You are on page 1of 111

Delft University of Technology

Department: Geoscience & Engineering

Correlating CPT data to stiffness


parameters of sand in FEM

Author:
Stef Engels

October 12, 2016


Delft University of Technology

Department: Geoscience & Engineering

Correlating CPT data to stiffness


parameters of sand in FEM

Graduation Committee:
Prof. dr. K.G. Gavin
Ing. H.J. Everts (Tu Delft)
Ir. J. Rindertsma (Van Oord)
Dr. ir. K.J. Bakker (Tu Delft)

Author:
Stef Engels
student id: 4095359

To be publicly defended on October 27, 2016


Summary

Making foundation designs is one of the major tasks of a geotechnical engineer. To achieve
such a design, knowledge of the soil stiffness is required. Information about the soil stiffness
can be obtained by either laboratory tests or in situ tests. Laboratory testing is time
consuming, costly and sample disturbance is inevitable. In situ tests are done directly
on the soil encountered on the site and therefore are a good representation of the in situ
soil state. CPT’s are the most performed in situ tests in geotechnical engineering. CPT’s
are used to evaluate the subsurface based on the mechanical response translated to cone
resistance and sleeve friction. Therefore it is desirable to find a correlation between soil
stiffness and CPT results.

A lot of research has been done to correlate CPT results, such as the cone resistance, with
the stiffness of sand. It has been found that these correlations are highly variable and site
specific. State parameters as consolidation state have a high influence on the soil stiffness
but are difficult to evaluate from CPT results only.

In this thesis the correlation between cone resistance and stiffness parameters for sand is
investigated based on a series of Zone Load Tests done at a site in Kuwait. In a Zone Load
Test a footing is loaded in steps and the settlement of the footing is measured. First an
analytical settlement analysis is done with existing methodologies which use correlation
between cone resistance and soil stiffness. The predicted settlements according to these
methods are compared with settlements measured in the field. Afterwards, a new site
specific correlation between cone resistance and soil stiffness is proposed using regression
analysis.

A verification of the new proposed correlation is done with the finite element program
PLAXIS 2D. The numerical calculations were done with the Hardening Soil model using
an axisymmetric approach. Multiple Zone Load Tests are simulated with PLAXIS 2D
where the input parameters of the Hardening soil model are obtained from the proposed
correlation. The numerical calculated settlement of the footing are in agreement with the
measurements in the field and therefore it can be concluded that the correlation found for
this site is valid. The general application of the proposed correlation is not confirmed.

The research done in this thesis is related to direct settlement. Time dependent be-
haviour is excluded but could be of significant influence. Carbonate sands, as encountered
at the site, are sensitive for particle crushing. Particle crushing can lead to creep effects
and therefore it is advised to perform Plate Load Tests to get better insight in the creep
behaviour in the sand fill.

i
Acknowledgements

In February I started my research at Van Oord as a graduate student in Rotterdam. I am


grateful that I have got the opportunity to work in an organisation with big and interesting
projects and friendly colleagues. During the process I had a lot of help from my daily
supervisor ir. Rindertsma. When I was stuck in my research he always had ideas to set
me back on track again. Furthermore I would like to thank ir. Karreman who guided me
through the project and who gave me the opportunity to start at Van Oord.

Furthermore I would like to thank the rest of committee for their involvement in my
research. During the process dr. ir. Bakker showed a lot of interest in the subject and was
always full enthusiasm to help me improve my research. I would like to thank ing. Everts
for his faith in my abilities and sharing his experience with me. And finally I owe much
gratitude to prof. dr. Gavin. He recently joined the staff at Delft University of Technology
and he was very helpful right away. He always makes room in his schedule to get the best
out of his students.

Last but not least, I would like to thank my family and friends. My family always
supported me during my years as a student. Their love and support kept me going through
the years. I would like to thank my grandmother for her faith, love and financial support.
Furthermore I would like to thank all the friends I made during my studying period in
Delft. We share great moments and great times are still ahead. Finally I want to thank
Nienke for her love and understanding during my period as a student.

Stef Engels
October 2016

ii
iii
Contents

Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v

1 Introduction 1
1.1 Research context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Research project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Thesis outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Project information 4
2.1 Site overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Soil conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

3 Literature study 8
3.1 The Cone Penetration Test . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.1.1 The CPT procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.1.2 Results and interpretation . . . . . . . . . . . . . . . . . . . . . . . . 9
3.2 Theoretical background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.2.1 Discussion of soil moduli . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.2.2 Settling behaviour of soil . . . . . . . . . . . . . . . . . . . . . . . . 13
3.2.3 Compressive behaviour of sand . . . . . . . . . . . . . . . . . . . . . 14
3.2.4 Stress distribution underneath a shallow foundation . . . . . . . . . 15
3.2.5 Elastic settlement with constant Young’s modulus . . . . . . . . . . 16
3.2.6 Settlement analysis of a shallow foundation . . . . . . . . . . . . . . 17
3.3 CPT based methods for settlement of a shallow foundation . . . . . . . . . 18
3.3.1 De Beer and Martens (1957) . . . . . . . . . . . . . . . . . . . . . . 18
3.3.2 Schmertmann method (1978) . . . . . . . . . . . . . . . . . . . . . . 19
3.3.3 Modification Schmertmann suggested by Peck et al. (1996) . . . . . 22
3.3.4 Robertson (1990) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.4 Soil models for sand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.4.1 Mohr-Coulomb model . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.4.2 The Hardening Soil model . . . . . . . . . . . . . . . . . . . . . . . . 31
3.4.3 The Hardening Soil model with small-strain stiffness . . . . . . . . . 33

4 Zone Load Test procedure 35


4.1 Test set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.2 Loading procedure and test results . . . . . . . . . . . . . . . . . . . . . . . 35
4.3 Test uncertainties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.4 Interpretation and assumptions . . . . . . . . . . . . . . . . . . . . . . . . . 38

5 Analytical settlement analysis 39


5.1 Zone Load Test Criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
5.2 Processing the CPT data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
5.3 Overview of the procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.4 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.4.1 Comparison different site locations . . . . . . . . . . . . . . . . . . . 44

iv
5.4.2 Normality of the results . . . . . . . . . . . . . . . . . . . . . . . . . 45
5.4.3 Influence of the CaCO3 content . . . . . . . . . . . . . . . . . . . . . 49
5.4.4 Difference between analytical approach and reality . . . . . . . . . . 50

6 Site specific correlation 52


6.1 Compare back calculated secant modulus Es with ERob . . . . . . . . . . . 52
6.2 Regression analysis using one layer . . . . . . . . . . . . . . . . . . . . . . . 53
6.2.1 Correlation between qc and Es . . . . . . . . . . . . . . . . . . . . . 53
6.2.2 Correlation between Qtn and Es . . . . . . . . . . . . . . . . . . . . 56
6.2.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
6.3 Regression analysis using suggested layers . . . . . . . . . . . . . . . . . . . 57
6.3.1 Correlation between qc,weigthed and Es . . . . . . . . . . . . . . . . . 57
6.3.2 Correlation between Qtn,weigthed and Es . . . . . . . . . . . . . . . . 60
6.3.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
6.4 Regression analysis using 600 layers . . . . . . . . . . . . . . . . . . . . . . 62
6.4.1 Correlation between qc,weighted and Es . . . . . . . . . . . . . . . . . 62
6.4.2 Correlation between Qtn,weighted and Es . . . . . . . . . . . . . . . . 63
6.4.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
6.5 Interpretation of Es . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

7 Numerical verification with PLAXIS 2D 67


7.1 Used correlation for verification . . . . . . . . . . . . . . . . . . . . . . . . . 67
7.2 Modelling approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
7.3 Parameter determination . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
ref
7.4 Relating Es with E50 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
7.5 Overview of the numerical validation . . . . . . . . . . . . . . . . . . . . . . 73
7.6 PLAXIS calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
7.6.1 PLAXIS model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
7.6.2 Loading procedure and output . . . . . . . . . . . . . . . . . . . . . 76
7.6.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
7.7 Influence of ZLT procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
7.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

8 Conclusions and recommendations 83


8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
8.2 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
8.3 Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

Appendices 90
A Functions in Mohr-Coulomb model . . . . . . . . . . . . . . . . . . . . . . . 90
B Fitted nSBT chart Robertson . . . . . . . . . . . . . . . . . . . . . . . . . . 91
C Settlement analysis ZLT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
D Shapiro-Wilk test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
E Soil specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
F Comparison back calculated stiffness with Robertson stiffness . . . . . . . . 98
G Calculating standardized residuals . . . . . . . . . . . . . . . . . . . . . . . 99
Nomenclature

Symbols

Symbol Units Description


RC % Relative compaction
γd,f ield kg/m3 Dry field density
γd,max kg/m3 maximum dry density
Dr % Relative density
qc MPa Cone resistance
qt MPa Corrected cone resistance
q¯c,i MPa Average cone resistance in layer i
qc,weighted − Weighted cone resistance
fs MPa Sleeve friction
u MPa Pore pressure
u0 MPa Hydrostatic pore pressure
u2 MPa Measured pore pressure
Rf − Friction ratio
E kPa Young’s modulus
Es kPa Secant Young’s modulus
Eur kPa Unloading reloading Young’s modulus
Et kPa Tangent Young’s modulus
ref
E50 kPa Secant stiffness in standard drained triaxial
test
ref
Eoed kPa Tangent stiffness for primary oedometer
loading
Eur ref kPa Unloading/reloading stiffness
ERob kPa Stiffness obtained with Robertson method
s mm Settlement
si mm Settlement of layer i
se mm Elastic settlement
screep mm Creep settlement
v − Poisson’s ratio
B m Foundation width
L m Foundation length
Iρ − Settlement influence factor
Iz − Strain influence factor
0
Iz − Corrected strain influence factor
C − Compressibility coefficient
σv0 MPa In situ total stress
0
σv0 MPa In situ effective stress
0
σ0 kPa Effective overburden pressure
∆σ kPa Increase in pressure
H m Layer thickness
Cc − Compressibility index
e0 − Initial void ratio

vi
qb kPa Unit load acting on the base
C1 − Time dependent factor
C2 − Depth dependent factor
∆z m Thickness of a layer
n − Stress component
Qtn − Normalized cone resistance
Qtn,weighted − Weighted normalized cone resistance
Q¯tn,i − Average normalized cone resistance in layer
i
Fr − Normalized friction ratio
Bq − Normalized pore pressure
pa kPa Atmospheric pressure
Ic − Soil behaviour type index
Ic,rw − Soil behaviour type index (Robertson and
Wride)
Vs m/s Shear wave velocity
Vs1 m/s Normalized shear wave velocity
αvs m/s2 Shear wave velocity factor
ρ kg/m3 Mass density
G kPa Shear modulus
G0 kPa Small strain shear modulus
Gur kPa Unloading reloading shear modulus
KG − Small strain shear modulus number
α − Significance level
αG − Shear modulus factor
qult kPa Ultimate bearing capacity
KE − Young’s modulus number
αE − Young’s modulus factor
ϕ ◦ Friction angle
c kPa Cohesion
ψ ◦ Dilatancy angle
σt kPa Tensile strength
K0 − Coefficient of lateral earth pressure
m − Power for stress level dependency of stiff-
ness
vur − Poisson’s ratio for unloading/reloading
pref kPa Reference stress
K0nc − The value for K0 for normal consolidation
Rf − Failure ratio qf /qa
γ − Shear strain
γ0.7 − Reduction parameter when G reduces to
0.7G0
µ̂ − Estimated mean
σ̂ − Estimated standard error
W − Weighting factor
D m Diameter
De m Equivalent diameter
zbot,i m Bottom coordinate of layer i
ztop,i m Top coordinate of layer i
Abbreviations

Abbreviation Meaning
CPT Cone Penetration Test
ZLT Zone Load Test
UCSC Unified Soil Classification System
SBT Soil Behaviour Type
SBTn Normalized Soil Behaviour Type
CSL Critical State Line
PSD Particle Size Distribution
Chapter 1

Introduction

1.1 Research context

In geotechnical engineering, providing accurate predictions of foundation settlements is


one of the main challenges. The function of a foundation is to transfer loads from a given
structure to the subsurface. The compressive loads imposed on the subsurface will lead
to a permanent deformation of the soil, which is referred to as settlement. In order to
protect the integrity of the structure settlements should be kept within limits. It is the
responsibility of the geotechnical engineer to make an adequate foundation design in which
the safety of a structure is guaranteed. The requirements regarding to settlements can be
found in the Eurocode 7.

The earth’s subsurface is comprised of many different materials. Every specific type
of soil or rock has its own engineering properties with respect to stiffness and strength
parameters. Another difficulty which is typical for soils is that they are anisotropic and
heterogeneous. These factors indicate that making a geotechincal design is quite a chal-
lenging operation.

Fortunately experience in making a reliable design was gained over the last decades.
Designs are typically based on laboratory tests and in situ tests. For laboratory testing,
soil samples are obtained from the site and tested in the laboratory. Testing can be done
under controlled circumstances, but the soil samples may not represent the actual in situ
conditions due to sample distortions as stress relief. In situ test are done directly on the
soil as encountered in the field. Therefore these test are highly representative as the soil is
tested in the in situ state. However, a large amount of in situ test can significantly increase
costs. The cost increase are due to expensive equipment and skilled personnel that has to
be brought to the site.

The most used in situ test is the Cone Penetration Test (CPT). With this test a cone
penetrates the subsurface at a certain rate and measures the cone resistance, the sleeve
friction and the pore pressure. A CPT is used to classify the soil layering and get a better
understanding of the subsurface. CPT’s are widely used because the are cost-effective and
provide valuable soil information. Because of the availability of CPT’s, it would be useful
to develop a methodology to make a geotechnical design based on CPT results.

In the past research was done on making settlement analysis based on CPT results
(De Beer, E. Martens, 1957) (Schmertmann et al., 1978) (Peck et al., 1996) (Robertson,
1990). However, the methods proposed to correlate CPT results to soil stiffness differ and
the results are not consistent.

1
Correlating CPT data to stiffness parameters of sand in FEM

1.2 Research project


As discussed in the previous section, accurate derivation of stiffness parameters from CTP’s
would provide valuable information for foundation design. The idea of making an accurate
geotechnical design based on CPT’s is cost-effective and time efficient. To this end the
main research question can be formulated as:

Is it possible to predict stiffness parameters of sand with reasonable accuracy based on CPT
results?

A large amount of Zone Load Tests (ZLT’s) are performed on a hydraulic fill on a site in
Kuwait. These tests measures the settlement of a 3 m x 3 m footing on different locations.
This thesis will use the data from these ZLT’s to investigate the possibility to develop
a correlation between CPT results and stiffness parameters of sand. The results will be
verified by modelling ZLT’s in finite element software (PLAXIS 2D) and compare the
results with the measurements in the field.

To find a satisfying answer for this main research question, some other questions need to
be answered:

What kind of factors play an important role in the compressibility of sand?


How does the layering of the deposit affect the results?
Does the testing methodology influence the results?
Is it possible to distinguish creep settlements during a ZLT?
Is it possible to predict load-settlement behaviour based on CPT parameters?
Are the obtained results representative for other sites?

1.3 Thesis outline


The structure of this thesis can be summarized in seven main categories:

1. Site specific information


2. Theoretical background
3. Test procedure
4. Analysis of existing methods
5. Site specific correlations
6. Numerical verification with PLAXIS 2D
7. Conclusions and recommendations

In Chapter 2 the project at the site in Kuwait will be discussed. An overview of the site is
provided to give an idea of the locations of the tests and the size of the site. Chapter 3 starts
with the theoretical background with respect to foundation design. Afterwards a selected
amount of existing methods will be discussed that directly use CPT data for settlement
analysis. To conclude this chapter existing soil models for modelling the behaviour of
sandy soils will be presented. Chapter 4 evaluates the test procedure of a ZLT and the
interpretation of the output. This Chapter concludes with an overview of the assumptions
that are used for further analysis. In Chapter 5 the methods presented in the literature
study are used to make a settlement prediction of a selected amount of ZLT’s. A settlement
analysis is done and compared with the measured settlements. This chapter provides an
insight in how the existing methods perform at this specific site. In Chapter 6 a regression

2 Chapter 1 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

analysis will be done for a selected amount of representative ZLT’s. The objective in
this chapter is to obtain a site specific correlation between the cone resistance and the
secant Young’s modulus of sand. In Chapter 7 the representative ZLT’s are modelled
using FEM software (Plaxis 2D). The input parameters are based on the correlations that
are developed in Chapter 6. The objective to verify the correlations that are presented
earlier. In Chapter 8 the main conclusions are summarized and recommendations for future
research are made.

Chapter 1 Stef Engels 3


Chapter 2

Project information

2.1 Site overview


In Kuwait, a sand layer was constructed to serve as a suitable foundation layer for an oil
refinery. The site location is indicated in figure 2.1. At parts of the sites sabkha soil is
encountered. This type of soil usually consists of soft weak silts and clays cemented with
salts. The mechanical behaviour of sabkha can be compared with clay. In some parts
of the site the constructed sand layer is constructed on top of a sabkha layer. In other
areas the fill is constructed on a natural sand deposit. To reduce the settlements in the
subsurface, ground improvement techniques are applied. Several ground improvement
methods are used for the densification of the material. The improvement methods used
are Dynamic Compaction, Dynamic Replacement and Roller Compaction. Where sabkha
layers are present, sand columns can push through this layer with the Dynamic Compaction
technique. The arching effect between the sand columns needs to reduce the compression
of this sabkha layer. A closer view of the site with elevation levels of the fill is provided in
figure 2.3. The numbers in the figure indicate the height of the original ground level with
respect to the new ground level.

Figure 2.1: Location of the site

After the ground improvement is finished, several Zone Load Tests (ZLT’s) are performed
to measure the settlement of a 3 m x 3 m square footing. In the ZLT procedure, a concrete
footing is loaded in steps and the settlement of the footing is recorded during the process.
The settlements are measured on four locations on the footing. The ZLT’s are performed

4
Correlating CPT data to stiffness parameters of sand in FEM

over the entire site. The results of these tests are used to evaluate if the settlement
requirements (less then 25 mm) are met. Some ZLT’s run for an entire month to get
information about the long term settlement (creep) behaviour.

2.2 Soil conditions


After the compaction works are finished, Field Density Tests (FDT’s) are done. The
results of these tests are presented in table 2.1. With this test the maximum dry density
is determined in the laboratory. This can be done with for instance the Proctor test.
The maximum dry density is compared with the field density and this results in a new
parameter called the relative compaction (RC). The RC is defined as:
γd,f ield
RC = · 100% (2.1)
γd,max
Where:
RC: Relative compaction
γd,f ield : Measured dry field density after compaction
γd,max : Maximum dry density

Research by Gomaa and Abdelrahman (2007) concluded that there is a very accurate
positive correlation between relative compaction and relative density. Sands from 20
different sites were tested and the results are presented in figure 2.2. According to this
research the relative compaction is related to the relative density (Dr ) according to:

Dr = 5.5 · RC − 4.47 (2.2)

Figure 2.2: Correlation between relative compaction and relative density (Gomaa and
Abdelrahman, 2007)

A relative compaction between 95% and 96% corresponds with a relative density between
76% and 81%. Therefore it can be concluded that the compacted sand encountered on the
site can be classified as dense sand.

Chapter 2 Stef Engels 5


Correlating CPT data to stiffness parameters of sand in FEM

Table 2.1: Results of FDT tests

Parameter Unit FDT 1 FDT 2 FDT 3 FDT 4 FDT 5 Average


Wet density g/cm3 1.815 1.786 1.836 1.801 1.826 1.813
Dry density g/cm3 1.715 1.714 1.719 1.732 1.720 1.720
Max. dry density g/cm3 1.803 1.803 1.797 1.803 1.803 1.802
Relative compaction % 95.1 95.1 95.7 96.1 95.4 95.5

6 Chapter 2 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

3,5 km
7 km

Figure 2.3: Site location with elevation levels

Chapter 2 Stef Engels 7


Chapter 3

Literature study

In this chapter a theoretical background is given and previous research is summarized.


The main purpose of the literature study is to evaluate the available correlation methods
and the theoretical background that is used. Key is to understand why certain methods
perform better than others and learn what factors play a dominant role in correlating
CPT parameters to soil properties. In the first section of this literature study the general
CPT procedure is discussed. In the next section some theoretical background is given.
Afterwards available correlation methods will be evaluated. Some earlier developed methods
will be analysed first, because it is useful to see how correlation methods have developed
over time. Later on a more advanced method is explained. To finish the literature study
different soil models that can be used to model sand are briefly explained.

