You are on page 1of 22

xviii

LIST OF ABBREVIATIONS AND ACRONYMS

API American Petroleum Institute


CALTRANS California Transportation Department
CAM Conservatively Assessed Mean
CE Conservative Estimate
CHSL Cross Hole Sonic Logging
CL Clay (USCS)
CPT Cone Penetrometer Test
DPSH Dynamic Probe Super Heavy
FE Finite Element
FEA Finite Element Analysis
FEM Finite Element Model
FoS Factor of Safety
MAD Modified AASHTO Density
MAE Mean Average Error
MC Moisture content
NGL Natural Ground Level
OCR Overconsolidation Ratio
O-Cell Osterberg Loadcell
OMC Optimum Moisture Content
PI Plasticity Index
PLI Point Load Index
RMSE Root Mean Square Error
SANRAL South African National Roads Agency Limited
SC Clayey Sand (USCS)
SCD Saturated Consolidated Drained
SCU Saturated Consolidated Undrained
SICAP Screwed-In-Casing-Augered Pile
SLS Serviceability Limit State
SM Silty Sand (USCS)
SPT-N Standard Penetration Test Number
xix

SWCC Soil Water Characteristic Curve


UCS Uni-axial Compressive Strength
USCS Unified Soil Classification System
UU Unconsolidated Undrained
UUU Unsaturated Unconsolidated Undrained
CHAPTER 1 -INTRODUCTION 1

CHAPTER 1

INTRODUCTION
___________________________________________________________________________

1.1. BACKGROUND

Pile design capacity estimation techniques use assumptions regarding subsurface soil profiles
and soil-concrete interaction, to approximate the ultimate capacity of a single pile. The
accuracy of such assumptions is a function of the various parameters that is derived from
penetration tests, installation methodology followed and interface properties between pile
and soil. It is however, impossible to quantify the soil profile in its entirety as soil parameters
are known to vary significantly. Bond and Harris (2008) and Simpson et al (2009) proposed
various methods of estimating conservatively assessed mean/conservative estimate (CAM/CE)
values, with a 5% chance of overall worse value governing, for penetration tests. Nevertheless
these conservatively-assessed mean penetration values in combination with theories used to
model the pile and to derive shear and stiffness parameters, might result in a gross under-or
overprediction of ultimate load capacities of a pile and estimated load-deflection behaviour;
as most of the methods used to derive parameters have further conservatism built into them.
It is therefore recommended that full scale load tests be undertaken to verify design, especially
on structures that are sensitive to very small settlements. Undertaking of full-scale pile load
tests is further supported where a pile derives capacity from predominantly side shear whilst
the toe of the pile is not founded on rock, as this might result in excessive displacements if
under-designed.

The Mt Edgecombe Interchange site in Umhlanga, KwaZulu-Natal, where three Osterberg


loadtests were undertaken have two incrementally launched structures that will be part of the
largest four-level interchange in Africa once completed. A photo of the downscaled
architectural model is included in Figure 1, the largest bridge on the project, once completed
will be 947m long with a smaller bridge of 443m. These incrementally-launched bridges and
various smaller bridges have differential settlement tolerances of some 10mm and individual
settlement tolerances of some 20mm per pier. The site consists of soils in excess of 50m thick
before rock is encountered and therefore a friction pile-piled raft solution was considered the
most economic design alternative.
CHAPTER 1 -INTRODUCTION 2

These three Osterberg load tests, resulted in underestimated ultimate capacities significantly
different from those predicted by theories used by the author to predict ultimate capacity and
load-deflection behaviour of the piles. Initial or serviceability load-deflection behaviour, where
the design is crucial, was however similar in nature and therefore only minor adjustments were
required to designed pile lengths.

Figure 1. Photo of scale model of the Mt Edgecombe Interchange

1.2. MOTIVATION FOR RESEARCH

Field static pile load tests, using Osterberg Internally-jacked method, were undertaken to
establish how measured single pile behaviour compared with predicted capacities and
deflection behaviour. Ultimate single pile capacities measured in three Osterberg load cell tests
undertaken at the Mt Edgecombe Interchange, were significantly higher than predicted
ultimate design capacities.

It is a well-known fact that it is difficult to quantify a soil mass in its entirety and that some
conservatism applies in deriving parameters used to predict ultimate capacities from in-situ
CHAPTER 1 -INTRODUCTION 3

tests. Additionally, ultimate capacity estimation techniques, might have their own conservative
assumptions built into them.

