile value
method proposed by Ching & Phoon (2011) is able to mitigate this poor performance
associated with stretching one numerical value/partial factor to cover diverse
calibration design scenarios on the one hand and proliferation of numerical
values/partial factor arising from segmentation of calibration design scenarios on the
other hand. It is accurate to say that considerable practical challenges remain in the
development of a simplified reliability-based design format suitable for realistic
application in geotechnical engineering, even for fairly routine pile design.
1. Piles Installed in Homogeneous Clay
Consider a pile installed in homogeneous clay with an average undrained shear strength
= Su as shown in Figure 2. It is subjected to a dead load (DL) and a live load (LL).
The diameter (B) of the pile is 1 m and the length is L. The loadings DL and LL are
lognormally distributed with mean values (μDL, μLL) and coefficients of variation
(c.o.v.) (δDL, δLL). Suppose that μLL/μDL = 0.5, δDL= 10%, and δLL = 20%. The
undrained shear strength Su is lognormally distributed with mean value μsu and c.o.v.
δsu.
In this example, it is assumed that the pile is long and hence, it is reasonable to
estimate the pile capacity based on the side resistance alone. The side resistance is
computed using the α method = αSuπLB, in which the adhesion factor, α = 14.88 × εα
× Su-0.7, 14.88 is an empirical coefficient for the correlation, and εα characterizes the
model error. It is assumed that ln(εα) is normally distributed with mean = 0 and
standard deviation = 0.3. This probabilistic model for the adhesion factor is obtained
from calibration with instrumented load test data as shown in Figure 3 (Chen et al.
2012). It is important to note that the undrained shear strength is referenced with
respect to the direct simple shear test (DSS). Failure is defined as the state in which the
total load (DL+LL) exceeds the side resistance (tip resistance assumed to be minor) or
the safety ratio is less than one:
α(Su , εα )Su πLB (14.88εαS−u0.7 )Su πLB

SR(Y, θ) = = <1 (1)
DL + LL DL + LL
in which SR = safety ratio depending on a vector Y consisting of 4 lognormal random

variables (Su, εα, DL, LL) and a vector θ consisting of three design parameters (μsu, δsu,
L). In other words, the dimension of the parameter space or calibration domain is three.
The range of values for (μsu, δsu) depends on the site conditions, site investigation
practices, and soil property evaluation methodologies covered in the design code. The
parameter L allows a range of loads possibly corresponding to a range of typical
superstructures found in the locale governed by the design code to be considered. The
calibration domain D is defined as the rectangular volume formed by the following
representative ranges of values: 50 ≤ μsu ≤ 200 kN/m2, 0.1 ≤ δsu ≤ 0.5, and 10 m ≤ L ≤
50 m.
@seismicisolation
@seismicisolation
Fig. 2 Pile installed in homogeneous clay.

Fig. 3 α-su correlation for drilled shafts in clay (Chen et al. 2012).
In reliability-based design, the goal is to ensure that all design scenarios (discrete
points in domain D) will achieve a prescribed probability of failure, p T, i.e.:
⎡ (14.88Su−0.7 εα )Su πLB ⎤

P⎢ < 1 μsu , δsu , L⎥ ≤ pT (2)
⎣ DL + LL ⎦
@seismicisolation
@seismicisolation
Note that there are two Su terms in Eq. (2). The first term is used to calculate the
adhesion factor, α. A full probabilistic design method will require Eq. (2) to be
evaluated repeatedly for different trial values of L until the target probability of failure
is achieved. For routine pile design, all existing design codes prefer to apply a simpler
partial factor format such as:
[14.88(Scu / γ su ) −0.7 εcα (Scu / γ su ) / γ R ]πLB

≥1 (3)
γ DL DLc + γ LL LLc
in which (εαc, Suc, DLc, LLc) are the characteristic values of the random variables (εα,
Su, DL, LL), γsu is the partial factor for undrained shear strength, γR is the partial factor
for side resistance and (γDL, γLL) are load partial factors. In Eurocode 7 (British
Standards Institute 2004), Clause 2.4.7.3.4.2 “Design Approach 1”, it is recommended
that the ultimate limit state of axially loaded piles should be verified using two
combinations of partial factors. In Combination 1 (A1 “+” M1 “+” R1), Eq. (3) is
verified using
1. γDL = 1.35 and γLL = 1.5 (Table A.3, Set A1),
2. γsu = 1.0 (Table A.4, Set M1), and
3. γR = 1.0 [Table A.7, Set R1 for “Shaft (compression)”]
It is evident from Eq. (3) that Combination 1 applies a combined resistance factor γsu ×
γR = 1.0. In Combination 2, (A2 “+” M1 “+” R4), Eq. (3) is verified using
1. γDL = 1.0 and γLL = 1.3 (Table A.3, Set A2),
2. γsu = 1.0 (Table A.4, Set M1), and
3. γR = 1.3 [Table A.7, Set R4 for “Shaft (compression)”]
It is evident from Eq. (3) that Combination 2 essentially applies a combined resistance
factor γsu × γR = 1.3 with reduced load factors. Eurocode 7 (British Standards Institute
2004) is silent on how one should factor a soil parameter when it increases resistance.
An example is the adhesion factor, α. The adhesion factor computed from a factored
undrained shear strength will be higher than that computed from an unfactored
undrained shear strength. If an unfactored undrained shear strength is adopted for
evaluating the adhesion factor [i.e. α = α(Suc)] while a factored undrained shear
strength is adopted for evaluating the side resistance [i.e. α × Su = α(Suc) × Suc/γsu], one
could be criticized for postulating a physically unrealistic scenario. It is possible that
Eurocode 7 (British Standards Institute 2004) is cognizant of this difficulty and
essentially adopts a resistance factor approach for piles, i.e. γsu = 1.
2. Reliability Calibration
In this study, the characteristic values of the random variables (S u, DL, LL) are
assigned to their respective mean values, i.e. Suc = μsu, DLc = μDL, and LLc = μLL.
Eurocode 7 (British Standards Institute 2004) is silent on how to select a characteristic
value and partial factor for a model error such as εα. It is assumed that εαc = 1 and this
characteristic value of εα is unfactored as shown in Eq. (3). Load factors from Set A1
are adopted and the goal is to calibrate the following partial factor format:
@seismicisolation
@seismicisolation
(14.88μsu−0.7 )μsu πLB / γ R

≥1 (4)
1.35μ DL + 1.5μ LL
To be more precise, the goal is to calibrate a single numerical value or multiple

numerical values for the resistance factor γR so as to minimize the degree of deviation
from the target reliability index over the range of design scenarios in domain D (50 ≤
μsu ≤ 200 kN/m2, 0.1 ≤ δsu ≤ 0.5, and 10 m ≤ L ≤ 50 m). The target reliability index of
3.8 recommended by Eurocode 7 (British Standards Institute 2004) is adopted in this
study. Two reliability calibration methods are adopted: (1) optimization method and
(2) quantile value method.
2.1. Optimization Method
The optimization method is widely adopted in the literature (e.g., Phoon et al. 1995,
Phoon et al. 2003a, Sørensen 2002). The key step involves solving the following
optimization problem:
n
min ∑ [βi ( γ R ) − βT ]2 (5)
λR
i =1
in which βi is the actual reliability index for the i-th calibration case, which is a
function of the resistance factor, γR, and βT is the target reliability index = 3.8. In this
study, n = 1000 points are selected randomly from domain D for calibration. For each
trial value of γR (e.g., γR = 1.5) and the values associated with the ith calibration point in
D (e.g., μsu = 100 kN/m2, δsu = 0.3, and L = 30 m), it is possible to calculate the critical
mean dead load such that Eq. (4) is satisfied exactly under the hypothesis μLL/μDL = 0.5
and B = 1 m:
(14.88μ su−0.7 )μ su πLB / γ R

μ*DL =
(1.35 + 1.5μ LL / μ DL )
(6)
14.88 × 1000.3 × π × 30 /1.5
= = 1772.4 kN
1.35 + 1.5 × 0.5
Once the critical mean dead load μDL* is obtained, the reliability index of the ith
pile can be computed using Monte Carlo simulation. Note that this is possible because
all the random variables are now fully defined: undrained shear strength is lognormally
distributed with μsu = 100 kN/m2, δsu = 0.3, dead load is lognormally distributed with
μDL = μDL*, δDL = 0.1, and live load is lognormally distributed with μLL = 0.5×μDL*, δLL
= 0.2. The optimal value for γR is the one that minimizes the sum of the squares of the
deviation from the target reliability index. It is possible to obtain a single numerical
value for γR that covers the entire domain D. An alternate strategy is to divide D into
subdomains. The following subdomains are studied in this paper:
1. 50≤μsu≤100 kN/m2 and 0.1≤δsu≤ 0.3

2. 100≤μsu≤200 kN/m2 and 0.1≤δsu≤ 0.3
@seismicisolation
@seismicisolation
3. 50≤μsu≤100 kN/m2 and 0.3≤δsu≤ 0.5

