363

RELIABILITY BASED CALIBRATION OF FOUNDATION

STRENGTH FACTOR USING FULL-SCALE TEST DATA
A GUIDE FOR DESIGN ENGINEERS

Working Group
B2.07

December 2008
 WG B2.07

Reliability based calibration of foundation

strength factor using full-scale test data
A guide for design engineers

Prepared by Asim Haldar, on behalf of the WG

WG B2.07 members

Neil Cuer, WG Convenor (UK) and Eddie O’Connor, Secretary (Ireland)

Regular Members
Mike Vanner (UK), Anthony Digioia (USA), Branko Zadnik (Slovenia), Jean-Pierrre Chouteau
(Canada), Asim Haldar (Canada), Eugene Dembicki (Poland), Aureo Ruffier (Brazil), Sylvain
Muyard (France), Mike Buckley (Ireland), Maurizio Leva (Italy), Michael Pietschke (Germany), John
P. Sivertsen (Norway), Boud Boumecid (UK), Toshiyuki Ozaki(Japan)

Corresponding Members
Kari Sahla (Finland), Graeme Paterson (Australia), N Sabri (Switzerland), Willem Combrinck (South
Africa), Patrick Eychenne (France)

Copyright©2008

“Ownership of a CIGRE publication, whether in paper form or on electronic support only infers right
of use for personal purposes. Are prohibited, except if explicitly agreed by CIGRE, total or partial
reproduction of the publication for use other than personal and transfer to a third party; hence
circulation on any intranet or other company network is forbidden”.

Disclaimer notice

“CIGRE gives no warranty or assurance about the contents of this publication, nor does it accept any
responsibility, as to the accuracy or exhaustiveness of the information. All implied warranties and
conditions are excluded to the maximum extent permitted by law”.

ISBN: 978- 2- 85873-050-6

Reliability Based Calibration\Section 1 i SCB2-WG07, August 2008

 TABLE OF CONTENTS

TABLE OF CONTENTS .................................................................................................... ii

ABSTRACT......................................................................................................................... vi
ACKNOWLEDGEMENTS .............................................................................................. vii
EXECUTIVE SUMMARY .............................................................................................. viii
LIST OF FIGURES .......................................................................................................... xxi
LIST OF TABLES .......................................................................................................... xxiii
LIST OF SYMBOLS ........................................................................................................xxv
GLOSSARY OF TERMS............................................................................................... xxix
1.0 INTRODUCTION................................................................................................. 1-1
1.1 General .....................................................................................................1-1
1.2 Objective ..................................................................................................1-2
1.3 Section Overview.....................................................................................1-3
2.0 RELIABILITY BASED DESIGN (RBD)-BACKGROUND ............................ 2-1
2.1 General .....................................................................................................2-1
2.2 Sample Statistics.....................................................................................2-1
2.3 Estimation of Confidence Interval (CI) and Prediction Interval (PI) ....2-1
2.3.1 Estimation of Confidence Interval (CI) .......................................... 2-1
2.3.1.1 Example Problem............................................................................ 2-3
2.3.2 Estimation of Prediction Interval (PI)............................................. 2-4
2.4 Comparison of Predicted Capacity and Characteristic Capacity........2-5
2.4.1 Predicted capacity ........................................................................... 2-5
2.4.2 Characteristic capacity (IEC 60826)............................................... 2-5
2.5 Estimating Sample Size, n ......................................................................2-7
2.5.1 Estimation of sample size when σ is known (n >30) .................... 2-7
2.5.2 Estimation of sample size when σ is unknown (n < 30) ............... 2-8
2.6 Reliability Based Design Methodologies ..............................................2-9
2.6.1 Allowable stress design (ASD)....................................................... 2-9

Reliability Based Calibration\Section 1 ii SCB2-WG07, August 2008

 2.6.2 IEC 60826 (2003) ......................................................................... 2-10
2.6.3 Load and Resistance Factor Design (ASCE- LRFD) ................... 2-11
2.7 Summary ................................................................................................2-11
3.0 FOUNDATION CAPACITY ASSESSMENT.................................................... 3-1
3.1 General .....................................................................................................3-1
3.2 Uplift Failure Mode ..................................................................................3-4
3.2.1 Frustum model (inverted truncated cone) ....................................... 3-4
3.2.2 Shear model .................................................................................... 3-5
3.3 Determination of Strength Parameters..................................................3-6
3.4 Example Problem (Calculation of Geotechnical Design Capacity).....3-6
3.5 Full-scale Test - Field Program ..............................................................3-8
3.5.1 Selection of sample size.................................................................. 3-9
3.5.2 Selection of test sites....................................................................... 3-9
3.5.3 Subsurface investigation ................................................................. 3-9
3.5.4 Foundation installation.................................................................... 3-9
3.5.5 Loading scheme and sequence........................................................ 3-9
3.5.6 Instrumentation and data collection .............................................. 3-10
3.5.7 Data reduction............................................................................... 3-10
3.5.8 Test results .................................................................................... 3-10
3.6 Interpretation of Test Data....................................................................3-11
3.7 Test Data Collection-Field Programme, an Example .........................3-13
3.7.1 Site specific repeat test data for single geotechnical design capacity
( Rn ) ....................................................................................................... 3-13

3.7.2 Site specific test data along the line route..................................... 3-14
3.8 Summary ................................................................................................3-15
4.0 CALIBRATION OF FOUNDATION STRENGTH FACTOR ( m - MODEL
OF BUCKLEY)..................................................................................................... 4-1
4.1 General .....................................................................................................4-1
4.2 m - Model – No Correlation (Repeated Test Data)................................4-1
4.3 m - Model-With Correlation (Scattered Test Data)................................4-3
4.3.1 Discussion ....................................................................................... 4-5
4.4 Example Problem- ESB Data (Buckley, 1994).......................................4-6
4.4.1 Background ..................................................................................... 4-6
4.4.2 Defining foundation capacities ....................................................... 4-7

Reliability Based Calibration\Section 1 iii SCB2-WG07, August 2008

 4.4.3 Design models................................................................................. 4-8
4.4.4 Foundation data for uplift test (Category 2-9 tests) ........................ 4-8
4.5 Summary ..................................................................................................4-9
5.0 CALIBRATION Of FOUNDATION STRENGTH FACTOR (LINEAR
REGRESSION MODEL)..................................................................................... 5-1
5.1 General .....................................................................................................5-1
5.2 Linear Regression Model........................................................................5-1
5.2.1 General............................................................................................ 5-1
5.2.2 Confidence interval (general linear regression analysis) ................ 5-3
5.2.3 Prediction interval (general linear regression analysis) .................. 5-4
5.2.4 Correlation Issue ............................................................................. 5-5
5.2.5 Linearity Issue................................................................................. 5-5
5.2.6 Zero Intercept Issue......................................................................... 5-6
5.2.7 Variance issue (constant versus non constant)................................ 5-7
5.3 Foundation Strength Factor, φF ............................................................5-9

5.3.1 General............................................................................................ 5-9

5.3.2 CASE A - Intercept is non zero and the variance of residuals is
constant ..................................................................................................... 5-10
5.3.3 CASE B - Intercept is non zero and the variance of residuals is non
constant ..................................................................................................... 5-11
5.3.4 CASE C - Slope only with constant variance of residuals............ 5-11
5.3.5 CASE D - Slope only with non constant variance of residuals
(Similar to m -model) ............................................................................... 5-12
5.4 Flow Diagram .........................................................................................5-13
5.5 Limitation of Case D Model (Equation 5.24)........................................5-13
5.6 Example Problem (Re-analysis of 9-point ESB Data using Linear
Regression Model) ............................................................................................5-14
5.7 Design Problem Using ESB Data (Foundation Depth) .......................5-20
5.8 Summary ................................................................................................5-21
6.0 Criteria For Interpolation and Extrapolation.................................................... 6-1
6.1 General .....................................................................................................6-1
6.2 Interpolation.............................................................................................6-1
6.3 Extrapolation............................................................................................6-2
6.4 Foundation Strength Factor – Interpolation and Extrapolation ..........6-2
6.5 Characteristic capacity given two design capacity values ................ 6-3

Reliability Based Calibration\Section 1 iv SCB2-WG07, August 2008

 6.5.1 Interpolation.................................................................................... 6-3
6.5.2 Extrapolation................................................................................... 6-3
6.6 Summary ..................................................................................................6-4
7.0 SUMMARY AND CONCLUSIONS ................................................................... 7-1
7.1 Summary ..................................................................................................7-1
7.2 Conclusions .............................................................................................7-2
8.0 References.............................................................................................................. 8-1
9.0 APPENDIX............................................................................................................ 9-1
9.1 Example Problem (refer to Figure 3.13).................................................9-1
9.2 Calibration Methodology (Summary).....................................................9-3
9.3 Table 9.3- Scattered test data (Table 3.6 reproduced) .........................9-4
9.4 Compute mi values – (use equation 4.2) ..............................................9-4

9.5 Correlation check (Test No. 1- equation 4.1 and replace y by R)........9-5
9.6 Regression Analysis (variance of residuals is constant) ....................9-5
9.7 “Zero Intercept” Check – [Test No. 2 - equation 5.11] .........................9-7
9.8 Variance of residuals-Test (Test No. 3- Figures 5.4a and 5.4b) ..........9-9
9.9 Predicted Test Capacity (sample size effect included) ........................9-9
9.10 Comparisons of foundation strength factor values (Case C and Case
D) .................................................................................................................9-13
9.11 CIGRE (2002)..........................................................................................9-14

Reliability Based Calibration\Section 1 v SCB2-WG07, August 2008

ABSTRACT

This technical brochure presents a reliability based calibration methodology of foundation

strength factor using full-scale test data; it is shown that the simple model developed by
Buckley (1994), is a subset of the general linear regression model provided the intercept is
zero and the residual has a non constant variance. A simple statistical test is presented to
check for zero intercept condition and a graphical method to test for variance trend of
residuals. A flow diagram provides the necessary steps for the calibration process. A
foundation design problem is also presented as an example to demonstrate the
methodology

SOMMAIRE

Ce rapport présente une méthode de calibration du facteur de résistance des fondations

basée sur une approche fiabiliste en utilisant des données d’essais en vraie grandeur. On
montre que le modèle simple proposé par Buckley (1994) est un cas particulier du modèle
de régression linéaire général, lorsque la résistance à l'origine est nulle et que la variance
des résidus est variable. Un test statistique simple permet de déterminer la condition à
l'origine (nul ou non nul) tandis que la tendance de la variance des résidus est vérifiée par
méthode graphique. Le processus de calibration est présenté sous forme de réseau de
décision qui détaille les principales étapes. Finalement, la méthode est illustrée à l'aide
d'un exemple de conception type.

Key Words:

Reliability based design, probabilistic design, foundation strength factor, undercut

foundation, uplift capacity, calibration, linear regression model, extrapolation, correlation,
exclusion limit, characteristic capacity, frustum cone angle, shear model, confidence
interval, prediction interval.

Mots clé:

Conception basée sur la fiabilité, conception probabiliste, facteur de résistance de

fondation, fondation en redan, résistance à l'arrachement, calibration, modèle de
régression linéaire, extrapolation, corrélation, limite d'exclusion, résistance caractéristique,
angle de pyramide inversée, méthode du cisaillement vertical, intervalle de confiance,
intervalle de prédiction.

Reliability Based Calibration\Section 1 vi SCB2-WG07, August 2008

ACKNOWLEDGEMENTS

The author would like to acknowledge the contributions of the following individuals during
the preparation of this report. The author would like to thank Electricity Supply Board
(ESB), Ireland for providing the uplift test data on undercut pad and chimney foundation.
The author would also like to thank Dr. Jan Rogier from Belgium and Dr. Ruy Menezes
from Brazil for reviewing the report and providing valuable comments. Finally, the author
sincerely appreciates the long term support provided by the Senior Leadership Team of
Newfoundland and Labrador Hydro.

WG 07 members Non members

Mr. Jean-Pierre Chouteau (CA) Professor A.H. S. Ang (USA)
Dr. Mike Buckley (IR) Professor Gary Sneddon (CA)
Mr. Mike Vanner (UK) Dr. Enrique Bazans (USA)
Mr. Neil Cuer (UK) Ms. Cathy Murphy (CA)
Dr. Tony Digioia (USA) Mr. Kyle Tucker (CA)
Mr. Eddie O’Connor (IR) Ms. Connie Hilliard (CA)
Mr. Boud Boumecid (UK) Mr. Kyle Tucker (CA)

Reliability Based Calibration\Section 1 vii SCB2-WG07, August 2008

EXECUTIVE SUMMARY

1.0 Introduction

Overhead line support structures are subjected to the effects of meteorological loads as
well as to the longitudinal loads due to broken conductor and unbalanced ice shedding.
Forces transmitted at the foundation level are often uplift (pullout) forces with small
transverse shear forces for tangent lattice towers or large overturning moments with
corresponding shear forces for single pole, angle or dead end structures. Foundation
types are normally spread footings and steel grillages for lattice towers and drilled shafts
for single poles, angle or dead end structures. Figure 1 depicts a typical undercut pad and
chimney foundation under uplift load.

Backfill
G.L.

Native
W soil Soil

Frustum
Angle, Assumed
Failure
Surface

Figure 1 - A typical foundation under uplift load

Since the meteorological loads are probabilistic in nature, the transmitted forces acting on
the foundation are also probabilistic and are subjected to a high level of uncertainties. In
addition, spatial variation of soil strength parameters also contributes a high level of
uncertainty. Therefore, the foundation design must consider all these uncertainties
quantitatively.

2.0 Objective

The primary objective of this study is to present method(s) of undertaking the calibration of
the theoretical geotechnical design model and hence determine the characteristic capacity
of the foundation, based on full-scale test data. To develop this calibration methodology, a
framework to determine the prediction interval of the foundation test capacity based on a
limited number of full-scale test data is developed using a linear regression model. The
prediction interval provides a confidence on a new test capacity value predicted given a
single geotechnical design capacity value, Rn . The lower bound of the prediction interval is
directly linked to the foundation strength factor, φF which can be used in reliability based
design equation for a single foundation component. The foundation strength factor with an
exclusion limit is defined as the ratio of the lower limit of the prediction interval (predicted
test capacity) with a certain confidence level (analogous to characteristic capacity) and the

Reliability Based Calibration\Section 1 viii SCB2-WG07, August 2008

corresponding geotechnical design capacity. A 10% exclusion limit implies that the
confidence level on the prediction is 90%.

3.0 Methodology

To develop a reliability based calibration methodology for foundations, the four steps
identified here are followed:

• Reliability based design equation (IEC 60826, 2003)

• Assessment of foundation geotechnical design capacity

• Assessment of foundation test capacity from a field test programme

• Data analysis and calibration based on m - model (Buckley, 1994) and linear
regression model

An example problem is presented to illustrate the methodology which is followed by

summary and conclusions.

3.1 Reliability Based Design Equation (IEC 60826, 2003)

IEC 60826 provides a semi probabilistic design equation in terms of the limit load acting on
a component and the strength of the component. The design equation (1) relates the
characteristic capacity, RC to the appropriate load effects, QT having a specified return
period of T-years with a global strength factor, φ R .

φ R RC ≥ QT (1)

Area, α

L
s R = Standard Deviation

Rc Rn

Figure 2 - Characteristic capacity of the test data

The global strength factor φ R depends on strength coordination factor φ S , number of

components subjected to the maximum load intensity φ N and quality of installation φQ ; it
also includes the characteristic strength factor φC , which takes into account the variation

Reliability Based Calibration\Section 1 ix SCB2-WG07, August 2008

between the characteristic capacity for an exclusion limit of e% and the capacity RC with
an exclusion limit of 10%. For this study, the calibration is undertaken for a typical
foundation only (one component) and therefore, the global strength factor, φ R is not used
explicitly.

Figure 2 depicts the characteristic capacity based on e% exclusion limit for a normal
distribution. For a 10% exclusion limit the characteristic capacity, RC ,10% is

RC ,10% = R (1- kα v R ) (2)

where R and vR are the mean and the coefficient of variation (ratio of the standard
deviation, s R and the mean value, R ) of the test capacity values for a finite sample size,
n. The term kα is a measure of how far the point “L” is from the mean to obtain the 10%
exclusion limit ( kα =1.28). For a finite sample size, n < 30, kα should be corrected based
on the Student t-distribution. The exclusion limit 10% implies that the characteristic
capacity is for the 90% confidence level and that there is a 10% probability that the
capacity could fall below this value.

Within the framework of IEC 60826 (2003), the characteristic capacity RC should be
related to the geotechnical design capacity Rn and this could be done through calibration.
If the full-scale test data is available, the appropriate foundation strength factor φ F

RC
φF = (3)
Rn

can be used to relate the geotechnical design capacity Rn to the characteristic capacity
RC with a defined exclusion limit (5% for ASCE and 10% for IEC 60826). The
geotechnical design capacity Rn in figure 2 is derived from analytical and/or semi
empirical relationship or numerical models. Depending on the calibration assumptions, the
relationship could be linear or non linear.

3.2 Assessment of foundation geotechnical design capacity

The geotechnical design capacity Rn of an undercut pad and chimney foundation as

shown in figure 1 is obtained from the frustum cone model. In this particular model, it is
assumed that an inverted frustum starts at the base of the foundation and slopes at an
angle, ψ to the vertical to the surface. The weight of the soils within this frustum and the
foundation weight itself resist the uplift force.

3.3 Assessment of foundation test capacity

Once the geotechnical design capacity Rn has been determined, the foundation designer
is given guidance on how to undertake a full-scale foundation test programme. Several
steps need to be considered before a full-scale foundation test programme can be
commenced. These include: (a) selection of the test site(s), (b) selecting the applicable
foundation geotechnical design model(s) to determine the foundation’s geotechnical

Reliability Based Calibration\Section 1 x SCB2-WG07, August 2008

design capacity, (3) subsurface investigation to determine the various geotechnical
strength parameters for the foundation design model(s), (4) sample size requirement for
the test programme and finally the test results in terms of a typical load versus
displacement curve (figure 3).

Subsequently, a reference is made to IEC 61773 which outlines a number of

methodologies to determine the foundation’s test capacity in terms of damage limit as well
as the ultimate limit capacity based on the test result. A review of the information
presented shows that the estimated ultimate capacity could vary considerably depending
on the method selected. However, the foundation designer can also choose a test
capacity (damage limit capacity, IEC 60826, 2003) based on a prescribed displacement
(subjective). For a pad and chimney foundation used in the example, load at a 10 mm
displacement is defined as the test capacity (Buckley, 1994).

Typical load versus displacement data from an

uplift test.
600
500
Load (kN)

400
300
200
100
0
0 5 10 15 20 25 30 35

Displacement (mm)

Figure 3 - Typical load versus displacement plot

3.4 Data analysis and calibration based on m - model and linear regression
model

3.4.1 General

The uncertainty in the prediction of the characteristic capacity RC is introduced through

two factors: (i) model error and (ii) data error due to a finite sample size n. However, there
is a need to assess both the quality of the data set (correlation coefficient, r) and its
influence on the prediction interval in order to develop a general methodology.

The correlation coefficient, r of the dataset is determined from

∑ ( xi − x )( yi − y )
r= (4)
(n − 1) s x s y

where x and y are the mean values and s x and s y are the standard deviations of the
geotechnical design capacity values xi and the test capacity values yi respectively. A

low correlation value ( r 2 ) would imply that there is no linear relationship between the

Reliability Based Calibration\Section 1 xi SCB2-WG07, August 2008

foundation test capacity and the geotechnical design capacity values. Table 1 provides
some guidance on the correlation value r 2 and the linear association between x and y.
Figures 4a and 4b present two datasets with very different correlation properties.

Table 1 – Suggested ranges of correlation value

Linear association
Correlation Value, r 2
between x and y
weak
Low , r 2 ≤ 0.3
moderate
Moderate (0.3 ≤ r 2 ≤ 0.8)
strong
High, r 2 > 0.8
Test Capacity, y

Data Banded
Around The
Line With
High sxy

Geotechnical Design Capacity, x

Figure 4a - Low correlation

Test Capacity, y

Data Banded
Around The
Line With
Low s xy

Geotechnical Design Capacity, x

Figure 4b - High correlation

3.4.2 m - model - no correlation (repeated test data at one specific test site)

Since xi = x , for the repeated test data at one specific site, the correlation coefficient r=0.0
(equation 4) indicates that the data is unique to a specific test site. In this case, the

Reliability Based Calibration\Section 1 xii SCB2-WG07, August 2008

foundation strength factor φ F can be obtained directly from equation (2) dividing Rc by
the model predicted geotechnical design capacity Rn (figure 2). The foundation strength
factor cannot be extrapolated to other sites, unless the designer can ensure that the sites
have very similar soil characteristics and the foundation has a similar aspect
Depth
( ) ratio.
Base Width

3.4.3 m - model - with correlation (scattered test data along a line route)

Once the user has determined the test and geotechnical design capacities for the various
test sites (s) and foundation sizes, the user can choose a simple calibration model based
Test Capacity
on m = ratio(s) following (Buckley, 1994). Using this
Geotechnical Design Capacity
approach, it can be shown that the foundation strength factor, φ F for a large sample size
(n >30) is

RC
φF = = m (1- kα * vm ) (5)
Rn

where

m = mean value of the m-values for a finite number of tests and

s
v m = coefficient of variation of the m-values= m ( sm = standard deviation)
m

However, to use this model, the user needs to ensure that the correlation of the data is
high (Table 1) and the coefficient of variation of the model error, vm is low. Provided these
two conditions are met, it can be shown that the m -model is a subset of the general linear
regression model (figure 5). The specific conditions are: (1) zero intercept and (2) non
constant variance of residuals, where the residual error values vary linearly with
geotechnical design capacity values (x-values). The residual error value ε defines the
difference between the actual test capacity and the estimated mean test capacity value
from the linear regression model. For other situations where this is not the case, the
foundation designer should undertake a more detailed analysis based on a linear
regression model; either to ensure that the intercept effect can be neglected or
alternatively to improve the theoretical design model. If the intercept effect cannot be
neglected, then an alternate methodology based on a linear regression model is also
presented illustrating how to include this in the analysis and to estimate the predicted
capacity with the required confidence interval.

In the linear regression model, the designer determines the slope, intercept and the
variance of residuals. Figure 5 depicts this information. Figure 5 also presents the
confidence interval on the estimated mean test capacity value, as well as, the prediction
interval on a new test capacity value given a geotechnical design capacity value. This
figure is very similar to figure 2, except that in figure 2 the characteristic capacity is
computed for the test data only; while, in figure 5 this is determined from the linear
relationship between the test capacity and geotechnical design capacity values.

Reliability Based Calibration\Section 1 xiii SCB2-WG07, August 2008

 140

130
Prediction Interval
Estimated Mean Test Capacity,
120 y^ at x o

Test Capacity, y (kN)

110

Residual
Error, ε
*
100
A

Distance from Estimated

90 Test Capacity
B
80 Slope, b
Lower Limit of
Prediction Interval, yp , is
Confidence
Intercept, a Characteristic Capacity, R c
70 Interval
x o , Any Arbitrary Geotechnical
Design Capacity Value
15 20 25 30 35 40 45 50 55 60

Geotechnical Design Capacity, x (kN)

Figure 5 - Linear regression model

A simple statistical test is presented to check for zero intercept condition ("a" on figure 5)
with a slope, m (b on figure 5) and a graphical method to test for variance trend of
residuals (figure 6). For a non constant variance of residuals, an appropriate weight
1
( wi = ) can be used to make a correction to the original dataset (Ang and Tang, 1975).
xi 2
The non constant variance of residuals is defined as the residual error value, ε in figure 6b
increasing with the geotechnical design capacity value. Moreover, the effect of constant or
non constant variance of residuals is presented and its effect is emphasized in analyzing
the data.

C onstant V ariance
of R esiduals
ε) Residual Errors (

ε = 0.0

G eotechnical D esign C apacities, X

Figure 6a – Constant variance of residuals

Reliability Based Calibration\Section 1 xiv SCB2-WG07, August 2008

 Non Constant
Variance of
Residuals Fan

ε)
Shaped

Residual Errors (

Geotechnical Design Capacities, X

Figure 6b – Non constant variance of residuals

3.4.4 Foundation strength factor, φ F under linear regression model

The foundation strength factor, φ F under the linear regression model is defined as the
ratio between the lower limit of the prediction interval, y p (figure 5, predicted capacity)
and the geotechnical design capacity value, x0 used to determine this capacity.

yp
φF = (6)
x0

Since there are two options (constant versus non constant variance of residuals) under
each of these conditions (zero or non zero intercept condition), the user will encounter four
possible cases for analyzing the data. Table 2 can be used as a guide for calibrating the
foundation strength factor value for a specific foundation type. Each case in Table 2 will
provide a lower limit on the predicted capacity. The equation for each case is given in the
technical brochure. The test for zero intercept is definitive. However, the test for constant
versus non constant variance of residuals is not quantitative and is rather subjective;
because the designer needs to plot the residual error values against the geotechnical
design capacity values and derive a trend line similar to figures 6. A flow diagram is
developed to assist the foundation engineer in choosing a particular case.