Analytical methods used to predict load-deflection behaviour, might therefore be quite


different to results derived from static load tests. Finite Element modelling of these tests assist
in determining what effect various parameters have on establishing actual behaviour, by
undertaking to some extent, a back-analysis of these static load tests.

1.3. RESEARCH AND OBJECTIVES

The aim of this research is to do an analytical study to compare static pile load test results to
those predicted by analytical methods and derived by Finite Element modelling. Derivation of
parameters used and the effect of these on Finite Element output, will be discussed in detail
and compared to those derived from a) static load tests and b) those derived by analytical
methods.

Objectives of the research are listed below:

Assess effects of various parameters used in finite element models to back-analyse


static load test results,
Determine why analytical methods proposed by the American Petroleum Institute
(1987) and Everett (1991) fail to predict actual test results with a fair degree of
accuracy.

1.4. PROGRAMME OF THE STUDY

Objectives of the study were achieved by implementing the following programme:

A study of API (American Petroleum Institute) design methodology, Everett (1991) load-
deflection prediction method, two Finite Element Constitutive models and derivation
of parameters from penetration tests.
Analysis of piles in geotechnical finite element software with two constitutive models
to determine how accurately onsite conditions may be duplicated by these models.
CHAPTER 1 -INTRODUCTION 4

Understanding statistical methods to compare predicted and actual test results.


Compare best-fit comparisons between Finite element methods and API-predicted
results to determine where the API method appears to be conservative.

1.5. LIMITATIONS OF RESEARCH

The limitations of this research are as follows. The study utilises penetration data measured at
irregular intervals, predominantly at every 3m down a single borehole undertaken at each test
pile position. Therefore assumed mean values, over depth ranges, might in fact be much larger
than conservatively-assessed mean values. Water tables might additionally have varied
subsequent to those measured in the original subsurface investigation. This might lead to
reduced or increased overburden pressures.

Average energy ratios at which Standard Penetration Test, N-values, were measured are not
entirely known. This may lead to over- or underestimates of parameters used in finite
element models.

The Duncan Chang hyperbolic model describes failure via the Mohr-Coulomb failure criterion,
but dilatancy cannot be described properly due to the fact that the plasticity framework is not
properly formulated.

Installation methods play an important role in ultimate capacities, with actual hydrostatic
pressures exerted by pile concrete on adjacent soils varying significantly from those assumed
when considering duration of concrete placement or concrete head above a certain point
before removing portions of the casing. The coefficient of lateral earth pressures derived,
might therefore vary significantly from those predicted. Details of installation followed on site
for every pile is not exactly known to the author and could therefore not be assessed in detail.
CHAPTER 1 -INTRODUCTION 5

1.6. REPORT LAYOUT

Chapter 1 is an introduction of the report. This chapter starts with a brief background
discussion followed by a motivation for this research. The objectives are listed with a discussion
on limitations. Lastly, a chronological layout is presented.

Chapter 2 consists of a literature review, incorporating the API design methodology, Everetts
load deflection theory, and statistical methods to derive CAM values for penetration tests and
discusses the Mohr-Coulomb and Duncan-Chang constitutive models. This chapter further
discusses how to derive parameters required by the constitutive models and presents a brief
introduction to the Osterberg static load tests as well as matric suction theory for unsaturated
soils.

Chapter 3 discusses geology of the site, typical Berea Formation parameters and design criteria
set for piles as well as the preferred pile type. The chapter further summarises the
characteristic SPT-N values derived for the three test pile positions.

Chapter 4 discusses the original design based on the API method, sensitivity of calculations
when limiting side shear capacities and the Everett curves for both limited side shear and the
non-limited condition.

Chapter 5 contains Finite Element modelling of the Osterberg load tests and equivalent top-
down models. The chapter then details a procedure for deriving coefficients of lateral earth
pressure at removal and expected friction-concrete shear reduction factors. Incorporating
these derived facets and other conclusions, the API capacities and Everett curves are re-
analysed to assess what influence they would have on ultimate capacities.

Chapter 6 summarises findings of the analyses and concludes with recommendations for future
design assumptions on friction piles with analytical methods in particular with regards to the
API methodology.