3.1 The Cone Penetration Test


The Cone Penetration Test (CPT) is the most widely used field test in geotechnical
engineering and every geotechnical engineer should be familiar with the interpretation of
the results of this test. In this section the procedure and specifications of the CPT are
briefly explained.

3.1.1 The CPT procedure


In a CPT a cone, connected by rods, will be pushed into the soil with a certain constant
penetration rate. During the penetration continuous measurements are made of the
penetration resistance of the cone and the sleeve. When using a piezocone, measurements
of the pore pressure are registered as well. The definitions above are visualized in figure
3.1. Standard electronic cones have a 60 degrees apex angle and a cross-sectional area of
either 10 cm2 of 15 cm2 (Robertson and Cabal, 2015). Typical penetration rates are 1 to 2
cm/s. The standard length for a rod is one meter.

Rod

Friction
sleeve

Load cell

Apex angle
Cone

Diameter

Figure 3.1: Terminology of CPT components

8
Correlating CPT data to stiffness parameters of sand in FEM

For performing a CPT, pushing equipment is required. The pushing equipment on land
generally consists of specially built units which can be truck or track mounted. A drilling
rig can also be used as pushing equipment. When performing CPT’s on the seabed, different
types of equipment are required. In shallow water floating or jack-up barges can be used,
where in deeper water seabed systems are used.

3.1.2 Results and interpretation


As stated before the CPT provides a continuous profile of the cone resistance (qc ), the
sleeve friction (fs ) and the pore pressure (u) if a piezocone is used. The ratio between fs
and qc determined as a percentage provides another parameter called the friction ratio
(Rf ):
fs
Rf = · 100% (3.1)
qc
A typical CPT output profile is given in figure 3.2. In these results the readings of two
individual CPT’s are represented in the same graph. One of the main advantages of the
CPT is the possibility to determine the soil stratigraphy based on the output parameters
qc , fs and Rf . One must keep in mind that soil classification based on a CPT relates to
the mechanical response of the material and therefore is not necessarily the same as the
soil classification based on the USCS (Unified Soil Classification System). In the USCS
the soil is classified based on sieving results and Atterberg limits. Various methods for
soil classification based on CPT’s exist (Begemann, 1965) (Schmertmann et al., 1978)
(Robertson, 1990) (Ramsey, 2002). Some methods will be discussed in detail later in this
literature study.

Figure 3.2: Typical result of two CPT’s (Robertson, 2009)

Chapter 3 Stef Engels 9


Correlating CPT data to stiffness parameters of sand in FEM

Before going into detail to much, some general statements can be made about the relation
between the CPT output parameters and the soil type (Lunne and Robertson, 1997):

• Gravelly sand – Very low friction ratio and very high cone resistance
• Sand - Low friction ratio and high tip resistance
• Sandy silt or silty sand – Moderate friction ratio and moderate cone resistance
• Clays – High friction ratio and low tip resistance
• Peat and organic clays – Very high friction ratio and very low cone resistance

The rules of thumb summarized above are very general and should not be used without
supporting data. To provide an indication for the order of magnitude, friction ratios of 1-2
% are considered to be very low, whereas friction ratios of 10-12 % are considered very
high. With the cone resistance measured values of 1-2 MPa are in the low category and
values of 8-10 MPa are in the high category.

When interpreting CPT data there is another effect that influences the values of the
readings. During the penetration the cone tip induces passive soil failure (figure 3.3). The
recorded cone resistance is an average value across the influenced zone. When the cone
is penetrating towards another layer as indicated in figure 3.3, caution should be taken
by interpreting the readings. The instrumental cone senses soil resistance of about 21 cm
ahead of the advancing cone (Rogers, 2004). Due to this effect soil layers may be either
stiffer or softer than the CPT results indicate.
Zone of Distubance

Passive failure
zone due to
advancing
cone tip

Less stiff
layer

Stiffer
layer

Figure 3.3: Passive soil failure during a CPT

3.2 Theoretical background


3.2.1 Discussion of soil moduli
In geotechnical engineering the soil modulus is a complex parameter. Often the Young’s
modulus (E) parameter is used as the deformability parameter for soil. One must keep in
mind that the official term of the Young’s modulus relates to linear elastic behaviour for a
continuum material. Since soil is not a linear elastic material, the soil modulus is not the
equal to the slope of the stress strain curve. This indicates that the soil modulus is stress
dependent. Looking at a stress strain curve of a dense sand in a triaxial test, many soil
moduli can be obtained (figure 3.4). Furthermore an unloading reloading modulus can be

10 Chapter 3 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

distinguished. The modulus during unloading reloading will be significantly higher than
with normal loading. In figure 3.4 the different moduli are deliberately denoted with the S
instead of E. The slope in the stress strain diagram is not the same as the soil modulus.
An exception to this statement is the case where the confining stress in the triaxial test is
zero.

Sinitial
Stress Stangent

Ssecant

0 Strain

Figure 3.4: Stress-strain curve obtained from a triaxial test of dense sand

When drawing a slope from the origin to a arbitrary point in figure 3.4, a secant slope
Ss is obtained and the secant modulus Es can be obtained from it. This is the modulus
that can be used to predict the settlement of a footing which is loaded for the first time
on a normally consolidated deposit. When the same footing is loaded on a deposit which
has been subjected to higher loads in the past (overconsolidated), the unloading reloading
modulus Eur should be used. The tangent slope St relates to the tangent modulus Et and
can be used to determine incremental movement when an incremental load is applied. An
example is an expansion of a high existing building with an extra level. This illustrates
that the soil modulus which should be used depends on the application.

Influence of state factors


The following state factors have influence on the soil stiffness.

Packing of the particles: If the particles are packed close to each other, the value of the
modulus tends to be high. The state can be measured by means of dry density and porosity.

Arrangement of particles: This refers to the structure of the soil. It must be noted
that although the dry density may be the same for two deposits, the structure can de
different. Coarse grained soils for example may have a dense or a loose structure and fine
grained soils may have a dispersed or flocculated structure. Soils that are well graded
behave stiffer than poorly graded soil since the voids are filled with finer particles.

Water content: The water content plays an important role because it has a direct in-
fluence on the effective stress state of the soil. At low water contents (especially in fine
grained soils) the water can bind the particles through suction. This will lead to an
increased value for the modulus. However, because of the lubrication effect of water, the

Chapter 3 Stef Engels 11


Correlating CPT data to stiffness parameters of sand in FEM

compaction of coarse grained soils with very low water content is less efficient than the
compaction with higher water content (Briaud, 2001). This would lead to a lower modulus
for lower water contents. There is an optimum value for the modulus as the water content
increases.

Stress history: When the soil has been subjected to higher loads in the past, the soil is
in an overconsolidated state. The soil will respond stiffer than a normally consolidated
deposit. Overconsolidation can be the result of glaciers that where present during the ice
age. There are also soils that are still consolidating under there own weight. These soils
are in an underconsolidated state. This is the result of a higher deposition rate than the
rate that pore pressures can dissipate. These soils will have a lower modulus compared
with normally consolidated soils.

Cementation: Soil cementation can be seen as glue between the particles. The suc-
tion effect which is discussed earlier can also be seen as a glue acting between the particles,
although this is temporary since it disappears with increasing water content. Furthermore
there is the process of chemical cementation. This is defined as the filling of intergranular
pore space by deposition of a mineral cement brought in by circulating groundwater. Highly
cemented soil will have a higher modulus. Chemical cementation will occur as a result of
lithification of sediments. This is a very slow process and will not play a significant role in
a new constructed hydraulic fill as encountered at the site.

Influence of load factors


Furthermore different load factors influences the soil stiffness. These load factors are
discussed.

Confining pressure: Soils under high confining pressure will behave stiffer than soils
under lower confining pressure. The confining pressure is the mean of the principle stresses.
Commonly used models for quantifying the effect of the confining pressure are created
by Kodner and Zelasko (1963) and Duncan and Chang (1970) . These models relate the
modulus to the confining stress using a power law.

Stress level : Since soil behaviour is nonlinear, the stress level influences the stiffness.
In most cases the secant modulus will decrease with increasing strain level.

Strain level : At very small strains the soil respond stiffer than at larger strains. This
behaviour is captured in the stiffness degradation curve which will be explained in detail
in section 3.4.3.

Strain rate: Soils are viscous materials. The faster the loading is applied, the stiffer
the response. The strain rate is defined as the accumulated strain per unit of time. Due to
this effect, standard CPT procedures are done with a specified penetration speed.

Number of loading cycles: When the loading process is repeated a number of times,
the modulus of the soil will change. The larger the number of loading cycles, the smaller
the modulus becomes. This is consistent with the accumulation of strains with an increasing
number of cycles.

Drainage effect: Two extreme cases can be distinguished, drained loading and undrained

12 Chapter 3 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

loading. In drained loading pore pressures can dissipate immediately and no excess pore
pressure is generated. In coarse grained soils, drained loading conditions apply. Undrained
loading occurs when the pore pressures can not dissipate due to very low permeability.
Much of the stiffness is then related to the compressibility of water, which will result in a
much higher stiffness than with drained loading.

3.2.2 Settling behaviour of soil

Settlement is defined as the volume reduction as a consequence of an increase in effective


stress. Different settlements can be distinguished:

Elastic/immediate settlements: These settlements occur quickly and are usually small.
These settlements are related to the compression of the grain skeleton and the free gasses
in the voids. Generally, not all immediate settlements are pure elastic. However, it is
often referred to as elastic settlement since elastic theory is usually used for the computation.

Consolidation: Compression that is associated with the expulsion of water. In granu-


lar soils consolidation develop quickly. In cohesive soils where the permeability is low,
consolidation develops slowly.

Creep: Compression that occurs without an increase of effective stress, but is related
to the slow long-term compression of the grain skeleton.

In a ZLT, a footing is loaded to simulate the process of the settlement of a shallow


foundation. A footing is loaded up in steps by extending the jack between the footing and
the load. As a result, there will be a stress distribution in the soil which is dependent on
the interaction between the plate and the soil. Elastic theory is often used to estimate
the distributions of stresses in the subsurface due to footing pressure. Soils are generally
considered elastic-plastic materials. The use of elastic theory however, can be verified
because of a reasonable match between the boundary conditions for footings and elastic
solutions (Ismael and Vesic, 1981). Another reason to use this approach is the lack of
acceptable alternatives. For a loaded footing, the pure elastic settlements will generally be
small compared to the total settlement. The major components of the settlement will occur
due to change in void ratio, particle rearrangement or grain crushing. When this occurs,
little of the settlements will be recovered after the load is removed. This phenomena is
associated with the elasto-plastic stress-strain behaviour of the soil.

A lot of research is done to examine creep settlements in granular materials. Research


by McDowell and Khan (2003) concludes that one of the mechanisms that cause creep is
particle crushing and occurs within 24 hours. Carbonate sands are encountered on the site
and these sands are known as crushable. This means creep can be of significant influence
during a ZLT. The effect of grain crushing can be visualized by comparing the particle size
distributions (PSD’s) before and after loading. In sand however, the increase in fine content
is usually so small that this effect is hard to measure. Therefore McDowell and Khan
(2003) tested the creep behaviour of pasta shells and compared it with creep behaviour of
sand. When compressing pasta shells the increase in fine content is much more significant
and can clearly be seen when comparing the PSD’s. The research concluded that creep
in both pasta and sand behaves in a similar way and that creep strains are proportional
to logarithmic time. This result is consistent with the hypothesis that all brittle granular
materials behave essentially in the same way.

Chapter 3 Stef Engels 13


Correlating CPT data to stiffness parameters of sand in FEM

Figure 3.5: Type A compression behaviour of Ottawa sand (data from Roberts and
de Souza (1958)) (Mesri and Vardhanabhuti (2009))

3.2.3 Compressive behaviour of sand


The compressive behaviour of granular soils can be different with other types of deposits.
Compression leads to a denser packing of the material and increases particle locking. This,
with engaging surface roughness among soil particles, increases the stiffness of the material
(Vesic and Clough, 1968). On the other hand, particle damage and inter-particle slip are
unlocking mechanisms which lead to a decrease in stiffness. When compressing a granular
mass, both locking and unlocking mechanisms will act simultaneously. The most dominant
of these two mechanisms will determine the compressive behaviour.

Particle damage may be quantified in three categories (Chuhan et al., 2003) (Mesri
and Vardhanabhuti, 2009). Level I damage means abrasion or grinding of particle surface
asperities. When level II damage occurs, particle surface protrusions and sharp particle
corners crush or break. At level III particle damage, Particles split, fracture or shatter.

According to Mesri and Vardhanabhuti (2009) three different types of primary compression
responses can be distinguished for most granular soils. The responses are summarized in
terms of type A, B or C. The type of response can be determined by looking at the void
0
ratio versus effective stress relationship (e versus σv ). For type A behaviour, three stages
of compression can be distinguished. In the first stage, small particle movements enhance
inter-particle locking. In this stage minor level I and level II damage occurs. The improved
locking dominates the unlocking effects. This means that the stiffness increases with an
0
increase in σv . At the second stage level III particle damage occurs. Particles start to
fracture and the unlocking effects become dominant. At this stage the stiffness decreases
with an increase in stress. In the third and final stage, the stiffness gained from particle
packing exceeds the unlocking effect and the stiffness will increase continuously with an
increasing σv 0 . Type A behaviour is often observed for clean well rounded, strong, coarse
particles (Nakata et al., 2001) (Chuhan et al., 2002). An example of type A behaviour is
shown in figure 3.5.
0
With type B behaviour, the e versus σv relation also displays three stages. The first stage
0
is equivalent with type A behaviour, where the stiffness increases with an increase in σv .
In the second stage the improved packing and the unlocking effects (by level III particle
damage) balance. This results in a constant stiffness with an increasing σv 0 , which can
0
be seen by a constant slope in σv - e space. The third stage is equal to the third stage

14 Chapter 3 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

Figure 3.6: Type B compression behaviour of Ganga sand (data from Rahim (1989))
(Mesri and Vardhanabhuti (2009))

Figure 3.7: Type C compression behaviour of carbonate sand (data from Chuhan et al.
(2003)) (Mesri and Vardhanabhuti (2009))

of type A behaviour where the stiffness increases with an increase in σv 0 . A typical type
B response is presented in figure 3.6. In type C behaviour level I and level II particle
damage begin at low values for σv 0 . The locking effect of improved gradation and packing
is dominant over the unlocking effect due to particle damage and inter-particle slippage.
Level III particle damage may or may not occur at higher effective stresses (Chuhan et al.,
2003). The stiffness continuously increases with an increasing σv 0 . This type of response is
typically observed in for angular weak particles, such as carbonate sands in presence with
clay minerals, mica or very fine material. This type of behaviour is shown in figure 3.7.

3.2.4 Stress distribution underneath a shallow foundation


Boussinesq (1883) developed equations to determine the stress state for a point in the
subsurface due to surface loading. These equations are based on elastic theory and as-
sumes that the soil mass is elastic, isotropic, homogeneous and has semi-infinite depth.
Another assumptions is that the soil is weightless. Furthermore the stress state is also
dependable on the rigidity of the foundation and the type of soil. Various methods have
been developed to determine the stress state at any point in the subsurface due to a surface
load. An example of such a method is developed by Newmark (1942), who created an

Chapter 3 Stef Engels 15


Correlating CPT data to stiffness parameters of sand in FEM

influence charts for computing stresses in an elastic foundation. This method is derived by
intergration of Boussinesq’s equation for a point load. The pressure distribution according
to Newmark for a square foundation is shown in figure 3.8. The pressure isobars are
drawn underneath the footing. From this figure it can be seen that the stress increase
at a depth of two times the width of the footing, is only about 0.1 times the applied loading.

0
0.9

0.3 1
0.2
z/B

0.1 2

3
2 1 0 1 2
y/B

Figure 3.8: Newmark solution for stress distribution underneath a square footing

The solution of the stress distribution can be justified when the stress increase occurs in the
soil only. The real requirement to use the solution is not the pure elastic response of the
soil, but a constant ratio between stress and strain. If the stresses induced in the soil are
small in comparison with the shear strength of the material, the Boussinesq solution can be
used. In practice, the Boussinesq solution can be used safely in homogeneous deposits as
clay, man-made fills and in uniform sands with limited thickness. In these kind of deposits
the stiffness will be approximately constant with depth. When the stiffness is increasing
with the depth, the Boussinesq stress distribution will not be valid and nonlinear elastic or
elastic-plastic analyses should be done. A soil profile where the stiffness increases linearly
with depth is known as a Gibson soil profile. Another solution for the stress distribution in
the subsurface is provided by Westergaard (1938). This solution is similar to Boussinesq
but can be used when soils have alternating layers of material. This solution assumes
that there are only vertical deformations and no lateral deformations (one dimensional
compression).

3.2.5 Elastic settlement with constant Young’s modulus


When a constant modulus of elasticity of the soil over the depth is assumed with an evenly
divided pressure, which acts on a homogeneous infinite half space which is isotropic and
linear elastic, the elastic settlement can be calculated according to elastic theory as:

1 − v2
se = qb · B · Iρ · (3.2)
Es

16 Chapter 3 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

Where:
qb is the unit load acting at the base
B is the foundation width
Iρ is a settlement influence factor
v is the Poisson’s ratio
Es is the secant Young’s modulus

In this equation a settlement influence factor Iρ is introduced. This factor depends


on the shape and the rigidity of the foundation. The values of v and Es are characteristic
properties of the soil.

3.2.6 Settlement analysis of a shallow foundation


When the settlement of a shallow foundation on a dense cohesionless soil is monitored,
a load settlement curve as figure 3.9 is obtained. The sudden drop down in the curve
indicates that the ultimate bearing capacity is reached. From the curvature, it can be seen
that the soil does not respond linear elastic but that the stiffness is stress dependent and
decreases with increasing stress level.

Load (kPa) Ultimate bearing


capacity

Settlement
(mm)

Figure 3.9: Typical load settlement curve for a shallow foundation on dense sand

When the load settlement curve is plotted on a logarithmic scale for both the applied load
and settlement, the curve in figure 3.10 is obtained. Two straight lines can be distinguished.
At the intersection point of these two straight lines a sudden drop in stiffness is observed.
This point is known as the yield point. Before this point the soil response is dominated by
elastic behaviour and afterwards the response is dominated by plastic behaviour. There
should be noted that though this is indicated as the elastic region, some plastic straining
can be expected with every loading cycle. The location of the yield point depends on
the highest stress the soil has experienced in the past, the preconsolidation stress. Soils
that have experienced a higher load in the past are by definition overconsolidated. The
behaviour at stress levels below the preconsolidation stress is much stiffer than when this
point is exceeded. This is caused by rearrangement of particles during previous loading. In
figure 3.9, the steeper slope after the preconsolidation stress indicates a significant decrease
in stiffness, which is related to the original stiffness of the normally consolidated material.
The line in the elastic region is called the re-compression line and the line in the plastic
region is called the virgin compression line.

Chapter 3 Stef Engels 17


Correlating CPT data to stiffness parameters of sand in FEM

log (load)

yield point

log (settlement)

Elastic Plastic

Figure 3.10: Load settlement curve in log-log space

3.3 CPT based methods for settlement of a shallow founda-


tion
In engineering practice, many correlation methods to predict the settlement of a shallow
foundation have been developed over the years. In this section an overview of different
methods is given. For the scope of this investigation only CPT related methods are
considered. Two categories of methods can be distinguished. Methods based on observed
settlements and semi empirical methods. Semi empirical methods use a combination of
theoretical analysis and empirical relations.

3.3.1 De Beer and Martens (1957)


One of the methods that is based on observed settlement was created by De Beer, E.
Martens (1957). They proposed the following expression to calculate the settlement of a
shallow foundation in sand:
0
H σ + ∆σ
s = 2.3 · · log10 ( 0 0 ) (3.3)
C σ0
qc
C = 1.5 · 0 (3.4)
σ0
Where:
s is the settlement
C is the compressibility coefficient
0
σ0 is the effective overburden pressure at considered depth
∆σ is the increase of pressure at that depth due to foundation loading
H is the thickness of the layer considered

The strategy in this method is to divide the soil stratum in a convenient number of layers.
The settlement of each layer is determined individually according to equation 3.3 and
eventually the settlement of each layer is summed to evaluate the total settlement.
n
X
s= si (3.5)
i=1

Where si represents the settlement of an individual layer and n represents the number
of layers. The value of ∆σ can be determined using the Boussinesq stress distribution

18 Chapter 3 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

charts and should be determined at the center of each individual layer. De Beer (1965)
concluded that the above method is only appropriate for normally consolidated sand. For
overconsolidated sands a reduction factor needs to be applied which can be obtained from
cyclic Oedometer tests. According to Hough (1969) the value of Cc is determined:
Cc = a(e0 − b) (3.6)
Where:
Cc is the compressibility index

The compressibility index Cc and the compressibility coefficient C are related via:
2.3
C= · (1 + e0 ) (3.7)
Cc
Where:
e0 is the initial void ratio

The empirical parameters a and b can be obtained from table 3.1.