Table 2- Analysis combinations

Variance of Residuals
Intercept
Constant Non constant
Non Zero Case A Case B
Zero Case C Case D

Reliability Based Calibration\Section 1 xv SCB2-WG07, August 2008

 Flow Diagram For Computing Foundation Strength Factor
Tabulate Data Properties for the
Problem, xi (geotechnical design
capacity) & yi (test capacity) values

yi
Compute mi =
xi

Data Plot, xi vs. yi

Evaluate m, s m and vm .
.
Low Correlation
Test 1:
Correlation Check
Table 1

High or Moderate
Correlation

Regression Analysis (Figure 5)

Slope, Intercept, Estimated Test Capacity Value
and Variance of Residuals

Slope &
Intercept Test 2: Slope only
Test
Test 2:
2:
Zero Intercept
Eq.
Eq. 4.14
Check

Test 3: Test 3:
Chk Figs. 6(a) Chk Figs. 6(a)
& 6(b) & 6(b)

Case A Case B Case C Case D *

Calculate Foundation
Strength Factors to Provide
Conservative Values -
Equation 6
Test 1: Test for Correlation END
__
Test 2: Simultaneous Test for Intercept = 0.0 and Slope = m
Test 3: Visual Comparison to Test Constant Variance of
Residuals vs. Non Constant Variance of Residuals
*Constant COV - Special Case
(Residual values banded around zero line or "fan
- CIGRE (2002)
shaped" - use judgment)

Figure 7 - Flow diagram

Reliability Based Calibration\Section 1 xvi SCB2-WG07, August 2008

4.0 Example Problem

4.1 m -model

Electricity Supply Board (ESB), Ireland provided data for pad and chimney foundations
tested under various soils conditions. The nine uplift foundation tests carried out in
cohesionless soils are initially analyzed using the m -model. The mean values of the test
and the geotechnical design capacities are 378 kN and 295 kN respectively, while the
coefficients of variation are approximately 35%. The m and vm values are 1.30 and 0.156
respectively.

4.2 Linear Regression Model

Figure 8 presents the data for regression analysis. The correlation value r 2 is 0.81 which
is high (Table 1). The 300 frustum cone model in general, appears to be conservative
because of large bias, i. e. the model in general underestimates the test capacity value by
30%. A linear regression model (CASE A) is run first following the flow diagram in figure 7
and Table 2. The statistical test for intercept indicated that the zero intercept assumption
and the slope, m =1.30 ( m -model) can be accepted. Therefore, the designer has to
choose between Case C and Case D respectively (figure 7).

Uplift Test Data For Undercut Foundation- 30

Degree Cone Model, Correlation Value=0.81

600
Test Capacity (kN)

400
Least Square Fit

200

0
0.0 200.0 400.0 600.0
Geotechnical Design Capacity (kN)

Figure 8 - ESB uplift data plot

Next, figure 9 presents the residual plot. Two lines are drawn to see a trend line that the
variance of residuals is non constant (similar to figure 6b) and it is not fully definitive that
this is the case. Therefore, the designer should try both cases C and D for determining the
foundation strength factor values. Figure 10 depicts the comparison of foundation strength

Reliability Based Calibration\Section 1 xvii SCB2-WG07, August 2008

factor, φ F based on CASE C and CASE D respectively; it is interesting to note that for
CASE D the foundation strength factor is constant across all geotechnical design capacity
values.

Residual Plot To Assess Constant Versus

Non Constant Variance of Residuals
100
Lines to depict the "Fan
Shaped" Plot To Identify
Residual Errors (kN)

50 Non Constant Variance of

Residuals Trend-Refer to
0 figure 6(b)
0.0 100.0 200.0 300.0 400.0 500.0 600.0
-50

-100

-150
Geotechnical Design Capacity (kN)

Figure 9 – Residual errors versus geotechnical design capacity values

Comparison of Strength Factor Values For CASES C and D (90%

Prediction Interval-10% EL)

1.200
Foundation Strength Factor

1.000

0.800

0.600
Data range- 150kN Foundation strength
0.400
to 499kN factor(Constant),φF = 1.00
Case D
Foundation strength
0.200
factor(Variable),φF -Case C
0.000
0 100 200 300 400 500 600 700 800 900
Geotechnical Design Capacity (kN)

Figure 10 - Comparison of foundation strength factor values (sample size effect

included)

Reliability Based Calibration\Section 1 xviii SCB2-WG07, August 2008

However, if the foundation engineer uses CASE C, where the variation of residual errors
with respect to x-values in figure 9 is interpreted as constant (figure 6a), the foundation
strength factor becomes strongly dependent on the geotechnical design capacity value (x-
value) as the predictor values are further away from the mean capacity value (295 kN).
Therefore, in this situation, the foundation designer may want to consider the minimum of
the two foundation strength factor values. A foundation design problem presented in the
brochure shows that the difference in the design depth of foundation could be 9% or more
depending on the assumption the foundation engineer chooses.

5.0 Summary and Conclusions

This study shows clearly that there are a number of steps one needs to consider before a
foundation strength factor, for a particular foundation type, can be estimated based on a
limited number of full-scale test capacity data and the theoretical foundation model
predicted geotechnical design capacity. The foundation engineer also needs to understand
that each dataset is different because of various test conditions, as well as model
parameters imposed on them. The WG B2-07 recommends that a clear and consistent
methodology based on accepted statistical theories, supported by valid assumptions, will
provide a good basis for testing any database and its practical use for reliability based
design of foundations. Simple tests, as proposed in this document include: correlation
test, simultaneous test for zero intercept and slope, trend line for variance of residuals
variation etc.

Based on these tests, the foundation designer can clearly decide whether to use the m -
model (CIGRE, 2002) or to use the linear regression model (alternative method),
applicable to those databases where the results clearly show that the assumptions made
under the m -model may not be valid. A flow diagram is developed to assist the foundation
engineer in choosing a particular option.

Some important conclusions are drawn from this study:

• A review of the current m -model (Buckley, 1994, CIGRE, 2002) is presented;

which has examined the underlying assumptions and their limitations in a
systematic manner. The method is primarily based on the linear regression model
and it is shown that it is valid under very specific conditions and assumptions, as
detailed below.

• The Buckley method relies on two very specific conditions which imply that the
intercept should be zero (figure 5) and the residual error values should vary with
the geotechnical design capacity values in an increasing manner (figure 6b).

• To apply this CIGRE method to a wide range of reliability based foundation design
calibrations with various degrees of correlation, the designer must be aware that
unless these specific conditions are met, the prediction interval with a high degree
of confidence (90% for IEC 60826 and 95% for ASCE) will produce results that
may not be valid.

• A simple statistical test (simultaneous test) is proposed here to provide guidance to

the foundation designer so that the assumption of “zero intercept” with a slope, m ,
is validated properly.

Finally, it is shown that the general linear regression model with the unique test for “zero
intercept” is quite applicable to the analysis of a wide range of foundation test data with

Reliability Based Calibration\Section 1 xix SCB2-WG07, August 2008

various degree(s) of correlation, coefficient of variations and m values. The methodology
presented here is quite robust and provides a consistent basis for analyzing a wide range
of test data. The foundation designer is able to review the details of these calculations in
order to understand the intricate relationships of various parameters that contribute to the
estimation of the foundation strength factor, φ F values. The foundation designer can
make an informed decision to choose a particular option or options (Cases A to D) in
determining the predicted foundation test capacity with a specified confidence level.

6.0 References

Ang, A. H. S and Tang, W.H. 1975 Probability Concepts in Engineering Planning and
Design, Volume 1, John Wiley & Sons

Buckley, M 1994 Reliability Based Design of OHL Foundations, CIGRE Publication, SC

22-203, Overhead Lines, Paris Meeting, August 28-September 3,11p.

CIGRE 2002 The Design of Transmission Line Support Foundations- An Overview, WG

22.07, August, Technical Brochure No. 206

IEC 60826 2003 Design Criteria of Overhead Transmission Lines

Reliability Based Calibration\Section 1 xx SCB2-WG07, August 2008

 LIST OF FIGURES

Figure 1 - A typical foundation under uplift load.................................................................viii

Figure 2 - Characteristic capacity of the test data............................................................... ix
Figure 3 - Typical load versus displacement plot ................................................................ xi
Figure 4a - Low correlation ................................................................................................ xii
Figure 4b - High correlation................................................................................................ xii
Figure 5 - Linear regression model ....................................................................................xiv
Figure 6a – Constant variance of residuals........................................................................xiv
Figure 6b – Non constant variance .................................................................................... xv
Figure 7 - Flow diagram .....................................................................................................xvi
Figure 8 - ESB uplift data plot ...........................................................................................xvii
Figure 9 – Residual errors versus geotechnical design capacity values.......................... xviii
Figure 10 - Comparison of foundation strength factor values (sample size effect included)
................................................................................................................................. xviii
Figure 1.1 - Typical loads on an overhead line structure (Buckley, 2001) ........................1-1
Figure 1.2 - Resultant forces acting on a typical foundation (Buckley, 2001) ...................1-2
Figure 1.3 - Resistive forces on a typical foundation (Buckley, 2001) ..............................1-2
Figure 2.1 - Two sided confidence interval and significance level ....................................2-2
Figure 2.2 - One sided confidence interval, p and significance level α ...........................2-6
Figure 2.3 - Characteristic strength of a foundation based on test data ...........................2-6

E*
Figure 2.4 - Sample size versus ( ) for various confidence levels...............................2-8
sR
Figure 3.1 - Typical line route showing the test sites (Topographic Map).........................3-2
Figure 3.2 - Typical undercut pad and chimney foundation ..............................................3-3
Figure 3.3 - Failure in native soils .....................................................................................3-4
Figure 3.4 - Failure in backfill soils....................................................................................3-5
Figure 3.5 - Shear model (IEEE, 1985).............................................................................3-6
Figure 3.6 - Typical subsurface profile of a specific test site.............................................3-7
Figure 3.7 - Subsurface soil profiles with assumed failure surface ...................................3-7
Figure 3.8 - Grillage foundation under uplift load (Prasad & Haldar, 2001) ....................3-10
Figure 3.9 - Typical load displacement plot for a pad and chimney foundation ..............3-11
Figure 3.10 - Repeated test data for an undercut pad and chimney foundation-project
specific ....................................................................................................................3-14
Figure 3.11 - Uplift test -scattered test data along the line route (damage limit capacity at
10mm displacement) ...............................................................................................3-15

Reliability Based Calibration\Section 1 xxi SCB2-WG07, August 2008

Figure 4.1 (a) - Low Correlation ........................................................................................4-2
Figure 4.1 (b) - High Correlation .......................................................................................4-3
Figure 4.2 - Predicted test capacity based on m -model ..................................................4-5
Figure 5.1 - Typical regression plot with intercept and slope ............................................5-2
Figure 5.2 - Regression plot with confidence interval on mean test capacity values and
prediction interval for new test capacity values .........................................................5-5
Figure 5.3 - Residual errors banded around ε = 0.0 line .................................................5-6
Figure 5.4(a) - Constant variance of residuals ..................................................................5-8
Figure 5.4(b) - Non constant variance of residuals ...........................................................5-8
Figure 5.5 - Non constant variance of residuals trend for a larger dataset .......................5-9
Figure 5.6 - Flow diagram ...............................................................................................5-15
Figure 5.7 - ESB Uplift test data (CIGRE, 2002).............................................................5-16
Figure 5.8 - Residual errors versus geotechnical design capacity values ......................5-17
Figure 5.9 - Predicted test capacity with 90% confidence level (Case C, constant variance
of residuals).............................................................................................................5-18
Figure 5.10 - Predicted test capacity with 90% confidence level (CASE D, non constant
variance of residuals) ..............................................................................................5-19
Figure 5.11 - Comparison of strength factor (sample size effect included).....................5-19
Figure 5.12 - Geotechnical design capacity with respect to change in depth .................5-20
Figure 6.1 - Interpolation ...................................................................................................6-1
Figure 6.2 – Extrapolation .................................................................................................6-2
Figure 9.1 - 3-point Scattered test data plot......................................................................9-1
Figure 9.2 - Flow diagram (figure 5.6, reproduced)...........................................................9-2
Figure 9.3 - Linear regression plot ....................................................................................9-7
Figure 9.4 - Linear regression plot (with intercept and without intercept) .........................9-7
Figure 9.5 - Residual values versus Geotechnical design capacity ..................................9-9
Figure 9.6 - Limit of Prediction Interval (case C).............................................................9-11
Figure 9.7 – Lower Limit of Prediction Interval (Case D) ................................................9-13
Figure 9.8 - Comparison of foundation strength factor....................................................9-14
Figure 9.9 – Foundation Strength factor (Buckley, 1994 & CIGRE, 2002)......................9-14

Reliability Based Calibration\Section 1 xxii SCB2-WG07, August 2008

 LIST OF TABLES

Table 1 – Suggested ranges of correlation value................................................... xii

Table 2- Analysis combinations .............................................................................xv
Table 2.1 - kα / 2 & kα - Values for various confidence levels (Benjamin & Cornell,

1970) ............................................................................................................ 2-3

Table 2.2 - tα / 2,ν & tα ,ν Values for figures 2.1 and 2.2 (Benjamin & Cornell,

1970) ............................................................................................................ 2-4

Table 3.1- Typical dimensions (m) of an undercut pad and chimney foundation . 3-3
Table 3.2 - Foundation geotechnical design capacity (kN) under various frustum
angles (ψ ) assumptions ............................................................................... 3-8
Table 3.3 - Comparison of ultimate capacity of an undercut foundation (IEC
61773) ........................................................................................................ 3-12
Table 3.4 - Geotechnical properties at Site A (Figure 3.1, IEEE, 1985) ............. 3-13
Table 3.5 - Geotechnical properties at three sites (Figure 3.1,IEEE, 1985) ....... 3-14
Table 3.6- Test data versus geotechnical design capacities.............................. 3-15
Table 4.1 - φF Values from equations (4.7 & 4.8) ............................................... 4-5
Table 4.2 – Suggested ranges of correlation value.............................................. 4-6
Table 4.3 - ESB Test Data (CIGRE, 2001) .......................................................... 4-8
Table 4.4 - Data Statistics.................................................................................... 4-9
Table 4.5 – Foundation Strength Factor Values for 9-point dataset (ESB Data).. 4-9
Table 5.1 - F-distribution Values (Benjamin & Cornell, 1970) .............................. 5-7
Table 5.2 - Various Cases in Linear Regression Model ..................................... 5-10
Table 5.3 - Theoretical upper Limit of vm for various confidence levels and sample

size (Equation 5.25).................................................................................... 5-14

Table 5.4 - Data Statistics for Intercept Check (Equation 5.11) ......................... 5-16
Table 5.5 - (Case C-b=1.27, s xy =59.29, tα ,ν =1.40) .......................................... 5-17

Table 5.6 (Case D- m =1.30, s xy = sm =0.20, tα ,ν =1.40) .................................... 5-18

Table 5.7 –Foundation Strength Factor Values (Linear Regression Model) ...... 5-20
Table 5.8 - Typical Dimensions (m) of a Pad and Chimney Foundation (Figure
3.2) ............................................................................................................. 5-21
Table 9.1: Methodology to Follow the Flow Diagram ........................................... 9-3

Reliability Based Calibration\Section 1 xxiii SCB2-WG07, August 2008

Table 9.2: Reference Equation Numbers for Various Cases (Table 5.2) ............. 9-4
Table 9.3- Scattered test data.............................................................................. 9-4
Table 9.4 - Section 5.2.1 (Case A, x = 438.7 kN, wi = 1.0 ).................................. 9-5

Table 9.5 – Variance of residuals calculation...................................................... 9-6

Table 9.6 - Terms in equation (5-11).................................................................... 9-8
Table 9.7- Data statistics for 3 – point scattered test data problem ..................... 9-8
Table 9.8 - Variance of Residuals Calculations ................................................... 9-9
Table 9.9 - Predicted Test Capacity................................................................... 9-10
Table 9.10 – Variance of Residuals Calculations............................................... 9-11
Table 9.11- Predicted Test Capacity.................................................................. 9-12
Table 9.12 - Comparison of foundation strength factor (Case C & Case D) ...... 9-13

Reliability Based Calibration\Section 1 xxiv SCB2-WG07, August 2008

 LIST OF SYMBOLS

Symbol Unit Description Equation

Number/
Section
Number(§)
A m Foundation depth (Fig. 3.1) §3.1
a kN Intercept of the linear regression line 5.3
a* kN Arbitrary value of “a” 5.11
B m Foundation width at the bottom (Fig. 3.1) §3.1
b - Slope of the linear regression line 5.2
C m Width of excavation (Fig. 3.1) §3.1
CI kN Confidence interval for the mean test capacity 2.6
c kN/m2 Cohesion 3.4
D m Height of foundation pad (Fig. 3.1) §3.1
D0 m Equivalent diameter of a rectangular foundation §4.4.4
Dn kN Effects of dead load 2.14
E* kN Error margin in estimating the confidence interval 2.12
(to estimate the population mean)
E m Width of foundation chimney (Fig. 3.1) §3.1
e % Exclusion limit 2.11
F0 kN Shear capacity (Frictional uplift resistance) 3.3
F* - A value computed for “zero intercept” check 5.11
F m Height of foundation chimney above ground (Fig. §3.1
3.1)
Fα ,2,ν - A value obtained from F-distribution with 5.11
significance level, α and degrees of freedom,ν &
2
G m Width of frustum at ground level 3.7
K - Coefficient of lateral earth pressure 3.4
kα - Constant for determining exclusion limit for one 2.11
sided interval (n> 30)
kα/ 2 - Constant for determining exclusion limit for two 2.6
sided interval (n> 30)
m - Mean of mi values 4.3
mi - Ratio of test capacity, yi and geotechnical design 4.2
capacity, xi , for sample test, i
N - SPT value to define soil strength parameters § 3.3
n - Sample size 2.1
PI kN Prediction interval for a new test capacity 2.8
p - Confidence level 2.5
QT kN Effect of a climatic load with a T-return period 2.16
R kN Sample mean test capacity data (also noted as y ) 2.1
RC ,e% kN Characteristic capacity defined as the capacity 2.11
with a specified exclusion limit e %

Reliability Based Calibration\Section 1 xxv SCB2-WG07, August 2008

Symbol Unit Description Equation
Number/
Section
Number(§)
RC,10% kN Characteristic capacity defined as the capacity 4.6
with a 10% exclusion limit (IEC 60826, RC )
Ri kN Foundation test capacity for sample test, i (also 2.1
noted as yi )
Rn kN Geotechnical design capacity (also noted as xi ) 2.14
Rn kN Mean geotechnical design capacity (also noted as 5.3
x)
Rn +1 kN New foundation test capacity value for prediction § 2.3.2
r - Correlation coefficient 4.1
r2 - Correlation value § 4.3.1
S m2 Shear surface area §3.2.2
s xy kN or non- Standard deviation of the error in the linear 5.6
dimensional regression analysis; for wi =1.0, dimension is kN
1
and for wi = , this is non dimensional
xi 2
s xy 2 (kN or non- Square of the standard deviation of the error in 5.6
dimensional)2 the linear regression analysis; Variance of
residuals
sx kN Standard deviation of xi - data 2.2 & § 4.1
sy kN Standard deviation of yi - data 2.2& § 4.1
sR kN Standard deviation of the foundation test capacity 2.2
data
tα ,ν - Constant for determining exclusion limit for one § 2.4.2
sided interval (n< 30, Student t-distribution)
tα / 2,ν - Constant for determining exclusion limit for two 2.7
sided interval (n< 30, Student t-distribution)
U kN Uplift force 3.1
VC m3 Total volume of the concrete 3.1
VC,B m3 Volume of the concrete below the ground line 3.8

VS m3 Volume of the soils within the frustum 3.5

3
VS , F m Total volume within the frustum 3.6
vm - Coefficient of variation of m-values §4.2, 4.5
vR - Coefficient of variation of the foundation test 2.3
capacity data
WC kN Weight of the concrete of the foundation 3.2
WS kN Weight of the soil within the frustum 3.2
WI n kN Effects of wind and ice loads on conductors and 2.14
support

Reliability Based Calibration\Section 1 xxvi SCB2-WG07, August 2008

Symbol Unit Description Equation
Number/
Section
Number(§)
wi - Weight for sample test, i , in the linear regression 5.2
analysis ( wi = 1.0 for constant variance of
1
residuals and wi = for non constant variance
xi 2
of residuals)
xL kN Lower limit of the xi data (predictor values) § 6.5
xU kN Upper limit of the xi data (predictor values) § 6.5
xi kN Geotechnical design capacity for sample test, i 5.5
x0 kN Single geotechnical design capacity 5.10
x kN Mean geotechnical design capacity data 4.1
yi kN (Actual) test capacity for sample test, i 5.1
yp kN Predicted test capacity for a given single 5.10
geotechnical design capacity, x0
ŷ kN Estimated mean test capacity from the linear 5.17
regression analysis for a single geotechnical
design capacity, x0
ŷi kN Estimated mean test capacity from the linear 5.5
regression analysis for sample test, i
y kN Mean test capacity data 4.1

α - Significance level 2.5

β - Reliability Index § 2.6
γD - Load factor for the dead loads 2.19
γ WI - Load factor for the combined wind and ice loads 2.19
εi kN Residual error value from the linear regression 5.1
analysis (difference between the actual test
capacity, yi and estimated mean test capacity, ŷi )
η - Exponent of Mors (1965) equation 3.3
μ kN Population mean § 2.2
ν - Degrees of freedom (n-1 for one unknown; n-2 for § 2.3
two unknowns)
ρC kN/m3 Unit weight of concrete 3.1
ρS kN/m3 Unit weight of soil within the frustum, soil density 3.1
σ kN Population standard deviation § 2.2
φ - Capacity reduction factor 2.19
φ ° Angle of internal friction 3.4
φF - Foundation strength factor defined as the ratio of 2.18
RC and Rn
φC - Characteristic strength factor taking into account § 2.6.2
the variation between the actual exclusion limit

Reliability Based Calibration\Section 1 xxvii SCB2-WG07, August 2008

Symbol Unit Description Equation
Number/
Section
Number(§)
and the supposed exclusion limit of 10% of the
characteristic capacity RC (IEC 60826)
φN - Strength factor related to the number of § 2.6.2
components subjected to maximum load intensity
φQ - Strength factor related to the quality of installation § 2.6.2
φR - Global strength factor 2.17
φS - Strength factor related to the coordination of § 2.6.2
strength
ψ ° Frustum angle for uplift capacity calculation 3.7
ψ* - Resistance factor 2.15
τ kN/m2 Soil shear strength 3.3

Reliability Based Calibration\Section 1 xxviii SCB2-WG07, August 2008

 GLOSSARY OF TERMS

Support - structure to carry overhead line conductors and ground wires.

Foundation - a structure below the ground to transfer support loads to the soils.

Sample mean - the arithmetic mean of a sample data of finite size.

Standard deviation - the dispersion of a random variable from its mean.

Population mean – mean of a normal distribution population.

Test capacity - load corresponding to a prescribed foundation movement.

Geotechnical design capacity - computed based on theoretical and/or semi empirical

design model; for example, the frustum cone model is used in this study to determine the
foundation geotechnical design capacity under uplift.

Coefficient of variation - the ratio of standard deviation to the mean value.

Correlation coefficient - a measure for the linear association of y with x.

Correlation value - square of the correlation coefficient.

Variance - square of the standard deviation.

Null hypothesis -a proposition which is believed to be true; this could be that the line
passes through the origin while the alternative hypothesis could be the line does not pass
through the origin.

Significance level - a probability measure to reject a null hypothesis even though the
hypothesis is in fact true.

Exclusion limit - a variable defined in terms of percentage (%) to define the low side of
the mean value from a probability distribution.

Confidence interval - provides a probability measure of how close the population mean is
to the sample mean given a finite sample size; population mean is normally unknown and
is an unobservable quantity.

Prediction interval - provides a probability measure that a new test value drawn randomly
from a distribution will lie within a certain interval given a finite sample size.

Characteristic capacity - provides a capacity measure in probabilistic term with a

specified exclusion limit; capacity with 5% or 10% exclusion limit means that the capacity
will be higher than the specified value with 95% or 90% probability.

Normal distribution - a symmetric distribution commonly used in statistical analysis;

normally valid for infinite sample size where the tail extends to infinity.

Log normal distribution - a probability distribution of any random variable whose

logarithmic value is normally distributed.

Reliability Based Calibration\Section 1 xxix SCB2-WG07, August 2008

Student t distribution - a distribution derived from normal distribution when the sample
size is finite.

F-distribution -a joint probability distribution with two degrees of freedom.

Reliability index - a measure of structural reliability taking into account the variations in
load and strength.

Undercut foundation – a type of foundation where the soil above the base of the
foundation is undercut, such the foundation is in direct contact with the soil; very common
in European countries.