4. 100≤μsu≤200 kN/m2 and 0.3≤δsu≤ 0.5
In this strategy, a potentially distinct optimal value can be obtained for γR in each
subdomain.
Even within the calibration domain, one would not expect the target reliability
index to be achieved exactly for every point, regardless of whether the calibration is
undertaken over the entire domain or several subdomains. If this were to be the case,
the objective function shown in Eq. (5) will converge to a minimum value of zero. Of
course it is possible to achieve the target reliability index exactly if one customizes a
resistance factor for each point in D, but this extreme approach is clearly impractical
and actually identical to the fully probabilistic approach requiring an engineer to solve
Eq. (2) directly rather than perform a simple algebraic check such as Eq. (4). In other
words, if too many values are recommended for the resistance factor, the intention for
adopting the familiar “look and feel” partial factor format is undermined.
The more realistic expectation is to hope that an independent set of points in
domain D would achieve reliability indices sufficiently close to the target value. In this
study, 1000 independent validation points are randomly selected from domain D. The
reliability indices associated with these validation points, βV, can be computed
following the same procedure described above for the calibration points. The key
statistics for βV [mean value, coefficient of variation (c.o.v.), maximum value and
minimum value] are reported in Table 2. A partial factor format such as Eq. (4) can be
considered “perfect” if mean βV = maximum βV = minimum βV = βT and c.o.v. of βV =
0.
It is useful to note that the difference in the reliability index between two reliability
classes is 0.5 in the head Eurocode (Table B2, British Standards Institute 2002). Hence,
it is reasonable to propose that a partial factor format is satisfactory if the difference
between the mean of βV and the minimum of βV is less than 0.5. It is clear that one
may accidentally drop to a lower reliability class if this threshold were to be exceeded.
Note that the minimum of βV refers to the most unconservative design produced by the
partial factor format within the 1000 validation points and it is therefore more critical
than the maximum of βV. A less stringent yardstick for satisfactory performance is to
require the standard deviation of βV to be less than 0.5. Other yardsticks, such as
requiring the difference between the median of βV and the 25th percentile of βV to be
less than 0.5, are reasonable as well, but they are not explored in this study.
Based on the last two rows of Table 2, it would appear that the standard practice of
calibrating a single numerical value for γR is satisfactory for this simple example. The
subdomain strategy does not appear to be necessary. Nonetheless, it is useful to
observe that the overall uncertainty in the resistance (14.88×Su0.3×εα×πBL) is
dominated by the c.o.v. of the model error rather than the c.o.v. of the soil parameter
(δsu). In fact, the approximate c.o.v. of the resistance is [0.3 2δsu2 + 0.32]0.5. Under this
condition, the effect of δsu varying over a wide range is muted, because the c.o.v. of the
resistance only varies from 0.30 to 0.33 when δsu varies from 0.1 to 0.5. The upper
bound is only 10% higher than the lower bound. It is possible to amplify the effect of
δsu by artificially reducing the standard deviation of ln(εα) from 0.3 to 0.1. The c.o.v.
of the resistance varies from 0.10 to 0.18 in this case. The upper bound is now nearly
two times the lower bound! The results are shown in Table 3. A single value for γR
cannot cater for the range of δsu in this case.
@seismicisolation
@seismicisolation
The values of the resistance factor reported in Table 2 are about 2.5, which is
significantly higher than the recommended value of γR = 1 in Eurocode 7 (British
Standards Institute 2004). One possible explanation is the choice of defining the
characteristic value of the resistance around the mean value (S uc = μsu and εαc = 1). If a
5% quantile is adopted as the characteristic value, it can be shown that the γR values in
Table 2 should be divided by approximately 1.64 for δsu = 0.1 and 1.72 for δsu = 0.5. In
Table 3, the γR values should be divided by approximately 1.18 for δsu = 0.1 and 1.34
for δsu = 0.5. In other words, it appears that the combination [γR = 1, Suc = 5% quantile,
standard deviation of ln(εα) = 0.1] can approximately achieve a target reliability index
of 3.8 for this example. A more thorough examination of quantiles is presented below.
2.2. Quantile Value Method (QVM)
The quantile value method requires a different design format:
14.88 × (Sηu ) −0.7 × ε ηα × Sηu × πLB

1−η ≥1 (7)
DL1−η + LL
Table 2. Performance of optimization and quantile value methods for a single pile installed in clay [standard
deviation of ln(εα) = 0.3]
Quantile Optimization method

value
method No
subdomains Four subdomains
(QVM)
50≤μsu≤200 50≤μsu≤100 100≤μsu≤200 50≤μsu≤100 100≤μsu≤200
0.1≤δsu≤0.5 0.1≤δsu≤ 0.3 0.1≤δsu≤ 0.3 0.3≤δsu≤ 0.5 0.3≤δsu≤ 0.5
* *
γR or η η = 0.0092 γR = 2.43 γR = 2.41 γR = 2.41 γR = 2.64 γR = 2.64
βV mean 3.77 3.72 3.80 3.81 3.82 3.84

βV c.o.v. 0.04 0.04 0.01 0.01 0.02 0.02
βV max 3.94 4.01 3.85 3.85 3.94 3.94
βV min 3.50 3.49 3.69 3.69 3.65 3.65
mean - min 0.27 0.23 0.11 0.12 0.17 0.19

std deviation 0.15 0.15 0.04 0.04 0.08 0.08
Table 3. Performance of optimization and quantile value methods for a single pile installed in clay [standard
deviation of ln(εα) = 0.1]

value
method No
(QVM)
50≤μsu≤200 50≤μsu≤100 100≤μsu≤200 50≤μsu≤100 100≤μsu≤200
0.1≤δsu≤0.5 0.1≤δsu≤ 0.3 0.1≤δsu≤ 0.3 0.3≤δsu≤ 0.5 0.3≤δsu≤ 0.5
γR or η* η* = 0.065 γR = 1.30 γR = 1.26 γR = 1.26 γR = 1.43 γR = 1.44
βV mean 3.77 3.54 3.72 3.71 3.81 3.83

βV c.o.v. 0.02 0.12 0.05 0.05 0.07 0.07
βV max 3.89 4.75 4.11 4.11 4.26 4.26
βV min 3.55 2.85 3.44 3.42 3.36 3.39
mean - min 0.22 0.69 0.28 0.29 0.45 0.44

std deviation 0.08 0.42 0.19 0.19 0.27 0.27
@seismicisolation
@seismicisolation
in which Xη denotes the η-quantile of the random variable X. If η is less than 0.5 (η
generally of the order of 0.01), Φ-1(η) is negative. Hence, it is also clear that
parameters related to strength are factored downwards and parameters related to loads
are factored upwards. Eq. (7) is sensible and partly in line with our current practice of
adopting conservative parameters on top of applying a global factor of safety. The key
difference is that quantiles are well defined points in a cumulative distribution function,
while “conservative” parameters are selected based on data, physics, experience, and
engineering judgment or some combination thereof. The quantiles corresponding to εα,
Su, DL and LL are given below:
ε ηα = exp ⎡⎣ σln( εα ) ⋅ Φ −1 ( η)⎤⎦ (8a)

⎡ ⎛ μ ⎞ ⎤
Sηu = exp ⎢ ln ⎜ su
⎟+ ln (1 + δsu2 ) ⋅ Φ −1 ( η) ⎥ (8b)
⎢ ⎜⎝ 1 + δsu ⎟⎠
2
⎥
⎣ ⎦
⎡ ⎛ μ ⎞ ⎤
DL1−η = exp ⎢ln ⎜ DL
⎟ − ln (1 + δ2DL ) ⋅ Φ −1 ( η) ⎥ (8c)
⎢ ⎜⎝ 1 + δ2DL ⎟
⎠ ⎥
⎣ ⎦
⎡ ⎛ μ ⎞ ⎤
LL1−η = exp ⎢ ln ⎜ LL
⎟ − ln (1 + δ2LL ) ⋅ Φ −1 ( η) ⎥ (8d)
⎢ ⎜⎝ 1 + δ2LL ⎟
⎠ ⎥
⎣ ⎦
It is clear from Eq. (8) that quantiles are proper constants. For example, if η = 0.05, it
can be shown from Eq. (8b) that Su0.05 = 0.84μsu for δsu = 0.1 and Su0.05 = 0. 41μsu for δsu
= 0.5. Hence, Eq. (7) is an algebraic check. In terms of calculation cost, it is
comparable to the more familiar Eq. (4). The key difference is that a quantile value is
applied for design, rather than a resistance factor. It will be shown below that applying
a constant quantile value is mathematically equivalent to applying a variable resistance
factor.
In this study, a modified version of Eq. (7) is adopted to follow the standard load
factor format:
14.88 × (Sηu ) −0.7 × εαη × Sηu × πLB

≥1 (9)
1.35μ DL + 1.5μ LL
The rationale is that loads are established by structural engineers and it is convenient to
follow structural code formats. The characteristic loads are again assumed to be mean
loads.
The quantile η is determined from a different calibration procedure. One
thousand calibration points are selected randomly from domain D. At each point, the
critical mean dead load can be computed under the hypothesis μLL/μDL = 0.5 as follows:
14.88 × (Sηu ) −0.7 × εαη × Sηu × πLB

μ*DL = (10)
1.35 + 1.5 × 0.5
@seismicisolation
@seismicisolation
The probability of failure implied by Eq. (9) for a given quantile η can be computed
from the following equation using Monte Carlo simulation (Ching & Phoon 2011):
⎛ 14.88 × (Sηu ) −0.7 × ε ηα × Sηu × πLB ⎞