Table 3.1: Values for empirical constants (Hough 1969)

Values of constants
Type of soil
a b
Uniform cohesionless material
Clean gravel 0.05 0.50
Coarse sand 0.06 0.50
Medium sand 0.07 0.50
Fine sand 0.08 0.50
Inorganic silts 0.10 0.50
Well graded cohesionless soil
Silty sand and gravel 0.09 0.20
Clean, coarse to fine sand 0.12 0.35
Coarse to fine silty sand 0.15 0.25
Sandy silt (inorganic) 0.18 0.25
The value of b should be taken as
the minimal void ratio whenever the
latter is known or can conveniently
be determined

The method of De Beer (1965) was intended to provide a ”safe upper limit” with respect to
expected settlement. The values obtained from this method were compared with measured
settlements of several bridge abutments and piers. The conclusion of this analysis was that
the method overpredicts the settlement about two times.

3.3.2 Schmertmann method (1978)


To start this section, the basic framework for calculating settlement is presented. This
framework holds for the following described methods. The basic form of the equation is:
qb · B
s = Iz · (3.8)
Es
Which can also be written in the form:
Z z
Iz
s = qb · B · dz (3.9)
0 Es

Chapter 3 Stef Engels 19


Correlating CPT data to stiffness parameters of sand in FEM

Where:
z is the vertical strain
B is the footing width
Iz is the strain influence factor at depth z
Es is the secant Young’s modulus at depth z
qb is the unit load acting on the base

The advantage of the form 3.9 over 3.8, is that the integral takes the soil layering and
different soil properties into account. The Schmertmann method is based on a physical
model of settlement which has been calibrated with empirical data. In this equation the
strain influence factor is introduced. The factor takes into account for different strain levels
with increasing depth. Measurements of settlements at various depths suggest a vertical
distribution of vertical strain, that starts from a finite value at foundation level, increases
with depth to a maximum and then decreases with depth (Schmertmann, 1970)(Burland
and Burbridge, 1985). Schmertmann (1970) used a simplified diagram to determine the
variation of Iz with depth for square/circular and strip footing foundations. A revised
version of this diagram was provided by Schmertmann et al. (1978) and can be seen in
figure 3.11. The Young’s modulus varies with the value of the cone resistance qc .

Iz
0.1 0.2 Ipeak

1
L/B=1

z/B
2

3
z = depth below footing
B = width of footing
L = length of the footing
4
L/B > 10

Figure 3.11: Varying influence factor according to Schmertmann (1978)

For circular and rectangular footings holds:


Iz = 0.1 at z = 0
Iz(peak) at z = zp = 0.5B
Iz = 0 at z = z0 = 2B

For a footing where L/B > 10:


Iz = 0.2 at z = 0
Iz(peak) at z = zp = B
Iz = 0 at z = z0 = 4B

The peak value of the influence factor can be calculated according to:
qb 0.5
Iz(peak) = 0.5 + 0.1 · ( 0 ) (3.10)
σ0

20 Chapter 3 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

The approach is to divide the soil in multiple layers based on the measured value of the
cone resistance qc . Depth and time factors are applied and allowance for previous existing
in situ stresses are made. The obtained equation has the form:
n
X Iz,i · ∆zi
s = C 1 · C 2 · qb · (3.11)
i=1
Es,i
Where:
C1 and C2 are depth and time dependent factors
qb is the net effective pressure applied at foundation level
Iz,i is the average influence factor corresponding to the sublayer
∆zi is the thickness of the sublayer
Es,i is the average secant Young’s modulus for each sublayer

The suggested values for the coefficients are:


q0
C1 = 1 − 0.5 · (3.12)
qb
t(years)
C2 = 1 + 0.2 · log (3.13)
0.1
The value of q0 is determined by the overburden pressure at the location of the footing.
The secant Young’s modulus of each sublayer is determined with the average corresponding
cone resistance q¯c,i measured in that layer. The value according to Schmertmann et al.
(1978) is taken as:
Es,i = 2.5 · q¯c,i for a circular/squared foundation
Es,i = 3.5 · q¯c,i for a strip foundation

The method of Schmertmann et al. (1978) is based on a series of tests done in a cal-
ibration chamber from the University of Florida. The sand tested in the calibration
chamber was uniformly distributed and normally consolidated. The method has been
proven to be conservative since young sands were tested. The effects of “aging” and
overconsolidation were therefore neglected. Aging of soils refers to the observed phenomena
that soil properties change over time. Overconsolidated deposits will behave much stiffer
than normally consolidated deposits. The total procedure of the Schmertmann method is
given in figure 3.12.

Perform in situ tests


for defining subsurface
conditions

Consider the soil mass


from the base of the
foundation to the
influence depth

Divide this domain into Determine the stiffness


layers, based on the of each layer based on
varying cone resistance cone resistance
vs. depth

Compute the value of


Determine the strain the strain influence Calculate the
influence diagram diagram in the middle settlement
of each layer

Compute the correction


factors C1 and C2

Figure 3.12: Flowchart of the Schmertmann method

Chapter 3 Stef Engels 21


Correlating CPT data to stiffness parameters of sand in FEM

3.3.3 Modification Schmertmann suggested by Peck et al. (1996)


Peck et al. (1996) proposed a slightly different method. They suggested another variation
of the strain influence factor with depth. The modification for the value of the strain
influence factor can be seen in figure 3.13.

Iz
0.2 Ipeak=0.6

1
L/B=1

z/B
2
L/B=3

3
z = depth below footing
L/B > 10 B = width of footing
L = length of the footing
4

Figure 3.13: Varying influence factor according to Peck et al. (1996)

For the value of the influence factor holds:


Iz = 0.2 at z = 0
Iz(peak) = 0.6 at z = 0.5B
Iz = 0 at z = z0

The value of the influence depth z0 can be calculated as:

L
z0 = 2 · B(1 + log )≤4 (3.14)
B

When the foundation level is not at the ground surface but is embedded, another value
0
for Iz should be used. The method to determine the corrected strain influence factor Iz
is published by Peck et al. (1996). This method makes a distinction between elastic or
immediate settlements and creep settlements. The elastic and creep settlements can be
calculated according to:
n
X Iz · ∆z
se = qb · (3.15)
i=1
Es

0.1 t(days)
screep = · z0 · log (3.16)
q¯c 1(day)

The parameter q¯c is the weighted average cone resistance measured through a sublayer.
For the value of the stiffness Peck et al. (1996) suggest the following:
Es,i = 3.5 · q¯c for a circular/squared foundation
L
Es,i (L/B) = Ei (L/B = 1) · (1 + log B ) ≤ 1.4 for a rectangular foundation

22 Chapter 3 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

3.3.4 Robertson (1990)

Normalized CPT parameters

The method Robertson (1990) suggest to use normalized and dimensionless CPT parameters.
Dimensionless parameters correct for increasing soil stress with depth. The assessment of
soil strength from cone resistance can be incorrect when this influence is not taken into
account. The three parameters are the normalized cone resistance Qtn , the normalized
sleeve friction Fr and the normalized pore pressure Bq . They are defined as:

qt − σv0 pa
Qtn = ( ) · ( 0 )n (3.17)
pa σv0

fs
Fr = · 100% (3.18)
qt − σv0

u2 − u0
Bq = (3.19)
qt − σv0

Where:
qt is the corrected net cone resistance
σv0 is the in situ total stress
0
σv0 is the in situ effective stress
pa is the atmospheric pressure
n is the stress component
fs is the sleeve friction
u2 is the measured pore pressure
u0 is the hydrostatic pore pressure

The corrected net cone resistance qt is determined using the value qc that is corrected for
pore water pressure effect. When drained loading conditions apply there is no need for
this correction since excess pore pressures can dissipate immediately . Based on these
normalized CPT parameters, Robertson (1990) developed a chart in which the soil can be
classified according to there mechanical response. Other more recent charts are developed
by Eslami and Fellenius (2004). Both charts perform comparable accurate. A Qtn − Fr
SBTn (normalized Soil Behavior Type) chart and a Qtn − Bq SBTn chart were constructed
by Robertson (1990). The Qtn − Fr chart was concluded to be more reliable. This chart is
given in figure 3.14.

Chapter 3 Stef Engels 23


Correlating CPT data to stiffness parameters of sand in FEM

Figure 3.14: Qtn − Fr chart by Robertson (1990)

The normalized parameters can be used to calculate the so called Soil Behaviour Type
Index Ic . When no measurements of the pore pressure are registered and parameter B is
excluded, a special index Ic,rw can be calculated according to Robertson and Wride (1998):

q
Ic,rw = 3.47 − log Qtn 2 + (1.22 + log Fr )2 (3.20)

With the value of this index it is possible to classify the soil. The contours of this index
can be plotted on the SBTn chart. This is done in figure 3.15. It can be seen that the
contours of Ic,rw follows the Robertson chart more accurately for low values of Ic,rw which
correspond with sandy soils.

24 Chapter 3 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

Figure 3.15: Contours of Ic,rw plotted on the SBTn chart ((Robertson, 2009))

Appropriate stress component


The stress component n which was introduced earlier varies with the SBTn. A lot of
research is done on what this value n should be. Zhang et al. (2002) suggested that the
parameter could be estimated using the SBTn index (Ic ) and that Ic should be defined
using Qtn . Another approach is that n should vary with relative density (Boulanger and
Idriss, 2004). The value of n is close to 1.0 for loose sand and n is less then 0.5 for dense
sand. Large studies in calibration chambers and centrifugal testing with sands with a
constant relative density have shown that the cone resistance increases nonlinear with the
effective stress for coarse grained soils. The n value which captures this nonlinearity is
close to n = 0.5 (Baldi et al., 1989). The nonlinearity is more present in dense sands than
in loose sands. This results in a larger n value for loose sands.

Research showed that the secant peak friction angle decreases with an increasing ef-
fective stress (Bolton, 1986). The friction angle is essentially constant for very loose sands.
This angle is denoted as the constant volume friction angle, or the critical state friction
angle. It is implied that the value of n should be close to one for very loose sands and for
very dense sands at very high stresses. When a dense sand experiences a very high stress
state, dilatancy is suppressed and grain crushing occurs (contractive behaviour). The point
where grain crushing occurs depends on grain characteristics. Sand which consists of silica
rounded particles do not crush until a mean effective stress of about 2 MPa (Bolton, 1986).
Angular silica sand and silty sand reach this point at about 1 MPa, whereas crushable
sands as carbonate sands can experience grain crushing at a mean effective stress of 0.1
MPa. During a CPT the stresses directly underneath the cone these values are reached
and grain crushing can occur. At the rest of the circular failure surface during a CPT
these values are not reached.

Based on earlier research, a generalized critical state line (CSL) is developed by Boulanger
(2003). The CSL is presented in the void ratio - log effective stress space and is presented
in figure 3.16. At low mean effective confining stresses (lower than 200 kPa), the line is
flattened. The line gets steeper at higher effective stresses. When the slope is small, there
is a strong connection between the relative density and the state parameter. A value of
n = 1 is appropriate when the normally consolidated line is parallel to the CSL (Wroth,

Chapter 3 Stef Engels 25


Correlating CPT data to stiffness parameters of sand in FEM

Figure 3.16: Critical state line and state parameter from Bolton (1986) (Boulanger, 2003)

1984). In sands, the CSL is nonlinear over a wide stress range and the consolidation varies
with respect to the CSL. It can be concluded that critical state soil mechanics supports
the idea of a varying the n value to normalize penetration resistance. The n value varies
between 0.5 for at low stresses and tends to go to 1.0 for higher stresses where the CSL
gets straight and the consolidation line becomes parallel to the CSL.

The slope of the CSL can be linked to the SBT index Ic according to Jefferies and Been
(2006). Based on the discussion above the following recommendation is made by Robertson
(2009) for varying the stress component n with Ic and effective overburden stress:

σ˙v0
0

n = 0.38 · Ic + 0.05 − 0.15 (3.21)


pa

Correlations for stiffness parameters


Now a connection needs to be made between the discussed parameters and the stiffness.
This study focusses on coarse grained cohesionless soils where drained conditions apply.
Eslaamizaad and Robertson (1997) and Mayne (2000) have determined that the load
settlement response of a foundation can be accurately predicted by using the shear wave
velocity Vs . Although direct measurements of Vs are preferred, correlations between qc
and Vs can be used as an estimate for lower risk projects. It has been shown that the
value of Vs predominantly determined by the number and area of grain to grain contacts
(Schneider et al., 2004). Therefore Vs depends on relative density, aging, cementation,
effective stress state and arrangement of particles. The value for qc is also dominated by
relative density and stress state but to a lesser degree by age and cementation. A strong
relationship between Vs and qc exist, but some variability can be expected. Since Vs is a
direct measurement of the small strain shear modulus G0 , an improved linkage between
CPT parameters and soil stiffness can be obtained.

With the normalized CPT parameters it is possible to approximate the normalized shear
wave velocity Vs1 (Robertson, 2009):

Vs1 = (αvs · Qtn )0.5 (3.22)

26 Chapter 3 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

The factor αvs is defined as the shear wave velocity factor and has the unit m/s2 . The
values for αvs can be estimated using the obtained value for Ic as follows:

αvs = 100.55·Ic +1.68 (3.23)

The value of the shear wave velocity Vs can then be determined according to:

(qt − σv ) 0.5
Vs = [αvs · ] (3.24)
pa
This general relationship is recommended for most Holocene- to Pleistocene-age deposits
which are predominately silica-based. At low shear strain levels the shear modulus has a
constant maximum value G0 . This modulus is determined as:

G0 = ρ · Vs2 (3.25)

Where ρ represents the mass density of the soil (kg/m3 ). The small strain shear modulus
number KG is related to the small strain shear modulus as:
0
σ
G0 = KG · pa · ( v0 )n (3.26)
pa
The relationships between the soil modulus and the cone resistance usually have the general
form:
G0 = αG · (qt − σv0 ) (3.27)

Where αG is the shear modulus factor for the estimation of G0 . Because the stress
component for the derivation of Qtn and G0 is similar, it follows that:
KG
αG = (3.28)
Qtn
The contours of KG and αG can be plotted on the SBTn chart. This is done in figure 3.17
(Robertson, 2009).

Figure 3.17: Contours of αG and KG on the SBTn chart (Robertson 2009)

Chapter 3 Stef Engels 27


Correlating CPT data to stiffness parameters of sand in FEM

To determine the appropriate value of αG from Ic , the link with αvs can be used:
ρ
αG = · αvs (3.29)
pa

When an average unit weight of γ = 18kN/m3 is taken and equations 3.23, 3.27 and 3.29
are combined, G0 can be calculated as:

G0 = 0.0188 · [10(0.55Ic +1.68) ] · (qt − σv0 ) (3.30)

The Young’s modulus E and the shear modulus G are interrelated via the Poisson’s ratio
v as follows:
E = 2 · (1 + v) · G (3.31)

For most soils v varies between 0.1 and 0.3 in drained conditions. Hence, for most coarse
grained soils holds E ∼ 2.5G. The small strain shear modulus obtained in equation ??
needs to be reduced to an appropriate value of the shear modulus G. The amount of
softening is a function of the stress level (Eslaamizaad and Robertson, 1997). A simple
approach to estimate the amount of softening according to Fahey and Carter (1993) is:

G q g
=1−f ·( ) (3.32)
G0 qult

Where q represents the applied load and qult the failure load. The constants f and g
depends on soil type and stress history. According to Fahey and Carter (1993) and Mayne
(2005) values of f = 1 and g = 0.3 are appropriate for uncemented soils which are not
highly structured. For many design application the stress level ranges from 0.2 to 0.3 in
comparison with the stress level at failure. The ratio of G/G0 ranges then from 0.3 to 0.38.
The simplified elastic solution for the Young’s modulus is approximately:

E ∼ 0.8G0 (3.33)

When a similar procedure is followed as with the shear modulus, the following relations
can be obtained for the Young’s modulus number KE and the modulus factor αE :
0
σv0 n
E = KE · pa · ( ) (3.34)
pa

E = αE · (qt − σv0 ) (3.35)

KE
αE = (3.36)
Qtn
When the equations 3.30 and 3.33 are combined, the appropriate value for αE and E can
be made:
αE = 0.015 · [10(0.55Ic +1.68) ] (3.37)

E = 0.015 · [10(0.55Ic +1.68) ] · (qt − σv0 ) (3.38)

It should be reminded that this value only applies for uncemented, predominately silica-
based coarse grained soils of either Holocene- or Pleistocene age. Furthermore this prediction
is only valid for specified assumptions for the Poisson’s ratio and stress level. The contours
for KE and αE are given in figure 3.18.

28 Chapter 3 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

Figure 3.18: Contours of αE and KE on the SBTn chart (Robertson 2009)

When the stress level increases (exceeds 0.25), the values for αE will be decreasing. If
the relation for the stiffness degradation curve according to Fahey and Carter (1993) and
Mayne (2005) is used, the stress level can be taken into account according to:
q 0
E = 0.047 · [1 − ( )0.3 ] · [10(0.55Ic +1.68) ] · (qt − σv0 ) (3.39)
qult

When this methodology is used, the ultimate bearing stress qult needs to be determined.
This value can be obtained by various analytical methods. With this methodology a
stiffness parameter can be calculated for each registered value of cone resistance and sleeve
friction. The settlement can be calculated using this stiffness in the framework which is
evaluated in section 3.3.2.

3.4 Soil models for sand


Before a qualitative numerical analysis with PLAXIS 2D can be done, some theory about
soil models needs to be evaluated. In this section, appropriate soil models for sand are
briefly explained.

3.4.1 Mohr-Coulomb model


Framework
The Mohr-Coulomb model is a linear elastic perfectly-plastic model. Since soil behaviour
is nonlinear and depends at least on stress level, stress path and strain level, this is a
simplified approach. Since this model is rather simple compared to the advanced models,
it can be used to make a first estimation of soil behaviour. The linear elastic part is based
on Hooke’s law of elasticity. The perfectly plastic part is based on the Mohr-Coulomb
failure criterion. This failure criterion is an extension of the friction law of Coulomb to
general stress states. The occurrence of plastic strains can be determined using a so called
”yield function” which is denoted with f . The full Mohr-Coulomb yield criterion consists
of six yield functions when formulated in principle stresses (Appendix A). Plastic yielding
will occur when the condition f = 0 is satisfied. When f < 0 all generated strains are
purely elastic. In principle stress space, the yield surface can be seen in figure 3.19. In the
Mohr-Coulomb model the yield surface is fixed in space.

Chapter 3 Stef Engels 29


Correlating CPT data to stiffness parameters of sand in FEM

Figure 3.19: The Mohr-Coulomb yield surface in principle stress space (Brinkgreve and
Vermeer, 2016)

For determining the direction and magnitude of plastic straining, a plastic potential func-
tion g is introduced. In plasticity analyses a distinction can be made between associated
and non-associated plasticity. With associated plasticity the yield function and plastic
potential function are the same and the direction of plastic strains is normal to the yield
surface. As a result of this assumption a symmetric elasto-plastic material stiffness matrix
is obtained. This reduces the calculation time. However, a non-associated plasticity frame-
work is used in this model because theory of associated plasticity will overestimate dilatancy.

Furthermore it should be noted that Mohr-Coulomb model have the tendency to over
predict the shear strength in undrained behaviour.

Parameters
The Mohr-Coulomb model uses six soil parameters:
E: Young’s modulus
v: Poisson’s ratio
c: Cohesion
ϕ: Friction angle
ψ: Dilatancy angle
σt : Tension cut off and tensile strength

The stiffness parameter E is already discussed in detail. A suitable value for the Poisson’s
ratio needs to be chosen. This value is directly related to the value of the coefficient of
0 0
lateral earth pressure K0 (ratio between σh and σv ) according to:
v
K0 = (3.40)
(1 − v)
In the Mohr-Coulomb model v is evaluated by matching a realistic K0 value. In many
cases this value will vary between 0.3 and 0.4 for v. For unloading reloading cases the
values for v in the range of 0.15 and 0.25 will be more appropriate. With this methodology
it is impossible to create K0 values that exceed 1, as is observed in highly overconsoli-
dated stress states. The cohesion c or undrained shear strength su has the dimension
of stress (kPa). In undrained loading cases the cohesion parameter in combination with
ϕ = 0 can be used to model undrained shear strength. The advantage of using this
method to model undrained shear strength is that the user has control over the shear

30 Chapter 3 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

strength, independent of the stress state and stress path followed. The Mohr-Coulomb
criterion then reduces to a Tresca failure criterion. The unit of the friction angle ϕ is degrees.