Frustum angle - an angle used to define the failure surface in the frustum model for
determining the uplift capacity.

Limit load - climatic load corresponding to a return period, T-years without imposing any
load factor.

Elastic limit capacity - the load at which no significant (reversible) displacement is

observed.

Damage limit (working limit) - the load at which some permanent (irreversible)
displacement is observed and is similar to serviceability limit state in building codes.

Ultimate capacity - the load at which the foundation has undergone a significant
movement and a complete failure will happen if this load is exceeded.

Residual - difference between the actual test capacity value and the estimated mean test
capacity value from the linear regression line.

Constant variance of residuals - square of the residuals provide a constant value across
all values along x-axis (geotechnical design capacity).

Non constant variance of residuals - square of the residuals vary with values along x-
axis (geotechnical design capacity).

Linearity trend - provides a basis to predict y from x using linear relationship.

Foundation Strength factor - provides a measure of the relationship (linear or non linear)
between the characteristic capacity and the geotechnical design capacity using full-scale
test data.

Standard error of estimate (Variance of residuals) - a term in linear regression analysis

to measure the dispersion from the estimated mean test capacity value (sum of the square
of residuals).

Weighted residual - a term used in regression analysis to correct the data when the
variance of residuals is non constant.

Reliability Based Calibration\Section 1 xxx SCB2-WG07, August 2008

1.0 INTRODUCTION

1.1 General

Foundations for overhead line supports are frequently subjected to the effects of
various types of meteorological loads, i.e. due to wind, ice and combined wind and
ice loads on conductors and the support. Apart from these loads, supports are also
subjected to broken conductor loads as well as longitudinal unbalanced loads due to
ice shedding. The support foundation design must take into account the effects of all
these forces.

V1

T1
L1

T2 T2
L2 L2
V2 V2

T3 T3
L3 L3
V3 V3

T4 T4
L4 L4
V4 V4

T5

T6

T7

L = L on g itud in al lo ads du e to T = T ra n sv e rse loa d s d u e to w ind on V = V e rtica l loa d s d u e to condu ctors,

un b ala n ce d co n du c to r tension s c o nd u cto rs , tow ers, fittin gs, ice ic e , fittings , to w er w eigh t
& w in d o n tow er & ice a n d u n ba la nc e d c o nd u ctor ten sio ns

Figure 1.1 - Typical loads on an overhead line structure (Buckley, 2001)

Forces transmitted at the foundation level are often uplift (pullout) forces with small
transverse shear forces for tangent lattice towers or large overturning moments with
corresponding shear forces for single pole, angle or dead end structures. Foundation
types are normally spread footings and steel grillages for lattice towers and drilled
shafts for single poles, angle or dead end structures. Figure 1.1 depicts a typical
double circuit tower with applied forces at the conductor attachment points. Figure
1.2 presents the resultant loads on a typical foundation, while figure 1.3 presents the
various resistive forces acting on this foundation.

Since the variable forces acting on the conductor and tower are probabilistic in
nature, the transmitted forces acting on the foundation are also probabilistic and are
subjected to high levels of uncertainties. In addition, soil conditions and
characteristics such as density, shear resistance, water level, vary also significantly
from one site to the other and also contribute a high level of uncertainty. To resist the
tower forces, the foundation must be designed based on the geotechnical design
capacity which is normally obtained from a theoretical design model or an empirical
method.

Reliability Based Calibration\Section 1 1-1 SCB2-WG07, August 2008

 L1
L1V D1V
D1
L1H
D1H

L2V
L2H

L2

Figure 1.2 - Resultant forces acting on a typical foundation (Buckley, 2001)

Applied
forces
Applied
forces

Soil force on chimney

Soil Soil
weight weight

Shear Soil force on pad

force
Shear force
Soil/foundation friction

Foundation = Resisting
weight forces

Figure 1.3 - Resistive forces on a typical foundation (Buckley, 2001)

Correspondingly, it is necessary to perform full-scale foundation tests to evaluate the

actual capacity of the foundation compared to the theoretical design capacity.

1.2 Objective

The primary objective of this guide on reliability based calibration of foundation

strength factor is to present method(s) of undertaking the calibration of the theoretical
geotechnical design model and hence determine the characteristic capacity of the
foundation, based on full-scale test data. To develop this calibration methodology, a
framework to determine the prediction interval of the foundation test capacity based
on a limited full-scale test data is developed based on a linear regression model. The
prediction interval provides the upper and lower limits of a new test capacity
predicted with a specified confidence level given the sample information. A direct
relationship is established between the lower bound of the prediction interval and the
foundation strength factor, φF which can be used in reliability based design equation
of a single foundation component. The foundation strength factor with an exclusion
limit is defined as the ratio of the predicted test capacity with a specified confidence
level and the corresponding geotechnical design capacity. To assist readers, the
report will present a step-by-step procedure to ensure that the foundation engineer is

Reliability Based Calibration\Section 1 1-2 SCB2-WG07, August 2008

aware of all the assumptions made and the corresponding limitations in applying the
calibration model(s). Specific recommendations are made to mitigate some of the
limitations of the earlier model developed in CIGRE (2002).

1.3 Section Overview

For clarity, this report is divided in the nine sections

• Section 1 presents the introduction and outlines the objective of the study.

• Section 2 provides the information on confidence intervals, prediction

intervals for a new test capacity, characteristic capacity (strength), sample
size estimation and reviews the reliability based design with particular
reference to allowable stress design (ASD), ASCE load and resistance factor
design (ASCE- LRFD) and IEC 60826 (2003) design methodology.

• Section 3 reviews various theoretical models for computing uplift capacity of a

pad and chimney foundation based on frustum angle and shear models. The
necessary geotechnical parameters are identified for calculating the uplift
capacity. The foundation designer is given guidance on how to undertake a
full-scale foundation test programme. A number of steps are outlined how to
carry out a field test programme and to obtain the full-scale test data. Two
scenarios are considered. These are: (1) repeated test data for one specific
site and (2) scattered test data at three sites along a transmission line route.

• Section 4 presents the determination of foundation strength factor based on

the m -model developed by Buckley (1994) and presented in CIGRE (2002).
Various assumptions of this model are presented. This model is applied to the
above two test data. This is followed by an analysis which includes the actual
test data provided by Electricity Supply Board (ESB) in Ireland.

• Section 5 presents an alternative methodology to calibrate the foundation

strength factor based on a linear regression model; it is shown that the m -
model is a subset of this generalized linear regression model. A flow diagram
is included to provide guidance, to the foundation designer, for analyzing the
data. A design problem based on ESB test data (section 4) is reanalyzed
using the linear regression model.

• Section 6 discusses the general criteria for extrapolation and its limitation
within the context of a linear regression model. A design example for
foundation characteristic capacity (predicted) is presented to show the effects
of interpolation versus extrapolation.

• Section 7 presents the summary and conclusions.

• Section 8 presents the references.

• Section 9 presents an APPENDIX with an example problem. A 3 point

scattered test data is analyzed to show how the foundation engineer can
calculate all the necessary parameters manually for determining the
foundation strength factor.

Reliability Based Calibration\Section 1 1-3 SCB2-WG07, August 2008

2.0 RELIABILITY BASED DESIGN (RBD)-BACKGROUND

2.1 General

This section presents the background information on confidence interval and

prediction interval on full-scale foundation test data only. The confidence interval on
mean foundation test capacity is linked to an estimation of the sample size, while the
prediction interval is linked to characteristic capacity with a defined exclusion limit. A
reference is also made to the estimation procedure that can be used in the context of
linear regression model when the test capacity data is related to model predicted
geotechnical design capacity. This is followed by a brief review of the deterministic
and semi-probabilistic methods used in foundation design such as allowable stress
design (ASD), IEC 60826 (2003) and load and resistance factor design (ASCE-
LRFD).

2.2 Sample Statistics

The mean, ( R ), standard deviation, ( sR ), and the coefficient of variation, ( vR ), of a

full-scale data on foundation test capacity of sample size, (n), are calculated based
on the following formulae:

n
R = (∑ Ri ) / n (2.1)
1
1 n
sR 2 = ∑ ( Ri − R )2
n −1 1
(2.2)

s
vR = R (2.3)
R

where Ri , i = 1,...n are the data on foundation test capacity.

Normally the test capacity value (R), is a random variable and is assumed to follow a
distribution (normal, log normal etc.). The population mean resistance ( μ ), is defined
as the estimated mean test capacity value of this random variable with a standard
deviation ( σ ), which are normally unknown. The variance of residuals is defined as
the square of the standard deviation and is a measure of the dispersion around the
estimated mean test capacity value.

2.3 Estimation of Confidence Interval (CI) and Prediction Interval (PI)

2.3.1 Estimation of Confidence Interval (CI)

The mean of a sample drawn randomly from the population is a point estimate of the
population mean ( μ ), and will vary for different samples. Given that the sample
mean is unlikely to be equal to population mean, how reliable R is the measure of μ .
The confidence interval provides this measure. In theory, as the number of samples
approaches infinity ( n → ∞ ), there will be no difference between the sample
statistics and population statistics. However, a large sample size is not practical for

Reliability Based Calibration\Section 3 2-1 SCB2-WG-07, August 2008

full-scale foundation testing and therefore, a confidence interval is often sought to
provide an accuracy of the estimate on the mean test capacity value, ( R ).

Figure 2.1 depicts a typical normal probability density function of R, as f (R) and is
described by the following equation:

1 (R−R ) 2
− [ ]
1 2 sR
f ( R) = e (2.4)
sR 2π

Figure 2.1 also presents the concept of two sided confidence interval (CI), confidence
level and the level of significance. The significance level is based on the area under
the shaded region. The total area under the curve is 1.0.

Probability Density
f (R)
Function of R-
α = Level of significance Equation (2.4)
Area = Confidence level = 1 - α

Area = α/2 Area = α/2

Error
Margin, E*
x
0 L U
(Lower Confidence Mean Value (Upper confidence
Limit) limit)
k α/2 s R from the Confidence Interval
mean to define
the interval width

Figure 2.1 - Two sided confidence interval and significance level (Rao, 1982)

The confidence interval provides an estimate where the true value may lie within the
upper and lower limits. For this case the confidence level is the area under the
probability density curve between the upper confidence (U) and the lower confidence
(L) limits marked on figure 2.1. Once the confidence level (p), is stated, the
probability of true mean (population mean) lying in the confidence interval is p.
Therefore, there is a probability of (1-p) % that the mean value will not lie within the
confidence interval. This quantity is also defined as the significance level, α (Rao,
1982) and is given by equation (2.5)

α = 1-p (2.5)

The significance level, α is very similar to exclusion limit (e %) which will be

discussed in section 2.4.1, kα / 2 sR in figure 2.1 provides the required offset from the
mean to define the confidence interval.

Reliability Based Calibration\Section 3 2-2 SCB2-WG-07, August 2008

Table 2.1 presents kα / 2 -values for various confidence levels. For example kα / 2 =
1.96 for 95% confidence level while kα / 2 = 1.645 for 90% confidence level.

Table 2.1 - kα / 2 & kα - Values for various confidence levels (Benjamin &
Cornell, 1970)

kα / 2 in Figure 2.1 kα in Figure 2.2

Confidence level in % (Two Sided) (One Sided)
99.9% 3.291 3.090
99% 2.576 2.326
95% 1.960 1.645
90% 1.645 1.282
80% 1.282 0.842

If the sample standard deviation, sR is the same as the population standard

deviation, σ (normally for large sample size, n > 30) and is known, the two sided
confidence interval is given as

s
CI= R ± kα / 2 R (2.6)
n

where kα / 2 is obtained from Table 2.1.

The population standard deviation ( σ ), is not known with certainty when the sample
size is less than 30 (n < 30). In this case, the two sided confidence interval (CI) is
obtained from equation (2.7)

sR
CI= R ± tα / 2,ν (2.7)
n

where tα / 2,ν replaces kα / 2 in equation (2.6) and is obtained from Student-t

distribution, ν is the degrees of freedom and in this case, ν = n-1.

Table 2.2 provides some typical tα / 2,ν values for two confidence levels, p= 90%
and 95% respectively. This is normally the case when the sample size, n is less than
30.

2.3.1.1 Example Problem

If we assume that five full-scale-foundation tests (n = 5) have been undertaken for a

specific foundation type and the test results are 200 kN, 250 kN, 275 kN, 280 kN, and
300 kN respectively. Following equations (2.1), (2.2) and (2.3), the mean, standard
s
deviation and the coefficient of variation are R =261 kN, sR =38.5 kN, and vR = R
R
=0.148, respectively.

Reliability Based Calibration\Section 3 2-3 SCB2-WG-07, August 2008

The 90% confidence interval (CI) on the population mean, neglecting the effect of
38.5
small sample size (n=5), is obtained from equation (2.6) as 261 ± 1.645 x = 261
5
± 28kN. The term kα / 2 in equation (2.6) is obtained from Table 2.1. This means
that the populations mean, ( μ ), will lie within ± 28kN of the sample mean ( R ), with
a p= 90% probability. The engineer may state that the true mean value of the
foundation uplift capacity lies between 233kN and 289kN with a 10% significance
level. The two limits of the confidence level (CI) are 233kN (L) and 289kN (U). In
figure 2.1, the value of E* = ± 28 kN is the error margin and will vary depending on
the confidence level sought p, sample size n, and the sample standard deviation sR .

Table 2.2 - tα / 2,ν & tα ,ν Values for figures 2.1 and 2.2 (Benjamin &
Cornell, 1970)

tα / 2,ν tα ,ν
Degrees of
Freedom, ν 90% 95% 90% 95%
( α / 2 = 0.05) ( α / 2 = 0.025) ( α = 0.1) ( α = 0.05)
1 6.32 12.7 3.08 6.32
2 2.92 4.30 1.89 2.92
3 2.35 3.18 1.64 2.35
4 2.13 2.78 1.53 2.13
8 1.86 2.31 1.40 1.86
10 1.81 2.23 1.37 1.81
11 1.80 2.20 1.36 1.80
30 1.70 2.04 1.31 1.70
∞ 1.64 1.96 1.28 1.64

Considering the effect of sample size n = 5 , equation (2.7) is used with tα / 2,ν =
2.13 for 90% confidence level (table 2.2). The revised calculation shows that the
38.5
lower and upper values are 224 kN and 298 kN respectively. ( 261 ± 2.13 x =
5
261 ± 37 kN); however, it is noted that the error margin ( ± E*) has now increased by
33% (37 kN versus 28 kN) because of the sample size effect. For the case when the
sample size is small, the sample size and the confidence interval width, ± E* =
sR
± tα / 2,ν are interrelated because tα / 2,ν depends on ν which is also
n
dependent on n (table 2.2). This will be discussed further in section 2.5 in
determining the sample size for foundation tests.

2.3.2 Estimation of Prediction Interval (PI)

Similar to confidence interval (CI), the foundation designer can also consider an
interval such that there is a p% probability that a new test capacity drawn randomly
from the population will occur within the interval. This is called the prediction interval.
The confidence interval is for a population parameter (unobservable quantity and
unknown), while the prediction interval is for a single test capacity value (future

Reliability Based Calibration\Section 3 2-4 SCB2-WG-07, August 2008

observation) given the sample information. The prediction interval (PI) for a new test
capacity value, ( Rn +1 ), for a given sample size, (n) is

1
PI= R ± kα / 2 sR [1 + ]0.5 (2.8)
n

The prediction interval (PI) for a new test capacity value, Rn +1 for a small sample
size (n less than 30) is

1
PI= R ± tα / 2,ν sR [1 + ]0.5 (2.9)
n

The prediction interval (PI) is wider than the confidence interval (CI in equation 2.6)
because Rn +1 is a variable quantity while the sample mean, ( R ), is a constant
value. For large sample size n, equation (2.8) can be approximated as

PI= R ± kα / 2 sR (2.10)

This implies that a future observation will lie between ( R + kα / 2 sR ) and

( R − kα / 2 sR ) with a significance level, ( α ).

2.4 Comparison of Predicted Capacity and Characteristic Capacity

2.4.1 Predicted capacity

However, in certain applications, the prediction interval (PI) can be used to derive a
guaranteed minimum test capacity with a specified confidence level. If the designer is
only interested in the minimum value (lower limit of the prediction interval) then this
becomes R − kα sR (one sided). Equation (2.10) can be used to provide this capacity
considering the lower limit only and figure 2.2 presents this case as a one sided
interval and kα / 2 is replaced by kα (table 2.1).

2.4.2 Characteristic capacity (IEC 60826)

The guaranteed minimum test capacity is analogous to characteristic capacity, ( Rc ) ,

an engineering term, defined as the low side of the estimated mean test capacity with
a specified exclusion limit and is depicted in figure 2.3 (IEC 60826, Melchers, 1999).
For one sided prediction interval, the significance level can also be interpreted as the
lower exclusion limit. For a large sample size, the predicted capacity and the
characteristic capacity will be almost identical. However, for a small sample size, the
difference could be 3% as shown below.

For example, if we have foundation test data of a large sample size with mean, ( R ),
s
and a coefficient of variation, ( vR = R ), then the characteristic foundation capacity
R
with an exclusion limit, e % (similar to significance level, α ) can be estimated from
equation (2.11)

Reliability Based Calibration\Section 3 2-5 SCB2-WG-07, August 2008

 RC ,e% = R (1- kα vR ) (2.11)

where kα is a measure from the mean value to the point L, on the x-axis where a
specified exclusion limit is sought. The exclusion limit e=10% implies that the mean
characteristic foundation capacity is guaranteed for 90% confidence level and there
is a 10% probability that the mean capacity could fall below this minimum value. This
implies that the area under the curve from left of point “L” is 0.10 and the area under
the curve to the right of point “L” is 0.90. The geotechnical design capacity, Rn in
figure 2.3 is calculated based on analytical and/or semi empirical relationship or
numerical models (Buckley, 1994, Digioia et al, 1993).

f (R)

α = Level of significance

Area = Confidence level = 1 - α

sR

x
0 L
(Lower Confidence
Limit)
k α sR from the One Sided Confidence Interval
mean to define
the interval
width

Figure 2.2 - One sided confidence interval, p and significance level α (Rao, 1982)

Area, α

L
s R = Standard Deviation

Rc Rn

Figure 2.3 - Characteristic strength ( RC ) of a foundation based on test data

Reliability Based Calibration\Section 3 2-6 SCB2-WG-07, August 2008

For sample size n greater than 30, kα value is 1.28 for 90% confidence level (10%
exclusion limit) and can be obtained from Table 2.1. However, for a finite sample
size, n < 30, kα sR should be corrected based on Student t-distribution (table 2.2).

The characteristic capacity with a 10% exclusion limit for the five tests presented
earlier is obtained from equation (2.11).

RC ,10% = R (1- kα vR ) = 261 (1-1.28* 0.148) = 212kN

This capacity is based on a kα value of 1.28 when the sample size effect is not
included. Since the number of test is less than 30 (n < 30), tα ,ν value is used to
correct for the small sample size effect. Using tα ,ν value = 1.53 from Table 2.2 (for
α = 0.10 and ν = n-1 = 4), the revised estimate of the characteristic capacity is:

RC ,10% = R (1- tν ,α vR ) = 261(1-1.53*0.148) = 202kN

This shows that the characteristic capacity is overestimated in equation (2.11) by 5%

when the effect of small sample size is not taken into account. The minimum
guaranteed capacity based on the prediction interval concept would be 196 kN (3%
less) compared to 202 kN. The characteristic capacity is applicable to single variable
data (test capacity only), while the concept of prediction interval can be used for the
problem where more than one variable is involved, such as the calibration
methodology involving the test capacity versus geotechnical design capacity.

2.5 Estimating Sample Size, n

Quite often, the question is asked “what sample size do I need to estimate the
sample mean test capacity value which will be accurate to within ± E* of the
population mean value”. ± E* defines the error margin in Figure 2.1.

2.5.1 Estimation of sample size when σ is known (n >30)

In order to match the accuracy between the sample mean and the population mean
by, ± E* the required sample size is:

k s
n = ( α / 2 R )2 (2.12)
*
E

However, if the sample size, n becomes less than 30 then a correction needs to be
made based on Student t-distribution. This will be discussed in the next section.

Therefore, to estimate the sample size, n we need to know the confidence level
(Figure 2.1), error margin and the standard deviation. Often we need to estimate the
standard deviation from prior knowledge or from the sample data.

For example, the engineer wants to estimate how many tests need to be done in
order to get an accuracy of E* = ± 28 kN with a 90% confidence level (p=0.9) when

Reliability Based Calibration\Section 3 2-7 SCB2-WG-07, August 2008

 E*
the standard deviation of the test data, sR is known to be 38.5 kN. In this case,
sR
28
= = 0.73. Therefore, the sample size is estimated as:
38.5

k s 1.64 x38.5 2
n = ( α / 2 R )2 = ( ) = 4.98 ≅ 5. This was the sample size given in
*
E 28
Section 2.2.

Estimation of Sample Size (Equation 2.12)

600
550
500
kα =1.28, 90%
450
confidence Level
400
Sample Size,n

350 90%
300 kα =2.33, 99% 95%
250 confidence Level 99%
200
150
100
50
0
0 0.1 0.2 0.3 0.4 0.5
(Error/Standard Deviation)

E*
Figure 2.4 - Sample size versus ( ) for various confidence levels
sR

However, if the engineer decreases the accuracy of the prediction by increasing the
E* value to E * = ± 40kN, the estimated sample size becomes:

k s 1.645 x38 2
n = ( α / 2 R )2 = ( ) = 2.44 ≅ 3
*
E 40

Therefore, the sample size, n starts decreasing as we increase the error margin.

2.5.2 Estimation of sample size when σ is unknown (n < 30)

Reliability Based Calibration\Section 3 2-8 SCB2-WG-07, August 2008

In this case, we need to replace kα / 2 in equation (2.12) by tα / 2,ν (Table 2.2) and
the sample size becomes:

tα / 2,ν s R 2
n=( ) (2.13)
E*
The problem arises in solving this equation (2.13) for n because tα / 2,ν is not known
a-priori; since tα / 2,ν is dependent on sample size through ν because ν = n-1.
However, a trial and error procedure can be used in solving equation (2.13).

Let us assume that tα / 2,ν initially is equal to kα / 2 which is 1.645 for 90%
confidence level. Equation (2.13) provides a value of n=5 which gives tα / 2,ν = 2.13
from Table 2.2. Using this value in equation (2.12), we find n=8.57 ≅ 9. Therefore,
the revised tα / 2,ν = 1.86 and we obtain n=8.54 in the second iteration. Therefore
we need at least n=9 tests to maintain the same level of error margin of ± 28 kN.

Figure 2.4 presents this when the standard deviation is known (n > 30) and the
sample size estimate is based on equation (2.12). The foundation engineer can also
estimate sample size for (n < 30) following the iterative scheme presented before.

2.6 Reliability Based Design Methodologies

2.6.1 Allowable stress design (ASD)

Current practices for design of overhead line support foundations are primarily based
on deterministic assumptions where the load effects on the foundation and the
foundation capacity are both defined in terms of single values. The load effects and
the capacity could be either defined in terms of the geotechnical design capacity or
the ultimate capacity with appropriate factors of safety. In simple form, this is
expressed as:

Rn
≥ ( D n +WI n ) (2.14)
FS

or

ψ *Rn ≥ ( Dn + WI n ) (2.15)

where:

Rn = geotechnical design capacity,

FS = factor of safety,
Dn = the effects of dead load such as tower weight, conductor weight etc;
WI n = the effects of wind and ice loads on conductor and tower and
ψ * = resistance factor =1/FS.

Reliability Based Calibration\Section 3 2-9 SCB2-WG-07, August 2008

If ice is not present WI n can be replaced by wind load effect only as, Wn and Rn
could be calculated based on semi empirical formulae or numerical models.

2.6.2 IEC 60826 (2003)

IEC 60826 (2003) provides a basic framework where the semi probabilistic design
equation is given in terms of the climatic load acting on a component and the
strength of the component. The basic design equation relates the characteristic
strength (capacity) value to the appropriate load effects having a specified return
period of T years. In simple form, this can be expressed as:

RC ≥ QT (2.16)

where RC is the characteristic strength (capacity) of the foundation with e%

exclusion limit and QT is the load effect with a T-year return period.

The design equation is based on the assumption that the distribution of load is
Gumbel type 1 and the strength follows a normal distribution. The design equation is
valid with a constant annual failure probability between (1/T to 1/2T) provided the
coefficient of variation of strength is 0.05 ≤ vR ≤ 0.20 and of the load effect is
0.20 ≤ vQ ≤ 0.50 respectively.

Since IEC 60826 recommends the use of annual failure probability of (1/T to 1/2T), it
is important that the coefficient of variation of strength be kept under 0.3 to meet the
above criteria. For foundation, this may not be always possible. The capacity is
normally based on the available test data with a typical exclusion limit of 10% with a
50 year return period normally assumed for the load. IEC 60826 also recommends
larger return periods for higher reliability levels.