⎜ ⎟
1.35μ DL + 1.5μ LL
P⎜ > 1 | μ Su , δ Su , L ⎟ = pT (11)
⎜ 14.88 × Su−0.7 × ε α × Su × πLB ⎟
⎜ DL + LL ⎟
⎝ ⎠
Because the mean dead load appears in the numerator and denominator of Eq. (11)
(note that DL = μDL exp{-0.5ln(1+δDL2)-[ln(1+δDL2)]0.5ZDL} in which ZDL is a standard
normal variable), it is not necessary to compute the critical mean dead load in this
example. The average of the 1000 η values is the recommended quantile for design
and it is denoted by η* in Tables 2 and 3. It suffices to note here that the optimization
method requires the solution of Eq. (5) to obtain the resistance factor γR, while QVM
requires the solution of Eq. (7) to obtained the average quantile η*.
The reliability indices associated with 1000 randomly selected independent
validation points, βV, can be computed via Monte Carlo simulation once the critical
mean dead load is obtained from Eq. (10) with η = η*. The statistics of βV are reported
in Tables 2 and 3. For this simple example, the following observations are clear:
1. It is better to calibrate a single value of η rather than a single value of γR.
2. QVM is less sensitive to the potentially wide range of resistance
coefficients of variation in domain D.
3. QVM satisfies mean βV – minimum βV < 0.5 or standard deviation of βV
< 0.5 without dividing D into subdomains.
4. In Table 2, QVM actually outperforms the resistance factor approach
even when it is calibrated locally in each subdomain.
The partial factor format based on QVM [i.e., Eq. (9)] and the partial factor format
based on γR [i.e., Eq. (4)] are the same if we set:
μ su
μsu−0.7 = (Sηu ) −0.7 × εηα × Sηu
(12)
γR
or
γ R = exp ⎡0.15ln (1 + δsu2 ) − 0.3 ln (1 + δsu2 ) × Φ −1 ( η) − σln( εα ) × Φ −1 ( η)⎤ (13)
⎣⎢ ⎦⎥
Based on Eq. (13), it is clear that the resistance factor implied by the QVM format
is not a constant. The superior performance observed above can be attributed to this
variable QVM-based resistance factor.
3. Piles Installed in Clay and Sand
Consider a pile installed in a layer of clay with an average undrained shear strength =
Su overlying a layer of sand with an average SPT N-value = N60 as shown in Figure 4.
The side resistance in sand is correlated to the Standard Penetration Test (SPT) as
2.71×εN×N60×πLsB (kN/m2), in which Ls (m) is the shaft length supported in sand, 2.71
@seismicisolation
@seismicisolation
is an empirical coefficient for the correlation, ln(εN) is normally distributed with mean
= 0 and standard deviation = 0.63, and N60 is lognormally distributed with mean value
μN and c.o.v. δN (Ching & Phoon 2012). This probabilistic model for the SPT N model
is obtained from calibration with instrumented load test data as shown in Figure 5
(Chen et al. 2012).
Fig. 4 Pile installed in clay and sand.

Fig. 5 fs-N correlation for drilled shafts in sand (Chen et al. 2012).
@seismicisolation
@seismicisolation
The safety ratio for this example is:
(14.88εαS−u0.7 )Su πLc B + 2.71ε N N 60 πLs B

SR(Y, θ) = (14)
DL + LL
in which Lc = shaft length supported in clay. The safety ratio depends on a vector Y
consisting of six lognormal random variables (Su, εα, N60, εN, DL, LL) and a vector θ
consisting of five design parameters (μsu, δsu, μN, δN, Lc/L, L). The domain D is
defined by the following representative ranges of values: 50 ≤ μsu ≤ 200 kN/m2, 0.1 ≤
δsu ≤ 0.5, 10 ≤ μN ≤ 50 blows/30 cm, 0.1 ≤ δN ≤ 0.5, 0 ≤ Lc/L ≤ 1, and 10 m ≤ L ≤ 50 m.
The rest of the assumptions are identical to those adopted in the previous example: (a)
shaft diameter (B) is 1 m and shaft length is L = Lc + Ls, (b) ln(εα) is normally
distributed with mean = 0 and standard deviation = 0.3, (c) Su is lognormally
distributed with mean value μsu and c.o.v. δsu, (d) DL and LL are lognormally
distributed with mean values (μDL, μLL) and coefficients of variation (c.o.v.) (δDL, δLL),
and (e) μLL/μDL = 0.5, δDL= 10%, and δLL = 20%.
The optimization method [Eq. (5)] is applied to calibrate two partial factor formats
for this example:
1. Load Resistance Factor Design (LRFD)
[(14.88μ su−0.7 )μ su πLc B + 2.71μ N πLs B] / γ R

≥1 (15a)
1.35μ DL + 1.5μ LL
2. Multiple Load Resistance Factor Design (MRFD) (Phoon et al. 2003b)
(14.88μ su−0.7 )μ su πLc B / γ c + 2.71μ N πLs B / γ s

≥1 (15b)
1.35μ DL + 1.5μ LL
The difference between LRFD and MRFD is that the former factors the overall
resistance with a resistance factor γR, while the latter factors each resistance component
separately with a factor γc for side resistance in clay and a factor γs for side resistance in
sand.
For both partial factor formats, optimization can be performed over the entire
domain D (50 ≤ μsu ≤ 200 kN/m2, 0.1 ≤ δsu ≤ 0.5, 10 ≤ μN ≤ 50 blows/30 cm, 0.1 ≤ δN ≤
0.5, 0 ≤ Lc/L ≤ 1, and 10 m ≤ L ≤ 50 m) or restricted to subdomains. As mentioned
previously, the latter strategy will produce a table of values for each resistance factor,
rather than a single value. As shown in Table 2, the c.o.v. of the undrained shear
strength (δsu) has a minor effect on the resistance factor. Hence, the range of δsu is not
segmented into two smaller parts. Rather than segmenting the ranges of values for μsu,
μN, and Lc/L individually, a combined parameter called the mean resistance ratio is
segmented into two smaller parts: 0 ≤ Rc ≤ 0.5 and 0.5 ≤ Rc ≤ 1. The parameter Rc is
defined as:
@seismicisolation
@seismicisolation
(14.88μsu−0.7 )μsu πLc B

Rc = (16)
(14.88μsu−0.7 )μsu πLc B + 2.71μ N πLs B
It is clear that Rc denotes the ratio of the nominal side resistance in clay to the total
nominal side resistance. When Rc ≤ 0.5, the total side resistance is dominated by sand
(Ls large and/or μN large) and vice-versa. Based on the above considerations, it is
possible to define four concise subdomains:
1. 0 ≤ Rc ≤ 0.5 and 0.1≤δN≤ 0.3

2. 0.5 ≤ Rc ≤ 1 and 0.1≤δN≤ 0.3
3. 0 ≤ Rc ≤ 0.5 and 0.3≤δN≤ 0.5
4. 0.5 ≤ Rc ≤ 1 and 0.3≤δN≤ 0.5
For the quantile value method (QVM), the following partial factor format is
calibrated:
14.88 × (Sηu ) −0.7 × εαη × Sηu × πLc B + 2.71 × εηN × N 60

η
× πLs B
≥1 (17)
1.35μ DL + 1.5μ LL
It is worth emphasizing that the average quantile (η*) is computed over the entire
domain D, i.e. the subdomain strategy is not implemented for QVM.
The same calibration and validation approaches are adopted. The only
computational difference is that the reliability indices for the calibration points are
computed using the First-Order Reliability Method (FORM), rather than Monte Carlo
simulation. The target reliability index in Eq. (5) is 3.8. The reliability indices for the
validation points, βV, are computed using Monte Carlo simulation.
Table 4 compares the performance of QVM with LRFD. It is clear that a layered
soil profile poses considerable challenges to partial factor formats in general. If one
adopts the more relaxed criterion requiring the standard deviation of βV to be less than
0.5, the performance of QVM can be considered to be reasonable. The standard LRFD
that attempts to cover all design scenarios (which include piles supported in varying
proportions of clay and sand) with a single resistance factor taking a single numerical
value is not satisfactory, which is hardly unexpected. It is rather interesting that the
four subdomains adopted in this study are not satisfactory as well, particularly for the
subdomains where sand is dominant (0≤Rc≤0.5). A likely reason is that the model
uncertainty associated with the evaluation of side resistance in sand using SPT-N is
large. Obviously the subdomain strategy can be refined to perfection if one is willing
to reduce the size of subdomains indefinitely.
Table 5 compares the performance of QVM with MRFD. It is useful to note that
the numerical values for γc and γs are distinctly different. This example nicely
demonstrates in a concrete numerical way what many practitioners may have already
suspected, namely a single resistance factor is too simplistic for design scenarios
involving significantly different resistance components (in the statistical sense). For
this example, MRFD is comparable to QVM. MRFD with subdomains is most
effective in achieving a consistent reliability index across domain D.
@seismicisolation
@seismicisolation
Table 4. Performance of optimization (LRFD) and quantile value methods for a single pile installed in clay
and sand.

value
method No
(QVM)
0.1≤δN≤0.5 0.1≤δN≤0.3 0.1≤δN≤0.3 0.3≤δN≤0.5 0.3≤δN≤0.5
0≤Rc≤1 0≤Rc≤0.5 0.5≤Rc≤1 0≤Rc≤0.5 0.5≤Rc≤1
γR or η* η* = 0.043 γR = 2.84 γR = 3.45 γR = 2.35 γR = 3.82 γR = 2.42
βV mean 3.81 3.74 3.74 3.91 3.68 3.91

βV COV 0.10 0.21 0.19 0.04 0.21 0.05
βV max 4.42 4.42 4.42 4.17 4.42 4.17
βV min 2.51 1.67 2.22 3.40 2.02 3.39
mean - min 1.30 2.07 1.52 0.51 1.66 0.52

std deviation 0.38 0.79 0.71 0.16 0.77 0.20
Table 5. Performance of optimization (MRFD) and quantile value methods for a single pile installed in clay
and sand.

value
method No
(QVM)
0.1≤δN≤0.5 0.1≤δN≤0.3 0.1≤δN≤0.3 0.3≤δN≤0.5 0.3≤δN≤0.5
0≤Rc≤1 0≤Rc≤0.5 0.5≤Rc≤1 0≤Rc≤0.5 0.5≤Rc≤1
γc, γs γc = 1.91 γc = 1.44 γc = 2.26 γc = 1.47 γc = 2.26
η* = 0.043
or η* γs = 5.84 γs = 6.95 γs = 2.65 γs = 9.13 γs = 3.06
βV mean 3.81 3.86 3.93 3.91 3.90 3.90