The dilatancy angle ψ also has the unit degrees. The dilatancy is most dominant in
dense sands. It is dependent on friction angle and relative density. For quartz sands, the
dilatancy can be estimated according to Brinkgreve and Vermeer (2016):

ψ ≈ ϕ − 30 (3.41)

For ϕ ≤ 30 the dilatancy angle is usually zero. A small negative value for ψ is only observed
in extremely loose sands.

3.4.2 The Hardening Soil model


Framework
The Hardening Soil model is a more advanced soil model compared with the Mohr-Coulomb
model. Instead of a fixed yield surface, it allows the yield surface to expand in principle
stress space due to plastic straining. Two different types of hardening can be distinguished:

Shear hardening: This type of hardening is used to model irreversible strains due to
deviatoric loading.

Compaction hardening: This type of hardening is used to model irreversible strains due to
compression in oedometer loading and isotropic loading situations.

The 2-D representation of the yield surface in mean - deviatoric stress space is repre-
sented in figure 3.20. Note that the cohesion is zero in this figure. The Hardening Soil
Model is suitable for simulating the behaviour of different soil types. Some basic charac-
teristics of the model are the stress dependency of stiffness (according to a power law),
plastic straining due to deviatoric loading, plastic straining due to compression, elastic
unloading/reloading and the use of the Mohr-Coulomb failure criterion. This model does
not account for time dependent behaviour and softening.

Mohr-Coulomb
Deviatoric failure line
stress

Shear
hardening

Compaction
hardening
Elastic region

Mean stress

Figure 3.20: Yield surface of the Hardening Soil model in mean-deviatoric stress space

Parameters
Some of the parameters used by the Hardening Soil model are the same as for the Mohr-
Coulomb model. These parameters are:

Chapter 3 Stef Engels 31


Correlating CPT data to stiffness parameters of sand in FEM

ref ref for drained triaxial test results (Brinkgreve and


Figure 3.21: Definition of E50 and Eur
Vermeer, 2016)

c: Cohesion
ϕ: Friction angle
ψ: Dilatancy angle
σt : Tension cut-off and tensile strength

The following parameters correspond with basic parameters for soil stiffness:
ref
E50 : Secant stiffness in standard drained triaxial test
ref
Eoed : Tangent stiffness for primary oedometer loading
ref :
Eur Unloading/reloading stiffness
m: Power for stress level dependency of stiffness

Furthermore some advanced parameters are defined:


vur : Poisson’s ratio for unloading/reloading
pref : Reference stress
K0nc : The K0 value for normal consolidation
Rf : Failure ratio qf /qa
σt : Tensile strength

Instead of the basic soil stiffness parameters, PLAXIS also accepts a compression in-
dex Cc , swelling index Cs in combination with an initial void ratio e0 . Although the
Hardening Soil model is a higher order approach for modelling soil behaviour, it is more
difficult to handle because of the large amount of input parameters. In contrast to the
Mohr-Coulomb model (elasticity based), the Hardening Soil model does not use a fixed
ref ref
relationship between the triaxial stiffness E50 and the oedometer stiffness Eoed . The
ref ref
default value given by PLAXIS for the Eur is three times the value of E50 . The definitions
ref ref are visualized in figure 3.21. Furthermore, the value of K nc is not simply
of E50 and Eur 0
a function of the Poisson’s ratio but an independent input parameter. Suggested is to use
the following correlation formula (Jaky, 1948):

K0nc = 1 − sin φ (3.42)

For the yield surface of the Hardening Soil model, both shear hardening and compaction
ref
hardening must be satisfied. The parameter E50 mainly controls the shear yield surface

32 Chapter 3 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

ref
and the parameter Eoed controls the cap yield surface. The magnitude of the yield cap is
determined by the isotropic pre-consolidation stress. The 3-D representation of the yield
contour in principle stress state is given in figure 3.22

Figure 3.22: 3-D representation of the yield contour of the Hardening Soil model in
principle stress space (Brinkgreve and Vermeer, 2016)

3.4.3 The Hardening Soil model with small-strain stiffness


The strain amplitude has influence on the value of the soil stiffness. When the strain
amplitude increases the stiffness reduces non-linearly. When the decrease in stiffness is
plotted against the logarithmic value of the strain, a typical S-shaped modulus stiffness
reduction curve is obtained (figure 3.23). With laboratory testing it is often not possible
to accurately determine the stiffness at very small strain levels. It turns out that with
conventional laboratory testing, the stiffness measured at minimal strains already has
decreased to less then half its initial value.

Figure 3.23: Stiffness reduction curve with increasing strain level (Atkinson and Sallfor,
1991)

Chapter 3 Stef Engels 33


Correlating CPT data to stiffness parameters of sand in FEM

To take this stiffer behaviour at very small strains into account, the Hardening Soil model
with small-strain (HS-small model) stiffness is developed. This model is implemented in
PLAXIS. The HS-small model uses the same parameters as the Hardening Soil model,
along with two additional parameters that describes the stiffness reduction curve. These
additional parameters are the small shear modulus (G0 ) and a parameter which is denoted
by γ0.7 . This parameter specifies the shear strain level where the secant shear modulus is
reduced to a value of 70% of the value of G0 . The degradation of stiffness can be described
by a hyperbolic law developed by Santos and Correira (2001):

Gs 1
= γ (3.43)
G0 1 + 0.385 · γ0.7

This relationship holds when the value of G lies between G0 and Gur (unloading reloading
shear modulus). Between these values there is a degradation of stiffness. When G decreases
ref
to Gur the behaviour is dominated by the Hardening Soil model parameters E50 and
ref
Eoed .

34 Chapter 3 Stef Engels


Chapter 4

Zone Load Test procedure

In this chapter the test procedure of the ZLT’s performed on the site are explained. First
some information is given about the test set-up. Next, the loading procedure will be
discussed and examples of some test results will be presented. Afterwards, uncertainties in
the test procedure will be elaborated. To conclude this chapter the interpretation of the
results and the assumptions that are done for further analysis will be evaluated.

4.1 Test set-up


In a ZLT, a 3 x 3 m squared footing is loaded in specified loading steps. The thickness
of the footing is 0.6 m. A schematic side view is presented in figure 4.1. Before the
loading starts, all the weight of the concrete blocks is transferred to the subsoil through
the supporting blocks. Loading up the footing will be done by extending the hydraulic
jack which is included with a load cell. The measurement devices are connected to the
reference beams. These beams are supported by piles that are installed further away from
the footing. It is not exactly clear at how deep these piles penetrate into the sand. A more
detailed representation of the footing and the measuring devices is presented in figure 4.2.
Concrete blocks for
loading

Steel bars

Reference beam Concrete support


blocks
Measering device
for settlement
Hydraulic jack Loaded footing

Figure 4.1: Schematized side view of a Zone Load Test

4.2 Loading procedure and test results


As stated before the loading procedure will be done in specified steps. The total duration
of the ZLT is 59 hours. The loading steps and measured settlements for one specific test
can be found in table 4.1. The load will not be adjusted until the rate of settlement is less
than 0.008 mm/min.

35
Correlating CPT data to stiffness parameters of sand in FEM

Figure 4.2: Close up from the loaded footing in a Zone Load Test

Table 4.1: Settlement registration of a zone load test

Duration of loading step Working load Working pressure Average settlement


(hours:minutes) (%) (kPa) (mm)
00:10 0 0 0.00
00:10 10 20 0.49
00:10 0 0 0.28
02:00 25 50 2.62
02:00 50 100 4.88
02:00 75 150 6.77
02:00 100 200 8.79
48:00 125 250 14.42
00:30 100 200 14.09
00:30 75 150 13.62
00:30 50 100 12.99
00:30 25 50 12.20
00:30 0 0 10.47

On the footing, four measuring devices are installed at different locations to measure the
settlement. Measurements are performed by four dial gauges that are placed in the middle
between the corners of the footing. The average settlement registered by all four devices
is given in the table above. Typical output consist of a diagram that plots the applied
pressure versus the displacement (figure 4.3) and a diagram that shows the evolution of
settlement in time (figure 4.4). After each loading step, the settlement increases during
the time the pressure is hold. When the applied pressure is increased to 250 kPa and held
for 48 hours, this phenomena can be clearly seen. This could occur due to the dissipation
of pore pressures (consolidation) from less permeable layers. Another possibility is that
creep occurs in the sand skeleton.

36 Chapter 4 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

Figure 4.3: Pressure displacement curve for zone load test DV-113

Figure 4.4: Time displacement curve with the applied pressure for zone load test DV-113

4.3 Test uncertainties


The basic idea of a ZLT is to measure the settlement of a footing which is subjected to
a certain load. In the ideal case, the test procedure should not influences the measured
settlement. However, there are reasons to believe this can be of influence. Possible uncer-
tainties related to the test procedure are summarized.

• Temperature influence: The difference in temperature at the site of Kuwait is


significant. Steel parts of the test set-up, like the reference beams are sensitive for
thermal expansion. Thermal expansion leads to deflections in the reference beams
and could result in inaccuracies of the measurements.
• Deflection of reference beam: The reference beams are supported at the ends. It
should be checked if the deformation due to its own weight is negligible with respect
to the measured settlement.
• Pre-loading of subsurface: All the load of the ZLT is present before the load is
transferred to the footing. Before the test begins this load is transferred into the
subsurface by the supporting blocks. This results in a stress distribution in the
subsurface. It should be checked if this procedure of testing is representative for the
in situ soil state.

Chapter 4 Stef Engels 37


Correlating CPT data to stiffness parameters of sand in FEM

4.4 Interpretation and assumptions


The various methodologies explained in the literature study can only be used to determine
the direct settlements of granular soils. To fairly examine the accuracy of these methods,
the settlement prediction needs to be compared to the direct measured settlement of the
grain skeleton. Therefore the large amount of extra settlement that is observed during 48
hours at the constant load of 250 kPa will be excluded from the analysis. However it is
interesting to know more about the physical processes related to these extra settlements.
It is possible that the extra settlement is related to creep in the sand fill. Time dependent
deformation can be the result of particle breakage (McDowell and Khan, 2003). They
showed that crushable materials can crush in time without an increase in load. Since
crushable carbonate sands are encountered on the site, this could be an explanation for the
extra observed settlement. Plate Load Tests can be done to determine whether the extra
settlement occurs in the sand fill. The diameter of the plate should be chosen as such, that
the extra stress increments in the subsurface only occur in the sand fill.

Another explanation could be the generation of excess pore pressure. In that case, the
extra settlement observed is the result of consolidation. For future projects, it is advised to
install pore pressure meters to make sure if fully drained loading conditions can be assumed.

Before the loading starts, the test set-up is fully build up. Because of the weight of
the concrete blocks, the subsurface is influenced before the loading starts. The stress in
the subsurface increases due to the load transfer from the supporting blocks to the sand.
It is assumed that this influence can be ignored. For future tests it is advised to minimize
this effect. The influence of the test set-up is examined numerically in Chapter 7.

The design value of the pressure is 200 kPa. Therefore the direct settlement at 200
kPa will be analysed. From table 4.1 can be seen that during the test an unloading step is
present. The plastic strains generated during the unloading reloading cycle, will not be
taken in to account. Therefore the settlement at 200 kPa in table 4.1 will be corrected
with 0.28 mm. The assumptions used for further analysis are summarized.

Assumptions
• The settlement measured will be due to compression of the sand
• The settlement at 200 kPa will be interpreted as direct settlement
• Fully drained loading conditions apply
• There is no suction above the groundwater table
• The test set-up does not influence the in situ soil conditions
• The Boussinesq stress distribution can be used
• The loaded concrete footing is rigid

38 Chapter 4 Stef Engels


Chapter 5

Analytical settlement analysis

The different methodologies presented in the literature study are tested on existing site
data in this chapter. The discussed methodologies are only applicable on sandy soils. To
make an accurate analysis, only the ZLT’s on sandy soils should be considered. As stated
in Chapter 2, the site consist of sandy, silty and sabkha soils. Therefore suitable locations
should be selected that can be used in the analysis. In the first section of this chapter the
criteria for a suitable ZLT are evaluated. Afterwards the different methods are applied
to make a settlement prediction for each of the selected ZLT. Finally the results of the
different methods will be compared with the measured settlement of the footing in the
field.

5.1 Zone Load Test Criteria


Underneath every ZLT, five CPT’s are done before the testing procedure starts. The depth
of the CPT depends on the site conditions. At some parts of the site, the CPT’s get stuck
after a few meters. In softer areas, the CPT can reach a depth that exceeds 12 meters
below ground level. CPT’s can get stuck for different reasons. When the cone resistance
exceeds 40 MPa the pushing equipment is not able to push the cone further into the soil.
Another possibility is that the cone hits a rock and tries to push it into the soil. When
performing a CPT there is always an amount of deflection (lateral movement) of the rods.
When this deflection exceeds a certain limit, the CPT is stopped to prevent damaging
the equipment. The level of the site after the fill is placed is denoted as the ground level.
The preferred subsurface underneath a ZLT consists of sandy soils only. However, a lot
of tests can not be used when such a high criteria is set. Besides, the upper limit of
information available of the subsurface is the depth of the CPT. Whatever soil conditions
are encountered underneath is anyone’s guess. A criteria should be set as such, that a
sufficient amount of tests can be used in the analysis and that the compressive behaviour
is dominated by sandy soils.

To extend the amount of tests for the analysis, some ZLT locations where sand is overlying
sabkha are also taken into account. Looking at the Boussinesq stress distribution theory,
the stress increase in the subsurface due to a uniform loaded area decreases with depth.
For a square footing of 3 x 3 m the stress increase at a depth of 5 m is about 0.15 times
the applied pressure on the footing. In this analysis a sabkha layer is accepted when it is
located at 5 m or deeper with respect to the new ground level. Due to the low permeability
of the sabkha undrained behaviour is expected. This means that all the load is taken by
the water in the voids and excess pore pressures can not dissipate. If this indeed is the
case the layer can be considered incompressible.

Another criterion that needs to be determined is when a CPT can be used in the analysis.
The analytical methods discussed all use the Boussinesq stress distribution theory. In the
method of the De Beer, E. Martens (1957) the incremental stress is calculated according to

39
Correlating CPT data to stiffness parameters of sand in FEM

this theory. In the other methods the strain influence factor is introduced which is based
on the same theory. The simplified strain influence factor reaches a value of zero at a depth
of 2 times the width for square footings. Therefore it is decided that the CPT must at
least penetrates 6 m into the subsurface before it is taken into account in the analysis.

5.2 Processing the CPT data


For the methodologies suggested by De Beer, E. Martens (1957), Schmertmann et al.
(1978) and Peck et al. (1996) the subsurface should be divided into a convenient amount of
layers. To determine the layering underneath each ZLT, the normalized soil behaviour type
(SBTn) chart is used (Robertson, 1990). This chart is discussed earlier in Section 3.3.4.
Wang et al. (2013) showed that the Robertson chart can be estimated very accurately by
fitting quadratic functions for every contour of the chart. The accuracy of this approach is
visualized in figure 5.1. The best fitted functions and intersection points as indicated in
the figure are provided in Appendix B.

Figure 5.1: Best fitting of the Robertson chart by quadractic functions by Wang et al.
(2013)

The raw data of the CPT’s consist of measurements of the cone resistance and the sleeve
friction for every 1 cm of penetration depth. These parameters can be converted to
normalized parameters as shown in section 3.3.4. When using Matlab and the best fitted
curves, all the registered points during the CPT can be plotted in the SBTn chart to
visualize what kind of soil is encountered.

An example of the processing of CPT’s under one ZLT is evaluated. As stated before under
each footing of 3 x 3 m there are five CPT’s available. The raw data is used as input in
Matlab and all the five CPT’s are plotted in the same figure (figure 5.2). Important in
this figure is to note if the five profiles are consistent with each other. In the figure all the
profiles are quite similar, that indicates low variability in the horizontal direction between
the CPT’s. The high friction ratio and low cone resistance at approximately 5 m depth,

40 Chapter 5 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

indicate the presence of a sabkha layer. Furthermore the original ground level (before the
fill was constructed) is marked in the figure.

0 0

1 1

2 2

3 3
Depth (m)

Depth (m)
4 4

Original Ground Level Original Ground Level


5 5

6 6

7 7
0 10 20 30 40 0 1 2 3 4 5 6
Cone resistance (MPa) Friction ratio (%)

Figure 5.2: Five CPT’s underneath zone load test DD145

The following step in the procedure is to combine the profiles to one profile which represents
the mean values of the cone resistance and friction ratio (figure 5.3). In the criterion of the
individual CPT’s is stated that a profile should at least have a penetration depth 6 m in
order to be included in the analysis. As can be seen in figure 5.2 all the profiles satisfy this
criterion. This means that all the profiles are used to calculate the mean profile. When
an individual profile does not reach six meters penetration depth it is excluded from the
analysis and will not be taken into account when constructing the mean profile.
Based on the mean values of the cone resistance and friction ratio, the normalized parameters
are calculated according to the method of Robertson (1990) (Section 3.3.4). The formulas
to determine these parameters are introduced earlier but are recapped:
qt − σv0 pa
Qtn = ( ) · ( 0 )n (5.1)
pa σv0
fs
Fr = · 100% (5.2)
qt − σv0
q
Ic,rw = 3.47 − log Qtn 2 + (1.22 + log Fr )2 (5.3)

σ˙v0
0

n = 0.38 · Ic + 0.05 − 0.15 (5.4)


pa
Because the stress component n appears multiple times in this system, an iterative procedure
is used to determine the value of n. This is done using Matlab until an accuracy within
0.01 is reached. The result of the analysis are vectors with the normalized parameters for
every reading of the CPT. Afterwards the values for the normalized cone resistance and
the normalized friction ratio can be plotted in the Robertson SBTn chart. To do this the

Chapter 5 Stef Engels 41


Correlating CPT data to stiffness parameters of sand in FEM

0 0

1 1

2 2

3 3
Depth (m)

Depth (m)
4 4

Original Ground Level Original Ground Level


5 5

6 6

7 7
0 10 20 30 0 2 4 6
Cone resistance (MPa) Friction ratio (%)

Figure 5.3: The mean CPT profile underneath zone load test DD145

contours of the chart are estimated using the best fitting curve approach by (Wang et al.,
2013). The result can be seen in figure 5.4.

10 3

7 6 8 9

10 2
Tip resistance, Q tn

4
5 3
10 1 1=Sensitive fine grained
2=Organic soils
3=Clays
4=Silt mixtures
5=Sand mixtures
6=Clean/silty sand
7=(Gravelly) sand
8=Very stiff sand/clayey sand
9=Very stiff, fine grained 1 2

10 0
10 -1 10 0 10 1
Friction ratio, F (%)

Figure 5.4: The normalized CPT data points in the SBTn chart of Robertson

As seen in section 3.3.4 in figure 3.15, specific values of Ic,rw correspond with the contours
of the SBTn chart. Using these specific Ic,rw values make it possible to estimate the soil
classification based on the normalized parameters. A profile of Ic,rw can be constructed
with respect to the depth. This gives an indication of the soil type and the soil layering. In
Matlab a new layer is distinguished when the value of Ic,rw exceeds the indicated boundaries
which are typical for a specific soil type. Furthermore is programmed that the minimal
layer thickness is 25 cm. An example of the output is given in figure 5.5. The dotted

42 Chapter 5 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

horizontal lines give an indication of the layering.

With this approach the suggested layers are automatically generated from the raw CPT
data. Care must be taken when using this soil layering. First of all the layering is based
on multiple CPT’s. When the individual CPT profiles significantly differ from each other,
the approach of introducing horizontal layers is inaccurate because a high horizontal
heterogeneity is observed. Furthermore, the contours of Ic,rw on Robertson chart fit the
Robertson contours very well for low values of Fr and high values of Qtn , but the accuracy
decreases when moving to the right bottom of the chart (figure 3.15). This means that the
layering indication based on Ic,rw is less accurate with increasing values of Ic,rw . Finally
it should be remembered that this method of soil classification is based on CPT results
only, which may have inconsistencies with soil classification based on the USCS (unified
soil classification system) where soils are classified based on sieving results and Atterberg
limits.