In designing individual line components based on a system approach, IEC 60826

also recommends to apply a global strength factor, φR in equation (2.16)

φR RC ≥ QT (2.17)

where φR is a global strength factor which depends on strength coordination φ S ,

number of components subjected to maximum load intensity φ N and quality of
installation φQ ; it also includes the characteristic strength factor φC which takes into
account the variation between the characteristic capacity for an exclusion limit of e%
and the capacity RC with an exclusion limit of 10%

Within the framework of IEC 60826, the characteristic capacity should be related to
the model predicted geotechnical design capacity and this could be done through
calibration. If the full-scale test data are available, the appropriate foundation strength
factor φF can be used to relate the geotechnical design capacity to the full-scale test
capacity with a defined exclusion limit (5% for ASCE, 10% for IEC 60826).

Reliability Based Calibration\Section 3 2-10 SCB2-WG-07, August 2008

A multiplicative form of φF has been presented by Buckley (1994). The characteristic
capacity, RC in equation (2.16) for a single foundation can then be modified as:

φF Rn = QT (2.18)

where φF is a function of the statistical properties of the test and the geotechnical
design capacities. The details of computing the calibration factor, φF based on m -
model and linear regression model will be discussed in sections 4.0 and 5.0
respectively.

2.6.3 Load and Resistance Factor Design (ASCE- LRFD)

Although the ASD approach has served the utility industry reasonably well,
increasingly utilities’ design practices are moving towards a reliability based design
methodology, in order to quantify the foundation reliability in a predictable manner. A
simple reliability based design format is the use of load and resistance factor design
(LRFD) approach where the geotechnical design capacity and the load effects are
factored appropriately to account for a specified target reliability index β . The
reliability index β provides a measure of reliability and is directly related to the
failure probability which can be determined from normal tables (Melchers, 1999).

The LRFD format can be expressed as follows;

ϕ Rn ≥ γ D D n +γ WI WI n (2.19)

where

ϕ = the capacity reduction factor or resistance factor and is always less than
1.0,
γ D = load factor for dead load Dn ,
γ WI = load factor for the combined wind and ice loads WI n .

If ice is not present, γ WI can be replaced as γ W , to represent the load factor for wind
load effect, Wn . These factors are calibrated with respect to existing design
practices, which are often deterministic in nature or against a design format
prescribed, by the standard or code. For a simple log normal distribution of load
effects and strength values, these factors can be computed analytically (Melchers,
1999).

2.7 Summary

This section provided a basis for determining (1) the confidence interval on mean
value and (2) the prediction interval on a future test value. The lower limit of the
prediction interval can be used to determine the guaranteed minimum foundation
capacity, with a specified confidence level; while, the characteristic capacity is
defined as the low side of the mean foundation capacity with a specified exclusion
limit. For a large sample size, both will provide almost an identical value. The
characteristic capacity is applicable to single variable data (test capacity only); while,
the concept of prediction interval based on a linear regression model can also be
applied to the problem where more than one variable is involved, (such as the test

Reliability Based Calibration\Section 3 2-11 SCB2-WG-07, August 2008

capacity versus geotechnical design capacity). Since the method here is applied to a
set of test data only, it will be shown in section 5.0 that the same methodology can be
used also, to provide the confidence interval for the estimated mean test capacity as
well as the prediction interval on a new test capacity, for a given geotechnical design
capacity value. It will be also shown in section 5.0 that the foundation strength factor,
φF can be related directly to the lower limit on the prediction interval.

Reliability Based Calibration\Section 3 2-12 SCB2-WG-07, August 2008

3.0 FOUNDATION CAPACITY ASSESSMENT

3.1 General

This section of the report presents a brief review of various design models in
determining the geotechnical design capacity of a pad and chimney foundation under
uplift condition. The soil strength parameters are identified for capacity calculation
followed by the field data collection programme. A short description of the field test
programme is presented which outlines the various steps that need to be followed in
collecting the test data as well as the data interpretation and analysis using the
methodology described in IEC 61773.

The fundamental premise of the reliability based design is to determine the

uncertainty in the design process and to quantify this appropriately. To account for
this uncertainty in assessing both the test capacity as well as geotechnical design
capacity, the designer needs to test a number of foundations in various soil types
(strength parameters), to ensure that the data represents a wide range of foundation
conditions along the transmission line route. Figure 3.1 depicts a typical line route,
which covers a wide range of soil conditions usually encountered in a project.

The transmission line engineer can do these tests in three different ways. The
engineer can carry out a number of tests with a foundation of fixed dimensions, in
homogeneous soils, with fixed installation technique to ensure that all tests will
produce similar results in soils of similar categories. This would provide a good
coefficient of variation of a specific foundation type and can be compared with the
coefficient of variations of other line components such as the support, conductor, and
insulator to coordinate the strength as per IEC 60826.

Alternatively, the engineer can vary the dimensions and carry out these tests in
similar soils to assess the effect of the geometry and can estimate the coefficient of
variation of the capacity.

Finally, the engineer can test a typical foundation of fixed dimensions, in various soil
types with repeated tests, thereby obtaining a wide range of coefficient of variations,
which will be more representative of a typical transmission line route. Of course, the
last two approaches will cost the utility more compared to the test carried out in one
soil type.

The overall foundation capacity is primarily dependent on two basic parameters.

These are (1) geometrical parameters which define the foundation dimensions and
(2) soil strength parameters such as shear strength, cohesion, density, overburden
pressure, effective friction angle, ground water table, etc. Although the geometrical
parameters can be defined deterministically for a specific foundation type, there is
normally a considerable amount of scatter in soil strength data. Therefore, the overall
variation of the foundation capacity is a direct function of the variation of soil strength
parameters.

If we assume that the engineer has designed a typical foundation and wants to
determine the actual or characteristic capacity with certain confidence limits, then it is
necessary to test the foundation for a wide range of soil conditions to provide a good
level of confidence in the predicted capacity. In testing a specific foundation type, the
engineer needs to ask a few questions and these are listed as follows.

Reliability Based Calibration\Section 3 3-1 SCB2-WG-07, August 2008

 Figure 3.1 - Typical line route showing the test sites (Topographic Map)

Reliability Based Calibration\Section 3 3-2 SCB2-WG-07, August 2008

 • Is there any data available with respect to this foundation type?
• What design model(s) do I use?
• Is there a standard design model that I can use?
• What subsurface information do I need to collect?
• What sample size do I need to get a reasonable prediction?
• Do I need to collect additional data?
• What criteria should I use for capacity?

Assuming that the specific foundation type is an undercut pad and chimney
foundation, and the foundation is to be tested under uplift load, the first thing the
designer needs to ask is “what design model is appropriate for predicting the
geotechnical design capacity of this type of foundation”. Figure 3.2 depicts a typical
undercut pad and chimney foundation.

An undercut foundation is where the base of the foundation is slightly wider than the
width at the neck level. The foundation is excavated by machine to the width ‘C’ in
figure 3.2. The bottom of the excavation is widened to the width ‘B’ in figure 3.2 using
a handheld shovel. The undercut is usually 0.1 m on each side of the foundation.
Table 3.1 presents the dimension of a typical pad and chimney foundation.

F G.L.

C
Soil
E
A
D

B Undercut

Figure 3.2 - Typical undercut pad and chimney foundation

Table 3.1- Typical dimensions (m) of an undercut pad and chimney

foundation

Type A B C D E F
Square 2.2 2.0 1.8 1.0 0.8 0.1
Based

Reliability Based Calibration\Section 3 3-3 SCB2-WG-07, August 2008

CIGRE (2002) presents a review of a number of theoretical design models that can
be used to determine the geotechnical design capacity for this type of foundation.
Two models are discussed here. These are (1) frustum angle model (inverted
truncated cone) in Section 3.2.1 and (2) shear model in Section 3.2.2.

3.2 Uplift Failure Mode

The failure mode will often depend on the type of the installation used. For example,
if the foundation excavation is done in such a way that the excavated hole is sized
sufficiently to place the foundation vertically or cast in place vertically with the
surrounding space filled by compacted backfill, then the failure zone would most
likely be covered by both native as well as the backfill soils. Figure 3.3 depicts a
situation where the failure zone is in the native soils. However if the excavation is
“open pit type” and is sufficiently large then it is possible to have the failure only
within the backfill zone (Figure 3.4). In this case, the backfill soil properties would
determine the geotechnical design capacity. This is more appropriate for sites with
boulder, cobbles etc. where often the excavation cannot be done narrowly or where
the walls of the excavation cannot stand up sufficiently safely.

3.2.1 Frustum model (inverted truncated cone)

In this particular model, it is assumed that an inverted frustum starts at the base of
the undercut foundation and slopes at an angle, ψ to the vertical to the surface (refer
to figures 3.3 and 3.4) and the weight of the soils within this frustum and the
foundation weight itself resist the uplift force. Mathematically, this is described as
follows:

U = VS ρ S + VC ρC (3.1)

where VS and VC are the volume of the soils and the concrete within the frustum
while ρ S and ρC are the unit weight of the soils and the concrete respectively. U is
the uplift force.

Backfill
G.L.

Native
W soil Soil

Frustum
Angle, Assumed
Failure
Surface

Figure 3.3 - Failure in native soils

Reliability Based Calibration\Section 3 3-4 SCB2-WG-07, August 2008

IEEE (1985) provides information on the frustum angle, ψ based on soil type,
overburden pressure, construction method and the groundwater condition. The
submerged weight of the soil should be used where the portion of the foundation is
below the ground water line.

Backfill G.L.

Wsoil
Native
Soil

Frustum Assumed
WF Failure Surface
Angle,
Within Backfill

Figure 3.4 - Failure in backfill soils

3.2.2 Shear model

In this model, it is assumed that the uplift force is resisted by shear force acting along
the vertical shear surface circumscribing the base of the foundation, the weight of the
concrete foundation, WC and the weight of the soil above the concrete base within
the frustum, W S . The uplift force is determined as:

U = Sτ + WS +WC (3.2)

where S is the shear surface area, τ is the shear strength.

Although a constant shear strength value of 19.6 kN/m2 was used by Killer (1953),
Buckley (1994) recommended varying shear strength values with depth to ensure
that the predicted ultimate capacity is more realistic and comparable with the test
capacity. Mors (1965) presented an empirical expression for the frictional resistance
(shear capacity):

F0 = SτAη (3.3)

where η is a constant (1.5 to 2.0), A is the depth of embedment, S is the shear

surface area and τ is the shear strength which should be adjusted to 50% of the
value if there is groundwater. IEEE (1985) also provides the frictional resistance
value for a square foundation :

Reliability Based Calibration\Section 3 3-5 SCB2-WG-07, August 2008

 F0 = 4 cBA + 2K ρ S B A2 tan φ (3.4)
where c = cohesion, K= coefficient of lateral earth pressure and φ = angle of
internal friction. B and A are the width and depth of the foundation respectively.

FO WC WS FO
A

Figure 3.5 - Shear model (IEEE, 1985)

CIGRE (2002) presents a comprehensive summary of various methods available in

the literature for determining the uplift capacity of a spread footing.

3.3 Determination of Strength Parameters

The strength parameters required for an undercut pad and chimney foundation under
uplift load are :( 1) soil density, ( ρ S ), (2) concrete unit weight, ( ρC ), and (3) shear
strength, ( τ ), along the depth depending on the model used.

Once the number of sites along the line route (Figure 3.1) has been chosen, soil
densities can be obtained by using a sand cone test in the test pit(s) near the
installation site. Alternatively, a standard penetration test (SPT) can be carried out in
undisturbed soils, to measure the penetration resistance in terms of blow counts (N-
value); which can be related to soil characteristics and strength parameters. The
sand cone test can be done at various depths for easy profiling; while, the SPT test
provides a boring log which can be further interpreted for soil density and angle of
internal friction (Brink et al, 1982). The shear strength values, τ can be estimated
once the soil classification and the angle of internal friction, φ has been determined.
For more detail evaluation, sample needs to be tested in laboratory under triaxial
load condition. If the excavated hole is entirely backfilled, the soil properties of the
backfill should be obtained from historical data or specific test in the backfill soil.

3.4 Example Problem (Calculation of Geotechnical Design Capacity)

Reliability Based Calibration\Section 3 3-6 SCB2-WG-07, August 2008

Fig. 3.6 depicts a subsurface profile for a specific test site for a pad and chimney
foundation shown in Figure 3.2. The N-values from SPT test for native soils in two
layers are shown as 20 and 30, respectively; it is also assumed that the native soils
will be used as backfill materials with the compaction level better than medium dense
soils (N = 30).

Backfill
N=30
G.L.

Native
Soil
Layer 1, N=20

Native
Soil

Layer 2, N=30

Figure 3.6 - Typical subsurface profile of a specific test site

Based on the above information, the following properties (IEEE, 1985) are chosen in
figure 3.7 for the native soils and the backfill soils.

Backfill
(N=30)
ρS = 17.2 kN/m 3 G.L.

Native
(N=20)
Soil
Assumed Layer 1,
Failure ρS = 16.7 kN/m3
Surface
Native
Frustum Angle Soil (N=30)
= 25°
Layer 2,
ρS = 17.2 kN/m3

Figure 3.7 - Subsurface soil profiles with assumed failure surface

Reliability Based Calibration\Section 3 3-7 SCB2-WG-07, August 2008

Since the backfill soil is well compacted granular materials and assuming that the site
is well drained, we select the frustum angle, ψ = 250 (IEEE, 1985). Fig. 3.7 also
depicts the failure diagram based on the frustum model with appropriate unit weights
labeled. In the calculation, soil density was assumed as ρ s =16.7kN/ m3 (minimum
value in figure 3.7) while the concrete unit weight is assumed as ρC = 22.6 kN/ m3.
The calculated uplift capacity is based on equation (3.1) where VS and VC are
computed based on equations (3.5) and (3.9) respectively. The input data are taken
from Table 3.1 and figure 3.2 for a square base foundation.

VS = VS , F - VC , B (3.5)

where

A 2
VS , F = [G + B 2 + ( B * G )] (3.6)
3

G= B+ 2 (A*tan ψ ) (3.7)

and

D 2
VC , B = [ B + C 2 + ( B * C )] + (( A − D) * ( E 2 )) (3.8)
3

where VC , B represents the volume of concrete below the ground line and VS , F is
the volume of the frustum. The total concrete volume is given by equation (3.9).

D 2
VC = [ B + C 2 + ( B * C )] + (( A − D) * ( E 2 )) + ( E 2 * F ) (3.9)
3

The above formulae are approximate and are based on a square base foundation. If
the base is rectangular, the appropriate adjustment should be made to equations
(3.6) to (3.9). The calculated geotechnical design capacity for the foundation type
shown in figure 3.2 is 375 kN. A sensitivity study was done by varying the failure
angle and the result is presented in Table 3.2.

Table 3.2 - Foundation geotechnical design capacity (kN) under various

frustum angles (ψ ) assumptions

Frustum angle 15-degree 20-degree 25-degree 30-degree

Geotechnical 277 322 375 439
design capacity
(kN)

3.5 Full-scale Test - Field Program

The field testing program normally consists of the following steps

• Selection of sample size

Reliability Based Calibration\Section 3 3-8 SCB2-WG-07, August 2008

 • Selection of the test sites
• Subsurface investigation
• Foundation installation
• Instrumentation and data collection
• Loading scheme
• Loading sequence
• Data reduction
• Test results

Since this report describes the uplift test for a pad and chimney type foundation we
will describe the steps required to carry out this type of test only. Other foundation
tests such as drilled shaft under moment may require additional information and can
be obtained from the test report (EPRI 2197, 1982). However, the overall steps would
be very similar to those described here.

3.5.1 Selection of sample size

Section 2.5 has provided the required background information on how to estimate the
sample size for cases where the standard deviation of foundation test capacity data
is either known or unknown. The user can use this information as a guide to
estimate the sample size for specified confidence level on the estimated mean test
capacity value of foundation test capacity data (figure 2.4). For sample size less than
30, a correction needs to be made based on Student–t distribution.

3.5.2 Selection of test sites

As mentioned earlier, the user needs to know whether the tests will be done at one
specific site or at several sites along the transmission line route (Figure 3.1). The
key parameter for selecting a test site could be the access as well as its typical soil
characteristics and distributions with respect to line routing.

3.5.3 Subsurface investigation

Depending on the testing program, the subsurface investigation could vary between
simple to very complex. For example, the uplift test program requires the
determination of a number of soil strength parameters. These are: soil types
(cohesive versus. cohesionless), relative densities, unit weight and the c − φ
properties. This information can be obtained from a sand cone test, SPT test or any
other suitable test program (laboratory or in-situ).

3.5.4 Foundation installation

The foundation is excavated by machine to the width ‘C’ in Figure 3.2. The bottom of
the excavation is widened to the width ‘B’ using a handheld shovel. Once the
foundation is in place, the surrounding hole is backfilled with well compacted granular
materials. The strength parameters of the compacted backfill soil at each layer
(300mm to 450mm) can be determined from a nuclear densiometer test, sand cone
test, etc. The detailed installation procedure for various foundation types can be
found in CIGRE (2006) technical brochure.

3.5.5 Loading scheme and sequence

Figure 3.8 presents a loading test arrangement of a grillage foundation under inclined
uplift load. The test was arranged to determine the anchor capacity of a grillage

Reliability Based Calibration\Section 3 3-9 SCB2-WG-07, August 2008

foundation in cohesionless soils for a 230 kV suspension guyed V-tower. The
excavation was an “open pit type” and the capacity was primarily based on the
backfill soil properties (Prasad and Haldar, 2001).

Figure 3.8 - Grillage foundation under uplift load (Prasad and Haldar, 2001)

The loading sequence can be simply one load cycle where the load is applied until
the foundation fails or it could be cyclic where it follows a loading-unloading
sequence. The rate of loading can also be varied. In this study, a cyclic loading was
used.

3.5.6 Instrumentation and data collection

For typical uplift test, the user needs to record the load and displacement of the
foundation. The load can be measured either using a dynamometer or electronically
through a load cell. Similarly, the displacement can be measured using a survey
transit, dial gages or using LVDT. The engineer can use any one of these or all
depending on the accuracy and the verification desired in the measurement process.

3.5.7 Data reduction

This will include taking out any offsets or any other transformations required in the
load and displacement measurements.

3.5.8 Test results

Once the data has been reduced, the final test result can be plotted in terms of load
versus displacement as shown in Figure 3.9. In the next section, the user will be
presented with the methodology to determine the ultimate capacity of the foundation.
The capacity is performance based and the method follows IEC 61773.

Reliability Based Calibration\Section 3 3-10 SCB2-WG-07, August 2008

 Typical load versus displacement data from an
uplift test.
600
500
Load (kN)

400
300
200
100
0
0 5 10 15 20 25 30 35

Displacement (mm)

Figure 3.9 - Typical load displacement plot for a pad and chimney foundation

3.6 Interpretation of Test Data

The performance criterion of a foundation is normally based on three limit capacities.

These are: (1) elastic limit capacity, (2) working (damage) limit capacity and (3)
ultimate limit capacity. The ultimate limit capacity of a foundation is defined when it
has reached its full plastic limit, such that little or no additional load is sufficient to
produce considerable deflection (Davidson, 1981, EPRI EL 2197). However, the
ultimate limit capacity may not be realized in many full-scale tests because of the
various constraints imposed by the test set-up and the in-situ soil conditions.
Therefore, some practical methods are required to determine the ultimate capacity.

For a given load displacement data set, the utility engineer can determine the elastic
limit capacity and the ultimate limit capacity following the procedures described in the
IEC 61773 entitled “Testing of Structure Foundations for Overhead Lines”. However,
the definition of working limit (damage limit) capacity is more subjective.

Buckley (1994) carried out a number of foundation tests in various soil conditions in
Ireland. The elastic limit load was defined as the load when the foundation movement
upon unloading was insignificant. The working limit load was defined as the load
associated with a 10 mm foundation movement. Details on the selection of 10 mm
displacement are given in section 4.4.2. This is similar to “Damage load” defined in
IEC 60826. The ultimate load was defined as the final pull out of the foundation.
Since one can get a large movement under ultimate condition, it was reported that
this load is not practical to use because the tower may not be able to sustain such a
large movement. Buckley (1994) has recommended a 10 mm displacement as the
determination of the working limit capacity for spread footing. For directly embedded
single pole structure, the capacity associated with a 20 rotation can be used to define
the damage limit (IEC 60826, 2003).

To illustrate the various methodologies for determining the test capacities, an

example is presented. Figure 3.9 depicts the typical load displacement curve
obtained from a full-scale test carried out in Ireland (ESB report). The elastic limit
capacity at 2.5 mm displacement is 253 kN while the working limit capacity at 10 mm

Reliability Based Calibration\Section 3 3-11 SCB2-WG-07, August 2008

displacement is 388 kN. However, the ultimate limit capacities are determined based
on the various methodologies described in the IEC 61773 document, which can vary
considerably depending on the specific method used.

The same is also true if the user chooses to use different displacement criteria for the
working limit capacity or damage limit as per IEC 60826 (2003). IEC Standard
provides various methods to determine the limit capacities. Table 3.3 presents the
comparison of ultimate capacities based on various methods described in IEC 61773.

Table 3.3 - Comparison of ultimate capacity of an undercut foundation

(IEC 61773)

Slope
Tangent
90% Tangent
Methods Intersection Parabolic Hyperbolic
Criteria (ST)
(TI)
4 10
mm mm
Estimated
Ultimate
323 542 620 625 362 450
Capacity
(kN)

In a study conducted by EPRI (EPRI EL 2870), a 4 mm displacement was used in

determining the ultimate capacity. The estimated mean test capacity based on a 4
mm displacement is 362kN (figure 3.9) while that using a 10 mm displacement is
450kN.

In the above sections, a specific example (figure 3.9) is presented to determine the
various test capacities of an undercut pad and chimney foundation tested by ESB in
Ireland. However, depending on the method used, the estimated ultimate capacity of
this foundation can vary considerably; it could range from 323kN to 625kN. The
mean value is 487kN with a coefficient of variation of 26.5%.

The calculated geotechnical design capacity can also vary significantly depending on
the specific failure model used. As stated earlier, the overall variation of the
foundation capacity is a direct function of the variation of soil strength parameters.
The uncertainty in the model predicted geotechnical design capacity, Rn can be
accounted theoretically only through the joint density function of all these parameters
in a probabilistic sense. However, in practice this is not possible because the joint
density function of all these parameters is often complex and is unknown. Obviously,
Test Capacity
the designer can obtain different ratios of m =
Geotechnical Design Capacity
depending on what methods are used in determining these limit capacities.

Reliability Based Calibration\Section 3 3-12 SCB2-WG-07, August 2008

3.7 Test Data Collection-Field Programme, an Example

Let us assume that the design engineer wants to test a pad and chimney foundation
(figure 3.2) to determine the characteristic capacity with a defined exclusion limit. The
engineer has a number of choices. The engineer can test the foundation either in one
specific site (Site A in Figure 3.1), or can arrange a testing programme along the line
route at different site locations. The test can be repeated at each site to obtain a
dispersion of the test capacity (Sites A, B and C in Figure 3.1), which can then be
used in determining the confidence level. Also, by testing at various locations along
the line route, the engineer will obtain the limit capacities, which are more
representative of the entire line route.

3.7.1 Site specific repeat test data for single geotechnical design capacity
( Rn )

The geotechnical characteristics of SITE A are obtained by SPT tests and Table 3.4
presents the data. The dimensions of this test foundation are presented in Table 3.1.
Using equations (3.1) and (3.5 to 3.8), the calculated geotechnical design capacity
based on a ψ =300 frustum angle (Section 3.2.1) is 439kN. To calculate this
capacity, the minimum value of the soil density used is 16.7kN/m3 (figure 3.7). The
concrete unit weight, ρC is assumed to be 22.6kN/m3. The engineer performed three
repeated tests at this site (Site A in figure 3.1) and obtained working limit capacities
at 10 mm displacement as 500kN, 540kN and 600kN respectively. These values can
be obtained from actual tests similar to the plot presented in Figure 3.9. Figure 3.10
depicts the plot, which shows the scatter in the test capacity with respect to one fixed
geotechnical design capacity value.

Table 3.4 - Geotechnical properties at Site A (Figure 3.1, IEEE, 1985)

Relative Angle of
N-Values Internal Unit Weight
Density Density, Dr in
(SPT) Friction, φ (kN/m3)
Percent

Medium 10-30 35-65 350 - 400 16.7-19.6

Since the test capacities are representative for one single geotechnical design
capacity value, there is no correlation between the test capacity and the geotechnical
design capacity. A correlation defines the degree of measure of the linear
relationship between the test capacity y and the geotechnical design capacity x. In
this case, it relates the linearity between the test capacity and the geotechnical
design capacity. In statistical data analysis, the degree of predictability will often
depend on the quality of the dataset (mutual dependence between test capacity and
geotechnical design capacity) and this is often measured by the statistical correlation
coefficient r. This will be discussed in detail later in sections 4.0 and 5.0.