Figure 2 shows a flow chart summarising the methodology adopted in this study.
CHAPTER 1 -INTRODUCTION 6

Problem Statement

Literature Review

API design methodology


Everett Load-deflection theory
Characteristic values
Derivation of shear, stiffness and
Duncan-Chang parameters
Osterberg loadtest method
Matric Suction

Data Collection

Field Investigation
Geology
Design criterion
SICAP methodology
Statistical analysis of SPT-N data

Review of original design

Capacity prediction API method


Load-deflection predictions
Sensitivity of critical depth
concept

Parametric Study Analysis of field measured results

Derive Mohr-Coulomb and Duncan-Chang SPT data


parameters Measured Displacements
Load-deflection behaviour from Mohr-Coulomb England top-down derived curves
and Duncan-Chang on raw test data (Cell-up and Strain gauge measurements
down
Deriving top-down load curves for the best
constitutive model based on comparison
between Mohr-Coulomb and Duncan-Chang
Reanalysis of API predictions based on findings
from FEM

Conclusions

Comments and recommendations

Figure 2. Methodology flow chart


CHAPTER 2 LITERATURE REVIEW 7

CHAPTER 2

LITERATURE REVIEW

This literature review will comprise a review of the API pile design methodology used by the
author to design the piles, and this will be followed by a review of the load-deflection
prediction methods as presented by Everett (1991). As the design is based on penetration test
results, a detailed review on how to derive cautious estimate SPT-N values for the profiles at
the test piles will also be undertaken. Finite element constitutive models used by the author
will be discussed thereafter to establish which parameters are required as input for these
models. Derivation of these parameters from penetration tests will also be discussed. To
conclude this chapter a brief review will be given of the Osterberg internally- jacked static load
test and the derivation of an equivalent top-down load settlement curve and the possible
effects of matric suction.

2.1. PILE CAPACITY API METHOD (1987)

Ultimate pile capacity for a single pile is derived from both a frictional component derived from
the shear on the circumference of the pile and by an end bearing component based on the
resistance provided by the soil at the toe of the pile on the cross-sectional area as depicted in
Figure 3. Components of the ultimate capacity of a single pile

The ultimate bearing capacity of a pile may be determined by the following equation according
to API (1987):

Qt = Qf + Qp = fAs + qAp (Eq.1)

where: Qf = skin friction resistance;

Qp = end bearing resistance;

f = unit skin friction capacity;

As = D = circumference of pile;

D = pile diameter;
CHAPTER 2 LITERATURE REVIEW 8

q = unit end bearing capacity; and

Ap = r2 = gross sectional area of pile.

Figure 3. Components of the ultimate capacity of a single pile

The way in which ultimate capacity may be calculated for a saturated clayey/cohesive material
and for a sandy or gravelly/non-cohesive material varies quite significantly and will be
discussed in the following sections.

2.1.1. PILES IN COHESIVE SOILS

2.2.1.1. SHAFT SHEAR

For piles installed in saturated cohesive soils, shaft friction at any point along the pile may be
calculated by:

f = cu (Eq.2)

where: = a dimensionless factor; and


CHAPTER 2 LITERATURE REVIEW 9

cu = the un-drained shear strength of the soil at the point in question.

The factor can be computed by the equations:

= 0.5-0.5 for 1.0; (Eq.3)

or = 0.5-0.25 for >1.0; (Eq.4)

with the constraint that should be smaller than 1.0.

where:

cu
= (Eq.5)
p'o

where po = effective overburden stress at the point in question.

2.2.1.2. END BEARING

The unit end bearing may be computed from the equation:

q = 9cu (Eq.6)

The cu-value may be derived from triaxial or shear box tests, from visual descriptors and/or from
the results of Standard Penetration Test (SPT) values. Table 1 extracted from Guideline for soil
and rock logging (Brink and Bruin, 2002) gives UCS-values to be assigned to saturated clays of
varying consistencies. These correlations were used by the author in the original design. Its
however the authors belief that these values are not a true representative of actual UCS values
as Brink and Bruin do not state the energy at which SPTs are undertaken or if the SPT-N results
have been corrected for overburden pressures. Bowles (1996) recommends that the SPT-N
values should be corrected for both energy and overburden to derive an (N1)70 value. ASTM
D1586 (2011) and CIRIA CP7 (Clayton, 1993) advocates using SPT-N values corrected to an Energy
Ratio, Er = 60%, Bowles (1996) however advocates using an Er of 70% as he states that this is
close to the actual energy ratio obtained in the United States of America. Corresponding cu values
are given in Table 2. Derivation of (N1)70 values will be discussed later in this document.
CHAPTER 2 LITERATURE REVIEW 10