0
Gravelly sand to dense sand

1
Clean sand to silty sand

Silty sand to sandy silt

Clayey silt to silty clay

Clay to organic soils


2

3 Silty clay to clay


Depth (m)

7
0 0.5 1 1.5 2 2.5 3 3.5 4
Soil Behaviour Type index

Figure 5.5: Variation of soil behaviour type index over the depth corresponding to zone
load test DD145

For each discussed methodology, a function is programmed in Matlab that evaluates the
settlement based on the CPT input data. For the De Beer, E. Martens (1957), Schmertmann
et al. (1978) and Peck et al. (1996) methods, different layers needs to be distinguished.
This is done based on the automated layering profile in combination with engineering
judgement. Most of the methods use the strain influence diagram. Note that the Peck et al.
(1996) method uses a modified strain influence diagram which is illustrated in figure 3.13.

5.3 Overview of the procedure

In this section a flowchart is presented that summarizes the procedure of the analytical
analysis. The intermediate parameters are calculated with a Matlab program that calculates
all the parameters based on the 5 CPT’s that are done beneath one ZLT. The flowchart is
presented in figure 5.6.

Chapter 5 Stef Engels 43


Correlating CPT data to stiffness parameters of sand in FEM

Analytical analysis
5 CPT´s at each
ZLT location

Calculate mean q c , fs Exclude CPT´s where


and plot the results depth < 6m

Plot all the Calculate normalized Calculate in situ stress


measurements on Robertson parameters Q tn state
Robertson SBTn chart and Fr

Plot SBT index over Calculate the SBT index Calculate stiffness
the depth parameters according to
for every measurements Robertson

Suggest layering
based on SBT index

Determine the average Determine the strain


cone resistance for influence diagram
each layer

Calculate the stress


increase in the middle
of the layer
(Boussinesq)

Calculate settlement Calculate settlement Calculate settlement Calculate settlement


with De Beer & Martens with Schmertmann with Peck et al. with Robertson method
method method method

Figure 5.6: Flowchart of the procedure of the analytical analysis

5.4 Results and discussion


In this section the results of the analysis of the selected ZLT’s are presented. All of these
ZLT’s met the criteria set in Section 5.1. In the first part of this section different site
conditions are compared. From the previous section it is mentioned that in this analysis
some locations are included where sabkha and silt are encountered. A separate analysis
is done for these locations to check whether this leads to inaccuracies. Afterwards, the
settlements are evaluated using all the described methods and compared to the measured
settlement in the field. The error is calculated according to:

(scalculated − smeasured )
Error = · 100% (5.5)
smeasured

When the error is calculated as in the equation above, negative values of the error indicate
that the calculated settlement is smaller than the measured settlement. This means an
under-prediction of the settlement and it should be noted that this is an unsafe situation.

5.4.1 Comparison different site locations


The methodologies that are used to evaluate the settlement during a ZLT are applicable to
sandy soils only. Since the fill is constructed over a weaker layer and varies in thickness,

44 Chapter 5 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

other soil types can be encountered as well in the influenced area underneath a ZLT. In
Section 5.1 is explained which criteria concerning the ZLT’s are used for this analysis. The
criteria set in that section are now examined.

To do such an examination, all the ZLT’s can be split up into three groups.

• Zone Load Test locations where only sand is encountered


• Zone Load Test locations where sand and silt are encountered
• Zone Load Test locations sand, silt and sabkha are encountered

For each group, the settlements are calculated using the proposed methodologies. Af-
terwards, the error between the measured settlement and the calculated settlement is
evaluated. The mean value of the error and the standard error of the error is determined.
The results are summarized in table 5.1.
Table 5.1: Accuracy of different analytical methods

Error
Method
mean (%) standard error (%)
De Beer and Martens 28 16
Schmertmann 42 13
Peck et al. -2 23
Robertson 0 20

The results of each individual test can be found in Appendix C. It can be seen that the
mean value and the standard error for each method of each group is quite similar. This
means that the error of each group is in the same order of magnitude. It can be concluded
that it is justified to extend the sample size with sabkha and silt locations when the criteria
of section 5.1 is met.

5.4.2 Normality of the results


It would be useful to recognize a certain pattern in the output. If the error of each method
follows a certain well known distribution, it is possible to make a safe evaluation of the
settlement. Therefore the error of each method is evaluated and tested for normality. With
this analysis it is possible to check whether it is likely that the values of the error are
normally distributed. When the error of a method is normally distributed, it is possible to
give a worst case scenario regarding to the settlement based on an analytical calculation
only. There are certain statistical tests to check whether it is likely to assume a normal
distribution to a set of variables. The test that will be used here is the Shapiro-Wilk test
(Shapiro and Wilk, 1965). The procedure is elaborated in Appendix D. The results of this
test will be combined with a graphical fit of the distribution to support the conclusion.

De Beer, E. Martens (1957)


The parameter that will be tested on normality is the error as defined in equation 5.5. As
stated earlier, the error indicates the accuracy of the expected settlement with respect to the
measured settlement in the ZLT. Negative values of the error indicate an underestimation of
the settlement. Since 43 tests are evaluated, the sample size is equal to 43. The histogram
plot with a fitted normal distribution is given in figure 5.7. The Shapiro-Wilk test utilizes
the null hypothesis principle. The null hypothesis states that the sample is normally
distributed. A p-value is calculated and this value is compared to a significance level α.
The significance level that is used is α = 0.05. The Shapiro-Wilk test is programmed in

Chapter 5 Stef Engels 45


Correlating CPT data to stiffness parameters of sand in FEM

14

12

10
Frequency
8

0
−30 −18 −6 6 18 30 42 54
Error (%)

Figure 5.7: Histogram of the error of the De Beer and Martens method with fitted normal
distribution

Matlab and a p-value of 0.047 is calculated. Since 0.047 > 0.05 the null hypothesis can
be rejected. It can be stated that the data is not normally distributed with a confidence
level of 95%. Looking at figure 5.7 the sample does look normally distributed except for
one test with an error of around -30%. It should be checked if this test procedure in this
ZLT is done correctly and if measurement inaccuracies are likely. The author beliefs that
there it is possible that the sample is normally distributed but there is insignificant proof
to state that the error normally distributed.

Schmertmann et al. (1978)


The same procedure is followed for the Schmertmann et al. (1978) method. The histogram
plot with the fitted normal distribution is given in figure 5.8. With Matlab, a p-value of
0.25 is calculated. Because 0.25 > 0.05 the null hypothesis can not be rejected and retains.
Looking at the fitted normal distribution it can be seen that there is indeed reason to
believe that the sample is normally distributed.

When this distribution is assumed, a characteristic value for the error can be calculated.
The estimated normal parameters (µ̂ and σ̂) are calculated with Matlab and are indicated
in the figure. Looking at the 95% confidence interval, it can be stated that the characteristic
error is:
Errorchar = µ̂ − 1.645 · σ̂ = 42.4 − 1.645 · 13.3 = 20.5% (5.6)
It can be concluded that when the settlements are calculated with the Schmertmann
method, in 95% of the cases, the calculated settlement will overestimate the measured
settlement with 20% or more.

Peck et al. (1996)


Since the method of Peck et al. (1996) follows a very similar procedure as Schmertmann
et al. (1978) it is expected that the output has a similar distribution. The histogram with
the fitted normal distribution can be seen in figure 5.9. The calculated p-value is 0.27.

46 Chapter 5 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

12
µ̂ = 42.4%
10 σ̂ = 13.3%

8
Frequency

0
10 18 26 34 42 50 58 66
Error (%)

Figure 5.8: Histogram of the error of the Schmertmann method with fitted normal
distribution

The null hypothesis is retained and looking at the figure a reasonable fit with the normal
distribution curve can be observed. There can be concluded that a normal distribution of
the error is a reasonable assumption. Looking at the 95% confidence interval, it can be
stated that the characteristic error is:

Errorchar = µ̂ − 1.645 · σ̂ = −1.8 − 1.645 · 23.2 = −40.1% (5.7)

12
µ̂ = −1.8%
10 σ̂ = 23.2%

8
Frequency

0
−80 −60 −40 −20 0 20 40 60 80
Error (%)

Figure 5.9: Histogram of the error of the Peck et al. method with fitted normal distribution

It can be concluded that when the settlements are calculated with the Peck et al. (1996)

Chapter 5 Stef Engels 47


Correlating CPT data to stiffness parameters of sand in FEM

µ̂ = 1.4%
8
σ̂ = 19.8%

6
Frequency

0
−35−25−15 −5 5 15 25 35
Error (%)

Figure 5.10: Histogram of the error of the Robertson method with fitted normal distribution

method, in 95% of the cases, the calculated settlement will underestimate the measured
settlement with 40% or less.

Robertson (1990)
Finally, the results of the Robertson (1990) method are analysed. The histogram of the
error and the fitted normal distribution are presented in figure 5.10. A p-value of 0.16 is
calculated. Therefore the null hypothesis retains and according to the Shapiro-Wilk test
there is insignificant evidence to state that the sample is not normally distributed. Looking
at figure 5.10 however, this hypothesis can be questioned.

When a normal distribution is assumed, the characteristic error at the 95% confidence
interval can be calculated as:

Errorchar = µ̂ − 1.645 · σ̂ = 1.4 − 1.645 · 19.8 = −31.2% (5.8)

It can be concluded that when the settlements are calculated with the Robertson method,
in 95% of the cases, the calculated settlement will underestimate the measured settlement
with 31% or less.

Conclusion
To summarize the previous analysis some concluding remarks are made. The method
proposed by De Beer (1965) does not pas the Shapiro-Wilk test and therefore the error
of this method is assumed not to be normally distributed. Therefore it is hard to predict
the error between the expected settlement and the observed settlement. The error in the
method of Schmertmann et al. (1978), Peck et al. (1996) and Robertson (1990) is assumed
to be normally distributed.

48 Chapter 5 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

To guarantee that the settlement in the sand layer is not underestimated, with 95%
confidence, the author advices:

• Reduce the calculated settlement with 20% in the Schmertmann et al. (1978) method
• Increase the calculated settlement with 40% in the Peck et al. (1996) method
• Increase the calculated settlement with 31% in the Robertson (1990) method

It has been shown that the Schmertmann et al. (1978) method is very conservative. The
methods of Peck et al. (1996) and Robertson (1990) give good results. The mean value of
the error is very close to zero. The Robertson (1990) method is the most favourable since
the mean value of the error is the closest to zero and the standard deviation is the smallest.
The Schmertmann et al. (1978) method is the safest method, since the settlement is never
underestimated. There should be noted that the results of this analysis are specific for this
site.

5.4.3 Influence of the CaCO3 content


As indicated before, the sands that are encountered on the site have a certain calcium
carbonate (CaCO3 ) content. The presence of CaCO3 is due to biological processes from
various forms of life like coral and shellfish. One of the characteristics of sands with a high
CaCO3 content is that grain crushing occurs at lower stresses than silica based sands. In
this section the influence of the CaCO3 content on the settlement behaviour of sand is
investigated.

Test locations

Over the site, various tests are done to determine the CaCO3 content in the sand. The
measurements for each test location are summarized in Appendix E. It can be seen that
not on every location information about the CaCO3 content is available.

For this analysis two test groups are created. A test group in which the CaCO3 content
is considered high (CaCO3 > 70%) and a test group in which the CaCO3 content is
considered low (CaCO3 < 15%). The test groups are give in table 5.2.

Table 5.2: Test locations with high and low CaCO3 content

Samples 1 2 3 4 5 6
CaCO3 > 70% DR107 DV113 DI121 DV128 DW129 DZ116
CaCO3 < 15% EZ105 DP154 DT159 BP150 DA117 DQ138

The influence of the CaCO3 content will be evaluated by looking at the error of the
calculated settlement with respect to the measured settlement. The distribution of the
errors will be compared between samples with high CaCO3 content and samples with
low CaCO3 content. Since the method of De Beer (1965) does not indicate a distributed
output of the error, it is not analysed.

The estimated mean µ̂ and the estimated standard deviation σ̂ of the error for both
test groups are given in table 5.3.

Chapter 5 Stef Engels 49


Correlating CPT data to stiffness parameters of sand in FEM

Table 5.3: Estimated normalized parameters for the error

Estimated normalized parameters µ̂(%) σ̂(%)


Schmertmann 42 14.07
CaCO3 > 70% Peck et al. -2 24.5
Robertson 10 18.72
Schmertmann 44 10.47
CaCO3 < 20% Peck et al. 0 18.21
Robertson -3 22.91

Conclusion
For the Schmertmann et al. (1978) and the Peck et al. (1996) method, the author believes
there is no reason to assume the CaCO3 content has a big influence on the expected error
of the calculations. The estimated mean and standard deviation are very comparable for
both high and low CaCO3 content (table 5.3).

For the Robertson (1990) method, the estimated mean of the high CaCO3 content samples
is significantly higher. This means that the expected settlement is higher than the observed
settlement samples with high CaCO3 content. In other words, the stiffness of the sand
body is higher than expected. This could be due to an increase in soil gradation during
grain crushing. This conclusion is consistent with the type C compressive behaviour which
is explained in Section 3.2.3

It should be noted that the sample size of six is quite small due to lack of data available.
Therefore the results of this analysis should be used as a guideline of what to expect rather
than a definite conclusion. To come to a stronger conclusion more tests need to be done to
determine the CaCO3 content on more locations.

5.4.4 Difference between analytical approach and reality


In this section, some concluding remarks are made which summarizes the assumptions in
the analytical analysis which can differ from the reality. Most of the analytical methods
proposed use a simplified strain influence diagram which are explained in section 3.3.2 and
section 3.3.3. The influence diagrams are a simplification of the reality.

Furthermore the analytical methods are derived to predict direct settlement. Using
these methods it is assumed that during a ZLT direct settlements are measured in the field.
This is not necessarily the case. Looking at the evolution of the load-settlement curve of a
ZLT in time there is reason to believe that creep settlements can play a role. The main
reason that supports this hypothesis is the generation of strains when no incremental load
is applied.

Drained loading conditions are assumed, because mostly sand is encountered. How-
ever, there are spots in the subsurface that contain silt. To make sure that no excess pore
pressures are generated, the pore pressures at these spots should be measured during a
ZLT. When there are excess pore pressures, the extra settlement can also be related to the
dissipation of pore pressures in less permeable layers.

The analytical methods used in this thesis are all based on correlations between CPT
parameters and stiffness parameters for sand. It is stated again that this methodology is
highly empirical and site specific. There are many uncertainties in correctly determine the

50 Chapter 5 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

state and strength parameters of the soil based on two CPT variables (qc , fs ) only.

Chapter 5 Stef Engels 51


Chapter 6

Site specific correlation

In this chapter the results of ZLT’s will be used to obtain a correlation between a secant
Young’s modulus and the cone resistance for the sand encountered at the site of Kuwait.
The correlations in this chapter will be obtained by doing a linear regression analysis. In
the first section the secant Young’s modulus will be evaluated and compared with the
stiffness obtained from the method proposed by Robertson (1990).

During a ZLT a stress bubble develops underneath the loaded area (Section 3.2.4). From
this stress bubble it can be seen that the stress increments underneath the footing decreases
with increasing depth. This means that the deeper layers have less influence than the
layers close to the footing. To take the effect of the stress distribution into account, the
simplified strain influence diagram as proposed by Schmertmann et al. (1978) will be used.
Three separate analysis will be done.

In the first analysis, the compressing soil strata will be generalized as one layer with
the thickness equal to the influenced depth. For the second analysis the layering will be
implemented as suggested by the SBTn chart according to Robertson (1990). In the final
analysis every measurement during the CPT will be considered a separate layer. For all
these cases the obtained correlations will be evaluated and compared.

6.1 Compare back calculated secant modulus Es with ERob


In this section back calculated values for the secant Young’s modulus are determined.
These values are calculated using the measurements from the ZLT’s. According to Atkinson
(2000) the settlement underneath a footing can be evaluated as:

qb · B · (1 − v 2 ) · Iρ
s= (6.1)
Es

Where:
qb is the unit load acting on the base
B is the footing width
v is the Poisson’s ratio
Iρ the settlement influence factor
Es is a secant Young’s modulus

The parameter Es represents an average secant Young’s modulus of the compressing


soil strata. Since soil stiffness is stress dependent the value of Es changes with stress level.
For dense sand under drained loading conditions, a value of v = 0.35 can be used (Das,
2010). The value of influence factor Iρ depends on the shape and rigidity of the footing,
the embedded depth and the thickness of the foundation layer. Considering a rigid 3 x 3
m squared footing at ground level, a value of Iρ = 0.89 can be used (Mayne and Poulos,
1999).

52
Correlating CPT data to stiffness parameters of sand in FEM

14

12

10
Frequency

0
−40 −20 0 20 40 60 80 100 120
Percentage difference between Es and ERob (%)

Figure 6.1: Percentage difference between the back calculated value Es and the Robertson
stiffness parameter ERob

When using the measured direct settlement of the ZLT’s it is possible to find a value of
Es . This value is calculated for each of the selected ZLT’s. The calculated Es is compared
to the average stiffness obtained from Robertson’s method over the influenced area. The
results are presented in Appendix F. The difference between the Es and ERob is visualized
in figure 6.1. There is no indication that this difference follows a well known distribution.
Positive values of the percentage difference indicate that ERob is higher than Es . It can be
concluded that for the site in Kuwait, the Robertson (1990) method gives an indication of
the secant Young’s modulus of the soil mass, but the accuracy is very limited. In extreme
cases the stiffness will be overestimated up to almost 100%.

6.2 Regression analysis using one layer


6.2.1 Correlation between qc and Es
Many research is done to find a direct correlation between the cone resistance qc and the
secant Young’s modulus Es . This direct correlation often has the form:
Es = α · qc (6.2)
Many researchers have tried to find a suitable value for the parameter α. The obtained
values of α usually vary between 2 and 5 (Lunne and Christofferson, 1983) (Schmertmann
et al., 1978). Although the concept of relating Es to qc with one constant parameter sounds
convenient, care must be taken using such a correlation. Lehane et al. (2008) showed that
there is a weak dependence between α and qc at small strain levels. At these strain levels
care should be taken using one constant value for α.

An analysis for this site is done to determine a suitable value for α. The compres-
sion soil strata is generalized as one layer. This is first done for all the selected ZLT’s and

Chapter 6 Stef Engels 53


Correlating CPT data to stiffness parameters of sand in FEM

afterwards for the ZLT’s where only sand is encountered. The results are presented in
figures 6.2 and 6.3.

Regression analysis using one layer


7

5
α = -0,1585qc + 6,2634
R² = 0,2957
4
α

0
5 7 9 11 13 15 17 19 21 23 25
qc (MPa)

Figure 6.2: Regression analysis generalized as one layer

Regression analysis using one layer


7

4
α = -0,3123qc + 8,0552
α

R² = 0,6162
3

0
5 7 9 11 13 15 17 19 21
qc (MPa)

Figure 6.3: Regression analysis where only sand is encountered generalized as one layer

The dependence between α and qc can be clearly seen. This indicates that it would be
inaccurate to use a constant value for α. The suggested linear regression is indicated
in both figures. The value of R2 indicates how well the scattered data points fit on the
suggested regression line. A value of R2 = 1 represents a perfect fit of the data points on
the regression line. In regression analysis the quality of the obtained relationship is not
only captured in the parameter R2 . For a correct statistical analysis also the residuals
need to be examined. The residual is defined as:

Residual = M easured value − P redicted value (6.3)


The predicted value is obtained by using the relationship given by the regression line. In
a qualitatively good regression model the residuals do not show any correlation with the
predicted value and are normally distributed. This can be visualized in a residual plot. To
get a better understanding of the magnitudes of the residuals the standardized residuals

54 Chapter 6 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

Standard residuals using one layer


2,5

1,5

1
Standard residual
0,5

0
2 2,5 3 3,5 4 4,5 5 5,5 6
-0,5

-1

-1,5

-2

-2,5
Predicted value for α

Figure 6.4: Standardized residuals of α, sand generalized as one layer

can be calculated. The residuals are then scaled to values which can be compared to
values belonging to a standard normal distribution. The procedure of calculating these
values is evaluated in Appendix G. The plot of the standardized residuals of figure 6.3
is given in figure 6.4. It can be seen that the residuals look randomly distributed. The
horizontal lines in the figure represents the 95% interval. This means that about 95% of
the standard residuals should be in within this limits. Both criteria hold, which means
that the regression analysis is correct.

It is clearly visible that the regression line is more accurate for the ZLT’s where only sand
is encountered. Looking at the stiffness properties of sand at this site, figure 6.3 can be
used. The values for α are roughly between 2 and 5, which is consistent with previous
research. It is advised at these small strain levels to take the dependence between α and
qc into account. When the dependency of Es with qc is plotted figure 6.5 is obtained. In
this figure it seems that Es does not depend on the value of qc . It can be concluded that
generalizing the strata as one layer, the value of Es is independent of qc at this specific
strain level. The average value of Es = 50000 kPa can be used.