Reliability Based Calibration\Section 3 3-13 SCB2-WG-07, August 2008

 Repeated Test Data at One Specific Site
620

600

580
Test Capacity (kN)

560

540

520

500

480
0 100 200 300 400 500

Geotechnical Design Capacity (kN)

Figure 3.10 - Repeated test data for an undercut pad and chimney foundation-
project specific

3.7.2 Site specific test data along the line route

Alternatively, the engineer decides to test this foundation type at three different
locations along the line route considering that the capacities will be more
representative of the soil conditions along the line route depicted, in Figure 3.1.
These sites are marked in figure 3.1 as Site A, Site B and Site C; respectively. Based
on SPT test data, Table 3.5 provides the soil characteristics and the estimated soil
densities.

Table 3.5 - Geotechnical properties at three sites (Figure 3.1-IEEE, 1985)

Relative Angle of Unit

Site N-Values Internal
Density Density, Dr Weight
Locations (SPT) Friction, φ
in Percent (kN/m3)
Site A Loose 4-10 15-35 300 - 350 13.7-16.7
Site B Medium 10-30 35-65 350 - 400 16.7-19.7
Site C Dense 30-50 65-85 400 - 430 19.7-21.6

The model foundation shown in Figure 3.2 is again used to calculate the geotechnical
design capacity for the three different soil types. Using the frustum angle model, the
calculated geotechnical design capacities are 379kN, 439kN and 498kN respectively
for loose, medium and dense soil conditions. The damage capacities at 10 mm
displacement are assumed to be 450kN, 540kN and 680kN respectively. Figure 3.11

Reliability Based Calibration\Section 3 3-14 SCB2-WG-07, August 2008

depicts the plot for these tests. Table 3.6 presents the data with mi values calculated
for these three data points where

Test Capacity
mi = (3.10)
Geotechnical Design Capacity

Scattered Test Data along a Route

700

600
Test Capacity (kN)

500

400

300
450 degree
200 line

100

0
0 100 200 300 400 500 600

Geotechnical Design Capacity (kN)

Figure 3.11 - Uplift test -scattered test data along the line route (damage limit
capacity at 10mm displacement

Table 3.6- Test data versus geotechnical design capacities

Foundation Capacity (kN) Site Site Site

A B C
Geotechnical design capacity 379 439 498
Test Capacity (at 10mm displacement) 450 540 680
Test Capacity 1.19 1.23 1.37
mi =
Geotechnical Design Capacity

3.8 Summary

This section presented a review of the various steps involved in the assessment of
foundation capacities from test data as well as from theoretical model predicted
design capacities. The latter is defined as geotechnical design capacity value. This
includes: (a) determination of the geotechnical design capacity based on specific
model(s), (b) field test programme to determine the test capacity and associated
strength parameters for the model, (c) sample size requirement for the test
programme, and (d) testing programme and interpretation of data according to IEC

Reliability Based Calibration\Section 3 3-15 SCB2-WG-07, August 2008

61773. Finally two typical databases, which the designer may encounter, are
described. In the first case, three repeated tests are done at one specific site for a
single geotechnical design capacity value while in the second case the tests are
carried out at three different sites along the transmission line route (scattered test
data). The next two sections will outline the calibration methodology and the
appendix will use the scattered test data to demonstrate the methodology of
computing the foundation strength factor based on a linear regression model.

Reliability Based Calibration\Section 3 3-16 SCB2-WG-07, August 2008

4.0 CALIBRATION OF FOUNDATION STRENGTH FACTOR
( m - MODEL OF BUCKLEY)

4.1 General

This section of the report presents a framework to calibrate the foundation strength
factor based on a limited number of full-scale test data. The methodology is based
on a simple mathematical model where the statistical properties of mi values (Table
3.6) are used following Buckley (1994). Later, various assumptions of this model are
examined and a linear regression model is proposed for general application. The
correlation effect is discussed and its importance in the data analysis is emphasized.

The uncertainty in the prediction is primarily introduced through two factors: (i) model
error and (ii) data error due to finite sample size. However, to develop a general
methodology for the calibration of the foundation test data against the model
predicted geotechnical design capacity (geotechnical foundation design model), the
foundation designer needs to assess the quality of the dataset (correlation) and its
influence on the predicted test capacity. The predicted test capacity is analogous to
the characteristic capacity described in section 2.4.

In statistical data analysis, the degree of predictability will often depend on the quality
of the dataset (mutual dependence between test capacity and geotechnical design
capacity) and this is often measured by the statistical correlation coefficient, r.

The correlation coefficient of the dataset is determined from equation (4.1):

∑ ( xi − x )( yi − y )
r= (4.1)
(n − 1) s x s y

where s x , s y are the standard deviations of the geotechnical design capacity values
xi and the test capacity values yi ; x and y are the mean values of the geotechnical
design and test capacities respectively. The last two symbols are also analogous to
Rn (mean geotechnical design capacity value) and R (mean test capacity value).
The variables x (or y ) and s x (or s y ) are computed from equation (2.1) and
(2.2) replacing Ri and R by xi (or yi ) and x (or y ) respectively. For convenience,
x and y will represent geotechnical design capacity and test capacity values. For
example, figures 4.1 (a) and (b) present two datasets with very different correlation
properties.

Two scenarios are considered. In the first scenario, it is assumed that the tests have
been carried out only at specific site with one type of foundation (repeated test data)
while in the second scenario the data is available for multiple test sites along the
transmission line route. This is identified as the scattered test data (refer to section
3.7).

4.2 m - Model – No Correlation (Repeated Test Data)

The m - model of Buckley is based on the ratio of the test capacity and the
geotechnical design capacity.

Reliability Based Calibration\Section 5 4-1 SCB2 WG-07, August 2008

 Test Capacity R y
mi = = i = i (4.2)
Geotechnical Design Capacity Rn xi

n
y
∑ ( xi )
i
m= 1 (4.3)
n

The standard deviation and the coefficient of variation of the mi values are
computed from equations (4.4) and (4.5)

1 n
2
sm = ∑
n −1 1
(mi − m ) 2 (4.4)

s
vm = m (4.5)
m

Since xi = x , for the repeated test data presented in Section 3.7.1, the correlation
coefficient is zero ( r = 0.0 ) from equation (4.1) indicating that the data is unique.
Therefore, the exclusion limit for the repeated test data is computed from the test
data only. In this case, the three data points are 500kN, 540kN and 600kN
respectively (figure 3.10). The geotechnical design capacity for all three tests is
439kN (reference to section 3.7.1).

Using equations (2.1-2.3), the mean test capacity R = 556 kN and the coefficient of
sR 115.9
variation, vR = = = 0.20
R 556

Therefore, the characteristic capacity with a 10% exclusion limit (90% confidence
level, one sided), is obtained from equation (2.11):

RC ,10% = R (1- k α vR )
Test Capacity, y

Data Banded
Around The
Line With
High s xy

Geotechnical Design Capacity, x

Figure 4.1 (a) - Low Correlation

Reliability Based Calibration\Section 5 4-2 SCB2 WG-07, August 2008

 Test Capacity, y
Data Banded
Around The
Line With
Low sxy

Geotechnical Design Capacity, x

Figure 4.1 (b) - High Correlation

According to equation (2.18), the foundation strength factor, φF is defined here as

the ratio between the characteristic capacity, RC and the geotechnical design
capacity, Rn . Based on the repeated test data, R = 556kN, Rn = 439 kN and kα is
1.28 from Table 2.1. Therefore, RC ,10% = 0.744 R , where R = 1.266 Rn and the
foundation strength factor, φF = 0.744 x 1.266 = 0.942. This foundation strength
factor value can be used directly in equation (2.18) following IEC 60826 (2003).

For a finite sample size n = 3, kα in equation (2.11) should be replaced by tα ,ν

and

RC ,10% = R (1- tα ,ν vR ) (4.6)

where tα ,ν = 1.90 from Table 2.2. Therefore, φF = 0.785 considering sample size
effect; correspondingly the foundation strength factor is decreased by 20% when the
sample size effect is considered.

Since the data for the repeated tests (reference to figure 3.10) are only valid for one
specific test site and for one size of foundation, it has no correlation with respect to
other test site locations along the line route unless the foundation designer can be
sure of another location where the subsurface information is similar to the test
location and the foundation has a similar aspect ratio (depth/width). For undercut
foundation with rectangular base, an equivalent diameter D0 may be considered as
alternative for the width.

4.3 m - Model-With Correlation (Scattered Test Data)

In the second case when the data is scattered and the tests are undertaken at three
different locations along the transmission line route (sites A, B and C in figure 3.1),
the foundation designer can expect that the data from various test sites will have a
degree of correlation. However, the degree of correlation will depend on how well the
geotechnical design model can predict the foundation capacity at the various test

Reliability Based Calibration\Section 5 4-3 SCB2 WG-07, August 2008

sites. Using equation (4.1), the statistical correlation coefficient is r = 0.90 for the
data presented in Table 3.6.

Based on figure 3.11 and the mi data presented in table 3.6, it is obvious that the
geotechnical design capacity based on a 300 frustum angle model is under-predicting
the test capacity (working limit at 10 mm), since all three geotechnical design
capacity values are below the 450 line passing through the origin. The mean, m
(slope) and the coefficient of variation, vm of these tests are obtained from equations
(4.2 – 4.5) and are 1.26 and 0.085 respectively. Figure 4.2 presents the average
slope, m . This means that on average, the model predicted geotechnical design
capacity underestimates the test capacity by 26%.

The foundation strength factor, φF is obtained from equation (4.3)

Rc
φF = = m (1- kα * vm ) (4.7)
Rn

The above equation was presented by (Buckley, 1994, m -model) and is

recommended in CIGRE (2002).

The characteristic capacities for 5% (ASCE) and 10% (IEC 60826) exclusion limits
can be obtained initially first from equation (4.7) neglecting the sample size (CIGRE
2002). The kα values for 5% and 10% exclusion limits (95% and 90% confidence
levels) are obtained from Table 2.2 and these values are 1.645 and 1.282
respectively.

RC , =1.26(1 - 1.645*0.085) Rn = 1.0838 Rn ; φF =1.084

5%

and

RC , =1.26(1 - 1.282*0.085) Rn =1.1226 Rn ; φF = 1.122

10%

For finite sample size, (n less than 30, Section 2.1), kα should be replaced by tα ,ν
in equation (4.7) and

Rc
φF = = m (1- tα ,ν * vm ) (4.8)
Rn
The tα ,ν value is obtained from Table 2.2. Table 4.1 summarizes the foundation
strength factor values with the plot shown in figure 4.2. Correspondingly, the
foundation strength factor values are reduced by 6% and 15% when the sample size
effect is considered (ν = n − 1 = 3 − 1 = 2) .

RC , =1.26(1 - 2.92*0.085) Rn = 0.947 Rn and φF =0.947

5%

and

RC , =1.26(1 - 1.90*0.085) Rn = 1.056 Rn and φF = 1.056

10%

Reliability Based Calibration\Section 5 4-4 SCB2 WG-07, August 2008

 Table 4.1 - φF Values from equations (4.7 & 4.8)

Sample Size φF ,5% φF ,10%

n is large 1.084 1.122
n = 3,ν = 2 0.947 1.056

Lower Limit on Prediction (90% lower confidence level-

10% exclusion limit)
90% lower
confidence level-
700 Average slope, 10% exclusion limit,
Test Capacity (kN)

1.26
600 kα =1.28, sample
size neglected
500
400 90% lower confidence
300 level-10% exclusion
200 limit, tα,ν =1.90,
100 sample size included
0
0 200 400 600
Geotechnical Design Capacity(kN)

Figure 4.2 - Predicted test capacity based on m -model

4.3.1 Discussion

The simplicity of the above model (equation 4.3) is that once the relationship
between the characteristic capacity, RC and the geotechnical design capacity
value, Rn is established through the foundation strength factor, φF , the geotechnical
design capacity is easily determined from the characteristic capacity with a defined
exclusion limit (e %). However, the method is based on the following five key
assumptions.

• The relationship between the foundation test capacity versus model predicted
geotechnical design capacity is derived from a linear regression model.

• The intercept is assumed to be zero, i.e., the regression line passes through
the origin.

• The residual error, ε i (the difference between the estimated mean test
capacity and the actual test capacity), is non constant, i.e. it increases as the

Reliability Based Calibration\Section 5 4-5 SCB2 WG-07, August 2008

 model predicted geotechnical design capacity ( Rn values along the x-axis)
also increases. The determination of residual error is explained in Section
5.1.

• The sample size, n is large and the data follow a normal distribution.

• Correlation of the data is high and the coefficient of variation, vm is small (<
0.10).

A low correlation value, r2 would imply that there is no linear relationship between the
foundation test capacity and the geotechnical design capacity. Table 4.2 presents
some suggested correlation values and the degree of linear association between the
test capacity, y and the model predicted geotechnical design capacity, x.

Table 4.2 – Suggested ranges of correlation value

Linear association
Correlation Value, r 2
between x and y
weak
Low, r 2 ≤ 0.3
moderate
Moderate (0.3 ≤ r 2 ≤ 0.8)
strong
High, r 2 > 0.8

Each of the above assumptions requires to be clearly understood. There are cases
where these assumptions may not apply. For example, the assumption that the line
passes through the origin needs to be validated. The assumption that the residual
errors vary in a non-constant manner also needs to be validated. Without these
validations, the lower limit on the prediction interval of the test capacity (analogous to
characteristic capacity with exclusion limit, section 2.4) for a single geotechnical
design capacity value, Rn and the extrapolation of the data across any geotechnical
design capacity values (data along x-axis) could be problematic and could lead to
erroneous conclusions. This will be discussed in detail in section 5 under “linear
regression model” and in section 6 under “interpolation versus extrapolation”.

4.4 Example Problem- ESB Data (Buckley, 1994)

4.4.1 Background

Electricity Supply Board (ESB), Ireland carried out a total of 68 full-scale foundation
tests during the period 1984 to 1989. Fifty four (54) of these foundations were of the
pad and chimney type and cast, where possible against undisturbed soil and
undercut where feasible. These tests were carried out on foundations newly installed
primarily for research purposes; while the remaining fourteen (14) old foundations
were also pad and chimney type and made available for testing because of the
removal of an existing 40 year old line. These were installed in the disturbed soils (c-
φ ). All of the fifty four (54) foundations tested were installed in various types of soils
and using different installation techniques. Buckley (1994) has reported in detail the
foundation test results and the next paragraph is a direct quote from his recent
communication (Buckley, 2007).

Reliability Based Calibration\Section 5 4-6 SCB2 WG-07, August 2008

“The tests were designed to provide generic test parameters for the vast majority of
tower foundations on the ESB system. The vast majority of tower foundations in ESB
system were set in consolidated glacial till (c- φ ) soils and a substantial minority in
cemented sand and gravel ( φ ) soils. There were also areas overlain, with peat and
under laid with glacial till at shallow (1-2 meters) and occasionally deep (up to 6
meters) depth. The water levels in the glacial till were generally high and the
cohesionless soils were usually dry. The selection of test sites was based on this
spread of ground conditions and on the availability of the sites for extensive and
disruptive foundation setting”.

Except for the existing 14 foundations, all the research foundations were installed in
cohesive and/or cohesionless soils. Nine (9) of these installed foundations were in
granular soils where the foundations were above the ground water table; while 22 of
the research foundations were in soils partly below the ground water table and the
soils were overconsolidated glacial tills (c- φ soils). The soil conditions at the existing
fourteen foundation sites varied from cohesive to silty soils (c- φ ). The majority of
these foundations was pad and chimney type (figure 3.2) and is classified as
undercut foundations.

All tests were completed to determine the ultimate capacity of the foundations under
uplift condition. Uplift loads were applied using high capacity cranes, while the
horizontal (shear) loads were applied via a tractor mounted winch. The typical testing
period was 20 minutes and in some instances loads were cycled. The tests were
carried out until the foundation failed, although in some instances the ultimate
foundation capacity could not be reached due to limitations of the test equipment.

4.4.2 Defining foundation capacities

The ESB study defined the foundation failure capacity in terms of the foundation
movement. These were: (i) elastic limit load, (ii) working limit load and (iii) ultimate
limit load. The elastic limit load was defined as the load when the foundation
movement upon unloading was insignificant. The working limit load was defined as
the load associated with a 10 mm foundation movement (similar to “Damage load”
defined in IEC 60826) and was based on the following criteria:

• The load was normally 60% -70% of the ultimate load determined based on
the other criteria such as the tangent intersection, slope tangent etc (IEC
61773).

• A number of these foundations initially tested with the lower capacity crane
equipment (50 tonnes) experienced movement more than 10 mm but did not
fail. These were later tested again to failure with the larger equipment (200
tonnes) and showed the same results under the repeat tests.

• Earlier work by (Mors, 1964) indicated that the tower can sustain a 10 mm
differential leg movement without any damage to the tower.

The ultimate load was defined as the final pull-out capacity of the foundation. Since
usually the foundation displacement under ultimate condition is large, it was reported
that is not practical to use this load because the tower would not be able to sustain
such a large displacement. Therefore, in all subsequent data analyses, a working
limit load at 10 mm is used as the ultimate capacity.

Reliability Based Calibration\Section 5 4-7 SCB2 WG-07, August 2008

 4.4.3 Design models

In calculating the geotechnical design capacity of each of the foundations tested,

Buckley (1994) used three different geotechnical design models. These were (1) the
frustum cone model, (2) German design model (DIN EN 50341-3-4, 2001) and (3)
the shear model (Killer, 1953). Details of these models are given in this paper.
According to Buckley (2007), none of these models accurately reflects the failure
mechanisms of spread footings. Also, the assumed soil density and the concrete unit
weight used to determine the geotechnical design capacities for various foundation
types did not always reflect actual in-situ properties but are simple to use, provided
the designer understands the limitations. The geotechnical design capacities
determined for ESB data are based on a fixed soil density and a concrete unit
weight. The progressive failure model introduced in 1994 is more realistic but
requires more detailed soil knowledge which is seldom available; it provides the
ultimate load. The foundations were arranged into various categories based on their
actual dimensions and the soil types in which they were installed. A total of 10
categories were used to subdivide all foundations groups tested. The primary
objective of the study was to compare foundation strength factor values for various
foundation types based on the different models.

4.4.4 Foundation data for uplift test (Category 2-9 tests)

The nine data points for the “Category 2” foundation test are analyzed here using the
m -model (equations 4.7 and 4.8). Table 4.2 provides the data, i.e. the geotechnical
design capacity based on a 300 frustum angle model (section 3.1.1) and the
measured test capacity at 10 mm displacement (working limit capacity) for each test.
This data is taken directly from ESB report. The unit weights used are 22.56kN/m3
for concrete and 15.7kN/m3 for soil respectively. Table 4.3 also presents two
dimensions for each foundation, the depth and the equivalent diameter at the base.
The equivalent diameter was used since all of the tested foundations had a
rectangular base (Buckley, 2001).

Table 4.3 - ESB Test Data (CiGRE, 2001)

Foundation Type ML1A ML1B ML1C ML1D ML2A ML2B ML4B ML8A ML8B
Depth-A (m) 1.5 1.7 1.9 2.1 2.3 2.3 2.7 2.2 2.3
Equivalent Base Diameter-D0 (m) 1.53 1.34 1.63 1.49 1.54 1.52 1.65 1.66 1.59
A 0.98 1.27 1.65 1.41 1.49 1.51 1.63 1.32 1.45
Aspect Ratio- ( )
D0
Model Predicted –Geotechnical 154** 163 249 270 332 328 489 332 345
design capacity (kN)
Test Capacity at 10 mm 204* 248 364 416 389 306 680 383 418
displacement (kN)
Test Capacity 1.32 1.52 1.46 1.54 1.17 0.93 1.39 1.15 1.21
mi =
Geotechnical Design Capacity
* Numbers have been rounded to nearest digit

Table 4.4 presents the data statistics. The mean values of the test and the
geotechnical design capacity are 378kN and 295kN respectively, while the coefficient
of variation values are approximately 35%. The m and vm values are 1.30 and

Reliability Based Calibration\Section 5 4-8 SCB2 WG-07, August 2008

0.154, respectively. These values can be calculated from equations (4.2) to (4.5).
Similarly the mean value and the coefficient of variation of the aspect ratio are also
given in Table 4.4.The model in general, appears to be conservative because of
large bias, i.e., the model generally underestimates the test capacity value by 30%.

Table 4.4 - Data Statistics

Average Coefficient of
n x sx y sy m vm Aspect Variation of
Ratio Aspect Ratio

9 296 102 378 135.4 1.3 0.154 1.36 0.15

Table 4.5 presents the foundation strength factor values for 5% and 10% exclusion
limits (prediction with 95% and 90% confidence levels respectively).

Table 4.5 – Foundation Strength Factor Values for 9-point dataset (ESB
Data)

Confidence Level 90% 95%

n 9 9
m 1.30 1.30
vm 0.154 0.154
kα -values 1.28 1.64
φF for kα -values (equation 4.7) 1.04 0.97
tα ,ν where ν = n-1=8 (Table 2.2) 1.40 1.86
φF for tα ,ν (equation 4.8) 1.02 0.93

4.5 Summary

This section of the report presented a simple mathematical model to compute the
foundation strength factor following Buckley (1994). The mathematical model
requires the statistical properties of the m-values defined as the ratio of the
foundation test capacity and the geotechnical design capacity. The various
underlined assumptions of the model are presented and a need for a more robust
linear regression model has been identified when the specific assumptions are not
met. This is presented in Section 5 under the linear regression model (“alternative
method”). An example design problem for a pad and chimney foundation is used in
section 5 to demonstrate the applications of simple model as well as the ‘alternative
model’ based on a linear regression analysis.

Reliability Based Calibration\Section 5 4-9 SCB2 WG-07, August 2008

5.0 CALIBRATION OF FOUNDATION STRENGTH FACTOR
(LINEAR REGRESSION MODEL)

5.1 General

As described in Section 4, the foundation designer could initially analyze the

foundation test data, based on the assumptions that the intercept is zero and the
variance of residuals is non constant. This will lead directly to the m - model
(Buckley, 1994) and will provide a single foundation strength factor for a given
dataset. However, in order to validate these two assumptions the foundation
designer should consider using a linear regression model where the foundation
strength factor can be obtained directly from the lower limit of the prediction interval.

The prediction interval is very similar to the confidence interval estimate for the
mean test capacity value (reference to section 2.3); except that the prediction interval
is used to predict a foundation test capacity (new value) for a given single foundation
geotechnical design capacity value. The lower limit on the prediction interval in the
linear regression model is analogous to the exclusion limit on characteristic capacity
described in section 2.4.2. This linear regression model can be applied to a wide
variety of foundation test data, where the correlation value may not be high (table
4.1) and the intercept from the linear regression equation may have an effect.

This section of the report correspondingly provides an alternative methodology to

calibrate the foundation strength factor of one foundation based on a linear
regression model. The regression model developed is for general application; it is
also shown that the equation (4.7) is a subset of the generalized linear regression
model. In addition, a simple test is provided that can be used to simultaneously
accept or reject the “hypothesis” that the intercept is zero or can be neglected and
v v
the slope is m (equation 4.3), and as such validate the m - model. This is a major
finding of this study and will validate the use of equation (4.3) across the data range,
provided that the hypothesis is acceptable. An example problem (3 point test data),
using manual calculations is presented in section 9.0 (APPENDIX) to illustrate the
methodology. A design problem based on ESB test data (section 4.0) is presented
using the linear regression model (“alternative method”).

5.2 Linear Regression Model

5.2.1 General

In the linear regression model, the test capacities (y-values) are plotted against the
foundation model predicted geotechnical design capacities (x-values) and a straight
line is fitted through the data to ensure that the residual error, ε i , is minimum. Figure
5.1 depicts a least square fit of a typical test dataset. The residual error, ε i given in
equation (5.1), is defined as the difference between the estimated mean test capacity
)
value, yi (given in equation 5.5), and the actual test capacity value, yi and is shown
in Figure 5.1.
)
ε i = yi - yi (5.1)

This residual error, ε i , should be minimized to ensure that there is a ‘best fit’ line to
the dataset. The regression model presented is for the case where each test

Reliability Based Calibration\Section 5 5-1 SCB2 WG-07, August 2008

capacity has a corresponding model predicted geotechnical design capacity. This
means that each pair of data ( xi , yi ) is one test. If there are n-sample data plotted, it
can be shown that the slope, b, and the intercept, a, are given by the following
formula:

b=
∑ wi ( xi − x ) yi (5.2)
∑ wi ( xi − x )2
a = y − bx (5.3)

where wi is the appropriate weight applied to each data point (Ang and Tang, 1975).
Normally, the variance of residuals in the linear regression model is assumed to be
constant and wi =1.0, unless the residual varies in a non constant manner; this point
will be discussed later. The sum is carried out based on a sample size of n.