βV COV 0.10 0.10 0.05 0.04 0.06 0.04
βV max 4.42 4.42 4.26 4.17 4.42 4.17
βV min 2.51 2.60 3.19 3.28 3.10 3.29
mean - min 1.30 1.26 0.74 0.63 0.80 0.61

std deviation 0.38 0.39 0.20 0.16 0.23 0.16
4. Conclusions
A partial factor format involves multiplying a nominal resistance such as the mean
bearing capacity by a resistance factor or dividing a characteristic strength such as the
lower 5% undrained shear strength by a partial factor. The purpose of a partial factor
format is to implement reliability-based design without requiring the engineer to
perform reliability analysis. Hence, the performance of a partial factor format must be
measured by its ability to produce designs achieving a desired target reliability index
within an acceptable error margin. Based on the concept of a reliability class proposed
in BS EN1990:2002, two yardsticks for an acceptable error margin are recommended:
(1) mean βV – minimum βV < 0.5 and (2) standard deviation of βV < 0.5, in which βV
refers to the actual reliability index produced by a partial factor format design. The
recommendations imply that a sufficiently large sample of representative designs
should be evaluated. Fundamental to this performance evaluation is the explicit
definition of the range of design scenarios covered by a partial factor format called the
design domain. The error margin should be acceptable within this design domain.
@seismicisolation
@seismicisolation
Two partial factor formats are studied. They are distinguished by the nature of the
factor associated with the resistance or strength. The first format factors the overall
resistance (LRFD) or factors each resistance component separately (MRFD). The
second format factors the strength indirectly using a quantile. These factors can be
calibrated over the entire design domain or within subdomains. It is obvious that the
deviation from the target reliability index must be smaller in a subdomain, because the
design scenarios are more uniform. The former strategy will produce a single
numerical value for each factor. The latter strategy will produce a table of values – the
number of values is equal to the number of subdomains.
Based on the examples studied in this paper (pile installed in homogeneous clay
and pile installed in clay and sand), one may venture to conclude that:
1. If a design guide is insistent on recommending a single numerical value in
association with a partial factor format, it is more sensible to recommend a
constant quantile value, rather than a constant resistance factor.
2. If a design guide is willing to recommend more than one numerical value, it is
worthwhile evaluating MRFD and subdomain calibration. The former
recommends a different resistance factor for each resistance component. The
latter recommends different numerical values for the LRFD/MRFD factors in
different subdomains
References
British Standards Institute (2002). Eurocode: Basis of Structural Design. BS EN 1990:2002, London.
British Standards Institute (2004). Eurocode 7: Geotechnical Design – Part 1: General Rules. BS EN 1997-
1:2004, London.
Chen, J.R., Chuang, B.Y., Ching, J., Yang, Z.Y. & Shiau, J.Q. (2012). “Axial behavior of dilled shafts in
multiple strata – Taipei database”, Soils and Foundations, under review.
Ching, J.Y. & Phoon, K. K. (2011). “A quantile-based approach for calibrating reliability-based partial
factors”, Structural Safety, 33(4-5), 275-285.
Ching, J. Y. & Phoon, K. K. (2012). “Quantile value method versus design value method for calibration of
reliability-based geotechnical codes”, Structural Safety, under review.
Phoon, K.K., Kulhawy, F.H. & Grigoriu, M.D. (1995). “Reliability-based design of foundations for
transmission line structures”, Rpt. TR-105000, Electric Power Research Inst., Palo Alto (CA), 380 p.
[available online at EPRI.COM]

Phoon, K.K., Kulhawy, F.H. & Grigoriu, M.D. (2003a). "Development of a reliability-based design
framework for transmission line structure foundations", Journal of Geotechnical and Geoenvironmental
Engineering, ASCE, 129(9), 798-806
Phoon, K. K., Kulhawy, F. H. & Grigoriu, M. D. (2003b). "Multiple Resistance Factor Design (MRFD) for
spread foundations", Journal of Geotechnical and Geoenvironmental Engineering, ASCE, 129(9), 807
– 818.
Phoon, K.K. & Kulhawy, F.H. (2008). “Serviceability limit state reliability-based design”, Chap. 9 in
Reliability-Based Design in Geotechnical Engineering: Computations & Applications, Ed. K.K. Phoon,
Taylor & Francis, London (U.K.), 344-384.
Sørensen, J. D. (2002). Calibration of partial safety factors in Danish Structural Codes. Workshop on
Reliability Based Code Calibration, Swiss Federal Institute of Technology, ETH Zurich, Switzerland,
March 21-22, 2002.
@seismicisolation
@seismicisolation
IOS Press, 2013
doi:10.3233/978-1-61499-163-2-311
Target Reliabilities and Partial Factors for

Flood Defenses in the Netherlands
Timo SCHWECKENDIEKa,b,1, Ton VROUWENVELDERa,c,, Ed CALLEb,
Wim KANNINGa and Ruben JONGEJANa,d
a
Delft University of Technology
b
Deltares, Unit Geo-engineering
c
TNO Built Environment and Geosciences
d
Jongejan Risk Management Consulting
Abstract. Modern codes of practice such as Eurocode strive to provide design

rules including partial factors with an appropriate level of safety for a wide range
of applications. The target reliability levels and partial factors in such codes are
not equally efficient for all applications, since they are calibrated to work well in
both typical and unusual situations. For large engineered systems like flood
defense systems with large potential consequences and substantial investments in
improvement and maintenance, it is worthwhile to develop tailor-made solutions.
This paper describes the approach adopted in the Netherlands to develop safety
requirements to flood defenses such as partial factors for dikes within an
acceptable risk framework accounting for system reliability aspects. The main
steps herein are to define a risk-informed target reliability for the whole system
(i.e., including all elements and failure modes), to derive target reliabilities for
specific elements (e.g., dike sections) and failure modes and to calibrate partial
factors on the latter. After describing those steps, the paper provides an application
example for the uplift and piping failure mode (i.e., internal erosion).
Keywords. flood defenses, LRFD, dikes, failure modes, system reliability, length-
effect, spatial variability, partial factors, code calibration, target reliability
Introduction
Modern codes of practice such as Eurocode use consequence or reliability classes to

assign target reliabilities and corresponding partial factors to different types of
structures or structural members. Flood defense systems are large engineered systems
protecting flood-prone areas from inundation. Since the consequences of failure of
flood defenses can be significant as are typically the investments in increasing the
reliability of such systems, it is worthwhile define tailor-made target reliabilities
instead of using the reliability classes from general geotechnical codes of practice,
which may be inefficient (i.e., either too low or too high) for a given flood defense
system. Furthermore, system reliability aspects need to be paid special attention. Flood
defenses are typically long linear structures, for the so-called length-effect is significant.
That is the effect that the reliability of a linear structure decreases with increasing
length. The paper describes the proposed steps to derive partial factors for geotechnical
failure mechanisms of linear flood defenses from high level risk-based safety
requirements.
@seismicisolation
@seismicisolation
312 T. Schweckendiek et al. / Target Reliabilities and Partial Factors for Flood Defenses
Figure 1. Outline Step-wise Procedure and Corresponding Sections

As outlined in Figure 1, sections 1 to 4 describe the steps from high level risk and
reliability targets to partial factors for individual dike reaches. Section 1 illustrates how
risk-informed target reliability values at the highest level, that is for a flood defense
(sub-)system with relatively homogeneous consequences in case of failure (i.e. where
the consequences are of similar magnitude regardless of the mode or position of failure
in the (sub-)system), can be chosen. Individual risk, group risk and economic risk
provide valuable input, however, the ultimate decision remains a political one. A flood
defense (sub-)system is composed of different types of structures, which are prone to
different failure modes. Section 2 describes how target reliabilities for a particular
failure mode and structure can be derived or motivated. The so-called length effect is
treated in section 3, in which we account for the spatial variability in load and
resistance, especially in ground properties. The final step from target reliability per dike
section to a set of calibrated partial factors is elaborated in section 4. Furthermore, it
will be shown that it is more efficient to take the last two steps together by using an
alternative calibration criterion. Section 5 illustrates the workings of the approach for
the derivation of safety factors for an assessment rule for the piping failure mode.
It is emphasized that the described method and example are based on a safety
assessment framework, rather than on a design situation (i.e., design of a new structure).
For design situations, the changes of loading conditions and resistance properties
during the system’s lifetime should be taken into consideration. The overall approach
remains the same, however.
1. From Acceptable Risk to System Target Reliability
Target reliabilities for engineered systems or structures are often risk-informed. The
higher the potential consequences of failure, the higher the target reliability. Below we
discuss three widely used types of criteria for evaluating the risks related to floods and
major industrial hazards. From such acceptable risk-criteria, we can derive an
acceptable probability of failure or a target reliability for a given system.
@seismicisolation
@seismicisolation
T. Schweckendiek et al. / Target Reliabilities and Partial Factors for Flood Defenses 313
1.1. Economic Criteria
Economically optimal protection levels can be obtained from cost-benefit analyses. In

such analyses, risk is typically defined as (valued at) the expected value of economic
damage (which is the actuarially fair insurance premium). The economically optimal
reliability of an engineered system can be found by equating marginal costs with
marginal benefits, or by minimizing the sum of the present value of the cost of
strengthening flood defenses and the present value of the economic risk, see e.g. Van
Dantzig (1956) for an application to a major levee system (a present value is today’s
value of future cash-flows). This optimization procedure is shown schematically in
Figure 2.
Cost
total cost
cost of increasing
system reliability
expected loss
(present value)
optimal reliability System Reliability
Figure 2. The optimization of a system’s reliability: total cost equals the cost of increasing the system
reliability plus the present value of expected loss (PV = present value)
1.2. Individual Risk Criteria
Individual risk criteria are reference levels for evaluating individual exposures. These
are often defined as a maximum allowable probability of death. The stringency of
individual risk criteria is often related to considerations regarding the voluntariness of