Table 1. SPT-N derived cu value (Brink and Bruin, 2002)

Consistency SPT N-value UCS (kPa) (cu = 1/2 of UCS)

Very soft <5 < 50

Soft 5-10 50 - 125


Firm 11-25 126 - 250

Stiff 26-50 251 -500

Very stiff 51-80 > 500

Table 2. SPT-(N1)70 derived cu value (Bowles, 1996)

Consistency (N1)70 UCS (kPa) (cu = 1/2 of UCS)

Very soft 0-2 <25

Soft 3-5 25 -50

Firm 6-9 50 -100


Stiff 10-16 100 -200

Very stiff 17-30 200 400

Hard >30 >400

The moist unit weight of a clay may be derived from (N1)70 values from Bowles (1996) as shown
in Figure 4 and by the derived equation:

m = -5E-05(N1)704+0.0022(N1)703-0.0448(N1)702+0.6576(N1)70+15.734 (Eq.7)

where m = moist unit weight; and

(N1)70 = SPT-N number corrected for energy and overburden.


CHAPTER 2 LITERATURE REVIEW 11

Figure 4. Unit weight versus (N1)70 for a clay

2.1.2. PILES IN COHESIONLESS SOILS

2.1.2.1. SHAFT SHEAR

Shaft friction may be calculated by the following equation:

fs = Kpotan (Eq.8)

where: K = coefficient of lateral earth pressure at removal(ratio of horizontal


to vertical normal effective stress);

po = effective overburden stress at the point in question;

= friction angle between soil and pile wall.

The values of f should be limited as per Table 3 due to the critical depth concept incorporated in
the API design method.
CHAPTER 2 LITERATURE REVIEW 12

Table 3. Bearing Capacity factors and Soil-Pile friction

Density Soil Description Soil-Pile Limiting Nq Limiting end


friction shaft friction bearing q
() f (kPa)* (MPa)

Very loose Sand 15 47.8 8 1.9


Loose Sand-Silt (tan15
Medium Silt =0.26)
Loose Sand 20 (0.36) 67.0 12 2.9
Medium Sand-Silt
Dense Silt
Medium Sand 25 (0.46) 81.3 20 4.8
Dense Sand-Silt
Dense Sand 30 (0.57) 95.7 40 9.6
V Dense Sand-Silt
Dense Gravel 35 (0.70) 114.8 50 12.0
V Dense Sand
*Based on critical depth concept (Vesic, 1965 from le Tirant,1992)

Bhengu, Kalumba & Chebet (2013) undertook large shearbox tests to determine interface
properties between South African sands on concrete, with some results highlighted in Table 4.
They found on average that the interface coefficient is some 0.85 for medium to rough
concrete which is much higher than the 0.66 normally quoted in literature.

Table 4. Soil-Pile friction reduction factors (Bhengu, Kalumba & Chebet (2013))

/
Cape flats sand Peak Cape flats sand Klipheuwel sand
Residual
Smooth concrete 0.76 0.84 0.82
Medium to Rough 0.78 0.90 0.84
Concrete
CHAPTER 2 LITERATURE REVIEW 13

Rough Concrete 0.85 0.97 0.92

Limiting of values of skin friction and end-bearing resistance as stipulated in the API method as
taken from Vesics critical depth concept is depicted in Figure 5. The critical depth concept is a
subject of considerable debate, with Fellenius (1995) suggesting it does not exist as the
phenomenon can rather be explained by residual loads in piles prior to testing. Limiting side
shear capacity can evidently result in a gross underestimation of ultimate capacity derived from
the API theory if the argument goes in favour of the fact that the critical depth concept is a
misconception.

Figure 5. Expected trend in variation of side shear with depth, compared to API (le Tirant, 1992)

If only SPT results are available, Table 5 extracted from Brink and Bruin (2002), gives descriptors
and typical dry densities to be assigned to saturated sand of varying relative density. These
Brink and Bruin values were used in the original design by the author. SPT-N values are
however assumed to be uncorrected as Brink and Bruin do not clarify SPT energy. This is a very
crude way of estimating the unit weight, and can lead to dry unit weights in orders of
magnitude larger than the actual unit weights, especially at great depth, as the overburden
CHAPTER 2 LITERATURE REVIEW 14

pressure at the specific depth at which an SPT is undertaken would have a significant impact
on the measured SPT-N value.