The slope, b, can also be computed from the following relationship if we know the
correlation coefficient, r from Equation 4.1:

sy
b=r (5.4)
sx

The estimated mean test capacity value along the regression line corresponding to
)
each xi , is yi , and can be calculated from equation (5.5) as:

)
yi = a + b( xi ) (5.5)

110

100 Actual y Value

Residual
Test Capacity, y (kN)

Error, ε
90

Estimated, y
80

70

Intercept, a Slope, b
15 20 25 30 35 40 45 50 55 60

Geotechnical Design Capacity, x (kN)

Figure 5.1 - Typical regression plot with intercept and slope

Reliability Based Calibration\Section 5 5-2 SCB2 WG-07, August 2008

 2
The conditional variance, s xy about the regression line often called the Standard
Error of Estimate (SEE) of yi and xi is assumed to be constant within the predictor
values (data along the x-axis) and called hereafter, the variance of residuals and is
obtained from equation (5.6):

n
)
∑ wi ( yi − yi )2
s xy 2 = 1 (5.6)
n−2
)
where yi is the estimated mean test capacity value of yi , and is based on Equation
(5.5) and wi =1.0. For non constant variance of residuals where the residuals vary
with xi values, wi is dependent on the predictor values along the x-axis (Ang and
Tang, 1975). Depending on the weight, the unit of s xy could be dimensional or non-
dimensional.

The variance of residuals can be also estimated directly from the data without doing
a regression analysis and is given by equation (5.7):

n −1 2
s xy 2 = [ ]s y (1 − r 2 ) (5.7)
n−2
For large n,

s xy 2 → s y 2 (1 − r 2 ) (5.8)

All these calculations can be done routinely in a spreadsheet to determine the

intercept, slope, variance of residuals and the correlation value.

5.2.2 Confidence interval (general linear regression analysis)

)
The two-sided confidence interval (CI) on the estimated mean test capacity value, y
given a geotechnical design capacity value, x0 in equation (5.5) is given as:

) 1 ( x0 − x ) 2
CI = y ± kα / 2 * s xy [ + ]0.5 (5.9)
n n
∑ wi ∑ wi ( xi − x ) 2
1 1

In this equation, wi = 1.0 because the variance of residuals is constant. For finite
sample size, kα / 2 should be replaced by tα / 2,ν where ν = n-2 degrees of
freedom. The equation is very similar to equation (2.6) except that the estimated
)
mean test capacity y is a conditional mean given a geotechnical design capacity
value x0 . The term s xy is the standard deviation of the residual error of the combined
data (square root of equation 5.6), which includes both the actual test capacity and

Reliability Based Calibration\Section 5 5-3 SCB2 WG-07, August 2008

geotechnical design capacity values. Figure 5.2 depicts the confidence interval along
the regression line.

For repeated test data at a specific site, x = x0 (sections 3.7.1 and 4.1), r = 0.0
(equation 4.1), b= 0.0 (equation 5.4), a= R (mean value of the test data itself,
equation 5.3), s xy = sR (square root of equation 5.8) and the equation (2.6) is
obtained from equation (5.9). Therefore, the confidence level on the mean value of
the test capacity data referred in section 2.3 (equation 2.6, for test capacity data
only) is a subset of equation (5.9) under these specific conditions.

5.2.3 Prediction interval (general linear regression analysis)

2
The prediction interval is not only dependent on the variance of residuals s xy , but
)
also on the uncertainty in the estimated mean test capacity value, y given a
geotechnical design capacity value, x0 (equation 5.5). The lower limit of the
prediction interval of the predicted test capacity y p , based on a single geotechnical
design capacity x0 , (same as Rn ) is obtained from equation (5.10)

) 1 ( x0 − x ) 2
y p = y - kα s xy [ 1 + + ] 0.5 (5.10)
n n
∑ wi ∑ wi ( xi − x ) 2
1 1

The estimated mean test capacity value ŷ is dependent on the regression

coefficients such as the intercept a and slope b. These additional factors introduce
larger error in the prediction interval estimate when the predictor data x0 are far from
their mean value, x i. e., ( x0 − x ) in equation (5.10) is large. The variance of
residuals for the prediction must take into account the fact that the quantity being
predicted is a random variable (section 2.4). The quality of the prediction will depend
on how small is the difference between ŷ and y p and the error ( ŷ - y p ) ², which is
called the mean squared error of prediction.

The equation (5.10) is very similar to equation (2.8) except that the prediction is
based on the lower limit of the interval given a geotechnical design capacity value x0 .
This is depicted in figure 5.2 where it is shown that for a specified geotechnical
design capacity value x0 , the lower limit of the prediction interval point B, is similar to
point L, in figure 2.3 (the characteristic capacity RC ). The only difference is that the
computation of the lower limit of the prediction interval in figure 5.2 is based on a
given conditional geotechnical design capacity value x0 ; while in the latter case, it is
done on the test data directly (figure 2.3).

For repeated test data at a specific site x = x0 (sections 3.7.1 and 4.1),
r = 0.0 (equation 4.1), b= 0.0 (equation 5.4), a= R (mean test capacity value of the
test data itself, equation 5.3), s xy = sR (equation 5.8) and the lower limit of equation

Reliability Based Calibration\Section 5 5-4 SCB2 WG-07, August 2008

(2.8) is obtained from equation (5.10). Therefore, the prediction interval of a new test
capacity referred in section 2.4 (equation 2.8, one sided for test capacity data only) is
a subset of equation (5.10) under these specific conditions.

Details of the calculation of intercept, slope, correlation value, variance of residuals,

confidence interval and the prediction interval for a three point scattered data (figure
3.11, table 3.6) are given in Section 9 (APPENDIX). The next four subsections
discuss various issues that need to be considered in analyzing the data using a
linear regression model.

140

130
Prediction Interval
Estim ated M ean Test Capacity,
120 y^ at x o
Test Capacity, y (kN)

110

Residual
Error, ε
*
100
A

Distance from Estim ated

90 Test Capacity
B
80 Slope, b
Lower Lim it of
Prediction Interval, yp , is
Confidence
Intercept, a Characteristic Capacity, R c
70 Interval
x o , Any Arbitrary G eotechnical
Design Capacity Value
15 20 25 30 35 40 45 50 55 60

G eotechnical Design Capacity, x (kN)

Figure 5.2 - Regression plot with confidence interval on mean test capacity
values and prediction interval for new test capacity values

5.2.4 Correlation Issue

Equation (5.8) indicates that the variance of residuals is dependent on the variance
2
of the test data s y (test capacity data along y-axis) and the correlation value r 2
between the test data and geotechnical design capacity. The correlation value can
be obtained from equation (4.1). If a dataset has a high degree of correlation ( r 2 →
1.0), then every test data point is almost equal to the model predicted geotechnical
design capacity (refer to figure 4.1b). In this case, the slope of the line b → 1.0
and y ≅ x and s y ≅ s x . The intercept will be close to zero (equation 5.3). In this
case the user can assume that the line passes through the origin. The correlation
value will reflect the realistic characteristic of a design model if the data covers
geotechnical design capacity values both in the low (near x =0.0) and high ranges
within a sample size.

5.2.5 Linearity Issue

To validate the linearity issue, the user needs to plot the residual errors ε i (equation
)
5.1) against the estimated mean test capacity values yi obtained from equation
(5.5). A typical graph for the residual error values plotted against the estimated mean
test capacity values ( ŷi values) is shown in figure 5.3. The residual errors should

Reliability Based Calibration\Section 5 5-5 SCB2 WG-07, August 2008

produce a band centered along ε = 0.0 line, to demonstrate a linear regression
model.

Linearity Test

300
200
Residual Errors (kN)

100
0
-100 0.0 200.0 400.0 600.0 800.0

-200
-300
-400
Estimated Mean Test Capacity (kN)

Figure 5.3 - Residual errors banded around ε = 0.0 line

5.2.6 Zero Intercept Issue

As the correlation value tends to decrease ( r 2 ≤ 0.8, table 4.2) with increasing
coefficient of variation of the model predicted error ( vm ≥ 0.1), consideration should
be given to include the intercept effect in the prediction interval calculation. For
foundation calibrations where the dataset has moderate correlation ( r 2 ≤ 0.8 ), the
)
intercept has a significant influence in the estimated mean test capacity value y i . In
this case, by forcing the line to pass through the origin arbitrarily, the foundation
designer will make the prediction optimistic (less conservative); while in reality the
designer will not be able to validate the predicted foundation test capacity with a very
high confidence level.

For example, if a database has a poor correlation ( r → 0.0 ) and the second
assumption (zero intercept) is used arbitrarily, this will make the results acceptable
when in reality the test capacity values along the y-axis have no linear relationship
with corresponding geotechnical design capacity values along x –axis, due to a lack
of correlation (Table 4.1). If the foundation design engineer considers the feasibility
) )
in reducing the model from yi = a + b( xi ) to yi = m ( xi ), where the line passes
through the origin, then in principle a null hypothesis ( H 0 ) should be tested
simultaneously at a specified significance level, α (Benjamin and Cornell,1970). A
null hypothesis implies that the proposition

H 0 : a=0.0 and b= m

Reliability Based Calibration\Section 5 5-6 SCB2 WG-07, August 2008

is believed to be true. The significance level ( α ), provides a probability measure of
rejecting the null hypothesis when in fact it is true. A significance level 0.05 implies
that the engineer could reject the hypothesis 5% of the time in the long run when in
fact it is true.

A specific test can be used to reject simultaneously a null hypothesis that the
intercept, a = a* (arbitrary value) and the slope b = m ( H 0 : a= a* and b= m ) using
equation (5.11).

F*= [n * (a − a* ) 2 + 2 * n * x * (a − a* )(b − m ) + (∑ xi 2 )(b − m ) 2 ] / (2 * s xy 2 ) (5.11)

*
If the calculated F value is greater than the allowable value Fα ,2, n − 2 obtained from
F-distribution, with a significance level, ( α ), and ν = n − 2 degrees of freedom then
the hypothesis is rejected (Bowker and Lieberman, 1972, Sneddon, 2003). In this
case, the intercept should be included in the analysis and restrictions imposed on
how far the data can be extrapolated. This is particularly important for the
database where the correlation is moderate ( 0.3 ≤ r 2 ≤ 0.8, table 4.2) and the data
has large coefficients of variation, i. e. all values of vm , vx , v y are considerably
greater than 0.1. This situation can also arise when the database does not contain
enough points near x=0.0 (low end). It is important that the foundation design
engineer collects test data which at least will cover the low as well as high ranges to
ensure that the correlation value truly represents the main characteristics of the
design model.

The foundation designer should validate that equation (4.7) is satisfied for a* = 0.0
and b = m for the m model presented by Buckley (1994). This will be discussed
later in the example problem presented in section 5.6.

In equation (5.11), xi , (i = 1,...n) = model predicted geotechnical design capacities for

various test data. Table 5.1 provides some typical values for Fα ,2, n − 2 for two
significance levels.

Table 5.1 - F -distribution Values (Benjamin & Cornell, 1970)

Fα ,2, n − 2 for Fα ,2, n − 2 for

Sample size. n ν = n−2
α =10% α =5%
3 1 49.50 199.5
9 7 3.26 4.74
12 10 2.92 4.10

5.2.7 Variance issue (constant versus non constant)

In the linear regression model, it is normally assumed that the residual errors across
all geotechnical design capacity values remain fairly constant as depicted in figure
5.4a. However for many datasets, it is possible for the residual errors to vary linearly
or non linearly with the geotechnical design capacity values along the x-axis. In this
case, it is known as non constant variance of residuals. Figure 5.4(b) presents this

Reliability Based Calibration\Section 5 5-7 SCB2 WG-07, August 2008

case, which is one of the basic assumptions of the m -model recommended in
CIGRE (2002) and presented by Buckley (1994).

Constant Variance
of Residuals

ε)
Residual Errors ( Y

ε = 0.0

Geotechnical Design Capacities, X

Figure 5.4(a) - Constant variance of residuals

Non Constant
Variance of
Residuals Fan
ε)

Shaped
Residual Errors (

Geotechnical Design Capacities, X

Figure 5.4(b) - Non constant variance of residuals

For non constant variance of residuals, the assumption is that the variance of
residuals is proportional to the square of the x-values and an appropriate weight wi =
1
is applied to the original dataset and the regression analysis is carried out on the
xi 2
modified data (Ang & Tang, 1975). Figure 5.5 presents a typical residual plot against
the geotechnical design capacity values for a large dataset.

Reliability Based Calibration\Section 5 5-8 SCB2 WG-07, August 2008

 Residual Plot To Assess Constant versus Non
constant Variance of Residuals
300

200
Residual Errors(kN)

100

0
0 200 400 600 800
-100

-200

-300

-400
Geotechnical Design Capacity (kN)

Figure 5.5 - Non constant variance of residuals trend for a larger dataset

5.3 Foundation Strength Factor, φF

5.3.1 General

The basic linear regression model can be used to provide the confidence interval (CI,
equation 5.9) on the estimated mean test capacity ŷ (equation 5.5 for xi = x0 ) , as
well, as the lower limit of the prediction interval of a new test capacity y p (equation
5.10), given a single geotechnical design capacity value x0 (figure 5.2)

The foundation strength factor φ F is defined as the ratio between the lower limit of
the predicted test capacity y p and the geotechnical design capacity value x0 used to
determine this capacity (figure 5.2, Haldar, 2001, Ang, 2002). If the predicted test
capacity with a specified confidence level is y p and the geotechnical design
capacity value is x0 , then the foundation strength factor is:

yp
φF = (5.12)
x0

So far we have discussed two scenarios; these are: (1) the line passes through the
origin and (2) the line does not pass through the origin and therefore has an
intercept. Since there are a further two options (constant versus non constant
variance of residuals), under each of these conditions (zero or non zero intercept
condition), the foundation designer will end up with four possible cases for analyzing
the data. This is described in Table 5.2. Each case in Table 5.2 will provide a lower

Reliability Based Calibration\Section 5 5-9 SCB2 WG-07, August 2008

limit on the predicted test capacity with certain confidence level based on linear
regression model outlined in section 5.2.

Table 5.2 - Various Cases in Linear Regression Model

Intercept Variance of Residuals

Constant Non constant
Non Zero Case A Case B
Zero Case C Case D

The foundation designer can still obtain the foundation strength factor value using
equation (5.12) provided the predicted test capacity is known from any one of the
cases in Table 5.2. Equation (5.11) provides a definite test whether to include the
intercept in the analysis or not. However, the test for constant variance of residuals
versus non constant is not quantitative and is rather subjective because the designer
needs to plot the residuals against the geotechnical design capacity values and to
derive a trend line similar to figures 5.4a and 5.4b respectively. If it is not clear that
the residuals vary either in a constant or non constant manner then the designer will
end up with Cases A and B as one group or Cases C and D as the other group
respectively. This will provide the upper and lower limits on the foundation strength
factor values or in the predicted test capacity values. An example design problem
based on ESB data is shown later to indicate how the designer can select a
particular option.

The following section summarizes formulae for the four cases that the user may face
in analyzing the data. The formulae are presented to predict the lower limit of the
foundation test capacity y p for a given geotechnical design capacity value x0 for
conditions with or without zero intercept and with or without constant variance of
residuals. All these formulae also consider the correlation issue and the coefficients
of variation of the data. The various parameters such as slope b, variance of
2
residuals s xy , estimated mean test capacity ŷi , confidence interval (two sided) on
mean test capacity values CI and the lower limit of the predicted test capacity for a
single geotechnical design capacity value x0 with a given confidence level y p are
presented for each case. Often, the predicted test capacity value will depend on the
2
intercept a, slope b, variance of residuals s xy , the correlation coefficient r and the
weight wi used to estimate the variance of residuals. Since the intercept can be
either zero or can have a finite value and the variance of residuals can vary either in
a constant or a non constant manner (figures 5.4a and 5.4b), the designer will likely
encounter at least two of the four cases as presented in Table 5.2. A flow diagram is
also presented later to provide a guidance to the design engineer in the data
analysis process.

5.3.2 CASE A - Intercept is non zero and the variance of residuals is

constant

This case is applicable to problems where the intercept requires to be included and
the residuals plotted against the geotechnical design capacity values produce a band
around the ε i = 0 line (Figure 5.4a, constant variance of residuals). The slope b,
intercept a, estimated mean test capacity value ŷi , variance of residuals

Reliability Based Calibration\Section 5 5-10 SCB2 WG-07, August 2008

s xy 2 (standard error of estimate), confidence interval (CI) on the estimated mean
test capacity value ŷ and the lower limit on the prediction interval (PI) of the
predicted test capacity y p given a single geotechnical design capacity x0 can be
obtained from equations (5.2), (5.3), (5.5), (5.6), (5.9) and (5.10) respectively (figure
5.2).

5.3.3 CASE B - Intercept is non zero and the variance of residuals is non
constant

This case is applicable to problems where the intercept needs to be included and the
residual does follow a non constant variation according to Figure 5.4b. In this case,
there is a need to include both intercept and slope in estimating the prediction
interval. The slope b, intercept a, estimated mean test capacity value ŷi , variance of
2
residuals s xy , confidence interval (CI) on the estimated mean test capacity value
ŷ and the lower limit on the prediction interval (PI) of the predicted test capacity y p
given a single geotechnical design capacity x0 can be obtained from equations
(5.2),(5.3), (5.5),(5.6), (5.13) and (5.14) respectively (figure 5.2) with the appropriate
1
weight, wi = .
xi 2
1 ( x0 − x ) 2
Confidence interval CI = ŷ ± kα / 2 * s xy [ + ] 0 .5 (5.13)
n n
∑ wi ∑ wi ( xi − x ) 2
1 1

2
x 1 ( x0 − x ) 2
Predicted test capacity y p = ŷ - kα * s xy [ 0 + + ] 0.5 (5.14)
1 n n
∑ wi ∑ wi ( xi − x ) 2
1 1

5.3.4 CASE C - Slope only with constant variance of residuals

This case is applicable to problems where the intercept can be neglected based on
equation (5.11) and the residual does not increase with geotechnical design capacity
values (Fig. 5.4a). In this case, we only need to include the slope in estimating the
prediction interval. The weight wi = 1.0 , because the variance of residuals is
2
constant. The slope b, variance of residuals s xy , estimated mean test capacity
value ŷi , confidence interval (CI), on estimated mean test capacity value ŷ and the
lower limit on the prediction interval (PI) of the predicted test capacity y p , given a
single geotechnical design capacity value x0 with a specified confidence interval can
be obtained from equations (5.15), (5.16), (5.17), (5.18) and (5.19) respectively
(figure 5.2 with the intercept being zero).

b=
∑ wi ( xi ) yi (5.15)
∑ wi ( xi )2

Reliability Based Calibration\Section 5 5-11 SCB2 WG-07, August 2008

 ∑ wi ( yi − yˆi )
2
s xy 2 = (5.16)
n −1

ŷi = b xi where xi = x0 (5.17)

( x0 ) 2
Confidence interval CI = ŷ ± kα / 2 * s xy [ ] 0.5 (5.18)
n
∑ ( xi ) 2
wi
1
( x0 ) 2
Predicted test capacity y p = ŷ - kα * s xy [ 1 + ] 0.5 (5.19)
n
∑ ( xi )2 wi
1

5.3.5 CASE D - Slope only with non constant variance of residuals (Similar
to m -model)

This case is applicable to problems where the intercept is neglected based on

equation (5.11) and the residual does strictly follow a conical (fan shape) variation
(Figure 5.4b). In this case there is only a need to include the slope in estimating the
1
prediction interval. This is obtained using the weight wi = . The slope b, variance
xi 2
2
of residuals s xy , estimated mean test capacity value ŷi , confidence interval (CI) on
the estimated mean test capacity value ŷ and the lower limit on the prediction
interval (PI) of the predicted test capacity y p , given a single geotechnical design
capacity x0 can be obtained from equations (5.20) to (5.24).respectively (figure 5.2
with the intercept being zero)..
n
∑ yi / xi
b=
∑ wi ( xi ) yi = ∑ (1 / xi )2 ( xi ) yi =
∑ (1/ xi ) yi = 1 = m (5.20)
∑ wi ( xi )2 ∑ (1 / xi )2 ( xi )2 ∑i n

∑ wi ( yi − yˆi )
2
2
s xy = = sm 2 (5.21)
n −1

ŷi = m xi where xi = x0 (5.22)

1 0 .5
Confidence interval CI= ŷ ± kα / 2 * sm x0 [ ]
n
v
= m (1 ± kα / 2 m ) x0 (5.23)
n
1 0 .5
Predicted test capacity y p = ŷ - kα * sm x0 [ 1 + ]
n

Reliability Based Calibration\Section 5 5-12 SCB2 WG-07, August 2008

 1
= m (1- kα * vm [1+ ] 0.5 ) x0 (5.24)
n

Equation (5.24) is similar to equation (4.3) presented by Buckley (1994) but includes
1 0.5
a factor [1+ ] which was omitted in the earlier study. For small sample size, this
n
may create differences in the foundation strength factor values. For large sample
size, the term under the second bracket in equation (5.24) is close to 1.0. This is
similar to the difference in the predicted test capacity value and the characteristic
capacity presented in section 2.4. The foundation strength factor obtained from
equation (5.24) will be constant across all model predicted geotechnical design
capacity values (Sneddon, 2003). If this is not the case, then the foundation designer
may find that the foundation strength factor will be a function of the geotechnical
design capacity since the predicted test capacity is also a function of the
geotechnical design capacity (Cases A, B and C in table 5.2).

In all the above equations, kα values should be obtained from Table 2.1 for one
sided confidence level or use tα ,ν when the sample size is finite (n < 30, Table 2.2).
For cases A and B, ν = n- 2 while for cases C and D, ν = n-1.

Equation (5.23) is similar to equation (2.6) where the confidence level is sought on
mean test capacity data only. Equation (5.23) provides the confidence interval on the
estimated mean test capacity (conditional mean value of ŷ given a geotechnical
design capacity value x0 (or Rn ), ŷ I x0 as long as they are related through a
regression line with a slope b= m and a coefficient of variation vm . Equation (5.24)
similarly provides the lower limit on the prediction for a single geotechnical design
capacity value x0 This also corresponds to equation (2.8), where a minimum
strength (characteristic capacity with e% exclusion limit) is obtained from the test
data directly.

5.4 Flow Diagram

In order to analyze the data in a systematic manner, a flow diagram is presented in

Figure 5.6, where all the required validation tests and the corresponding equation
numbers are identified.

5.5 Limitation of Case D Model (Equation 5.24)

The foundation designer needs to consider two questions before the equation (5.24)
(also applicable for the simple model, equation 4.7), can be applied generally. These
are:

• What is the upper limit of vm beyond which the method is not valid?
• How far do we go along the x-axis to extrapolate the data in providing a
prediction interval for the test capacity?

To address the first question, the coefficient of variation cannot be greater than
1
vm ≤ [ ] (5.25)
1
tα ,ν [1 + ]0.5
n

Reliability Based Calibration\Section 5 5-13 SCB2 WG-07, August 2008

since m > 0.0 (equation 5.24).

This will ensure that the foundation strength factor is positive and provides a
reasonable value. Of course it can be argued that even with a very high m value, it
is always possible to achieve a positive foundation strength factor value when the
term inside the bracket (equation 5.24) is very low. However, in such a case the
foundation designer should not use the geotechnical design model, because it is
underestimating significantly the test data and correlation will be low (table 4.2).
Table 5.3 presents some typical values of vm to ensure that the foundation strength
factor does not become negative. For other intermediate value, use equation (5.25).

Table 5.3 - Theoretical upper Limit of vm for various confidence levels

and sample size (Equation 5.25)

90% Confidence level 95% Confidence level

Sample
Size, n tα ,ν 1 tα ,ν 1
vm ≤ vm ≤
1 1
tα ,ν [1 + ]0.5 tα ,ν [1 + ]0.5
n n
3 1.90 0.45 2.92 0.29
5 1.53 0.59 2.13 0.42
12 1.36 0.70 1.80 0.53
30 1.28 0.76 1.64 0.59

If the test data and the model predicted geotechnical design capacity produce a vm
value greater than the allowable limit shown in Table 5.3, the foundation strength
factor φF could become negative. The foundation design engineer needs to be
careful when the data is extrapolated outside the range. Even when the vm values
are close to the values shown in Table 5.3, the foundation strength factor value will
be low and probably unrealistic indicating that the error is introduced because of the
large coefficient of variation. Similar criteria can also be used for other cases from
the above equations or plotting the data graphically. Section 6 will provide some
guidelines on interpolation and extrapolation.

5.6 Example Problem (Re-analysis of 9-point ESB Data using Linear

Regression Model)

In this section, the 9-point ESB test data (table 4.3) is re-analyzed using the linear
regression model where two major assumptions that the intercept is zero and the
variance of residuals is non constant are examined. Figure 5.7 presents the
regression plot of the data and the correlation value.

Table 5.4 presents the data statistics. The model, in general, appears to be
conservative because of large bias, i. e. the geotechnical design model on average
underestimates the test capacity value by 30%. The correlation value, r 2 is 0.81 is
high (equation 4.1 - Test 1 in flow diagram, table 4.2). A linear regression model
(CASE A) is run first following the flow diagram in figure 5.6. The standard deviation

Reliability Based Calibration\Section 5 5-14 SCB2 WG-07, August 2008

of error s xy is 63kN based on constant variance of residuals assumption (CASE A,
equation 5.6 with wi = 1.0) ). The geotechnical design capacity of data ranged from
150kN to 500 kN.