exposure, and the degree to which an exposed person benefits from the hazardous
activity (e.g. Vrijling et al., 1998), for Western countries typically ranging from a
probability of 10-2 per year for voluntary activities with relatively high direct benefit
like mountaineering to 10-6 for involuntary activities without direct benefit, the latter
being frequently used in industrial safety.
1.3. Societal Risk Criteria
When individual exposures are low, there is still a chance that disaster strikes affecting
a vast population. Psychometric studies have shown that risk perception is strongly
influenced by “dread”, or catastrophic potential (e.g. Slovic, 1987). The assessment of
multi-fatality disasters often involves societal risk criteria. One common type of
societal risk criterion is the FN-criterion (e.g. Jonkman et al., 2003). An FN-curve
shows the exceedance probabilities (F) of different numbers of fatalities (N), plotted on
double logarithmic scale. The FN-curve should not cross the criterion line, see Figure 3.
@seismicisolation
@seismicisolation
For the definition and motivation of FN-acceptance criteria, the reader is referred to e.g.
Jongejan (2008) , Ball and Floyd (1998) and Stallen et al. (1996).
Exceedance
probability FC/nD
(F) on log
scale FN-criterion exceeded
Number of fatalities (N) on log scale

Figure 3. Schematic overview of different types of FN-criteria..
2. Target Reliabilities per Failure Mode
The first step in going from system target reliabilities (previous section) towards
concrete safety requirements is to break down the system in terms of elements and
failure modes as discussed below.
2.1. Failure Modes of Flood Defense Systems

Figure 4. Fault tree for Flood Defense Failure Modes

A flood defense system is a series of dikes, dunes, retaining walls, higher grounds,
storm surge barriers, sluices and so on, protecting a flood-prone area. Dikes and dunes
are usually subdivided into statistically homogeneous sections in terms of relevant
geometrical, material and loading parameters. They are typically between, say, 100
meters to several kilometers long. The most common failure mechanisms for dike
sections are shown in Figure 4. For detailed descriptions of these mechanisms in terms
of limit state functions, reference is made to Steenbergen and Vrouwenvelder (2003).
For dune sections usually only dune erosion is considered and for structural
components (sluices, locks, quay walls) a set of mechanisms such as overflow, piping,
structural integrity and incorrect closure operations may be relevant.
@seismicisolation
@seismicisolation
2.2. System Reliability
In order to calculate the inundation probability for a protected area, one needs a set of
mathematical models to describe the various failure mechanisms as well as the
statistical properties for all basic random variables involved. A proper statistical model
requires the distribution type and parameters (e.g., mean and standard deviation) and
descriptions of the mutual as well as auto-correlations in space and time. Note further
that it is essential to include all uncertainties, not only the natural variability in dike
heights, soil properties or water levels, but also epistemic uncertainties like the
inaccuracies in resistance and load models as well as the uncertainties in the statistical
parameters as well as distribution types.
The first step in estimating the failure probability of a flood defense system is to
calculate the failure probability for each individual failure mode of each individual
element or section:
Pf ,i , j P Z i, j ( X ) 0 ³[ f X
Z i , j ( )0
([ ) d [ (1)
where Zi,j is the limit state function for failure mode i (1..m) and element or section
number j (1..n) (i.e., Zi.j < 0 corresponds to failure), X is the vector of stochastic
variables and fx([) their n-dimensional probability density. The frequently used
reliability index is related to the probability of failure by:
E ) 1 Pf (2)
in which is standard normal distribution.

The second step is to aggregate the failure probabilities of all modes and sections
or elements to the probability of (sub-)system failure. In the case of a simple series
system of m independent modes and n independent components, the failure probability
of the system may (for small probabilities) be written as:
m n
Pf , sys ¦ ¦ Pf ,i j (3)
i 1j 1
This equation holds equally for summation over sections, mechanisms, wind directions
or time periods, provided that the contributing probabilities of failure, Pf,i,j, are small
and mutually independent. In reality, however, the contributing probabilities are
(slightly) mutually dependent, due to (spatial) correlation of the random variables
involved in the computation of the failure probabilities (see section 3) and due to
common random variables in the limit state models (e.g. hydraulic load). Eq. (3) is
therefore an (upper bound) approximation.
@seismicisolation
@seismicisolation
2.3. Target reliabilities per failure mode
Assume that acceptable risk considerations have led to an acceptable annual probability
of system failure expressed in terms of PT,,sys (T = target). The task of the engineer is to
design (and maintain) the flood protection system in such a way that it meets this
reliability target in the most efficient way. For the sake of practicability one may start
by allocating failure probabilities to each failure mode and each section or element in
such a way that the top (system) requirement is met. The simplest way is to set:
PT , sys
PT ,i. j (4)
mn
This, however, maybe very inefficient as we ignore correlations between failure modes
and sections, as well as differences between cost functions. In principle, an economic
optimum could be determined using the philosophy described in section 1.1. A more
pragmatic way would be to allocate ‘failure probability space’ to different
elements/failure modes based on current practice, as one may assume that in decades or
centuries of trial and error have led to more or less efficient designs. An advantage of
such an approach would be that new approaches are aligned with current practice. It is
well known that introducing too many changes at a time is error-prone and unlikely to
be easily accepted.
Another important aspect in allocating target probabilities of failure (PT,i,j) is to
account for the spatial variability. This will be discussed in the next section.
3. The Length-Effect
Spatial variability or statistical dependence between random variables (load and/or

resistance properties), between and within a system’s components (e.g. dike sections)
have significant impact on its reliability. When it comes to flood protection systems,
the main aspect to consider is the so-called length-effect.
3.1. What is the length-effect?
The length-effect refers to the increase of the failure probability with the length of a
dike due to imperfect correlations and/or independence between different cross sections
and/or elements. The underlying reason is a lack of full correlation between the
different load and resistance variables as a result of spatial variability. Typically, the
load variables are highly correlated over different dike sections, whereas resistance
properties of dike sections exhibit little correlation due to the small horizontal
correlation distances. This results in a partially correlated limited state function in the
length direction and thus an increase of failure probability in the length direction. It
must be noted that only horizontal correlation distances are affecting the length-effect,
the effect of vertical spatial variability is assumed to be incorporated in the analytical
models. As a general rule one may say that the smaller the correlation distances (see
3.2) of important random variables, the higher the increase of the probability of failure
with the length. Two types of spatial variability contribute to the length-effect:
continuous fluctuations and discontinuities (e.g. anomalies).
@seismicisolation
@seismicisolation
For flood defenses, the uncertain soil conditions relevant to geotechnical failure
modes are normally the main contributors to the length-effect. This is because the
uncertainties are often considerable and correlation distances are relatively short. The
spatial correlation (lengths) of load parameters (e.g., typically water levels or wave
conditions) are relatively high. Below we explain and illustrate the length-effect due to
continuous fluctuations in soil properties.
3.2. Outcrossing Approaches
Spatial variability can be modeled by random fields (e.g. Vanmarcke, 1977). A

Gaussian autocorrelation function can be used to describe the decay of spatial
correlation with distance x:
2
§ x ·
¨ ¸
© G0 ¹
U ( x) U x 1 U x e (5)
in which x models the part of the variance that does not decay in space (non-ergodic,
see Vrouwenvelder, 2006) and 0 is the correlation distance. In order to incorporate the
length of the dike (L) in the reliability calculation, we are interested in the probability
of the limit state function Z being less than 0 in the domain [0, L], for which
outcrossing approaches (or first passage see e.g. Vanmarcke (1975)) provide the
following generic solution:
³
h ([ )d [
P (0, L) 1 e 0
(6)
where h() is the conditional failure rate. For long L, h() can be approximated by the
mean crossing rate (see Rice, 1952), which is the mean rate at which level is crossed
by the considered random variable. If we make several assumptions (Pcs is small, L is
long, the upcrossings, which are the crossing from below to above the threshold, are
independent, the crossing rate is constant), the probability of failure P(0,L) can be
approximated by eqn. (7), (see e.g. Calle, 2010). These assumptions are valid failure
probabilities are in the order of 10-3 or lower and dike section lengths are larger than
horizontal correlations distances, which is usually the case.
§ L·
P (0, L) | Pcs ¨1 ¸ (7)
© l¹
Where P(0) is the probability of failure of a cross-section and l is the independent

equivalent length (for ”(0)>0 and thus for one of x,i < 1) according to:
2S 1
l| (8)
ET ,cs U ''(0)
@seismicisolation
@seismicisolation
where ”(0) is the second derivative of the autocorrelation function (Eq. 5):
N 2D i2 (1 U x ,i )
U "(0) ¦
i 1 G 0,2 i
(9)
where is the FORM sensitivity factor, which is the result from a reliability analysis
(forward analysis) or an average of several dike analyses for the derivation of safety
factors.
3.3. Accounting for the length effect
In order to account for the length-effect, we have to find the target reliability of the
cross-section (ET,cs) as a function of the dike ring characteristics for the considered
failure mechanism. Hence, we need to use Eq. (8) in an inverse way where P(0)
corresponds to PT,cs and P(0,L) to PT,sec, as discussed in section 2.3. So we end up with:
PT ,sec
PT ,cs (10)
(1 L / l )
Several assumptions have to be made to solve equation (11), notably to find a

representative value of l, which can be found by taking the average of several dike
rings determined in (forward) reliability analyses. Furthermore, one needs to realize
that L should actually be interpreted as the dike length of sections in the system that
contribute significantly to the failure probability. For example, if only 10% of the
system contributes to the probability of piping (e.g., some dikes may not have a piping
sensitive layer), only 10% of the system length should be taken into account. In fact,
this is part of the determination of target reliability per failure mode and dike section;
see e.g. equation (4).
6
Cross Section Target Reliability ET,cs
5.5
4.5 ET,sys = 4.26