Regression analysis using one layer


70000

60000
Es = -32,81qc + 49974
R² = 9E-05
50000

40000
Es (kPa)

30000

20000

10000

0
5 7 9 11 13 15 17 19 21
Average qc (MPa)

Figure 6.5: Regression analysis between Es with qc , sand generalized as one layer

Chapter 6 Stef Engels 55


Correlating CPT data to stiffness parameters of sand in FEM

The regression line in figure 6.2 is too limited in accuracy for practical applications.
Therefore only the sand locations will be evaluated in further analysis.

6.2.2 Correlation between Qtn and Es


In the previous section the cone resistance is evaluated and correlated to the stiffness
parameter Es . In this section the same analysis is done with the normalized cone resistance
Qtn . It is investigated if a direct correlation exist between Qtn with Es of the form:

Es = β · Qtn (6.4)
Since Qtn is a dimensionless parameter the factor β will have the unit kPa. The regression
plots for both β and Es are plotted in figures 6.6 and 6.7.

Regression analysis using one layer


450

400

350

300
β (kPa)

250
β = -1,429Qtn + 539,64
R² = 0,4434
200

150

100

50

0
100 120 140 160 180 200 220 240
Average Qtn

Figure 6.6: Regression analysis between β and Qtn , sand generalized as one layer

Regression analysis using one layer


70000

60000
Es = 14,778Qtn + 46774
R² = 0,0025
50000

40000
Es (kPa)

30000

20000

10000

0
100 120 140 160 180 200 220 240
Average Qtn

Figure 6.7: Regression analysis between Es and Qtn , sand generalized as one layer

It can be seen that the results are quite similar to the previous analysis. Factor β is
negatively correlated with Qtn . The value of R2 is lower than in the previous regression
analysis. This means that the regression line fits the data better when qc is correlated to
α. The horizontal regression line in figure 6.7 indicates that stiffness does not depend on

56 Chapter 6 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

Qtn and therefore the variables are uncorrelated. Be aware that this is the case for this
specific stress level and only applies when the soil is generalized as one layer.

To check the validity of the correlation between Qtn and β the standardized residu-
als are plotted in figure 6.8. The standard residuals plot between the 95% boundaries and
look random. Therefore the correlation is valid.

Standar residuals using one layer


2,5

1,5

1
Standard residuals

0,5

0
200 220 240 260 280 300 320 340 360
-0,5

-1

-1,5

-2

-2,5
Predicted values for β (kPa)

Figure 6.8: Standardized residuals of β, sand generalized as one layer

6.2.3 Conclusion
The soil strata is considered as one layer and a mean value of the cone resistance is used to
obtain a direct correlation. When this approach is used, the correlation between qc and α
is favourable over the correlation between Qtn and β. When this generalization is done, the
value of the stiffness is uncorrelated with both qc ad Qtn . This indicates that a constant
value of Es = 50000 kPa can be used for this specific site where the applied pressure on
the footing is 200 kPa.

6.3 Regression analysis using suggested layers


6.3.1 Correlation between qc,weigthed and Es
In the analytical analysis (Section 5.2), the layering is evaluated using the SBTn chart
developed by Robertson (1990). In this analysis the suggested layers are used. This means
that the average qc of each layer is calculated. Each qc will be weighted by a certain factor.
This factor needs to take into account that the upper layers have more influence on the
settlement then the deeper layers. The weight factor will be determined using the strain
influence factor Iz proposed by Schmertmann et al. (1978). The weight factor is evaluated as:
R zbot,i
z Iz dz
Wi = Rtop,i
2B
(6.5)
0 Iz dz
By definition the summation of the weighting factors should equal a value of one, since the
entire influenced area is taken into account.
n
X
Wi = 1 (6.6)
i=1

Chapter 6 Stef Engels 57


Correlating CPT data to stiffness parameters of sand in FEM

Regression analysis using suggested layers


6

5
α = -0,0945qc + 5,1173
R² = 0,0953

4
α
3

0
5 7 9 11 13 15 17 19
Weighted average qc (MPa)

Figure 6.9: Regression analysis between α and qc,weigthed using the suggested layering

The average weighted cone resistance that can be used for the analysis is then calculated
as: n X
qc,weighted = Wi · q¯c,i (6.7)
i=1
Where:
n is the amount of layers
Wi is the weight factor for layer i
ztop,i the top coordinate of layer i
zbot,i is the bottom coordinate of layer i
q¯c,i is the average cone resistance of layer i
qc,weighted is the weighted cone resistance

In this analysis the correlation of the following form is investigated:

Es = α · qc,weighted (6.8)

The results of the regression analysis for both α and Es with qc,weigthed are plotted in figure
6.9 and 6.10.

The correlation between α and qc,weigthed is very weak. The low value of R2 indicates
that the fit does not accurately match the data and therefore this correlation should not
be used. The correlation between Es and qc,weigthed is better, but still the value of R2 is
relatively low. The standardized residuals of Es are plotted in figure 6.11. In this figure
it can be seen that one data point is clearly plotting outside the 95% interval. Looking
at the magnitude of the standardized residual of this point in comparison with the other
standardized residuals, this data point can be considered an outlier. Outliers can occur
due to mistakes in measurements or an experimental error. Since this outlier influences the
results significantly and is seems inconsistent with the rest of the data, it will be removed in
the following analysis. There is no ’rule of thumb’ that always apply dealing with excluding
outliers. Some statistical engineers drop them out of the analysis when the standardized
residual is removed two standard deviations from zero.

When excluding the outlier, the obtained regression line for Es is represented by figure
6.12. The value of R2 is much higher and a better correlation is obtained. This shows that
the outlier has a significant effect on the accuracy of the regression line.

58 Chapter 6 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

Regression analysis using suggested layers


70000

60000

50000 Es = 2907,2qc + 12427


R² = 0,3948
40000
Es (kPa)

30000

20000

10000

0
5 7 9 11 13 15 17 19
Weighted average qc (MPa)

Figure 6.10: Regression analysis between Es and qc,weigthed using the suggested layering

Standard residuals using suggested layers


2,5

1,5

1
Standard residual

0,5

0
30000 35000 40000 45000 50000 55000 60000
-0,5

-1

-1,5

-2

-2,5
Predicted value for Es (kPa)

Figure 6.11: Standardized residuals of Es using the suggested layering

Regression analysis using suggested layers


70000

60000
Es = 3129,2qc + 8401,8
R² = 0,53368
50000

40000
Es (kPa)

30000

20000

10000

0
5 7 9 11 13 15 17 19
Weighted average qc (MPa)

Figure 6.12: Regression analysis between Es and qc,weigthed using suggested layering and
exclude outlier

Chapter 6 Stef Engels 59


Correlating CPT data to stiffness parameters of sand in FEM

6.3.2 Correlation between Qtn,weigthed and Es


In this section the correlation between Qtn,weighted and Es will be evaluated when the
suggested layering based on the SBTn chart is used. Qtn,weighted is defined as:
n
Wi · Q¯tn,i
X
Qtn,weighted = (6.9)
i=1

Where the weighting factors are the same as in equation 6.5 and Q¯tn,i is the average
normalized cone resistance in layer i. The obtained regression line for both β and Es are
given in figures 6.13 and 6.14.

Regression analysis using suggested layers


400

350
β = -0,1372Qtn + 289,75
R² = 0,0073
300

250
β (kPa)

200

150

100

50

0
100 120 140 160 180 200 220 240 260
Weighted average Qtn

Figure 6.13: Regression analysis between β and Qtn,weigthed using suggested layering

Regression analysis using suggested layers


70000
Es = 234,5Qtn + 5450,8
R² = 0,4101
60000

50000

40000
Es (kPa)

30000

20000

10000

0
100 120 140 160 180 200 220 240 260
Weigthed average Qtn

Figure 6.14: Regression analysis between Es and Qtn,weigthed , using suggested layering

No correlation is found between β and Qtn,weighted . A better correlation is found between


Es and Qtn,weighted . This will be further investigated.

To check for outliers, the standardized residuals are plotted in figure 6.15. Also in
this plot one of the values lies outside the 95% confidence interval and deviate from the
other values. This value will be considered as an outlier and is excluded from the analysis.

60 Chapter 6 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

When the outlier is excluded the regression plot in figure 6.16 is the result. This correlation
has a higher value for R2 and can be used for technical design applications.

Standard residuals using suggested layers


3

2,5

1,5

1
Standard residual

0,5

0
35000 40000 45000 50000 55000 60000
-0,5

-1

-1,5

-2

-2,5
Predicted values for Es (kPa)

Figure 6.15: Standardized residuals of Es using suggested layering

Regression analyisis using suggested layers


70000

60000
Es = 275,01Qtn - 3570
R² = 0,6377
50000

40000
Es (kPa)

30000

20000

10000

0
120 140 160 180 200 220 240 260
Qtn, weighted

Figure 6.16: Regression analysis between Es and Qtn,weigthed using suggested layering and
exclude outlier

6.3.3 Conclusion
The obtained correlation between Qtn,weigthed and Es can be used when the SBTn based
layering is used.
Es (kP a) = 275 · Qtn,weighted − 3570 (6.10)

This correlation is more accurate than the one obtained between qc,weighted and Es . The
best correlation using the suggested layers is slightly more accurate than when the soil
strata is generalized as one layer. Another difference is that when the soil is divided into
multiple layers, the stiffness positively correlate with the cone resistance. When the soil
was generalized as one layer the stiffness and cone resistance were uncorrelated.

Chapter 6 Stef Engels 61


Correlating CPT data to stiffness parameters of sand in FEM

6.4 Regression analysis using 600 layers


In the next analysis, every single measurement of the CPT will be considered as a separate
layer. In this analysis all the measured cone resistances will be used and weighted with
the influence factor proposed by Schmertmann et al. (1978). After the analysis it can be
concluded if the accuracy of the correlations increases when dividing the soil into more
layers. Since the strain influence diagram of Schmertmann et al. (1978) is used only the
first 6 m of the sand is assumed to compress. Since measurements are registered every cm
the first 6 m will be divided into 600 layers. The weighting of the cone resistance will be
done in the same way as for the ”suggested layering” approach.

6.4.1 Correlation between qc,weighted and Es


A regression analysis is done to look for a direct correlation between the average weighted
cone resistance qc,weighted and the stiffness parameter Es . The obtained regression lines for
α and Es are presented in figures 6.17 and 6.18.

Regression analysis using 600 layers


6

5
α = -0,0692qc + 4,9815
R² = 0,0429

4
α

0
5 7 9 11 13 15 17 19
Weighted average qc (MPa)

Figure 6.17: Regression analysis between α and qc,weigthed using 600 layers

Regression analysis using 600 layers


80000

70000 Es = 3520,3qc + 6151,1


R² = 0,56384
60000

50000
Es (kPa)

40000

30000

20000

10000

0
5 7 9 11 13 15 17 19
Weighted average qc (MPa)

Figure 6.18: Regression analysis between Es and qc,weigthed using 600 layers

qc,weighted seems uncorrelated (or very weakly correlated) with the parameter α and

62 Chapter 6 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

is therefore not further analysed. qc,weighted shows positive correlation with Es . The
standardized residuals are plotted to search for possible outliers (figure 6.19).

Standard residuals using 600 layers


2,5

1,5

1
Standard residual

0,5

0
32000 37000 42000 47000 52000 57000 62000
-0,5

-1

-1,5

-2

-2,5
Predicted value for Es (kPa)

Figure 6.19: Standardized residuals of Es using 600 layers

In this figure, it can be seen that one point plots outside the 95% confidence interval and
deviate from the rest of the data points. This point will be considered as an outlier and
will be excluded in a new regression analysis. The regression line obtained is represented
by figure 6.20. A useful correlation exist between qc,weighted and Es due to a relatively high
value for R2 .

Regression analysis using 600 layers


80000

70000 Es = 3520,3qc + 6151,1


R² = 0,56384
60000

50000
Es (kPa)

40000

30000

20000

10000

0
5 7 9 11 13 15 17 19
Weighted average qc (MPa)

Figure 6.20: Regression analysis between Es and qc,weigthed using 600 layers and exclude
outlier

6.4.2 Correlation between Qtn,weighted and Es

In this section, the same analysis is performed with the normalized weighted cone resistance.
The regression lines are presented in figures 6.21 and 6.22.

Chapter 6 Stef Engels 63


Correlating CPT data to stiffness parameters of sand in FEM

Regression analysis using 600 layers


400

350
β = 0,0208Qtn+ 267,83
R² = 0,0002
300

250

β (kPa) 200

150

100

50

0
100 120 140 160 180 200 220 240 260
Weighted average Qtn

Figure 6.21: Regression analysis between β and Qtn,weighted using 600 layers

Regression analysis using 600 layers


70000
Es = 260,69Qtn+ 2006,1
R² = 0,4775
60000

50000

40000
Es (kPa)

30000

20000

10000

0
100 120 140 160 180 200 220 240 260
Weighted average Qtn

Figure 6.22: Regression analysis between Es and Qtn,weighted using 600 layers

No correlation is found between β and Qtn,weighted and therefore this will not be further
analysed. The positive correlation between Es and Qtn,weighted looks promising and
therefore the standardized residuals are evaluated (figure 6.23). The standard residuals
have a random pattern but one outlier can be spotted. Therefore a new regression analysis
will be done excluding this outlier. The result can be found in figure 6.24. A high value for
R2 is obtained which means that the data is well represented by the fitted regression line.

64 Chapter 6 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

Standard residual using 600 layers


2,5

1,5

1
Standard residual
0,5

0
35000 40000 45000 50000 55000 60000 65000
-0,5

-1

-1,5

-2

-2,5
Predicted value for Es (kPa)

Figure 6.23: Standardized residuals for Es using 600 layers

Regression analysis using 600 layers


80000

70000

60000

50000
Es = 280,43Qtn - 2814,5
Es (kPa)

R² = 0,6438
40000

30000

20000

10000

0
120 140 160 180 200 220 240 260
Qtn,weighted

Figure 6.24: Regression analysis between Es and Qtn,weighted using 600 layers and exclude
outlier

6.4.3 Conclusion
Dividing up the soil stratum into 600 layers increases the value of R2 slightly compared
to the suggested layering. However since the limited data points in the data set and the
values of R2 are almost the same, both methods can be considered equally accurate in
this analysis. It is advised to use the correlation between Es and Qtn,weighted over the
correlation between Es and qc,weigthed since this relationship is more accurate. The author
believes that correlations between Es and Qtn,weighted can be used in both the ”suggested
layering” approach and ”600 layers” approach. The best correlation for the ”600 layers”
approach is:

Es (kP a) = 280.43 · Qtn,weighted − 2814.5 (6.11)


It should be emphasized that regression models are highly empirical and not necessarily
correct. Furthermore they are probably site specific and therefore should be checked on
other types of sand. Still it can be very useful for design purposes as long as one realize
what the limitations of such an analysis are. As the great statistician George Box ever
said: ”Essentially, all models are wrong, but some are useful.”

Chapter 6 Stef Engels 65


Correlating CPT data to stiffness parameters of sand in FEM

ZLT's

Figure 6.25: Stiffness reduction curve with strain level of ZLT’s (Atkinson and Sallfor,
1991)

6.5 Interpretation of Es
The previous sections proposed correlations between the secant Young’s modulus Es and
the (normalized) cone resistance. In this section some additional information is provided
on the interpretation of the parameter Es .

As stated earlier the obtained value for Es is site specific. The value of the secant
Young’s modulus varies with both stress and strain level. Therefore there is never a
unique value for the secant Young’s modulus. The parameter Es that is derived in the
previous sections corresponds with a stress level of 0.25. This means that the applied
foundation pressure is about 0.25 times the failure load (ultimate bearing capacity). When
a different pressure is applied, the stress level changes and therefore the value of Es changes.

Stiffness also depends on strain level. Typical strains measured during a ZLT are in
the order of 0.1 - 0.3%. The typical stiffness degradation curve is an S-shaped curve in
is presented in figure 6.25. In this figure the strain levels observed in the ZLT are also
indicated. At these strain levels the soil stiffness is higher than the residual soil stiffness at
ref
higher strains. Therefore the values of Es can be high in comparison with the E50 that is
ref
used in PLAXIS. How to relate Es to E50 is evaluated in section 7.4.

66 Chapter 6 Stef Engels


Chapter 7

Numerical verification with PLAXIS


2D

In this chapter, the problem of the settlement in the ZLT’s is analysed numerically. Only
the locations where sand is encountered are analysed. For this analysis PLAXIS 2D is
used. This program evaluates the settlement based on a finite element calculation. The
subsurface is modelled using the Hardening Soil model. The basic features of this models
are discussed in section 3.4. This model is chosen over the HS-small model due to high
uncertainties in evaluating the extra parameters (V0 and γ0.7 ). The objective of this analysis
is to check the engineering application of one of the correlations that is derived in the
previous chapter. The correlation is verified by checking if it is possible to model the ZLT
numerically based on input stiffness parameters that are derived from CPT’s. The reader is
reminded that this verification only holds for the site at Kuwait. Afterwards the influence
of the ZLT procedure is evaluated by using PLAXIS 3D. At the end of the chapter the
obtained conclusions are summarized.

7.1 Used correlation for verification


One of the correlations that is derived in the previous chapter shall be checked with multiple
numerical calculations. Three different approaches were distinguished:

• Soil mass generalized as one layer


• Suggested layering according to SBTn
• Soil mass divided into 600 layers

In PLAXIS a convenient amount of soil layers can be introduced in the geometry of the
model. Dividing the soil mass into 600 layers of 1 cm is highly impractical. Therefore
it is chosen to evaluate the ”suggested layering” approach. Besides the correlation that
is derived using the SBTn layering can be considered just as accurate as the ”600 layer”
approach. Therefore the correlation that will be evaluated is the correlation between the
weighted normalized cone resistance and the secant Young’s modulus (Section 6.3.3):

Es (kP a) = 275 · Qtn,weighted − 3570 (7.1)

7.2 Modelling approach


The general outline of the model that evaluates the settlement of a footing is drawn in
figure 7.1. A soil mass is defined and this body is supported by roller supports at the
bottom and at both sides. The roller supports on the sides prevent movements in the
x-direction and the supports on the bottom prevent movements in the y direction. The use
of PLAXIS 2D instead of PLAXIS 3D has the main advantage that the calculation time
reduces significantly. Modelling problems in PLAXIS 2D can be either with a plane strain

67
Correlating CPT data to stiffness parameters of sand in FEM

model or an axisymmetric model. Using a plane strain model means that the strains can
only take place in the x- and y-direction and the strains in the z-direction are zero. Such
an approach can be used when modelling a strip foundation (when the length is about 10
times bigger than the width of the foundation). For a square footing, significant straining
occurs in the z direction and the plane strain model is not suitable. In the axisymmetric
model the strains in all radial directions are equal. This means that the strains in the
x-direction are equal to the strains in the z-direction. Using an axisymmetric approach
implies that the structure is symmetrical along the vertical y-axis.

When modelling the footing in the axisymmetric model, the geometry of figure 7.1 changes.
Because of the symmetry along the vertical y-axis, only half of the geometry needs to be
defined. The half at the right hand side of the y-axis in figure 7.1 is drawn. In PLAXIS 2D
the problem is rotated around the y-axis. The consequence of such a modelling approach
is that the footing is now modelled as a circle instead of a square. The axisymmetric
geometry is visualized in figure 7.2. Because the square footing now is modelled as a circle,
an equivalent diameter is calculated and implemented in the model. The approach is that
the square footing is represent by a circular footing with an equal area (figure 7.3). The
equivalent diameter is calculated as:

2·D
De = √ (7.2)
π

For an accurate calculation it is advised that the width of the soil mass is modelled as four
times the width of the radius of the footing (figure 7.2). With this approach the stress
influenced area due to the loaded footing should be captured.

Figure 7.1: General outline of the PLAXIS model

68 Chapter 7 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

Rotation around
the y-axis

r 3xr

Symmetry axis

Figure 7.2: Axisymmetric geometry of the footing in the PLAXIS model

D De

Figure 7.3: Square footing modelled as equivalent circular footing

7.3 Parameter determination


Since the specific interest in this research is related to the mechanical behaviour of sand,
the Hardening Soil model is a suitable model. Detailed information of this model and
its parameters is provided in Section 3.4.2. Although this model has good features for
modelling realistic soil behaviour, it can be difficult to determine the right values for the
input parameters. This section evaluates the determination of these parameters.