Flow Diagram For Computing Foundation Strength Factor

Tabulate Data Properties for the
Problem, xi (geotechnical design
capacity) & yi (test capacity) values

yi
Compute mi =
xi

Data Plot, xi vs. yi

Evaluate m, sm and vm From
Equations (4.3), (4.4) & (4.5)
.

Test 1: Low Correlation

Eq. 4.1,
Table 4.2

High or Moderate
Correlation
Regression Analysis (Figure 5.2)
Slope (5.2), Intercept (5.3), Estimated Test Capacity
Value (5.5), and Variance of Residuals (5.6)

Slope &
Intercept Test 2: Slope only
Eq. 4.14
5.11

Test 3: Test 3:
Chk Figs. 5.4(a) Chk Figs. 5.4(a)
& 5.4(b) & 5.4(b)

Case A Case B Case C Case D *

(Eqn. 5.10) (Eqn. 5.14) (Eqn. 5.19) (Eqn. 5.24)

Calculate Foundation
Strength Factors to Provide
Conservative Values -
Equation 5.12
Test 1: Test for Correlation END
__
Test 2: Simultaneous Test for Intercept = 0.0 and Slope = m
Test 3: Visual Comparison to Test Constant Variance of
Residuals vs. Non Constant Variance of Residuals
*Constant COV - Special Case
(Residual values banded around zero line or "fan
- CIGRE (2002)
shaped" - use judgment)

Figure 5.6 - Flow diagram

Reliability Based Calibration\Section 5 5-15 SCB2 WG-07, August 2008

 Table 5.4 presents also the regression parameters for CASE A. The test for intercept
*
indicated (equation 5.11 –Test 2 in flow diagram) that F =0.164 which is less than
the allowable Fα ,2,ν = 3.26 (Table 5.1). Therefore, the null hypothesis that the
intercept is zero and the slope is equal to m =1.30 (Table 4.4) can be accepted.
Therefore, the designer has to choose between Case C and Case D respectively
(figure 5.6).

Uplift Test Data For Undercut Foundation- 30

Degree Cone Model, Correlation Value =0.81

600
Test Capacity (kN)

400 Least Square Fit-intercept,

a=26(equation5.3)
slope,b=1.19 (equation5.2)
s xy2 = 3934 (SEE-equation5.6)
200

0
0.0 200.0 400.0 600.0
Geotechnical Design Capacity (kN)

Figure 5.7 - ESB Uplift test data (CIGRE, 2002)

Table 5.4 - Data Statistics for Intercept Check (Equation 5.11)

s xy
n r2 x sx y sy a b ∑ xi 2 F*-value m vm

9 0.81 296 102 378 135.4 25.8 1.19 63 0.87E+05 0.164 1.3 0.154
< 3.26

Next, the residual values for the nine data points are plotted against the geotechnical
design capacity values and Figure 5.8 presents this. Two lines are drawn to see a
trend line where the variance of residuals is non constant (similar to figure 5.4 b) and
it is not fully definitive that this is the case. Therefore, according to the flow diagram
in figure 5.6, the designer should try both Cases C and D for determining the
foundation strength factor. Obviously under Case C the foundation strength factor
will not be constant and rather will be a function of the geotechnical design capacity;
because the predicted test capacity value is also dependent on the geotechnical
design capacity value (equation 5.19)

Reliability Based Calibration\Section 5 5-16 SCB2 WG-07, August 2008

 Residual Plot To Assess Constant Versus
Non Constant Variance of Residuals
100
Lines to depict the "Fan
Shaped" Plot To Identify
Residual Errors (kN)

50 Non Constant Variance of

Residuals Trend-Refer to
0 figure 5.4(b)
0.0 100.0 200.0 300.0 400.0 500.0 600.0
-50

-100

-150
Geotechnical Design Capacity (kN)

Figure 5.8 - Residual errors versus geotechnical design capacity values

Table 5.5 - (Case C-b=1.27, s xy =59.29, tα ,ν =1.40, wi = 1.0)

Geotechnical Estimated ( x0 ) 2 y p = ŷ -P Foundation

Design Mean Test P= tα ,ν * s xy [ 1+ ] 0.5 Strength
n (equation
Capacity Capacity yp
Value, x0 Value ∑
( xi ) 2 wi 5.19) Factor=
1 x0
yˆ = bx0
(second term in equation 5.19) (equation, 5.12)
(first term in
equation
5.19
200 254 85 169 0.847
400 509 90 419 1.046
600 763 99 664 1.107
800 1017 109 908 1.135

The lower limit of the prediction interval with (90%) confidence level is presented for
both cases, in tables 5.5 and 5.6 respectively. Figures 5.9 and 5.10 present the plots.
Figure 5.11 depicts the comparison of foundation strength factor, φF based on
CASE C and CASE D scenarios; it is interesting to note that for CASE D where the
foundation strength factor is constant across all geotechnical design capacity values,
the interpolation may not be a problem. However, for CASE C scenario, where the
residual is interpreted as constant, (figure 5.4a), the foundation strength factor
becomes strongly dependent on the geotechnical design capacity value (x-value)
chosen as the predictor values and further away from the mean value (297 kN).
Therefore, in this situation, the foundation designer may want to consider the lower
value of the two foundation strength factors to ensure that the foundation design is

Reliability Based Calibration\Section 5 5-17 SCB2 WG-07, August 2008

 conservative (Flow diagram-figure 5.6). Discussions on interpolation versus
extrapolation are presented in section 6.0.

ESB Uplift Test Data (Case C)-300 Frustum

Angle
1000
900 Regression line,
800 slope=1.27,
Variance of
Test Capacity (kN)

700
Residuals=3515
600
500 Original 90% Lower
400 Data Points Prediction Line -10%
300 Exclusion Limit
200
100
0
0.0 100.0 200.0 300.0 400.0 500.0 600.0 700.0 800.0 900.0 1000.0

Geotechnical Design Capacity (kN)

Figure 5.9 - Predicted test capacity with 90% confidence level (Case C, constant
variance of residuals)

1
Table 5.6 (Case D- m =1.30, s xy = sm =0.20, tα ,ν =1.40, wi = )
xi 2

Geotechnical Estimated 1 0.5 y p = ŷ -P Foundation

Design Test P= tα ,ν * sm [ 1+ ] x0 Strength
n (equation 5.24)
Capacity Capacity yp
Second term in Factor=
Value, x0 Value
equation (5.24) x0
yˆ = m x0
(equation, 5.12)
First term in
equation
(5.24)

200 260 59 201 1.00

400 520 119 1.00
401
600 780 178 602 1.00
800 1040 237 803 1.00

If the foundation designer wants to correct for the small sample size, then tα ,ν value
should be based on Student t-distribution (reference to Table 2.2) and the foundation
strength factor becomes

Reliability Based Calibration\Section 5 5-18 SCB2 WG-07, August 2008

 yp Re%
φF = = = m (1- tα ,ν * vm [1+ 1/n] 0.5) = 1.30*(1-1.40*0.146*(1+0.11) ^0.5) =
x0 Rn
1.00

This value is shown in Table 5.7. The value is slightly different from the one
1 0.5
presented in Table 4.5 because the term [1+ ] in equation (5.24) was not
n
included in the earlier analysis.

ESB Uplift Test Data (Case D)-30 Degree Cone

Angle
1000
900 Original
800 Data Points
Test Capacity (kN)

700
600
500
400 90% Lower prediction
300 line-10% exclusion limit,
Sample size included
200
100
0
0.0 100.0 200.0 300.0 400.0 500.0 600.0 700.0 800.0 900.0 1000.0

Geotechnical Design Capacity (kN)

Figure 5.10 - Predicted test capacity with 90% confidence level (CASE D, non
constant variance of residuals)

Comparison of Foundation Strength Factor Values For

CASES C and D (90% Prediction Interval-10% EL)
1.200
Foundation Strength Factor

1.000

0.800

0.600
Constant Foundation
0.400 Data Strength Factor,φ F =
R 1.004
Variable Foundation
0.200
Strength Factor Values,φ F
0.000
0 100 200 300 400 500 600 700 800 900

Geotechnical Design Capacity (kN)

Figure 5.11 - Comparison of foundation strength factor (sample size effect

included)

Reliability Based Calibration\Section 5 5-19 SCB2 WG-07, August 2008

 Table 5.7 –Foundation Strength Factor Values (Linear Regression
Model)

CASE D (equation 5.24)

Confidence Level
90% 95%
n 9 9
m 1.30 1.30
vm 0.154 0.154
kα -values 1.28 1.64
φF for kα -values (equations 5.24 & 1.03 0.97
5.12 )
tα ,ν where ν = n-1 (Table 2.2) 1.40 1.86
φF for tα ,ν (equations 5.24 & 5.12, 1.00 0.93
replace kα by tα ,ν in equation 5.24 )

5.7 Design Problem Using ESB Data (Foundation Depth)

A base case model for an undercut foundation (refer to figure 3.2 and table 5.8) was
used to obtain the uplift capacity based on a 300 cone model. The unit weights for
soil and the concrete were taken as 15.7kN/m3 and 22.6kN/m3 respectively. The
calculated capacity was 508kN. Next, a sensitivity study was undertaken by
changing the depth while keeping the base width constant. Figure 5.12 presents the
data and it is apparent that the tripling of the depth has a seven fold increase in the
foundation uplift capacity. The equation for computing the depth for a given
geotechnical design capacity is also given on the graph. The correlation value is
0.9999 (figure 5.12).

Geotechnical Design Capacity Based on

30 Degree Cone Model
800
Capacity (kN)

600
400
y = 73.586x2 + 8.2967x + 30.364
200
R2 = 0.9999
0
0 1 2 3 4
Change in Depth (A in m)

Figure 5.12 - Geotechnical design capacity with respect to change in depth

Reliability Based Calibration\Section 5 5-20 SCB2 WG-07, August 2008

 Table 5.8 - Typical Dimensions (m) of a Pad and Chimney Foundation
(Figure 3.2)

Type A B C D E F
Square 2.5 2.0 1.9 0.2 0.8 0.1
Based

The design equations (2.18) and (5.12) provide a relationship between the predicted
capacity y p (similar to RC ), and the geotechnical design capacity x0 , (same as Rn )

RC = φF Rn

For example, if the foundation design load due to extreme wind is QT = 500 kN, and
the foundation designer wants to provide a 90% confidence level based on nine ESB
tests, then the geotechnical design capacity required is (Table 5.5 , φF =1.00 for
90% confidence, 10% exclusion limit)

QT 500
Rn = = = 500kN
φFC
1.00

Once the geotechnical design capacity is known, the required foundation depth can
be estimated using figure 5.12. For QT = 500 kN, Case C provides a geotechnical
design capacity of 465 kN. This can be computed from equation (5.19) by
substituting y p = QT and solving for x0 . The foundation designer can also compute
the geotechnical design capacity from figure 5.12 for a predicted test capacity, y p =
500 kN following the lower limit of the prediction interval line (green line). The results
show a 7% difference in capacity between Case C and Case D respectively. The
corresponding design depth can be estimated either from figure 5.12 or from the
equation given on figure 5.12 for these two geotechnical design capacity values.
These are: design depths 2.37 m versus 2.47 m (5.0% difference).

However, if QT = 200kN, Case C will provide a geotechnical design capacity of

230kN (figure 5.9) while for Case D this would still be 200 kN (figure 5.10, 15%
difference in capacity). Figure 5.12 provides the design depth for Rn = 200 kN as
1.46 m while 1.6 m for Rn = 230 kN (9% difference in depth). Based on the analysis,
it is recommended that the designer uses the maximum depth for the design to be
conservative.

5.8 Summary

This section of the report presented a framework to determine the prediction interval
of the foundation test capacity based on a limited number of full-scale test data. The
lower limit of the prediction interval is analogous to characteristic capacity with a
specified exclusion limit (section 2.4) and can be linked to the foundation strength
factor (equation 5.12). The prediction interval is based on the calibration of data on
full-scale test capacity against model predicted geotechnical design capacity. The

Reliability Based Calibration\Section 5 5-21 SCB2 WG-07, August 2008

calibration procedure uses a linear regression model. Furthermore, It is shown that
the simple model (CIGRE, 2002) is a subset of the linear regression model when
certain assumptions are met. The example problem presented in section 4 is
reanalyzed using the linear regression model. It is shown that the designer can
obtain a comparable foundation strength factor value using the linear regression
model if the specific conditions are met.

Reliability Based Calibration\Section 5 5-22 SCB2 WG-07, August 2008

6.0 CRITERIA FOR INTERPOLATION AND EXTRAPOLATION

6.1 General

The foundation strength factor, φF , obtained from a linear regression model has a
valid range and is normally dependent on the predictor data range during the
calibration process (i.e. geotechnical design capacity values along the x-axis x0 or
Rn values). Any interpolation and/or extrapolation should be done properly to ensure
that the confidence level on the estimated mean test capacity value, ŷ from equation
5.5), and the prediction interval used to determine a new foundation test capacity y p ,
for a single geotechnical design capacity value x0 , are obtained accurately.

6.2 Interpolation

The relationship between the predicted test capacity y p for a given geotechnical
design capacity, x0 and the predictor (geotechnical design capacity, x0 ) used in the
calibration model applies strictly when the predictors have similar values that were
used in the calibration model. If this is the case, then it is called interpolation.

ESB Data (CASE D, Slope Only--Interpolation for 400kN nominal design

capacity,sample size effect included)

800
750 Original Data Points
700
Regression Line,
Test Capacity (kN)

650
600 slope=1.30
550
500 Interpolation Point, x=400,
450
400 Estimated Mean Test
350 Capacity=512, Equation 5.22
300
250
200 Predicted Capacity, 90%
150
100 confidence level, 401 -Equation
50 5.24
0
0 100 200 300 400 500 600 700 800 900 1000

Geotechnical Design Capacity (kN)

Figure 6.1 - Interpolation

Therefore, it is important to develop the linear regression model based on a wide

data range (predictor values, x0 ) to ensure that the large error is not introduced
during interpolation. In this example (9 points data set), the geotechnical design
capacity values along the x-axis (predictor values) range between 150kN and 500kN.

Reliability Based Calibration\Section 6 6-1 SCB2 WG-07, August 2008

Figure 6.1 presents a case where a geotechnical design capacity value of 400kN is
interpolated along the regression line for CASE D (equation 5.22). The estimated
mean test capacity is 512kN ( m =1.30, yˆ = m x0 = 400 x 1.30=512). On the same
graph, the predicted test capacity (90% confidence level - 10% exclusion limit) is also
depicted which has a value of 424kN (equation 5.24, tα ,ν = 1.90 ). All these
computations are done based on CASE D model where the line passes through the
origin and the variance of residuals is assumed to be non constant.

6.3 Extrapolation

The extrapolation process is making a prediction outside the range of values of the
predictor used in the sample to generate the model. Therefore, the further the value
of the predictor is moved from the mean value, the larger the error in the prediction
(reference to section 5.2.2). Besides, there is no guarantee that the linear
relationship holds outside the data range (Ang and Tang, 1975).

Figure 6.2 presents a case where the extrapolation is done for a geotechnical design
capacity of 800 kN along the regression line (CASE D model). The estimated mean
test capacity is 1040 kN (equation 5.22). On the same graph, the predicted test
capacity (90% confidence level-10% exclusion limit) is also depicted which has a
value of 849 kN (equation 5.24, tα ,ν = 1.90 ) where the slope is only considered (the
line passes through the origin) and the variance of residuals is assumed to be non
constant.

ESB Data (CASE D, Slope Only-Extrapolation for 800 kN Geotechnical design

capacity value, sample size effect included)

Extrapolation point for a

1200 geotechnical design capacity,
1100 800kN-Corresponding estimated
1000 mean test capacity,1040kN-
Original
900 Equation 5.22
Test Capacity (kN)

Data
800 points
700
600 Predicted capacity for a
500 geotechnical design
400 capacity, 800 kN with 90%
Regression
300 line,slope=1.30 confidence -Equation 5.24
200
100
0
0.0 100.0 200.0 300.0 400.0 500.0 600.0 700.0 800.0 900.0 1000.0
Geotechnical Design Capacity (kN)

Figure 6.2 – Extrapolation

6.4 Foundation Strength Factor – Interpolation and Extrapolation

Figure 5.11 presents the foundation strength factor plot for various geotechnical
design capacity values. The geotechnical design capacity values for ESB data varies

Reliability Based Calibration\Section 6 6-2 SCB2 WG-07, August 2008

between 150kN and 500kN. Between the limits of 300kN to 500kN, both cases
(CASE C and CASE D) provide similar foundation strength factor values. However,
as the predictor (geotechnical design capacity value) becomes less than 300kN, the
foundation strength factor in CASE C starts to decrease compared to the value
obtained in CASE D. Therefore, a judgment is needed when one tries to determine
the foundation strength factor based on CASE C.

However, if the foundation designer believes that the variance of residuals is fully non
constant based on figure 5.8, then a constant foundation strength factor value can be
used for the entire data range (150 kN to 500 kN). Of course to extrapolate outside
the data range, the assumption still is that the linearity is still valid in this range. As
mentioned earlier, this is only valid when the line passes through the origin (no
intercept) and the variance of residuals is strictly non constant. If this is not the case,
the foundation strength factor can be dependent on the geotechnical design capacity
values for certain data range.

6.5 Characteristic capacity given two geotechnical design capacity

values

The predicted test capacities, y p or the characteristic capacities, RC for two

geotechnical design capacity values, Rn are computed as follows:

6.5.1 Interpolation

RC = φ F Rn = 1.046 x 400 = 418kN (Table 5.5, Figure 5.10-Case C)

RC = φ F Rn = 1.00 0 x 400 = 400kN (Table 5.6, Figure 5.11-Case D)

The lower and upper ranges of the design value are 400kN to 418kN

6.5.2 Extrapolation

RC = φ F Rn = 1.135 x 800 = 908kN (Figure 5.10-Case C)

RC = φ F Rn = 1.000 x 800 = 800kN (Figure 5.11-Case D)

The lower and upper ranges of the geotechnical design capacity value are 800kN to
908kN respectively.

The user needs to be careful when extrapolating the data outside the range,
particularly when the variance of residuals is constant (Case A and Case C). Also,
there are no published guidelines how far one can go to extrapolate the predictor
value. Correspondingly, it is recommended here that for constant variance of
residuals case, we do not go normally beyond the xL - s x or xU + s x where xL and
xU are the lower and upper value of the predictor data range and s x is the standard
deviation of the geotechnical design capacity. The foundation designer therefore
needs to use judgment in extrapolating the data outside the range assuming that the
linearity relationship holds. As the engineer gets more data from future tests, this
data can then be added to the database and the analysis updated. The designer
should collect data which covers the low and high ranges, particularly data near the
low end (x=0.0); which could allow the intercept effect to be minimum provided the
correlation is high. On the other had if the designer collects data only in the high end
and the correlation is moderate, then most likely the intercept effect would be

Reliability Based Calibration\Section 6 6-3 SCB2 WG-07, August 2008

considerable and the extra caution should be used in extrapolating the data outside
the range (toward the low end). The economical impact of not having data that
covers both low and high ranges adequately could be increased cost (increased size
of the foundation) depending on the specific case chosen with particular reference to
calibration of foundation strength factor.

This section presented the limitation of the data extrapolation and some guidelines
how to do this extrapolation. An example problem is shown using the foundation
strength factor obtained in Section 5.0 (Tables 5.5 and 5.6).

6.6 Summary

This section presented a discussion on the issues related to interpolation versus

extrapolation. Whereby, it is shown that the designer needs to be aware the
foundation strength factor value can be dependent on the geotechnical design
capacity for constant variance of residuals case, while for a non constant variance of
residuals this is always constant. The designer also should be aware that in
extrapolating foundation strength factor data outside the data range, the linearity
assumption must hold. Furthermore, it is recommended here that the extrapolation
should be not done normally beyond the limits of xL - s x or xU + s x unless the
linearity assumption can be validated outside the data range.

Reliability Based Calibration\Section 6 6-4 SCB2 WG-07, August 2008

7.0 SUMMARY AND CONCLUSIONS

7.1 Summary

This report presents the various steps necessary to determine the lower limit of the
prediction interval of the foundation test capacity (similar to characteristic capacity)
and to calibrate the foundation strength factor based on a limited number of full-scale
foundation test data. The calibration is undertaken within the framework of a linear
regression model. Initially, a brief review is presented with respect to confidence
interval on the sample mean of the test capacity values and the prediction interval of
a new test capacity value given a sample data of finite size. The confidence interval
is used in the determination of required sample size for a foundation test programme
while the lower limit of the prediction interval is linked to the characteristic capacity
with a specified exclusion limit. This is followed by a brief review outlining the
deterministic as well as the semi-probabilistic design methodologies such as ASD
(Allowable stress design), IEC 60826 (2003) and ASCE (LRFD).

The foundation designer is given guidance on how to undertake a full-scale

foundation test programme. Several steps need to be completed before a full-scale
foundation test programme can be commenced. These include: (a) selection of the
test site(s), (b) selecting the applicable foundation geotechnical design model(s) to
determine the foundation’s geotechnical design capacity, (c) subsurface investigation
to determine the various geotechnical strength parameters for the foundation design
model(s) and (d) sample size requirement for the test program. Once the
foundation’s geotechnical design capacity has been determined, the foundation
designer is presented with a number of steps required to carry out a full-scale
foundation test programme. This generally includes the choice of test site locations,
installation, loading scheme and sequence (load cycles), instrumentation and data
collection, data reduction and finally the test results in terms of a typical load versus
displacement curve.

Subsequently, a reference is made to IEC 61773 which outlines the available

methodologies to determine the foundation’s test capacity in terms of damage limit as
well as the ultimate limit capacity based on the test result. A review of the information
presented shows that the estimated ultimate capacity could vary considerably
depending on the method selected. However the foundation designer can also
choose a damage capacity (IEC 60826) based on a prescribed displacement
(subjective). For a pad and chimney foundation used in the example, load at a 10
mm displacement is defined as the test capacity (Buckley, 1994).

Once the foundation’s test capacity and the geotechnical design capacity have been
determined for specific site(s), the foundation designer can choose a simple
Test Capacity
calibration model based on m = ratio(s). However, to
Geotechnical Design Capacity
use this simple model ( m -model) the foundation designer needs to ensure that the
correlation value of the data is high (table 4.2) and the coefficient of variation of the
m-values, v m (model error) is low.

For other situations where this is not the case, the foundation designer may need to
carry out a more detailed analysis (full linear regression model) to ensure that the
intercept (ordinate at origin) effect can be neglected. This can be undertaken using a
simple statistical test. If the intercept effect cannot be neglected, then the linear
regression model (an alternative methodology) is also presented on how to include

Reliability Based Calibration\Section 7 7-1 SC B2 WG-07 August 2008

this in the analysis to estimate the prediction interval of the test capacity given a
geotechnical design capacity value. Moreover, the effect of constant or non constant
variance of residuals is presented and its effect is emphasized in analyzing the data.

Finally a design example is presented based on ESB test data on undercut pad and
chimney foundation. A discussion on the limitation of data extrapolation is also
presented.

7.2 Conclusions

Some important conclusions are drawn from this report:

• A review of the current CIGRE method (CIGRE, 2002) is presented; which

has examined the underlying assumptions and their limitations in a systematic
manner. The method is primarily based on the linear regression model and it
is shown that it is valid under very specific conditions and assumptions, as
detailed below.

• The CIGRE Method relies on two very specific conditions which imply that the
intercept should be zero and the residual (the difference between the actual
test capacity and the estimated mean test capacity) should vary with the
geotechnical design capacity values in an increasing manner.

• To apply this CIGRE method to a wide range of reliability based foundation

design calibrations with various degrees of correlation, the designer must be
aware that unless these specific conditions are met, the prediction interval
with a high degree of confidence (90% for IEC 60826 and 95% for ASCE) will
produce results that may not be valid.

• A simple statistical test is proposed here to provide guidance to the designer

so that the assumption of “zero intercept” is validated properly.

• In addition, it is shown that the CIGRE method does not consider the error
due to small sample size in calculating the prediction interval. CIGRE method
is similar to Case D (equation 5.24) but does contain an additional
1
term [1 + ]0.5 . Therefore the CIGRE method is modified to include the effect
n
of small sample size. The difference in the foundation strength factor value
could be 2 % to 6% depending on the sample size, n and v m values.

• If the correlation value, r 2 is greater than 0.8 and the coefficient of

variation, v m is low, the intercept can be neglected without introducing
significant error in the analysis. Under this condition, CIGRE method (similar
to CASE D) is quite acceptable provided the variance of residuals is non
constant.