ET,sys = 4.06
4 ET,sys = 3.89
ET,sys = 3.78
3.5
0 20 40 60 80 100 120 140 160 180 200
Contributing Dike Ring Length L [km]
Figure 5. Relation between Target Reliability and Contributing Dike Length L (l = 282 m)
@seismicisolation
@seismicisolation
Figure 5 shows an example for piping with l = 282 m (based on a statistical analysis of
results for three Dutch dike rings from the VNK2 project (Lopez de la Cruz et al.,
2010); the values of ET,sys correspond to a failure probability that equals 10% of the
Dutch safety standard. For typical contributing lengths of between 10 and 100 km, the
difference between ET,sys and ET,cs (i.e., length effect) is about 0.5 to, which corresponds
to ratios of approximately 10 to 100 of the respective target probabilities of failure.
It is important to realize that the length-effect has implications for the allocation of
target reliabilities for dike sections or elements, as well as for translating these into
requirements for individual cross sections.
4. Derivation of Partial Factors
Load and resistance factors for geotechnical failure modes of flood defenses (e.g., slope
instability, uplift & piping etc.) are typically derived in an iterative fashion. First
preliminary factors are determined using standardized FORM sensitivity factors D.
Subsequently, their effect on the resulting reliability is investigated by reliability
analyses for a variety of representative cases, before final values are chosen.
Partial factors typically apply to loads and resistances, the uncertainty of which can
be modeled as random variables. Some uncertainties involved in geotechnical
engineering, however, lack the possibility of being adequately modeled this way. For
example, schematizations of the stratification of the subsoil, based on limited soil
investigation or assessment of the type of response to external actions (e.g., drained or
un-drained), may be highly uncertain. Uncertainties involved in this schematizations
are at least as important as uncertainties of resistances due to uncertain soil parameters
(Terzaghi, 1929). Yet, these type of uncertainties are not (explicitly) addressed in codes
of practice. It is trusted that they are adequately handled by professional judgment.
Only recently, Schweckendiek & Calle (2010) suggested more explicit ways of
handling this type of schematization uncertainties. However, in the present paper, the
focus is on partial factors for loads and resistances, which can adequately be modeled
as random variables, having normal distributions.
4.1. From Target Reliability to Partial Factors
To get a first impression of the partial factors, needed to fulfill the target reliability
concerning some potential failure mechanism, use can be made of the ISO standardized
sensitivity factors (Table 1).
Table 1. Standardized FORM-sensitivity factors (ISO 2394)
Random Variable Xi Di
Dominant resistance parameter 0.8
Other resistance parameters 0.8 x 0.4 = 0.32
Dominant load parameter - 0.7
Other load parameters - 0.7 x 0.4 = - 0.28
@seismicisolation
@seismicisolation
The idea is based on the design point notion in FORM type reliability analyses (ISO
2394). In such a first-order framework, the required partial safety factors for load S and
resistance R, assuming Gaussian probability distributions, can be expressed as:
1 ET D S VS 1 1.65VR
JS JR (11)
1 1.65VS 1 E T D R VR
where VS and VR are the respective coefficients of variation and the factors are used
with 5% and 95%-characteristic values (i.e., Sd = Sk JS and Rd = Rk / JR).
4.2. Calibration Using Representative Test Sets
In order to produce more accurate partial factors one may use representative data sets
and work in a trial and error fashion. A recent example for piping is given by Lopez et
al. (2011). The test set members are realistic cases, for example, draft designs of dike
cross sections, representative for the envisaged range of application. The line of
thought is that the test set members are first “designed”, using preliminary partial
factors, based on the target index of reliability. Next, probabilistic analyses of the
designs yield a set of actually ‘realized’ probabilities of failure and, equivalently the
actually realized indices of reliability. These are then compared with the target
reliability. This way an appropriate set of partial factors can be found iteratively.
Since a set of partial factors will not perform equally well on all test set members,
one needs to adopt a calibration criterion. An example would be that 95% of the
designs need to satisfy the target reliability, which, however, is a rather strict criterion
that would often be economically suboptimal. Somewhat more relaxed criteria have
been suggested in the literature (e.g., Ciria, 1977; Sørensen, 2001; Faber and Sørensen,
2002). For example, for a single safety factor, the calibration criterion may read
(Sørensen, 2001):
N
J : min W( J ) ¦w ( Pf , cs ,i ,k ( J ) PT ,cs , i )2 (12)
k
k 1
where is the safety factor to be determined, Pf, cs, i, k () is the realized probability of
cross section failure (considering failure mechanism i) of test set member k (k=1…N),
when applying the safety factor , and PT, cs, i is the target probability of failure. The wk
are weights, reflecting relative frequencies of occurrence of design situations in
practice, represented by each of the test set members. The weights sum up to 1. Instead
of probabilities of failure, actually realized and target reliability indices may be used in
eq. (13).
So far, these criteria refer to safety requirements for individual test set members.
For (flood defense) systems the calibration criterion may be related to the system target
reliability as discussed in section 4.3.
4.3. Calibration Criterion at System Level
An alternative way to determining local target reliabilities first (i.e., for a section or
element) is to calibrate partial factors directly on the basis of the system target
reliability, in this case of a (part of a) dike ring. From now on we consider an annual
@seismicisolation
@seismicisolation
target probability of failure for one mechanism for all elements in the system PT,i
Combining Eq. (3) and Eq. (8) yields the following approximation for the system
probability of failure of mode i (e.g., piping):
m § ·
P | ¦ P f , cs, j ¨1 L j / l j ¸ (13)
f , sys, i
j 1 © ¹
where Lj are the lengths of the individual sections and lj the independent equivalent
lengths of the failure mode, the latter often being assumed an average value per failure
mode in calibration analyses.
The calibration criterion is based on the premise that the average probability of
flooding over all consequence systems should be lower than the safety standard (in
practice, the probability of flooding is evaluated for each system separately). This is
approximately the case when the following condition is met:
P
T , sys, i
P d (14)
cs, avg
n §¨1 L /l ·¸
avg © avg avg ¹
where Pcs,avg = average failure probability of cross sections, Pt,sys,i = target failure
probability for mode i on system level, navg = average number of sections per system,
Lavg = average section length and lavg = average equivalent independent section length
(mode-dependent).
Hereafter, the criterion defined in Eq. (15) will be referred to as calibration
criterion type 1. When the test cases do not cover entire systems, this calibration
criterion may yield overly optimistic results. To deal with such situations, another type
of calibration criterion was developed, hereafter referred to as type 2. When the
reliability indices per section (that are designed according to a specific set of partial
factors) are normally distributed with a standard deviation of 0.5, it can be shown that
in order to meet the requirement stated by Equation (14), the following condition has
to be met:
§ ·
¨ P ¸
1 T , sys, i
E t ) ¨ ¸ (15)
cs, 20% ¨¨ n §¨1 L /l
·¸
¸¸
© avg © avg avg ¹ ¹
where cs,20% = 20th percentile of the reliability indices per cross section. Note that both
calibration criteria yield the same results when N is sufficiently large and the reliability
indices per cross section are normally distributed with a standard deviation of 0.5.
Furthermore, both calibration criteria assume sections to be independent. In case of
strong correlations (e.g. greater than 0.85), calibration criterion type 1 will yield
conservative design values, while the opposite holds true for calibration criterion type 2.
After all, when sections (and their cross-sections) are perfectly correlated, the
@seismicisolation
@seismicisolation
reliability index of a system will be equal to the lowest cross-sectional reliability index,
which is less than the 20th percentile.
5. Application Example: Semi-probabilistic Assessment Rule for Piping
5.1. Introduction
The example concerns the calibration of a semi-probabilistic safety assessment rule for
piping, an internal erosion mechanism elsewhere also referred to as under-seepage or
backward erosion. The calibration was carried out within the WTI2017-project, which
has the objective to develop a new framework for statutory safety assessments in the
Netherlands based on acceptable probabilities of flooding (i.e., dike ring system
failure).
The underlying limit state function is a revised version of the well-known
Sellmeijer model (Sellmeijer, 1988) and is given by (see e.g. Lopez de la Cruz et al.
2010; Sellmeijer et al. 2012):
Z i L §¨ h h 0.3d ·¸ (16)
c p © in ¹
where ic = critical (average) gradient for piping [-]; Lp = seepage length [m]; h = water
level [m+REF]; hin = hydraulic head at the exit point [m+REF] (i.e., h-hin is the
hydraulic head difference over the levee); and d = thickness of the impervious blanket
layer [m]. Notice that the product ic,p Lp represents the critical head difference [m].
5.2. Target Reliability
The system target reliability in the Netherlands were based on the present-day Dutch
flood safety standards, see Table 2 and Figure 6. For calibration purposes, these target
reliabilities had to be broken down to the different failure mechanisms. Based on
experience from the VNK2-project (Jongejan et al., 2013), a nation-wide flood risk
analysis, the so-called failure probability budget for piping (i.e. the allowable
contribution of the piping mechanism to the system’s overall probability of failure) for
all dike sections in a dike ring was set to 35% (see Table 2).
Table 2. Target Reliabilities (levees only), all probabilities are on an annual basis.
Safety standard/ Failure Target Probability Target reliability