Volumetric weight (γ)

Both the saturated and unsaturated volumetric weight of the soil needs to be deter-
mined because it has a direct influence on the in situ stress state of the soil. Obtaining
these values is relatively simple and can be done by weighting samples from the site. Field
Density Tests (FDT’s) are performed on the site. With these tests the dry density and wet

Chapter 7 Stef Engels 69


Correlating CPT data to stiffness parameters of sand in FEM

density are determined. When the maximum density is determined in the laboratory, the
relative compaction can also be determined. The results of the FDT performed on the site
are given in table 7.1.

Table 7.1: Results of FDT test

Parameter Unit FDT 1 FDT 2 FDT 3 FDT 4 FDT 5 Average


Wet density g/cm3 1.815 1.786 1.836 1.801 1.826 1.813
Dry density g/cm3 1.715 1.714 1.719 1.732 1.720 1.720
Max. dry density g/cm3 1.803 1.803 1.797 1.803 1.803 1.802
Relative compaction % 95.1 95.1 95.7 96.1 95.4 95.5

Cohesion (c)

Sands in general are cohesionless soils. Sands can also behave cohesive when the pore
pressures between the voids are negative (suction). An example where this phenomena can
be observed is with sand castles. In this analysis it is assumed that there is no suction.
PLAXIS can handle cohesionless soils but is it advised to enter a small value for the
cohesion to avoid numerical complications (Brinkgreve and Vermeer, 2016).

Friction angle (ϕ)

Specific test data from the site is available. Direct Shear Tests are done over the en-
tire site to determine the friction angle ϕ. Although many tests are done, it should be
noted that the friction angle can also vary over the depth. It is difficult to get an accurate
measurement because of sampling disturbance. This problem is of specific relevance with
sandy cohesionless soils. For the problem of the settlement of a loaded footing it is not
expected that a very accurate knowledge of the friction angle is required. The friction
angle determines the shape of the Mohr-Coulomb failure surface, but the ZLT is not a
failure test. This does not mean that the friction angle has no influence at all, because
some local failure can be expected but not an entire failure surface develops. Based on the
tests available, a value of ϕ = 38◦ is used.

Poisson’s ratio (v)

The Poisson’s ratio can be calculated with triaxial test data by measuring the verti-
cal and horizontal strains during the test. These tests are not available and therefore an
estimation of the Poisson’s ratio is done. The sand at the site is dense sand and a Poisson’s
ratio of v = 0.35 can be used as an estimation (Das, 2010).

Angle of dilatancy (ψ)

The dilatancy angle ψ can be estimated from the friction angle. This parameter is
predominant for dense sands. Since compaction works are performed over the entire site,
most of the sand encountered will be dense sand. The following formula can be used to
estimate the dilatancy angle (Brinkgreve and Vermeer, 2016):
ψ = ϕ − 30 (7.3)
Overconsolidation ratio (OCR)

The overconsolidation ratio is an important parameter. Overconsolidated soils will respond


much stiffer than normally consolidated soils. The overconsalidation ratio is defined as:

70 Chapter 7 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

0
σp
OCR = 0 (7.4)
σv0
In the HS-small model, the pre-consolidation stress is the points that marks the tran-
sition between elastic and plastic deformations. When the stress is smaller than the
pre-consolidation stress the response is dominated by the unloading reloading stiffness of
the soil mass. In the default settings of PLAXIS the unloading reloading stiffness of soil is
typical 3 times as high as the stiffness of the normally consolidated material.

The value of the OCR decreases with depth. PLAXIS can handle only one value for
the OCR in each layer. Therefore the value of the OCR in the middle of each layer is
determined and this value is used. The OCR is determined using two correlation formulas
(Mayne, 2007):
0
qt 0.22 σv0 −0.31
K0 = 0.192 · ( ) ·( ) · OCR0.27 (7.5)
σatm σatm
K0 = (1 − sin(ϕ)) · OCRsin(ϕ) (7.6)
The correlation in equation 7.5 is developed from calibration chamber tests on clean sands.
It is difficult to evaluate the stress history of sands. The use of the correlation given by
equation 7.5 is rather limited when it is used on its own. However, if a relation between
K0 and OCR is established (equation 7.6) it is possible to give a more accurate prediction
of the OCR (Mayne, 2007). The correlation in equation 7.6 holds for soils that are not
highly cemented nor structured.

The approach is to vary OCR until the same value of K0 is obtained. This iterative
procedure can be done using Matlab, where an initial value of the OCR is chosen. The
procedure starts with OCR = 1 and compares the values of K0 calculated according to
equations 7.5 and 7.6. When both values of K0 differ more than 0.01 an new value of
OCR = 1.01 is used. The procedure of increasing the OCR continuous until the difference
in both values of K0 is less than 0.01.

Exponent m

The exponent m is the parameter which captures the stress dependency of stiffness.
The value of this parameter varies between 0.5 and 1. In dense sand the nonlinearity is
more dominant than in loose sands. A value of m close to 0.5 would be appropriate to
model this nonlinearity. Therefore a values of m = 0.5 is used for the compacted sand
encountered at the site.
ref ref ref
Stiffness parameters E50 , Eoed and Eur

The most challenging parameters to determine are the stiffness parameters. For a re-
alistic simulation in a numerical model, extensive lab testing is needed for obtaining these
parameters. In this research however, the objective is to give a realistic indication of these
parameters based on correlation with CPT results. The procedure to relate Es obtained
ref ref is
from the correlation to E50 is explained in detail in the next section. The value for Eur
ref
determined as 3 times the value of E50 which is recommended by Brinkgreve and Vermeer
ref
(2016) when no extensive lab test results are available. The value for Eoed is taken the
ref
same as the value of E50 .

It is assumed that pore pressures can dissipate during the test, so drained conditions

Chapter 7 Stef Engels 71


Correlating CPT data to stiffness parameters of sand in FEM

apply. Therefore there is no need to specify parameters as void ratio or permeability


coefficients. More advanced parameters that are not discussed will be set as the default
values that are recommended by PLAXIS.

ref
7.4 Relating Es with E50
PLAXIS uses a specified value for the stiffness. The stiffness used in the Hardening Soil
model is defined as a secant Young’s modulus at a specified confining pressure. Typically
the reference confining pressure is equal to the atmospheric pressure which is 100 kPa. This
is the secant Young’s modulus that will be observed in a triaxial test at a stress level of 0.5
(50% of the failure load) where the cell pressure is 100 kPa. The definition is visualized in
figure 7.4.

pref pref
ref
E50

1 ✏
ref
Figure 7.4: Definition of E50

In the procedure suggested by Robertson (1990) also a secant Young’s modulus is obtained.
ref
This secant Young’s modulus cannot directly be compared with the E50 that is used in
PLAXIS. The stiffness Es obtained by Robertson (1990) follows from in situ measurements
and for a specified stress level of 0.25 (25% of the failure load). The secant Young’s modulus
that would be more appropriate for these conditions would be E25 at a confining pressure
which is equal to the in situ soil state (Es = E25 ). The parameter is visualized in figure
7.5.
0
v

0 0 3
4
h h
E25
1
4

v
0

Figure 7.5: Definition of E25

To compare both stiffness parameters, the parameter E25 need to be converted to a value
ref ref
E25 that corresponds with the same reference stress as E50 . This can be done using the
following formula:
0
ref c · cos ϕ − σh · sin ϕ
E25 = E25 · ( ) (7.7)
c · cos ϕ − pref · sin ϕ

72 Chapter 7 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

ref ref
Table 7.2: Calculated values for E25 with varying values for E50 at pref = 100 kPa
ref ref
E50 (kPa) E25 (kPa) Ratio
10000 13977 1.4
20000 27979 1.4
50000 69826 1.4
100000 139652 1.4

The horizontal in situ stress is evaluated as:


0 0
σh = K0 · σv (7.8)
ref ref
The parameters E25 and E50 both correspond to a triaxial test at the same cell pressure
(100 kPa). The two parameters can be related by looking at the stress strain diagram from
triaxial tests under this cell pressure. This is done by using the ”Soil Test” mode which is
available in PLAXIS 2D. The author suspects that there is a fixed ratio between the two
stiffness parameter. To check this hypothesis, an analysis is done with four different values
ref ref
of E50 . From the obtained stress strain curves, the values of E25 can be calculated. The
results are summarized in table 7.2. From this analysis can be concluded that a fixed ratio
between the two parameters exist.
ref
E25
ref
= 1.4 (7.9)
E50
There should be noted that this ratio only holds when the parameter m equals 0.5 and
when the reference pressure is equal to 100 kPa.

7.5 Overview of the numerical validation


In this section a flow chart is presented which summarizes the steps in the numerical
procedure. The intermediate steps are solved by a Matlab program which calculates all the
necessary parameters from the 5 CPT’s that are given as input. The flowchart is presented
in figure 7.6.

7.6 PLAXIS calculation


In this section, one of the ZLT will be implemented in PLAXIS. For all the other ZLT’s
that are included in the analysis, the same procedure is followed. Since the correlation is
based on 13 ZLT’s, 13 numerical calculations will be done. At the end, the analytical and
numerical results are compared with the field measurements.

7.6.1 PLAXIS model


First, the layering as determined in the SBTn will be implemented as different soil layers
in the PLAXIS model. The considerations with respect to the dimensions of the total
geometry in the x- and y-direction are already evaluated in section 7.2. The test that will
be evaluated as an example is ZLT with the code BC159. The SBTn chart and the layering
based on this chart are presented in figures 7.7 and 7.8.

The layering that is suggested based on the soil behaviour type index (Ic,rw ) is implemented
in PLAXIS. The ground water table in the field is measured and is included as well. The
obtained geometry that is used in PLAXIS is presented in figure 7.9. The load and the
footing are also visible. The footing is modelled as a stiff concrete footing.

Chapter 7 Stef Engels 73


Correlating CPT data to stiffness parameters of sand in FEM

Numerical analysis
5 CPT´s at each
ZLT location

Plot averaged
Exclude CPT´s where
profile
depth < 6m

Calculate normalized
Calculate the SBT index Robertson parameters Q tn Calculate in situ stress
for every measurement and Fr state

Determine the Calculate PLAXIS input


layering based on SBT parameters for the HS model

Calculate average stiffness for each Exclude the first 40 cm


layer with the correlation of the CPT

Convert the calculated stiffness to


PLAXIS stiffness parameter

Assign the stiffness to each layer


in PLAXIS and simulate the Zone
Load Test

Compare the displacements in


PLAXIS with the measurements

Figure 7.6: Flowchart of the process in the numerical analysis

10 3

7 6 8 9
Normalized cone resistance, Qtn

10 2

4
5 3
10 1

1=Sensitive fine grained


2=Organic soils
3=Clays
4=Silt mixtures
5=Sand mixtures
6=Clean/silty sand
7=(Gravelly) sand
8=Very stiff sand/clayey sand 1 2
9=Very stiff, fine grained

10 0
10 -1 10 0 10 1
Normalized friction ratio, Fr (%)

Figure 7.7: The normalized CPT data points in the SBTn chart of Robertson (ZLT BC159)

74 Chapter 7 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

Gravelly sand to dense sand

Clean sand to silty sand

Silty sand to sandy silt

Clayey silt to silty clay

Clay to organic soils


Silty clay to clay
2
Depth (m)

6
0 0.5 1 1.5 2 2.5 3 3.5 4
Soil Behaviour Type index

Figure 7.8: Suggested layering based on Ic,rw (ZLT BC159)

Figure 7.9: Geometry of ZLT BC159 in PLAXIS 2D using axisymmetric model

Chapter 7 Stef Engels 75


Correlating CPT data to stiffness parameters of sand in FEM

Table 7.3: Loading steps in PLAXIS analysis

Step Percentage Pressure (kPa)


1 0 0
2 25 50
3 50 100
4 75 150
5 100 200

7.6.2 Loading procedure and output


In the ZLT procedure the footing is loaded in a fixed amount of steps. The real ZLT
procedure includes an unloading reloading cycle which is not included. The loading from 0%
until 100% of the design pressure is simulated in the PLAXIS analysis. The loading steps
are specified in table 7.3. To specify these load steps in PLAXIS the staged construction
option is used. The different stages in the PLAXIS analysis are summarized in table 7.4.
In this table also the mesh is visualized. A fine mesh is used and local mesh refinements
are made near the surface of the footing and in the top layers.

When the calculation is performed by PLAXIS, the displacement underneath the footing is
known. The displacements are visualized in figure 7.10. The displacement of the footing in
ZLT BC159 according to PLAXIS is 9 mm. The measured displacement on the site was
9.1 mm. For this specific ZLT the PLAXIS simulation match reality with a high accuracy.

Figure 7.10: Displacements in the soil body according to PLAXIS (ZLT BC159)

76 Chapter 7 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

Stage Description

In the initial stage PLAXIS performs the


K0 procedure. The initial stresses are de-
termined based on the values K0 , the volu-
metric weight and the groundwater table.

The footing is activated and the load is


applied according to the loading steps in
table 7.3. The loading up till 100% is done
in four steps.

The deformed mesh at 100% of the load is


presented. The deformations are scaled up
20 times for a better visualization of the
deformed soil body.

Table 7.4: Overview of the different stages in PLAXIS

Chapter 7 Stef Engels 77


Correlating CPT data to stiffness parameters of sand in FEM

7.6.3 Results

The procedure explained in the previous section is done for every ZLT where only sand is
encountered. This results in 13 different analyses. The numerical results are compared
with the analytical results and the field measurements. The analytical method uses the
correlation to obtain a value for the secant Young’s modulus and uses elastic theory to
calculate the settlement:

Es = 275 · Qtn,weighted − 3570 (7.10)

qb · B · (1 − v 2 ) · Iρ
s= (7.11)
Es

The results of the analyses are compared in figure 7.11. The exact values of the displacement
are presented in table 7.5. It can be seen that the errors in the numerical method are
generally higher than in the analytical method. The settlements calculated by the numerical
method lie within a range of 30% of the measured settlement. Negative values for the
error indicate an underprediction of the settlement. At small deformations the numerical
method tends to underpredict the settlement. In the analytical analysis the settlements
calculated lie within range of 15% of the measured settlement.

16
Analytical method
14 Numerical method
Measured
12
Settlement (mm)

10

0
7

3
21

4
48

C 0

C 8
7
10

11

12

15

15

14

14

14

11

12

12
I1

I1
R

T
D

B
D

A
D

C
D

B
B

Figure 7.11: Comparison of different analysis

The errors of both methods are also presented in figures 7.12. In this figure the range of
error is visualized. No well known distribution can be discovered for both methods.

78 Chapter 7 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

Table 7.5: Settlements according to analytical and numerical method

Analytical Numerical Measured


Test ID Error (%) Error (%)
settlement (mm) settlement (mm) settlement (mm)
DR107 10.5 -2 10.1 -6 10.8
DV113 9.6 -3 7.9 -20 9.9
DI121 10.9 4 13.1 26 10.4
DD121 8.9 11 5.4 -33 8.0
BC159 9.6 6 9 -1 9.1
BK157 13.1 7 11.5 -6 12.3
AU146 13.1 -3 15.2 13 13.5
BD144 7.6 -2 6.3 -18 7.7
BI148 9.3 -2 7 -26 9.5
CH146 8.8 -3 6.1 -33 9.1
CU110 9.6 13 9.3 10 8.5
CW128 10.9 14 9.7 1 9.6
CT127 11.4 11 9.7 -6 10.3
Average 4 -8

6
6

4
Frequency

Frequency

2 2

0 0
−10 0 10 20 −40 −20 0 20
Error (%) Error (%)
(a) Analytical method (b) Numerical method

Figure 7.12: Histogram of the error for both methods

7.7 Influence of ZLT procedure

In this section a study is done to check the accuracy of the ZLT procedure. It is suspected
that the test set-up as described in section 4.1 influences the soil that is tested. The load
is transferred from the supporting blocks to the footing by extension of a hydraulic jack.
A top view of the geometry of the footing and the supporting blocks are given in figure 7.7.
No exact dimensions of the supporting blocks are given and therefore the dimensions are
approximated.

Chapter 7 Stef Engels 79


Correlating CPT data to stiffness parameters of sand in FEM

1,5 m 1,5 m 3m 1,5 m 1,5 m

Support Support
Footing 3m
Block Block

Figure 7.13: Top view of the geometry of the footing and supporting blocks

From figure it can be seen that the contact area of the footing is the same as the contact
area of both supporting blocks. The total pressure that is applied on the footing is 250 kPa.
At this pressure also load should be taken by the supporting blocks to ensure the stability
of the total structure. Therefore the total pressure caused by the load of the test set-up is
approximated as 300 kPa. The footing and supporting blocks are modelled using PLAXIS
3D. Since the problem is symmetrical and two symmetry axis can be distinguished, it is
possible to model 1/4 of the total geometry. The loading from the supporting blocks to
the footing is modelled with the stage construction option.

From the third stage in figure 7.14, it can be seen that the displacements underneath the
footing and supports interfere with each other. Therefore it can be stated that the stress
bubble underneath the footing, is influenced by the supporting blocks. The soil tested
by loading up the footing is preloaded by the load that was initially on the supporting
blocks. This will lead to stiffer behaviour than what is representative for the in situ soil
state. For this example for instance, the calculated settlement where the test-setup is taken
into account is 10.2 mm. When the supporting blocks are ignored and only the footing
is modelled, the settlement is 12.2 mm. This means that in this case, due to the test
set-up, the settlement is underestimated with about 17%. For future tests, it is advised to
increase the distance from the supporting blocks to the footing or change the geometry of
the supporting blocks.

7.8 Conclusion
In the numerical verification the correlation that is developed for the site in Kuwait is tested
using PLAXIS 2D. An axisymmetric approach was used to simulate the process of the
ZLT’s where only sand was encountered. The results consist of 13 numerical calculations
that evaluate the settlements at 200 kPa load. The results all fall within a range 30% of
the settlement measured in the field.

The numerical method performs less than the analytical method. The Hardening Soil
model requires input parameters that are difficult to accurately evaluate. Parameters of
high influence are the unloading reloading stiffness and the OCR. These parameters are
approximated by correlation methods which are limited in accuracy. Extensive lab testing
is necessary to accurately determine these parameters. Furthermore the PLAXIS model
assumes soil layers with constant soil parameters. In real life there is always soil variability
which is not taken into account in the PLAXIS model. This can be modelled with the use
of random field theory.

80 Chapter 7 Stef Engels


Correlating CPT data to stiffness parameters of sand in FEM

Stage Description

In the initial stage PLAXIS performs the


K0 procedure. The initial stresses are de-
termined based on the values K0 , the volu-
metric weight and the groundwater table.

All the load is transferred to the subsurface


through the supporting blocks. This load
is approximated as 300 kPa.

Part of the load is transferred from the sup-


port to the footing. The footing is loaded
up to 200 kPa. The load that carried by
the support is then 100 kPa.

Figure 7.14: Numerical analysis for transferring the loading from support to footing

Chapter 7 Stef Engels 81


Correlating CPT data to stiffness parameters of sand in FEM

The obtained correlation for this site is used to determine the stiffness parameters in
the Hardening Soil model in combination with the correlation developed by Mayne (2007)
to determine the OCR of each layer. It can be concluded that it is possible to make a
reasonable prediction of the settlement using PLAXIS 2D with the Hardening Soil model
(within 30% range) for this site.

It should be noted that the ZLT procedure can be improved. The process of trans-
ferring the load from the supporting blocks to footing compromises the in situ soil state.
Therefore the measurements during the ZLT indicate stiffer behaviour than reality. When
these tests are available and accurate knowledge of the shear wave velocity is known, it
is advised to use the Hardening Soil model with small strain stiffness. The shear wave
velocity can be measured during a CPT by using a seismic cone.

82 Chapter 7 Stef Engels


Chapter 8

Conclusions and recommendations

8.1 Introduction
For land works the CPT is a standard procedure and is often performed at sites to determine
the local geological conditions. It is desired to extract as much information as possible
from these CPT’s. Over the years researchers tried to obtain engineering parameters from
cone resistance and sleeve friction. In this thesis the objective was to extract stiffness
parameters from CPT results for sand. Previous studies are done on this subject, but
developed methodologies are limited in accuracy. The main research question was defined
in Chapter 1 as:

Is it possible to predict stiffness parameters of sand with reasonable accuracy based on CPT
results?

The main objective was to evaluate the correlation between CPT parameters and the
stiffness parameters of sand using an amount of ZLT’s done at a site in Kuwait. It has
been shown that a correlation exist between normalized cone resistance and secant Young’s
modulus for this specific site. To come to a satisfying result several sub-studies were
performed in this thesis.