• However; if the correlation value, r 2 is greater or less than equal to 0.8 (table
4.2) and the coefficient of variation v m starts to increase (> 0.10), the
designer needs to carry out a full regression analysis to choose a particular
option and/or options from the flow diagram (figure 5.6) to ensure that the
analysis result provides not only a predicted capacity which produces a
conservative design, but also that the design is optimal (economical).

Reliability Based Calibration\Section 7 7-2 SC B2 WG-07 August 2008

 • A simple example problem with 3-point test data is illustrated in Section 9
(“Appendix”) to show the designer how to calculate the various steps
manually to check the CIGRE method. The results based on the CIGRE
method are not only verified but the limitations are also identified.

• An example problem is illustrated for a typical pad and chimney foundation in

cohesionless soils. The comparison is excellent between the results obtained
from m - model and CASE D.

• Finally, it is shown that the general regression model with the unique test for
“zero intercept” is quite applicable to the analysis of a wide range of
foundation test data with various degree(s) of correlation, coefficient of
variations and m values. The methodology presented here is quite robust
and provides a consistent basis for analyzing a wide range of test data. The
foundation designer is able to review the details of these calculations in order
to understand the intricate relationships of various parameters that contribute
to the estimation of the foundation strength factor, φ F values. The foundation
designer can make an informed decision to choose a particular option or
options in determining the predicted test capacity of the foundation with a
specified exclusion limit.

Reliability Based Calibration\Section 7 7-3 SC B2 WG-07 August 2008

8.0 REFERENCES
Ang, A. H. S. and Tang, W.H. 1975 Probability Concepts in Engineering
Planning and Design, Volume 1, John Wiley & Sons

Ang, A. H. S. 2002 Review of the “Current Method” of Calibration and a Proposal

for an “Alternate Method” for Calibration of Foundation Strength factor- prepared by
Asim Haldar, September 2001, presentation to CIGRE WG07 meeting in Oslo,
Norway.

Ang, A. H. S. 2002 Personal communication

Benjamin, J. and Cornell, Allin 1970 Probability, Statistics and Decisions for
Civil Engineers, McGraw Hill

Bowker, A. H. and Lieberman, G. J. 1972 Engineering Statistics, Prentice Hall,

NY. 641 p

Brink, A.B.A, Patridge, T.C. and Williams, A.A.B. 1982 Soil Survey for
Engineering, Clarendom Press, Oxford, 378 p

Buckley, M. 1994 Reliability Based Design of OHL Foundations, CIGRE

Publication, SC 22-203, Overhead Lines, Paris Meeting, August 28-September
3,11p.

Buckley, M. 2001 Example of Probabilistic Foundation Design, CIGRE WG

meeting, Heidelberg, Germany, 8p

Buckley, M. 2007 Background Information on ESB Tests, e-mail communication,

2p.

CIGRE 2002 The Design of Transmission Line Support Foundations- An Overview,

WG 22.07, August, Technical Brochure No. 206

CIGRE 2006 Foundation Installation-An Overview, WG B2.07, May, Technical

Brochure No. 306

Davidson, H.L. 1982 Laterally Loaded Drilled Pier Research, Volumes 1 and 2,
EPRI EL 2197, RP 1280-1, prepared by GAI Consultant, USA

Digioia et al 1993 A comparison of Various Methods for Predicting the Response

of Drilled Shafts Subjected to High Moments, ELECTRA No. 149, 1993

Digioia Jr., A.M., 2000 Reliability Based Design and Assessment of Foundations
for Transmission Line Structures, IEEE T & D World Expo, April, 26-28, Cincinnati,
Ohio

EN 50341-3-4 2002 Overhead Electrical Lines Exceeding AC 45 kV-Part 3-4:

National Normative Aspects (NNA) for Germany

ESB Test Report on Overhead Line Foundations

Reliability Based Calibration\Section 8 8-1 SCB2 WG-07, August 2008

EPRI EL 2870 1983 Transmission Line Structure Foundations for Uplift-
Compression Loadings, Electric Power Research Institute, Palo Alto, California.

IEC 60826 2003 Design Criteria of Overhead Lines

IEC 61773 Testing of Structure Foundations For Overhead Lines.

IEEE 1985 Transmission Structure Foundation Design Guide

Killer, J. 1953 Economical Foundations of Towers for High Voltage Transmission

Lines, Third Soil Mechanics Conference, Zurich, Switzerland

Melchers, R. 1999 Structural Reliability Analysis and Prediction, John Wiley

Mors, H. 1964 Methods of Dimensioning for Uplift Foundations of Transmission

Line Towers, Paper No 210, CIGRE, Paris, France

Prasad, Yenumula and Haldar, Asim 2001 Full-scale Tests On Grillage Anchor
Foundations for High Voltage Lines, Proc. Canadian Geotechnical Conference,
Calgary

Rao, S.S. 1992 Reliability Based Design, McGraw Hill Publication

Sneddon, Gary 2003 Review of RBD Reports, Department of Mathematics,

Memorial University of Newfoundland, St. John’s, Newfoundland, Canada

Sneddon, Gary 2004 Personal Communications, Department of Mathematics,

Memorial University of Newfoundland, St. John’s, Canada

Reliability Based Calibration\Section 8 8-2 SCB2 WG-07, August 2008

9.0 APPENDIX
9.1 Example Problem (refer to Figure 3.13)

This section will provide the user a step-by-step procedure for analyzing the data
following the flow diagram in figure 5.6. Each step is shown clearly so that the user
can understand the calibration process in determining the foundation strength factor
for a specific foundation dataset. The user is still required to use some judgment in
selecting a particular option or a combination of options to ensure that the foundation
strength factor is not only conservative but also optimum. Details of the calculations
are presented here. All calculations are based on 90% confidence level (10%
exclusion limit). Figure 9.1 presents the three point scattered test data along a
transmission line route (section 3.6.2, reproduced figure 3.11). Figure 9.2 presents
the flow diagram (reference figure 5.6) which will be followed in computing the
foundation strength factor for this dataset. Table 9.1 presents the various parameters
that need to be computed to determine the foundation strength factor. A summary of
the methodology is presented in two steps.

Scattered Test Data along a Route

700

600
Test Capacity (kN)

500

400

300
450 degree
200 line

100

0
0 100 200 300 400 500 600

Geotechnical Design Capacity (kN)

Figure 9.1 - 3-point Scattered test data plot

In the first step, a routine linear regression analysis is carried out by minimizing the
residual errors across the all data points. Underlying assumption of the linear
regression analysis is that the regression considers an intercept and the variance of
residuals is constant. To ensure that the intercept can be neglected, a null hypothesis
is tested and a decision is made either to include or exclude the intercept. In the
second step, residual error values are plotted to test the assumption of constant
variance of residuals. If this is not validated, then an appropriate weight is applied to
each data point and the regression analysis is carried out on the transformed data,
with or without the intercept. Table 9.2 presents the necessary equation numbers for
four cases referred in Table 5.2.

Reliability Based Calibration\Section 9 9-1 SCB2 WG-07, August 2008

 Flow Diagram For Computing Foundation Strength Factor
Tabulate Data Properties for the
Problem, xi (geotechnical design
capacity) & yi (test capacity) values

yi
Compute mi =
xi

Data Plot, xi vs. yi

Evaluate m, sm and vm From
Equations (4.3), (4.4) & (4.5)
.

Test 1: Low Correlation

Eq. 4.1,
Table 4.2

High or Moderate
Correlation
Regression Analysis (Figure 5.2)
Slope (5.2), Intercept (5.3), Estimated Test Capacity
Value (5.5), and Variance of Residuals (5.6)

Slope &
Intercept Test 2: Slope only
Eq. 4.14
5.11

Test 3: Test 3:
Chk Figs. 5.4(a) Chk Figs. 5.4(a)
& 5.4(b) & 5.4(b)

Case A Case B Case C Case D *

(Eqn. 5.10) (Eqn. 5.14) (Eqn. 5.19) (Eqn. 5.24)

Calculate Foundation
Strength Factors to Provide
Conservative Values -
Equation 5.12
Test 1: Test for Correlation END
__
Test 2: Simultaneous Test for Intercept = 0.0 and Slope = m
Test 3: Visual Comparison to Test Constant Variance of
Residuals vs. Non Constant Variance of Residuals
*Constant COV - Special Case
(Residual values banded around zero line or "fan
- CIGRE (2002)
shaped" - use judgment)

Figure 9.2 - Flow diagram (figure 5.6, reproduced)

Reliability Based Calibration\Section 9 9-2 SCB2 WG-07, August 2008

9.2 Calibration Methodology (Summary)

Table 9.1: Methodology to Follow the Flow Diagram

Items Parameters Sections Equation

Numbers
Data Properties for the Geotechnical Design §3.2 to §3.4
Problem Capacities, xi

Test capacities, yi §3.5 to §3.7

Calculation of mean x , s x & vx §4.1 Use equations (2.1)

value, standard to (2.3) replacing
deviation and coefficient Ri and R by xi
of variation xi and
(or yi ) and x (or y )
yi values y, s y & v y §4.1
respectively.

Compute mi values; y §4.2 4.2

mi = i
xi

Compute the statistics of m , sm, & vm §4.2 4.3,4.4 and 4.5

mi values

Linear regression §4.1 4.1

analysis- ∑ ( xi − x )( yi − y )
¾ correlation value, r=
(n − 1) s x s y
r2
)
¾ residual error ε i = yi - yi §5.2.1 5.1 and 5.5
value, ε i

¾ variance of n §5.2.1 5.6

)
residuals, s xy 2: ∑ wi ( yi − yi )2
s xy 2= 1
n−2

Test 1 on correlation value, r2 according to Table 4.2

Test 2 on intercept, “a” according to Equation (5.11) and Table 5.1

Test 3 on variance of the residuals of errors, ε i according to Figures 5.4 a & b

Reliability Based Calibration\Section 9 9-3 SCB2 WG-07, August 2008

 Table 9.2: Reference Equation Numbers for Various Cases (Table 5.2)

Case ν wi b a ŷ (1) s xy y p (analogous to φ F

RC ) for a single
geotechnical
design capacity,
x0 ( or Rn )
A n-2 1.0 (5.2) (5.3) (5.5) (5.6) (5.10) (5.12)
B n-2 1 (5.2) (5.3) (5.5) (5.6) (5.14) (5.12)
xi 2
C n-1 1.0 (5.15) 0.0 (5.17) (5.16) (5.19) (5.12)
D n-1 1 (5.20) 0.0 (5.22) (5.21) (5.24) (5.12)
xi 2
(1) yˆ is yˆ i at xi = x0

9.3 Table 9.3- Scattered test data (Table 3.6 reproduced)

Table 9.3- Scattered test data

Foundation Capacity (kN) Site Site Site

A B C
Geotechnical design capacity 379 439 498
Test Capacity (at 10mm displacement) 450 540 680
Test Capacity 1.19 1.23 1.37
m=
Geotechnical Design Capacity

9.4 Compute mi values – (use equation 4.2)

n
450 540 680
m = (∑ mi ) / n = [ + + ] /3 = [1.19 +1.23+1.37]/3 = 1.261
1
379 439 498

Next we calculate s m and v m as using equations (4.3) and (4.4)

1 n 1
2
sm = ∑
n −1 1
(mi − m ) 2 =
3 −1
[(1.19 − 1.261) 2 + (1.23 − 1.261) 2 + (1.37 − 1.261) 2 ]

=0.009
s 0.093
v m = rm = = 0.0737
m 1.261

Reliability Based Calibration\Section 9 9-4 SCB2 WG-07, August 2008

9.5 Correlation check (Test No. 1- equation 4.1 and replace y by R)

To do the following calculations use equations (2.1-2.3);

n
R = (∑ Ri ) / n = (450+540+680)/3 = 556.7 kN
1
1 n 1
sR 2 = ∑
n −1 1
( Ri − R ) 2 =
3 −1
[(450 − 556) 2 + (540 − 556) 2 + (680 − 556) 2 ]

= (115.9)2 kN

s 115.9
vR = R = = 0.208
R 556.7

Similarly the mean, standard deviation and the coefficient of variation of x-Values
(Geotechnical design capacity) are x = 438.7 kN, s x =59.5 kN and v x = 0.136. The
correlation coefficient r , value is calculated directly from equation (4.1).
n
1 /(n − 1)[∑ ( xi − x )( yi − y )
r= 1
sxs y
1/ 2[(379− 438.7)(450− 556.7) + (439− 438.7)(540− 556.7) + (498− 438.7)(680− 556.7)]
=
59.5x115.9
= 0.992

Therefore, the correlation value r 2 = 0.983 is high and the linear association
between the test capacity y and the geotechnical design capacity x is strong (table
4.2). Therefore, the dataset passed the Test 1 according to the Flow diagram (Figure
9.2).

9.6 Regression Analysis (variance of residuals is constant)

Table 9.4 - Section 5.2.1 (Case A, x = 438.7 kN, wi = 1.0 *)

Geotechnical Damage
Test number design Limit wi ( xi − x ) yi wi ( xi − x ) 2
capacity, xi Capacity, yi
1 379 450 -26850 3560
2 439 540 180 0.11
3 498 680 40347 3520
Sum = ∑ 13676 7080
* For basic regression analysis, the variance of residuals is always constant unless it
is proven otherwise. Therefore, wi = 1.0

The slope b, is calculated from equation (5.2)

Reliability Based Calibration\Section 9 9-5 SCB2 WG-07, August 2008

 b=
∑ wi ( xi − x ) yi =
13676
=1.93
r
∑ wi ( xi − x )2 7080

The intercept is calculated from equation (5.3)

a = y − bx

a = 556.7- 1.9316x 438= -290kN

)
The estimated mean test capacity yi values are obtained based on regression
analysis from equation (5.5)
)
yi = a + b( xi ) = -290 + 1.9316 xi

Table 9.5 – Variance of residuals calculation ( wi = 1.0 )

Geotech
Damage Square of
nical )
Test design
Limit yi = a + b( xi ) Residual Residuals
Number Capacity = ( yˆi − yi ) xi 2
capacity = ( yˆ i − yi ) 2
yi
xi
1 379 450 441.4 8.58 73.65 143641
2 439 540 557.3 -17.31 299.65 192721
3 498 680 671.3 8.72 76.17 249004
Sum 1.7e-13 449.49 584366
= ∑
n
)
∑ wi ( yi − yi )2 [(450 − 441) 2 + (540 − 557) 2 + (680 − 671) 2 ]
s xy 2 = 1 = = 449.49;
n−2 (3 − 2)

Therefore, s xy = 21.1;

Reliability Based Calibration\Section 9 9-6 SCB2 WG-07, August 2008

 3-Point Scattered Test Data (Regression
Analysis) y = 1.9316x - 290.64
r^2 = 0.9833

600
Test Capacity (kN)

400

200

0
0.0 200.0 400.0 600.0
Geotechnical Design Capacity (kN)

Figure 9.3 - Linear regression plot

9.7 “Zero Intercept” Check – [Test No. 2 - equation 5.11]

3-Points Scattered test data

Regression line
600 Original Data
with slope
Test Capacity (kN)

Points
only,1.26 (m-bar)
400
Regression Line With
200 Intercept, a=-290 and
slope, b=1.93
0
0.0 200.0 400.0 600.0
-200

-400
Geotechnical Design Capacity (kN)

Figure 9.4 - Linear regression plot (with intercept and without intercept)

F*= [n * (a − 0.0) 2 + 2 * n * x * (a − 0.0)(b − m ) + (∑ xi )(b − m ) 2 ] / (2 * s xy 2 )

2

where

Reliability Based Calibration\Section 9 9-7 SCB2 WG-07, August 2008

 n= 3 (sample size),
a = -290kN (intercept),
b = 1.93 (slope),
m =1.26 (slope based on “zero intercept”)
∑ xi 2 = 584366
s xy = 21.1kN (standard deviation of the error in the linear regression analysis)
x = 438kN (mean value of 3 geotechnical design capacities)
Fα ,2, n − 2 = F0.10,2,1 =49.50 (Table 5.1)

The three terms in equation 5.11 are calculated below.

Table 9.6 - Terms in equation (5-11)

First Term-
n (a-0.0)^2
Second Term-
2n x (a-0.0)(b- m ) ∑
Third Term- ( xi 2 )(b − m ) 2 F* = ∑ Sum
3(-290.64)^2= 2*3*438(-290-0.0)(1.93-1.26) = - 584366*(1.93-1.26) 2 3378
253415 483114 = 233078

F* =
∑ Sum =
3378
= 3.78 < Fα ,2, n − 2 = 49.50 (Table 5.1)
2 * s xy 2 2 * 212

Therefore, we accept the Hypothesis that the intercept is zero and the slope is equal
to m = 1.261 ( m - model, Section 4); All results are summarized in Table 9.5.

Table 9.7- Data statistics for 3 – point scattered test data problem

n R2 x sx y sy a b s xy ∑ xi 2 F-value m vm

3 0.983 438 59 556 116 -290 1.93 21 584366 3.78 1.26 0.0737
< 49.50

Note:

Similar calculations can be done for ESB data and Table 5.4 can also be
produced following the above steps.

Reliability Based Calibration\Section 9 9-8 SCB2 WG-07, August 2008

9.8 Variance of residuals-Test (Test No. 3- Figures 5.4a and 5.4b)

Residual Plot To Assess Constant Versus Non

Constant Variance of Residuals
15 Envelop to show how a
"cone shaped hat" can
10 be drawn around the
Residual Values (kN)

data for non constant

5
variance. In this case it
is not obvious because
0
the sample size is too
0.0 100.0 200.0 300.0 400.0 500.0small
600.0
-5

-10

-15

-20
Geotechnical Design Capacity Values (kN)

Figure 9.5 - Residual values versus Geotechnical design capacity

In this example we assume the variance of residuals is not fully non-constant

because of limited data point. In actual foundation design problem, the user will have
more data points and therefore, a more rational decision can be made with respect to
residual variation. For details refer to Section 5.0 of the report.

9.9 Predicted Test Capacity (sample size effect included)

CASE C ( wi = 1.0) - equations (5.15) – (5.19)

Table 9.8 - Variance of Residuals-Calculations

Geotechnical
Test
Test design
Capacity, wi ( xi ) yi wi ( xi ) 2 yˆi = bxi ( yˆ i − yi ) 2
Number capacity, xi
yi (kN)
(kN)
1 379 450 170550 143641 484 1155
2 439 540 230760 192721 560 425
3 498 680 338640 248004 638 1940
Sum 746350 584366 3520

Reliability Based Calibration\Section 9 9-9 SCB2 WG-07, August 2008

 Following section 5.3.4 and table 9.2, the slope, b is calculated as

b=
∑ wi ( xi ) yi =
746350
=1.277
∑ wi ( xi )2 584366

The standard error of estimate (SEE) is obtained from equation (5.17)

s xy 2 =
∑ wi ( yi − yˆi )2 =
3520
=1760
n −1 2

Therefore, s xy = 41.9

The predicted test capacity is computed from equation (5.19) for kα = tα ,ν (small
sample size n=3)

( x0 ) 2
yp = ŷ - tα ,ν * s xy [1+ ] 0.5
n
∑ ( xi )2 wi
1

For a sample size n=3, tα ,ν = 1.90 and is obtained from Table 2.2 (90% value for one
sided interval). Therefore, the second term in the above equation for x0 =200 kN
becomes

( x0 ) 2 200 2 0.5
P= tα ,ν * s xy [ ] 0.5 = 1.90 * 42 *[1+ ] = 82.39
n 584366
∑ ( xi )2 wi
1
Table 9.9 presents the data for four geotechnical design capacity values.

Table 9.9 - Predicted Test Capacity

Geotechni Estimated ( x0 ) 2 y p = ŷ -P
cal design Mean Test P= tα ,ν * s xy [ 1+ ] 0 .5
n (equation 5.19)
capacity,
x0
Capacity
Value
∑ ( xi )2 wi
1
yˆ = bx0 (second term in equation 5.19)
(first term in
equation 5.19)
200 255 82 173
400 511 90 421
600 766 101 665
800 1022 115 907

Reliability Based Calibration\Section 9 9-10 SCB2 WG-07, August 2008

 Scattered 3-point Test Data (Case C- Constant
variance of residuals, slope only)
1200
1120
1040
960
880
Test Capacity(kN)

800
720 Regression
640
Line Predicted Capacity -
560
480 90% Confidence
400 level
320
240
160
80
0
0.0 100.0 200.0 300.0 400.0 500.0 600.0 700.0 800.0 900.0 1000.0

Geotechnical Design Capacity (kN)

Figure 9.6 - Limit of Prediction Interval (case C)

1
CASE D ( wi = ) – (equations 5.20 -5.24)
xi 2

Table 9.10 – Variance of Residuals Calculations

Test Geotechnical Test wi ( xi ) yi wi ( xi ) 2 yˆi = bxi wi ( yˆ i − yi ) 2

Number design Capacity,
capacity, xi yi (kN)
(kN)
1 379 450 1.19 1.0 478 0.0054
2 439 540 1.23 1.0 554 0.0010
3 498 680 1.37 1.0 628 0.0109
Sum 3.78 3.0 0.0173

The slope is calculated from Equation (5.20)

b=
∑ wi ( xi ) yi = m=
3.78
= 1.261;
∑ wi ( xi )2 3

The variance of residuals (standard error of estimate) is obtained from equation

(5.21)

Reliability Based Calibration\Section 9 9-11 SCB2 WG-07, August 2008

 s xy 2
=
∑ wi ( yi − yˆi ) 2
0.0173
==0.00865
n −1 2
s xy = 0.093; in this case s xy = s m (equation 5.21)

The “hat” on y-values represents the estimated mean test capacity and is obtained
from equation (5.21) as yˆ = m x0 where m is the slope. Therefore the coefficient of
s m 0.093
variation v m = = =0.0737. For a geotechnical design capacity value x0 =
m 1.26
200kN, the estimated mean test capacity value is obtained from equation (5.22) in
which m =1.26 and the estimated mean test capacity value is 252 kN. Equation
(5.24) provides

1 0 .5
yp = m x0 - tα ,ν sm x0 [ 1 + ]
n
1
Therefore, for x0 = 400kN, y p = {1.26 * 400- 1.90 * 0.0735* [1+ ]0.5 * 400} =
3
422.78kN

Table 9.11- Predicted Test Capacity

Geotechnic Estimated Mean 1 0 .5 y p = ŷ -P

Test Capacity P= tα ,ν * s m [ 1+ ] x0
al design n
Value (equation 5.24)
capacity, x0 Second term in equation
yˆ = m x0 (5.24)
First term in
equation (5.24)
200 252 41 211
400 504 82 422
600 757 122 635
800 1009 163 846

Reliability Based Calibration\Section 9 9-12 SCB2 WG-07, August 2008

 Scattered 3-Point Test Data-(Case D,Non Constant Variance of
residuals, Slope only)
Data Range 379kN to
1000
499kN
900
800
Regression Line
Test Capacity (kN)

700
600
500
400 Predicted Capacity
(90% Confidence
300
Level)
200
100
0
0.0 100.0 200.0 300.0 400.0 500.0 600.0 700.0 800.0 900.0 1000.0
Geotechnical Design Capacity (kN)

Figure 9.7 – Lower Limit of Prediction Interval (Case D)

9.10 Comparisons of foundation strength factor values (Case C and

Case D)

Sample size effect included

The foundation strength factor is obtained from equation (5.12)

yp
φF = ..
x0
For a geotechnical design capacity value x0 = 400 kN we obtained y p =423 kN
422.78
for CASE D; Therefore, φ F = =1.057
400

Table 9.12 - Comparison of foundation strength factor (Case C & Case D)

Geotechnical CASE C CASE D

Design Capacity Predicted Foundation Predicted Foundation
Value, x0 Test Strength Test Strength
Capacity Factor Capacity Factor
200 173 0.865 211 1.057
400 421 1.052 422 1.057
600 665 1.108 635 1.057
800 906 1.133 846 1.057

Reliability Based Calibration\Section 9 9-13 SCB2 WG-07, August 2008

 Comparison of Foundation Strength Factors for 10% Exclusion Limit
(Sample Size Considered)

tα,ν=1.90from
1.2
Foundation Strength

Table 2.2 for n=3

1
0.8
Factor

Constant φF
"Case C"
0.6
0.4 "Case D"
0.2 Variable φF

0
0 200 400 600 800 1000
Geotechnical Design Capacity (kN)

Figure 9.8 - Comparison of foundation strength factor

9.11 CIGRE (2002)

φ F = m [1- tα ,ν * vm ] = 1.261 (1-1.90 x 0.0737) = 1.084 (CIGRE, 2002);

φ F = m [1- tα ,ν * vm [1+1/n] 0.5] = 1.261 (1-1.90 x 0.0737[1.33] 0.5) = 1.057;

(CASE D)

Scattered test data

700

600
Test Capacity (kN)

500

400
φF =1.084
300 0
45
200
degree
100 line
0
0 100 200 300 400 500 600
Geotechnical Design Capacity (kN)

Figure 9.9 – Foundation Strength factor (Buckley, 1994 & CIGRE, 2002)

Reliability Based Calibration\Section 9 9-14 SCB2 WG-07, August 2008