acceptable probability probability of Failure PT,pip index ET,pip
of flooding1 PT budget
1/500 35% 1/1,400 3.20
1/1,250 35% 1/3,600 3.45
1/2,000 35% 1/5,700 3.58
1/4,000 35% 1/11,400 3.75
1/10,000 35% 1/28,600 3.98
1
The numbers in the Dutch safety standard actually refer to frequencies of the load events used in the
safety assessment. These, however, are frequently necessarily interpreted as acceptable probabilities of
flooding or system (i.e., dike ring) failure for sake of calibration of semi-probabilistic assessment rules.
@seismicisolation
@seismicisolation
5.3. Partial Factor and Safety Format
In semi-probabilistic safety assessments, characteristic values (subscript k) are used and

a partial factor is applied to the critical head difference:
Figure 6. Safety Standards for Flood Defenses in the Netherlands Interpreted as Acceptable Probabilities of
Flooding1
1
ic , k L p , k t h h 0.3d (17)
J k in, k k
p
where p = partial resistance factor. Note that ic,k is defined as the critical gradient from
Lopez de la Cruz (2010) with characteristic values for all input variables. A single
partial resistance factor was chosen in order to stay close to the currently used rule.
5.4. Test Set and Probabilistic Analyses
Table 3. The test set for the preliminary calibration exercise.
Levee system Safety standard/ acceptable No. of Total length

probability of flooding PT (a-1) sections (m)
No. 5 Texel island 1/4,000 13 16,626
No. 17 IJsselmonde 1/4,000 21 23,984
No. 36 Land van 1/1,250 33 25,345
Heusden/de Maaskant
@seismicisolation
@seismicisolation
The test set members were obtained from the VNK2-project (see Jongejan et al.,
forthcoming for more details on the VNK2-project). At the time of analysis, reliable
results were available for three levee systems (see Table 3). It contains a wide range of
characteristics in terms of water level (i.e., load) distributions, substrata thicknesses,
and grain size distributions etc. The test set members were analysed in an iterative
fashion as discussed in section 4.2.
The squared FORM sensitivity factors in Figure 7 indicate that the total
uncertainty in the limit state function was dominated by geotechnical uncertainties
(mainly the permeability together with other ground-related variables) in levee systems
no. 5 and 17 whereas the load uncertainty is much more important in no. 36. To avoid
inconsistencies with past performance, the inputs were modified with respect to the
original prior input from the VNK2 project. While this was done somewhat arbitrarily,
by reducing the variance of the permeability to raise the design point values of the
water level, Schweckendiek et al. (2012) demonstrate how Bayesian Inference can be
used to use actual data about past performance in a more formal and systematic fashion.
For each section, the required length of the seepage path was determined with the
semi-probabilistic assessment rule, for different values of p (Eq. 17). FORM analyses
(Hasofer & Lind 1974) were then carried out using the required seepage length to
analyse the “assessment reliability”. The results of these calculations as shown in
Figure 8.
Dike ring Dike ring Dike ring

no.5 no.17 no. 36
1
Squared FORM-influence coefficient
0,8
0,6 Other stochastic variables

Outer water level
0,4 Permeability
0,2
0
1 11 21 31 41 51 61
Testcase no.
Figure 7. Squared FORM-sensitivity factors per section after modifications based on past performance.
@seismicisolation
@seismicisolation
2.0
1.8
1.6
p
1.4
1.2
1.0
4 5 6 7 8
cs
Figure 8. Reliability index per cross-section (cs) as a function of the partial factor (R). The distinction
between the different levee systems, represented by the symbol shapes, is not elaborated in this figure, for it
is not essential to the example.
5.5. Calibration Criterion
According to calibration criterion type 1 (Equation 14, the ratio Pcs,avg/PT,,cs should be
equal to one 2 . If Pcs,avg/PT,,cs>1, the partial factor is too optimistic (too low); if
Pcs,avg/PT,,cs<1, it is too conservative (too high). Figure 9 shows the ratio Pcs,avg/PT,,cs as a
function of the partial factor (p), which should be 1.15-1.18 for the three levee systems
under consideration.
Figure 9. The ratio Pcs,avg/PT,c s as a function of the partial factor (p).
Calibration criterion type 2 is be based on the 20%-percentile of the cs-values as

shown in Figure 10. The partial factor p can be approximated by a linear function of
the target reliability T,cs (to offer flexibility for use for different reliability levels), see
Table 4. Notice that calibration criterion type 1 yields the same partial factors as
calibration criterion type 2.
2
For sake of conciseness, the failure mode index is dropped in the example and all values refer to piping.
@seismicisolation
@seismicisolation
1.8
1.6
1.4
p
Overall
1.2 Dikering 5
Dikering 17
Dikering 36
1.0
4.0 5.0 6.0 7.0 8.0
cs
Figure 10. Reliability index per cross-section (cs) as a function of the partial factor (R), together with the
20%-quantile values of cs.
Table 4. Calibrated Partial Factor R According to Calibration Criterion Type 2.
Levee Calibrated partial factor pas a ET,pip,cs Calibrated partial

system function of T,cs (annual basis) factor p
No. 5 p =0.30*T,pip,cs -0.24 4.73 1.15
No. 17 p =0.32* T, pip,cs -0.34 4.84 1.15
No. 36 p=0.59* T, pip,cs -1.48 4.57 1.18
6. Discussion
This paper has shown how partial factors for geotechnical failure modes of flood
defenses can be derived on the basis of top level risk acceptance criteria taking into
account system reliability and length effects.
The Water Act lays down safety standards for flood defenses, defined in terms of
the exceedance probabilities of the hydraulic loading conditions (i.e., design water
levels and wave conditions) that the flood defenses should be able to safely withstand.
These standards are currently being reviewed, in the light of new insights in the current
flood risk level in the Netherlands and acceptable risk criteria as discussed in section 1.
Apart from possible changes to the stringency of current changes, the type of standard
might also be changed: from exceedance probability to a failure probability. This
would allow for a closer link between the safety standards and the risks deemed
acceptable.
Breaking down top-level target reliabilities to local (i.e., element) requirements
requires a thorough understanding of the relations between the reliabilities of elements
or components and the reliability of the entire system. All failure mechanisms and
elements, as well as their interactions and correlations, need to be considered.
Especially for dikes, where geotechnical failure mechanisms often dominate the
probability of failure, the length-effect cannot be neglected.
There are several alternative methods for calibrating partial factors, ranging from
the use of standardized FORM sensitivity factors (D) as presented in ISO 2394 to
probabilistic computations for representative test data sets. The latter approach requires
@seismicisolation
@seismicisolation
some sort of calibration criterion. Different criteria have been proposed in literature;
there is no universally ‘best’ criterion. It is therefore recommend to develop calibration
criteria that are consistent with the acceptable risk criteria and system reliability
concepts that are used.
The example in section 5 has illustrated how partial factors were calibrated in the
Netherlands for a semi-probabilistic assessment rule for the failure mode piping. The
top-level requirements, i.e. the target reliabilities for each dike ring, were based on the
current safety standards. The target reliability for the piping mechanism was based on
the relative importance of the various failure mechanisms, as suggested by data from
the VNK2-project. The length effect was included in the analysis via probabilistic
calculations that related the top-level reliability requirement to requirements for
individual dike sections. Two different types of calibration criteria were used that both
led to similar values for the partial factor.
The example illustrates that a calibration exercise is by no means a straight-
forward to follow mathematical recipe, a considerable amount of judgment is involved.
Nevertheless, the presented framework provides a systematic and transparent approach
to the calibration of partial factors. At the highest level, a political/administrative
decision is needed as to the acceptability of risks. At the lower levels, various choices
have to be made on the basis of engineering judgment.
References
Ball, D.J. & Floyd P.J. 1998. Societal risks, Final report, commissioned by the Health and Safety Executive,
United Kingdom.
Calle, E.O.F. (2010). Lengte-effect en kalibratie van een toetsregel. Deltares Memo 15 April, 2010
Ciria 1977, Rationalization of Safety and Serviceability Factors in Structural Codes, Ciria report no 63,
London.
Faber, M..H. and Sørensen, J. D. 2002. Reliability based Code Design, Joint Committee of Structural Safety,
March 2002 (draft)
ISO 2394 – General Principles on Reliability for Structures.
Jongejan R.B. 2008. How safe is safe enough? Doctoral thesis. Delft University of Technology, Delft, The
Netherlands.
Jongejan, R.B., Vrouwenvelder, A.C.W.M. & Calle, E.O.F. 2012. The development of semi-probabilistic
rules for levee safety assessments, Proceedings of the Second European Conference on Flood Risk
Management (FLOODrisk 2012), Rotterdam, The Netherlands.
Jongejan, R., Stefess, H., Roode, N., Ter Horst, W. & Maaskant, B. 2013 (forthcoming). The VNK2-project:
a fully probabilistic risk analysis for all major levee systems in the Netherlands. IAHS Publ. From Risk
to Opportunity. (in preparation)
Jonkman, S.N., Van Gelder, P.H.J.M. & Vrijling, J.K. 2003. An overview of quantitative risk measures for
loss of life and economic damage. Journal of Hazardous Materials, vol. 99(1), 1-30.
Lopez de la Cruz, J., Schweckendiek, T., Mai Van, C. & Kanning, K. 2010. SBW Hervalidatie piping. HP8b
Kalibratie van de veiligheidsfactoren. 120213-002-GEO-0005. Delft: Deltares.
Lopez de la Cruz, J., Schweckendiek, T. & Calle, E.O.F. 2011. Calibration of Piping Assessment Models in
The Netherlands. Proc. of the 3rd Int. Symp. on Geotechnical Safety and Risk (ISGSR 2011), Munich,
Germany, 587–595.
Rice, S.O. 1952. Mathematical modeling of random noise. Bell telephone labs inc, New York.
Schweckendiek, T. and Calle, E. 2010. A Factor of Safety for Geotechnical Characterization . In Proc. of the
Seventeenth Southeast Asian Geotechnical Conference (17SEAGC) - Geo-Engineering for Natural
Hazard Mitigation and Sustainable Development., volume Vol. II: Plenary and Special Sessions., 227-
230, Taipei, Taiwan.
Sellmeijer, J.B. 1988. On the mechanism of piping under impervious structures. Dissertation. Delft: Delft
University of Technology.
@seismicisolation
@seismicisolation
Slovic P. 1987. Perception of Risk. Science vol. 236: 280-285.