Table 5. Typical Dry Densities from SPT results (Brink and Bruin, 2002)

Relative density Saturated SPT N- Typical Dry Density Dry Unit Weight
value (kg/m3) (kN/m3)

Very loose 0-4 < 1 450 <14.2

Loose 4-10 1 450-1 600 14.2 15.7

Medium dense 10-30 1 600-1 750 15.7 17.1

Dense 30-50 1 750-1 925 17.1 18.8

Very dense > 50 > 1 925 >18.8

Bowles suggests that (N1)70 values may be correlated against saturated unit weight as depicted
in Figure 6, by plotting Dry unit weight, assuming a difference of up to 20% (CALTRANS,2014).
This method is however also controversial as one would not have an estimated unit weight
with which to adjust SPT-N values, i.e. the (N1)70 is a function of the unit weight but
conservatively it is assumed a unit weight of 18kN/m3 can be used to derive these numbers for
dry conditions and 11kN/m3 for saturated conditions. When using this method one should
technically iterate until the unit weight used to adjust SPT-N numbers conforms to the unit
weight established. The effect is however considered to be very small. The equations derived
are as follows:

m = 2.7602ln((N1)70)+11.399 (Eq.9)

D = 2.208ln((N1)70)+9.1196 (Eq.10)
CHAPTER 2 LITERATURE REVIEW 15

Figure 6. Unit weight for sandy/granular soil versus (N1)70

2.1.2.2. END BEARING

Unit end bearing may be computed by the equation:

q = p0Nq (Eq.11)

where: po = effective overburden stress at the pile tip; and

Nq = dimensionless bearing capacity factor.

The values of q should be limited as per Table 3.

2.1.3. FoS

Fairly large safety factors normally apply to calculated ultimate end bearing capacity,
predominately due to the large load-deflection required in end bearing to derive full capacity. A
FoS of 1.5 is normally applied to the shaft friction of an auger pile founded in cohesionless
material to limit settlements. To correspond to an FoS = 1.5 on side shear at similar settlements
an FoS on end bearing would have to normally be in the order of 3 to 4, but this is dependent on
various factors. The author is however of opinion that the focus should rather be to review load-
settlement curves derived and base limiting loads on a specific corresponding settlement value
CHAPTER 2 LITERATURE REVIEW 16

as the serviceability condition is normally the governing factor on settlement-sensitive


structures.

2.2. LOAD-SETTLEMENT BEHAVIOUR FOR SINGLE PILES

The ultimate load capable of being carried by a piled foundation represents the limit state for
load carrying capacity. However the load bearing capacity at this ultimate state is of little
consequence if the integrity of the structure founded on the piles is compromised due to
excessive settlements. Thus load-settlement characteristics of the pile prior to the limit state
being reached are usually more important. This so-called serviceability state usually represents
working loads imposed on the pile.

When load is applied to the top of a pile it is transferred to the surrounding soil/rock either in
skin friction or end bearing. It has been found from interpreting results obtained from
instrumented test piles, that initial deformations induced by loads are usually absorbed in skin
friction. As the load increases, more of it is taken in end bearing. In fact the end-bearing portion
of the load may continue to build up long after peak side shear resistance is reached.

Equations for predicting load deformation curves for different pile types founded in a variety of
soils, are discussed below.

2.2.1. EVERETT (1991) LOAD TRANSFER FUNCTIONS

In 1991 Everett from Frankipile, now Keller Franki, proposed load-pile head deflection transfer
functions based on various full scale load tests conducted in South Africa. The so-called load
transfer functions employed by Everett are divided into two types; a side shear transfer
function and an end bearing one.

2.2.1.1. SIDE SHEAR TRANSFER FUNCTION

This function is mathematically represented by:

f = Cult[1-e-K1dps] + Fult[1 e-K2dps] (Eq.12)

where: f = unit side shear transfer (kPa);


CHAPTER 2 LITERATURE REVIEW 17

Cult = ultimate adhesion transfer (kPa);


Fult = ultimate intergranular transfer (kPa);
dps = pile shaft movement (mm); and
K1,K2 = constants.
From back-analyses of numerous pile load tests, the value of K1 was found to vary from about
1.5 for soft clay to about 2.4 for a stiff clay and in medium dense sand about 2.0 for an auger
pile.

The value of K2 was found to be fairly constant at around 0.2.