8.2 Conclusions
A correlation between normalized cone resistance and secant Young’s modulus is developed
for the site in Kuwait. The results where numerically verified using PLAXIS 2D with the
Hardening Soil model. The most important conclusions are summarized as follows:

• For the existing investigated methodologies, the method proposed by Robertson


(1990) is found to be the most accurate for determining stiffness parameters from
CPT results at the site of Kuwait. This methodology uses normalized cone resistance
and normalized sleeve friction.

• The error of each method is determined using the measurements of the ZLT’s. It can
be concluded that the method of Schmertmann et al. (1978) is the most conservative.
The error of each method is tested for normality with the Shapiro-Wilk test. The
error of the methods Schmertmann et al. (1978), Peck et al. (1996) and Robertson
(1990) can be assumed normally distributed.

• A site specific correlation is developed. It can be concluded that a workable correla-


tion can be obtained by dividing the sand into a convenient amount of layers. It is
advised to do this according to the SBT index. Increasing the number of layers does
not lead to better results. Therefore the author advises to not distinguish more than
6 layers.

83
Correlating CPT data to stiffness parameters of sand in FEM

• The strongest correlation with secant Young’s modulus is found with the normalized
cone resistance. The value of the normalized cone resistance is weighted with certain
factors that take the influence of the depth into account. These weighting factors are
based on Schmertmann’s modified strain influence diagram.

• The consolidation state of the material has a high influence on the stiffness. In sand
it very difficult to determine this parameter. Correlation methods can be used to get
an indication of OCR. It is advised to use the method developed by Mayne (2007)
which is based on an iterative procedure between two correlation formulas. It should
be emphasized that their is a limited amount of information that can be extracted
from two CPT parameters only.

• The obtained correlation formula is verified with PLAXIS 2D using the Hardening Soil
model. The results lie within 30% range of the observations in the field. The author
believes this accuracy is reasonable. The inaccuracy is mainly due to soil variability
and lack of information about the unloading reloading stiffness and consolidation
state of the deposit.

8.3 Recommendations
Based on the research done in this thesis several recommendations can be formulated for
future research. The results obtained in this thesis are calibrated for one site only and
several assumptions are made. The recommendations for further research are summarized.

• Check for creep settlements. In all ZLT’s deformations are measured when no
load increment is applied. This observation can be related with creep in the sand
body or with consolidation of a less permeable layer. For future works it is advised
to perform Plate Load Tests with a smaller diameter to monitor the behaviour of the
sand body. It is advised to perform Plate Load Tests with 1 m plate diameter.

• Change the ZLT set-up. It is proved that the set-up of the ZLT influences the soil
that is tested. Therefore the measurements are not fully representative for the in situ
soil state. It is advised to increase the distance between the supporting blocks and
the footing or to change the geometry of the supporting blocks to prevent interference
of the stress bubbles.

• Measure the pore pressure. For future projects it is advised to monitor the pore
pressures during a ZLT. This information is valuable to get a better understanding
of the settlements that are measured when no load increment is implied. Generation
of excess pore pressures indicates consolidation. When no excess pore pressures are
generated, creep is probably dominant.

• Measure shear wave velocity. The shear wave velocity provides valuable informa-
tion about the stiffness of the material. For the Robertson (1990) method the shear
wave velocity is estimated based on correlations with normalized CPT parameters,
but it is preferred to use in situ measurements. Furthermore an improved soil model
can be used for verification when this parameter is known. To get more accurate
knowledge of this parameter, it is advised to measure the shear wave velocity with a
seismic cone during a CPT.

84 Chapter 8 Stef Engels


• Improve soil model. The verification with PLAXIS is done with the Hardening
Soil model. When accurate knowledge of the shear wave velocity is known, it is
advised to use the Hardening Soil model with small strain stiffness. This model
is more accurate for modelling the stiffness of sand at very small strain levels but
requires two additional input parameters (Vs and γ0.7 ).

• Check general application. The proposed correlation in this thesis is based on


13 ZLT’s of one specific site in Kuwait. It would be interesting to test if the results
are also applicable for another site. Therefore it is recommended to test the accuracy
of the method using ZLT results from other sites.

85
Bibliography

J.H. Atkinson. Non-linear soil stiffness in routine design. Geotechnique, 5(50):487–508,


2000.
J.H. Atkinson and G Sallfor. Experimental determination of soil properties. In Proc. 10th
ECSMFE, pages 915–956, 1991.
G. Baldi, R Belotti, V.N. Ghionna, M Jamiolkowski, and D.F.C Lo Presti. Modulus of
sands from CPTs and DMTs. In Proceedings of the 12th International Conference on
Soil Mechanics and Foundation Engineering, pages 165–170, Rio de Janeiro, 1989.
H.K.S. Begemann. The friction jacket cone as an aid in determining the soil profile. In
Proc. 6th International Conference on Soil Mechanics and Foundation Engineering, pages
17–20, Montreal, 1965.
M.D. Bolton. The strength and dilatancy of sands. Geotechnique, 36(1):65–78, 1986.
R.W. Boulanger. High overburden stress effects in liquefaction analyses. Journal of
Geotechnical and Geoenvironmental Engineering, 129(12):1071–1082, 2003.
R.W. Boulanger and I.M. Idriss. State normalization of penetration resistance and the effect
of overburden stress on liquefaction resistance. In Proceedings of the 11th International
Conference on Soil Dynamics and Earthquake Engineering, pages 484–491, Berkeley,
2004. ASCE.
J Boussinesq. Application des potentials a l’etude de l’ equilibre et du mouvement des
solides elastiques. Gauthier-Villars, Paris, 1883.
J.L. Briaud. Introduction to Soil Moduli. Geotechnical News, 19(2):54–58, 2001.
R.B.J. Brinkgreve and P.A. Vermeer. Plaxis 2D - Material Models Manual. Balkema, 2016.
ISBN 905410.
J.B. Burland and M.C. Burbridge. Settlement of foundation on sand and gravel. In
Proceedings, Institution of Civil Engineers, pages 1325–1381, 1985.
F.A. Chuhan, A Kjeldstad, K Bjorlykke, and K Hoeg. Porosity loss in sand by grain crushing
- experimental evidence and relevance to reservoir quality. Marine and Petroleum Geology,
19:39–53, 2002.
F.A. Chuhan, A Kjeldstad, K Bjorlykke, and K Hoeg. Experimental compression of loose
sands: relevance to porosity reduction during burial in sedimentary basins. Canadian
Geotechnical Journal, 40(5):995–1011, 2003.
B.M. Das. Principles of Foundation Engineering. CL Engineering, 7 edition, 2010.
E De Beer. Bearing capacity and settlement for shallow foundations on sand. In Proceedings,
Symposium on bearing capacity settlement of foundations, pages 15–33, Durham, 1965.
Duke University.
A De Beer, E. Martens. Method of computation an upper limit for the influence of
heterogeneity of sand layers on the settlement of bridges. In Proceedings, 4th International
conference on soil mechanics and foundation engineering, pages 275–282, 1957.

86
J.M. Duncan and C.M. Chang. Nonlinear analysis of stess and strain in soils. Journal of
Soil Mechanics and Foundations Division, 96(SM5):1629–1653, 1970.

S. Eslaamizaad and P.K. Robertson. Evaluation of settlement of footings on sand from


seismic in-situ tests. In Proceedings of the 50th Canadian Geotechnical Conference, pages
755–764, Richmond, 1997. BibTech.

A Eslami and B H Fellenius. CPT and CPTu data for soil profile interpretation: Review
on methods and proposed new approach. Iranian Journal of Science & Technology,
Transaction B, 28, 2004.

M. Fahey and J.P. Carter. A finite element study of the pressuremeter in sand using a
nonlinear elastic plastic model. Canadian Geotechnical Journal, 30(2):348–362, 1993.

Y Gomaa and G Abdelrahman. Correlations between relative density and compaction


test parameters. In Twelfth International Colloquium on Structural and Geotechnical
Engineering, pages 1–8, 2007.

B.K. Hough. Basic soil engineering. Ronald Press, New York, 1969.

N.F. Ismael and A.S. Vesic. Compressibility and bearing capacity. International journal
of rock mechanics and mining sciences and geomechanics abstracts, 19(3):66, 1981.

J Jaky. Earth pressure in silos. In Proceedings of the 2nd International conference on Soil
Mechanics and Foundation engineering, pages 103–107, London, 1948.

M.G. Jefferies and K. Been. Soil liquefaction - A critical state approach. Taylor & Francis
Group, London and New York, 2006. ISBN 0-419016170-8.

R.L. Kodner and J.S. Zelasko. A hyperbolic stress strain formulation for sands. In Proc.
2nd Pan-American Conference on Soil Mechanics and Foundations Engineering, pages
284–324, 1963.

B.M. Lehane, J.P. Doherty, and J.A. Schneider. Settlement prediction for footings on sand.
In Proceedings of the Conference on Small-strain Stiffness in Practice, pages 133–150,
Atalanta, 2008.

T Lunne and H.P. Christofferson. Interpretation of Cone Penetrometer Data for Offshore
Sands. In Offshore Technology Conference, Houston, Texas, 1983. Offshore Technology
Conference.

T Lunne and P.K. Robertson. Cone Penetration Testing in geotechnical practice. 1997.

P.W. Mayne. Enhanced geotechnical site characterization by seismic piezocone penetration


tests. In Proceedings of the 4th International Geotechnical Conference, pages 95–120,
2000.

P.W. Mayne. Intergrated ground behaviour: In-situ and lab tests. In Proceedings of
the International Symposium on Deformation Characteristics of Geomaterials, pages
155–177, 2005.

P.W. Mayne. In-Situ Test Calibrations for Evaluating Soil Parameters. 2007.

P.W. Mayne and H.G Poulos. Approximate displacement influence factors for elastic
shallow foundations. Geotech. Geoenviron. Eng, 125(6):453–460, 1999.

G.R. McDowell and J.J. Khan. Creep of granular materials. Granular matter, 5(3):115–120,
2003.

87
G Mesri and B Vardhanabhuti. Compression of granular materials. Canadian Geotechnical
Journal, 46:369–392, 2009.
Y Nakata, M Hyodo, M Hyde, A.F.L Kato, and M Murata. Microscopic particle crushing
of sand subjected to high pressure one-dimensional compression. Soils and Foundations,
41:69–82, 2001.
N.M. Newmark. Influence charts for computation of stresses in elastic foundations. Technical
report, Univ of Illinois Bull, 1942.
R B Peck, G Mesri, and K Terzaghi. Soil Mechanics in Engineering Practice. John Wiley
and sons, New York, 3rd edition, 1996.
A Rahim. Effect of morphology and mineralogy on compressibility of sands. PhD thesis,
Indian Institute of Technology Kanpur, 1989.
N Ramsey. A calibrated model for the interpretation of cone penetration tests (CPTs)
in Nort Sea quaternary soils. In Offshore Site Investigation and Geotechnics ’Diversity
and Sustainability’: Proceedings of an International Conference, pages 341–356, London,
2002. society of Underwater Tech.
J.E Roberts and J.M. de Souza. The compressibility of sands. In Proceedings of the
American Society for Testing and Materials, pages 1269–1272, 1958.
P. K. Robertson. Soil classification using the cone penetration test. Canadian Geotechnical
Journal, 27(1):151–158, 1990. ISSN 0008-3674. doi: 10.1139/t90-014.
P. K. Robertson. Interpretation of cone penetration tests — a unified approach. Canadian
Geotechnical Journal, 46(11):1337–1355, 2009. ISSN 0008-3674. doi: 10.1139/T09-065.
P.K. Robertson and K.L. Cabal. Guide to Cone Penetration Testing for Geotechnical
Engineering. 6th Edition. Gregg Drilling and Testing, Inc., 2015.
P.K. Robertson and C.E. Wride. Evaluating cyclic liquefaction potential using the cone
penetration test. Canadian Geotechnical Journal, 35(3):442–459, 1998.
D. J. Rogers. Advanced Engineering Geology & Geotechnics - Notes on the Cone Penetration
Test, 2004.
J.A. Santos and A.G. Correira. Reference threshold shear strain of soil. its application
to obtain a unique strain-dependent shear modulus curve for soil. In Proceedings 15th
InternationalConference on Soil Mechanics and Geotechnical Engineering, pages 267–270,
Istanbul, 2001.
J.H. Schmertmann. Static cone to compute static settlement of sand. Journal of Soil
Mechanics and Foundations Division, 96(3):1011–1043, 1970.
J.H. Schmertmann, J.P. Hartmann, and P.R. Brown. Improved strain influence factor
diagrams. Journal of Geotechnical Engineering Division, 104(8):1131–1135, 1978.
J.A. Schneider, A.V. McGillivray, and P.W. Mayne. Evaluation of SCPTU intra-correlations
at sand sites in the Lowe Mississippi River valley, USA. In Proceedings of 2nd International
Conference on Site Characterization, pages 1003–1010, Porto, 2004. Millpress, Rotterdam,
the Netherlands.
S.S. Shapiro and M.B. Wilk. An analysis of variance test for normality (complete samples).
Biometrika, 52(3-4):591–611, 1965.
A.S. Vesic and G.W. Clough. Behavior of granular materials under high stresses. Journal
of the Soil Mechanics and Foundation Division, 94:661–688, 1968.

88
Y Wang, K Huang, and Z Cao. Probabilistic identification of underground soil stratification
using cone penetration tests. Canadian Geotechnical Journal, 50:766–776, 2013.
H.M. Westergaard. A problem of elasticity suggested by a problem in soil mechanics. In
Contribution to the mechanics of solids, Timoshenko anniversity Volume, New York,
1938.
C.P. Wroth. The interpretation of in-situ soil tests: Rankine Lecture. Geotechnique, 34(4):
449–489, 1984.
G. Zhang, P.K. Robertson, and R.W.I. Brachman. Estimating liquefaction-induced ground
settlements from CPT for level ground. Canadian Geotechnical Journal, 39(5):1168–1180,
2002. doi: 10.1139/t02-047.

89
Appendix

A Functions in Mohr-Coulomb model


In the Mohr-Coulomb model, perfectly plastic straining occurs when the yield criterion is
met. The yield conditions of the Mohr Coulomb are given in the following equations in
terms of principle stresses.
1 0 0 1 0 0
f1a = · (σ2 − σ3 ) + · (σ2 + σ3 ) · sin ϕ − c · cos ϕ ≤ 0 (1)
2 2
1 0 0 1 0 0
f1b = · (σ3 − σ2 ) + · (σ3 + σ2 ) · sin ϕ − c · cos ϕ ≤ 0 (2)
2 2
1 0 0 1 0 0
f2a = · (σ3 − σ1 ) + · (σ3 + σ1 ) · sin ϕ − c · cos ϕ ≤ 0 (3)
2 2
1 0 0 1 0 0
f2b = · (σ1 − σ3 ) + · (σ1 + σ3 ) · sin ϕ − c · cos ϕ ≤ 0 (4)
2 2
1 0 0 1 0 0
f3a = · (σ1 − σ2 ) + · (σ1 + σ2 ) · sin ϕ − c · cos ϕ ≤ 0 (5)
2 2
1 0 0 1 0 0
f3b = · (σ2 − σ1 ) + · (σ2 + σ1 ) · sin ϕ − c · cos ϕ ≤ 0 (6)
2 2
Other than the principle stresses, the yield criterion is a function of the friction angle ϕ
and the cohesion c. In addition to the yield functions, the Mohr-Coulomb model defines
six plastic potential functions. These functions are presented underneath.
1 0 0 1 0 0
g1a = · (σ2 − σ3 ) + · (σ2 + σ3 ) · sin ψ (7)
2 2
1 0 0 1 0 0
g1b = · (σ3 − σ2 ) + · (σ3 + σ2 ) · sin ψ (8)
2 2
1 0 0 1 0 0
g2a = · (σ3 − σ1 ) + · (σ3 + σ1 ) · sin ψ (9)
2 2
1 0 0 1 0 0
g2b = · (σ1 − σ3 ) + · (σ1 + σ3 ) · sin ψ (10)
2 2
1 0 0 1 0 0
g3a = · (σ1 − σ2 ) + · (σ1 + σ2 ) · sin ψ (11)
2 2
1 0 0 1 0 0
g3b = · (σ2 − σ1 ) + · (σ2 + σ1 ) · sin ψ (12)
2 2
In these functions, a third plasticity parameter is introduced which is called the dilatancy
angle ψ. This parameter is used to model positive plastic volumetric strain increments.
This behaviour is observed in dense sands.

When a soil is cohesive (c > 0) the Mohr-Coulomb model allows for tension. Because in
reality soils can not (barely) sustain tensile forces, three additional tensile cut-offs are
introduced in the yield criterion.
0
f4 = σ1 − σt ≤ 0 (13)
0
f5 = σ2 − σt ≤ 0 (14)
0
f6 = σ3 − σt ≤ 0 (15)

90
B Fitted nSBT chart Robertson
The Robertson nSBT chart can be accurately approximated using the quadratic functions
and the intersection points given in the table below (Wang et al., 2013).

Table 1: Best fitted parameters for the quadratic functions ln(Qtn ) = a·ln(Fr )2 +b·ln(Fr )+c
(Wang et al., 2013)

Function
a b c
ID
I -0.3703 -1.3625 1.5049
II 0.5586 -0.5399 0.3049
III 0.5405 0.2739 1.6959
IV 0.3833 0.7805 2.5718
V 0.2827 0.967 4.1612
VI 0.3477 1.4933 6.6507
VII 0.8095 -3.6795 8.1444
VIII 64.909 -187.07 139.2901

Table 2: Coordinates for the intersection points (Wang et al., 2013)

Intersection Coordinate Coordinate


point ln(Fr) ln(Qt)
A -2.3026 0
B 0.6569 0
C -2.3026 2.2268
D 2.3026 0
E 2.3026 2.0234
F 2.3026 0.1776
G 2.3026 3.9639
H 1.8687 4.0953
J 1.4505 4.5104
K -1.3334 2.2126
L 0.9622 5.3534
M -2.3026 3.4335
N 0.3655 6.9078
O 0.1658 6.9078
P -2.3026 5.0557
Q -2.3026 6.9078
R 2.3026 6.9078
S 1.6334 6.9078
T -0.5773 1.7179

91
C Settlement analysis ZLT
In this section the results of the analytical calculations of the settlement during a ZLT are
evaluated. The calculated settlements according to the different methods are compared
with the measured settlement at the site. The measured and calculated values represents
the settlement at 100 % of the design pressure (200 kPa).

Figure 1: Results of the calculations where only sand is encountered

Figure 2: Results of the calculations where sand and silt is encountered

Figure 3: Results of the calculations where sand silt and sabkha is encountered

92
D Shapiro-Wilk test

The following section describes the Shapiro-Wilk test for normality (Shapiro and Wilk,
1965). The test uses the principle of a null hypothesis to check whether a sample is normally
distributed. First, calculate SS as follows:

n
X
SS = (xi − x̄) (16)
i=1

Where:
(x1 , ..., xn ) is the sample that is checked for normality
xi is the ith order statistic, i.e. the ith smallest number in the sample
x̄ is the mean of the sample
n is the size of the sample

The next step is to calculate parameter m. When n is even, m = n/2. When n is


odd, m = (n − 1)/2. With this value, parameter b is calculated as:

m
X
b= ai · (xn+1−i − xi ) (17)
i=1

Where:
ai are weights that depend on the sample size. These weights can be found in table 3.

The test statistics are calculated according to:

b2
W = (18)
SS

From the test statistics W a p-value can be obtained. The p-value is found in table 4. If
the value can not be directly found in the table, linear interpolation is used. The null
hypothesis states that the data is normally distributed. If the p-value is bigger than a
certain significance level α, the null hypothesis can not be rejected. When p-value is smaller
than α the null hypothesis can be rejected and there is significant evidence to state the
sample is not normally distributed.

93
Table 3: Coefficients for the weight ai

94
95
Table 4: p-values

96
E Soil specifications

Table 5: Overview of the soil specifications

97
F Comparison back calculated stiffness with Robertson stiff-
ness

Figure 4: Comparison Es with Erobertson

98
G Calculating standardized residuals
In a regression analysis, the residuals of an accurate model do not show any correlation
with the predicted variable. To scale the residuals to a familiar magnitude the standardized
or studentized residuals can be calculated. In this approach the residuals are scaled to
values comparable to values generated from a standard normal distribution. Using the
familiar confidence interval, it can be seen of some data points deviate from the rest of the
data. This is useful for spotting so called ”outliers”.

The standardized residuals are calculated according to the following formula:


Ri
Rstandard = q Pn (19)
1 2
1−n i=1 Ri

Where:
Rstandard is the standardized residual
Ri is the ith residual
n is the sample size

99

You might also like