Sørensen, J.D. 2001. Code calibration and Timber Experience. Cost Action E24 “Reliability based Design of
Timber Structures”, Coimbra May 4-5.
Stallen, P.J.M., Geerts, R. & Vrijling, J.K. 1996. Three concepts of quantified societal risk. Risk Analysis,
16(5).
Steenbergen, H., Vrouwenvelder, A. 2003. Theoriehandleiding PC-Ring - Deel A: Mechanisme-
beschrijvingen (2003-CI-R0020). Technical report, TNO Bouw.
Terzaghi, K. (1929). E_ect of Minor Geological Details on the Safety of Dams. Bulletin of the American
Institute of Mining and Metallurgical Engineers, (TP 125):31-44.
Van Dantzig, D. 1953. Economic Decision Problems for Flood Prevention. Econometrica, 24(3):276{287.
Vanmarcke, E.H. 1977. Probabilistic modeling of soil properties. Journal of the geotechnical engineering
division – ASCE. November 1977, pp 1227 – 1246.
Vrijling, J.K., Schweckendiek T. & Kanning W. 2011: Safety Standards of Flood Defenses, Proc. of the 3rd
Int. Symp. on Geotechnical Safety and Risk (ISGSR 2011), Munich, Germany, 856–890.
Vrijling J.K., Van Hengel W. & Houben, R.J. 1998. Acceptable risk as a basis for design, Reliability
Engineering and System Safety, vol. 59, 141-150.
Vrouwenvelder, A.C.W.M. 2006. Spatial effects in reliability analysis of flood protection systems. Second
IFED Forum, April 26-29, 2006, Lake Louise, Canada.
@seismicisolation
@seismicisolation
IOS Press, 2013
Subject Index
AASHTO 183 limit state design (LSD) 25, 46, 128,
anchor 15 143
autocorrelation 87 limit state design/analysis 157
Bayesian modeling 214 limit states 243
bearing capacity 171, 228 load and resistance factor
bias 261 design (LRFD) 3, 183, 243,
calibration 60, 243 277, 295, 311
Canadian codes 277 LRFD calibration 261
characteristic value 36, 87 material strength design 3
code calibration 311 MRFD 295
code comparison 277 national annex 116
cohesion 195 numerical analysis 143
consequences classes 72 numerical modelling 157
consequences of failure 3 partial factor format 295
design approach 46, 116 partial factor method 171
dikes 311 partial factors 3, 25, 46, 72, 102,
discontinuity layout optimisation 195 128, 311
drilled shafts 183 performance-based design 183
EC7 implementation 25 pile foundation 46
effective friction angle 214 piles 295
Eurocode 7 25, 36, 46, 72, 87, piling 116
116, 195 plasticity 195
Eurocodes 15, 60, 128 Polish practice 25
factor calibration 214 Polish standards PN-B 25
failure modes 311 probabilistic methods 171
flood defenses 311 probability 214

foundation design 243 pullout 261
friction 195 reinforced soil walls 261
geotechnical code development 277 reliability 36, 72, 102
geotechnical design 36, 72, 102, 157 reliability calibration 295
geotechnical parameters 87 reliability theory 261
geotechnical structures 60 reliability-based geotechnical
geotechnical uncertainty 214 design 277
geotechnics 128 resistance factors 243
global safety factor 46 safety 102
gravity dam 143 safety factor 60
Hasofer-Lind method 171 sand strength 214
heterogeneity 36 scale of fluctuation 87
implementation 72 settlement 228
in-situ tests 214 shallow foundation 46, 171
length-effect 311 sliding 143
limit analysis 195 soil properties 36
limit state approach 228 spatial soil variability 87
@seismicisolation
@seismicisolation
330
spatial variability 311 target reliability 311

spread foundations 228 ultimate limit state 195
standards 102 uncertainties 87
steel strip 261 upper bound 195
system reliability 311
@seismicisolation
@seismicisolation
IOS Press, 2013
Author Index
Allen, T.M. 243, 261 Magnan, J.-P. 60
Bathurst, R.J. 261 Maranha das Neves, E. 143
Bond, A. 3 Mayne, P.W. 214
Burlon, S. 60 Misra, A. 183
Caldeira, L. 143, 171 Miyata, Y. 261
Calle, E. 311 Orr, T. 72
Cassidy, M.J. 214 Pereira, C. 171
Ching, J. 295 Phoon, K.-K. 295
Farinha, M.L.B. 143 Roberts, L.A. 183
Farrell, E.R. 15 Schneider, H.R. 87
Fenton, G.A. 277 Schneider, M.A. 87
Gajewska, B. 25 Schuppener, B. 102
Hannink, G. 128 Schweckendiek, T. 311
Hicks, M.A. 36 Simpson, B. 116
Huang, B. 261 Smith, C. 195
Jongejan, R. 311 Uzielli, M. 214
Kanning, W. 311 van Seters, A. 128
Lees, A. 157 Vaníček, I. 228
Lemos, J.V. 143 Vaníček, M. 228
Lesny, K. 46 Vrouwenvelder, T. 128, 311
@seismicisolation
@seismicisolation
@seismicisolation
@seismicisolation

Modern Geotechnical Design Codes of Practice Implementation, Application

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Modern Geotechnical Design Codes of Practice Implementation, Application

Uploaded by

Copyright:

Available Formats

MODERN GEOTECHNICAL DESIGN CODES

ISSN 2212-781X (print)

Amsterdam • Berlin • Tokyo • Washington, DC

ISBN 978-1-61499-162-5 (print)

Distributor in the USA and Canada

PRINTED IN THE NETHERLANDS

Implementation and Evolution of Eurocode 7 3

The Safety Concept in German Geotechnical Design Codes 102

Limit State Design of the Foundations of Concrete Gravity Dams –

Influence of Ground Water Level on Shallow Foundation Design. Application

AASHTO Geotechnical Design Specification Development in the USA 243

and Ruben Jongejan

Subject Index 329

Implementation and evolution of

• Design Approach 1 (DA1) requires two separate verifications to be performed

• 6 countries have chosen DA1 (Belgium, Iceland, Lithuania, Portugal,

• 1 country allows a choice of DAs 2 and 3 (France)

The division of Design Approaches indicates an almost equal number of countries

1.2. Reliability discrimination based on design situation

Table 1. Design situations defined in EN 1990

(γ F , persistent = γ F ,transient ) > ( γ F , accidental = γ F , seismic )

Two countries – Austria and Germany – have decided to specify alternative

γ F , persistent > γ F ,transient > ( γ F , accidental = γ F , seismic )

For example, in Germany, the partial factor applied to variable actions is

γ R , persistent > γ R ,transient > ( γ R , accidental = γ R , seismic )

1.3. Reliability discrimination based on consequences

Consequence Description Examples Associated minimum KFI

One way of introducing reliability discrimination is to introduce an extra partial

1.4. Summarizing reliability discrimination

1.5. Other changes to partial factors for ground properties

1.6. Some lessons learnt from implementation of Eurocode 7

The preceding review of the implementation of Eurocode 7 in various countries

Eurocode 7 Values given for DA3 1.25 1.25 1.4

In May 2010, the European Commission issued a ‘Programming Mandate’ M/466 to

2.1. Response to the Mandate

CEN’s initial response to the Commission’s Programming Mandate proposed new

CEN’s response also proposed various improvements to existing Eurocodes, including:

Nationally Determined Parameters allow the Member Countries of CEN to decide

2.2. How the evolution of Eurocode 7 is being organized

Figure 5. SC7’s Evolution Groups

2.3. Maintenance and simplification

In view of the above, TC250 agreed to ensure simplification in the future

Harmonisation of Anchor Design within

Keywords. Anchor, Eurocode.

be updated in the near future. No standard is currently available on the testing of

2 Limit states in anchor design

2.1 Ultimate limit state (ULS) design force to be resisted by an anchor

RULS;d ≥ EULS,d (1)

2.2 Serviceability limit state (SLS) design force to be resisted by an anchor

A separate validation of the SLS of an anchor is not required in some countries as it is

RSLS;d ≥ γF,SLS FServ,k (4)

2.3 Geotechnical ultimate limit state anchor resistance

RULS;k = RULS;m,min / ξULS (5)

where RULS,m,min is the lowest value of RULS;m measured in investigation and/or

RULS;d = RULS,k / γa,ULS (6)

where γa,ULS is a partial factor.

2.4 Geotechnical serviceability limit state anchor resistance

RSLS;k = RSLS;m,min / ξSLS (7)

RSLS,d = RSLS,k / γa,SLS (8)

3 Determining RULS,m and RSLS,m

PP ≥ ξULS γa,ULS EULS,d (9)

The difference between investigation and suitability tests is mainly that

RULS;m = Min { Rm (αULS or kl ULS) and PP} (10)

A similar approach is taken in the determination of RSLS;d, for those countries