2.2.1.2. END BEARING TRANSFER FUNCTION

This function is characterised by:

q = qult[ X(1-x)]K3 (Eq.13)

where: q = end bearing stress at toe deflection dtoe (kPa);

qult = ultimate end bearing stress (kPa);

X = ratio of the pile toe deflection dtoe to 10 percent of the pile base diameter;

K3 = a constant depending on the soil and pile type.

K3 was found to vary as follows:

Franki pile with enlarged base in sand 0.5;

Franki pile with enlarged base in clay 0.2;

Bored piles in soft rock 0.6;

Bored piles in sand below water table 1.5;

Precast and tube piles in sand 0.6; and

Precast piles in extremely dense sand 0.5.


CHAPTER 2 LITERATURE REVIEW 18

2.2.1.3. LOAD DEFLECTION CURVE

Based on the above transfer functions a typical load/deformation curve, as illustrated in Figure
7 will be derived. The method suggests that one should select the serviceability limit state load,
where the load/deformation curve changes from hyperbolic to asymptotic (indicated by the
orange circle). This point is defined as the failure point. Based on the failure point the FoS for end
3400
bearing will be some = 8.33 (where 600 is the corresponding end bearing capacity at the
600

serviceability limit state deflection for side shear). It also implies that application of factors of
safety to ultimate load alone as a means of controlling deformation is crude and illogical and
conformation should result by comparing of load/deformation curves with the factored SLS
design load.

3400

Figure 7. Load Deflection Curve (Derived from Everetts Load Transfer functions)

2.3. CHARACTERISTIC VALUES

Site investigations to classify underlying soil stiffness for geotechnical purposes, rely in many
cases on in-situ penetrometer tests to obtain appropriate information. Contradictory to
concrete and steel where properties are controlled in a laboratory scenario, these
penetrometer results vary significantly with increasing depth, resulting in a large scatter with
concomitant large co-efficients of variation . They are usually only undertaken in practice every
1.5-3.0m in a borehole. Bond and Harris (2008) advocate that pile design, using in-situ
penetrometer test results, involves a large volume of ground for which the average over depth
CHAPTER 2 LITERATURE REVIEW 19

property is of greater relevance than a 5% fractile typically used by structural engineers. It is


the authors opinion that in statistical terms, what must be calculated are 95% confidence
limits to the 50% fractile (average), i.e. the chance of a worse average value governing the
design would be less than 5%.

Schneider (1997) as discussed in Simpson et al. (2009) argued that a cautious estimate of the
mean value, with a 5% chance of the worse overall value governing, would lie some 0.5
standard deviations from the mean measured values. A comparison between these two
assessments can be seen in Figure 8.

Figure 8. Alternative derivations of characteristic values (Simpson et al, 2009)

Bond (2010), after asking numerous engineers to predict the cautious estimate from a set of
penetrometer results and receiving a broad spectrum of answers ranging from extremely
conservative to non-conservative, proposed a method to determine the characteristic value
for penetrometer tests by determining a best fit line through the data. This method may be
summarised as follows:

Step 1: Determine best fit curve through SPT data

The trendline function found in the Microsoft Excel spreadsheet software package can be used
to determine the best fit curve through a data set of SPT values, with the highest R2 number,
CHAPTER 2 LITERATURE REVIEW 20

as shown in Figure 9 below. The coefficient of determination, R2 is the square of the


Pearsons correlation coefficient R and is a statistical check to see how close actual data is to
the fitted curve, i.e. which percentage of the variation in y is explained by the variation in x.
The R2 cannot determine whether the prediction is subjective in terms of over or under
predicting the SPT value, which is why residual plots, which will show how the predicted values
varies from the actual values, must be assessed to detect any skew in the trendline.

Figure 9. Best fit linear regression through SPT data

Step2: Determine the residuals of each data point and the standard deviation of the residuals

The difference between the best fitted curve and the actual SPT-N plots, as given in Figure 9,
must be calculated and plotted as shown in Figure 10. It is important to assess these residuals
to determine if there is any skew in the trend line. From Figure 9, visually it can be seen, for
the example that actual results would vary significantly from the best fitted curve when the
depth exceeds roughly 35m. Orange lines in Figure 10 show standard deviations of the
residuals. Where residual plots tend to fall outside these standard deviation lines, the best fit
curve either tends to over-or underestimate actual values compared to the remainder of the
data. It is considered by the author of this document that these values should therefore be
treated as outliers and disregarded, and technically a revised best fitted curve should be
derived exclusively over these depths and the same steps should be followed as discussed in

You might also like