Reliability Based Serviceability Limit State Design of Driven Piles in Glacial Deposits

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 25

Geotech Geol Eng

https://doi.org/10.1007/s10706-022-02163-0

ORIGINAL PAPER

Reliability‑Based Serviceability Limit State Design


of Driven Piles in Glacial Deposits
Markus Jesswein · Jinyuan Liu 

Received: 11 June 2020 / Accepted: 28 April 2022


© The Author(s), under exclusive licence to Springer Nature Switzerland AG 2022

Abstract  The reliability-based design (RBD) was were evaluated to represent the observed correla-
investigated in this study on the serviceability limit tions between the two hyperbolic parameters. Lastly,
state of steel piles driven in glacial deposits. A data- the resistance factors in the RBD were calibrated
base of 42 high-quality pile load tests was used to with Monte Carlo simulations. In the end, a series
quantify the uncertainties of the standard penetration of resistance factors were developed for an allow-
test (SPT)—based design methods. A two-parameter able displacement of 5 to 25 mm for three SPT-based
hyperbolic model was adopted to represent the meas- design methods. Based on this study, the values of
ured load–displacement response of a pile, and the the resistance factors ranged from 0.04 to 0.49, which
load component of the model was normalized with were largely influenced by the design methods due to
the capacity identified from the pile load test. A statis- their capacity biases and variations.
tical analysis was conducted on the performances of
11 different failure load identification methods. The Keywords  Serviceability limit state · Reliability-
De Beer method was selected due to its low variations based design · Driven piles · Standard penetration
and relatively conservative estimation. The measured test · Glacial deposits
capacities were compared to predictions from three
SPT-based design methods. The statistical properties
of the capacity bias, a ratio of the measured to pre- 1 Introduction
dicted capacity, were evaluated for characterizing the
uncertainties of a pile design in glacial deposits. The Many kinds of uncertainties have been encountered
average of the capacity bias varied from 0.96 to 1.13 during the design process of foundations. These
with the coefficient of variations ranging from 36.4 to uncertainties are caused by varying material proper-
50.9%. From the pile load tests, probability distribu- ties, especially from the ground; inconsistencies from
tions were fitted to the collected capacity biases and the loads; and inaccuracies in the design formulas
hyperbolic model parameters, and several copulas or methods. The allowable stress design (ASD) phi-
losophy has combined all these uncertainties into a
global factor of safety (FOS), but a global FOS can
M. Jesswein · J. Liu (*)  be selected subjectively from previous experiences
Department of Civil Engineering, Ryerson University, 350
and judgements and correspond to varying levels of
Victoria St, Toronto, ON M5B 2K3, Canada
e-mail: jinyuan.liu@ryerson.ca risk (Becker 1996; Duncan 2000; Phoon et al. 2003).
On the contrary, a reliability-based design (RBD)
M. Jesswein 
e-mail: markus.jesswein@ryerson.ca considers the likelihood of failure with probabilistic

Vol.: (0123456789)
13
Geotech Geol Eng

methods and reliability theories and has been gradu- The focus is on axially loaded piles subjected to static
ally replacing ASD over the last few decades in geo- compression loads. A reliability analysis was con-
technical engineering (Paikowsky 2004; Phoon and ducted for the pile deformation, and the pile SLS was
Kulhawy 2008; Fenton et al. 2016). represented with an allowable load that corresponds
As a scheme of RBD, the load and resistance to a tolerable settlement. First, a database of pile tests
factor design (LRFD) has been adopted in North was collected with high-quality load–displacement
America by several design codes and specifications, responses along with soil measurements. Second, the
such as American Association of State Highway load–displacement curves were normalized by their
and Transportation Officials (AASHTO) (2014), pile capacities identified by the De Beer failure cri-
Canadian Highway Bridge Design Code (CHBDC) terion. Third, a hyperbolic model was selected to rep-
(CSA 2019), and National Building Code of Canada resent the normalized load–displacement responses,
(NRCC 2015). LRFD separates the failure modes into and the hyperbolic model parameters were collected
individual limit states, and two conditions need to be by fitting the model to the normalized load–displace-
met to offer a safe and suitable design for a pile foun- ment curves. Fourth, the hyperbolic model param-
dation: (1) the ultimate limit state (ULS) is related to eters were assessed on their statistical properties,
the load capacity; and (2) the serviceability limit state which includes their mean, coefficient of variation
(SLS) considers the tolerable movement. Currently, (COV), and probability distribution. Their correlation
the ULS has been calibrated for driven piles located structure was represented with copula theory. Lastly,
various regions by many researchers over the last few according to recommendations by AASHTO (2014),
decades (Paikowsky 2004; Abu-Farsakh et  al. 2009; resistance factors were computed for the SLS with
AbdelSalam et al. 2012; Dithinde et al.2011, Salgado Monte Carlo simulations (MCS). In all, this study
et al. 2011); however, the SLS has received less atten- intends to promote the application of RBD for pile
tion due to challenges in accurately representing the designs in glacial deposits.
load-deformation behaviours of a pile.
Recently, several curve-fitting methods have been
proposed to characterize the uncertainties for the SLS 2 Serviceability Limit State Design
of foundations (Misra and Roberts 2009; Uzielli and
Mayne 2011; Huffman et al. 2015, 2016; Reddy and The SLS considers the deformation, cracking, or
Stuedlein 2017; Tang and Phoon 2018a, b, c; Tang vibration of a structure (AASHTO 2014; CSA 2019),
et al. 2019; Wu and Xin 2019). In these studies, the but the SLS will be defined here as the maximum per-
soil conditions were simplified into either cohesive missible displacement,  sa , obtained at the pile head
or cohesionless soils. These ideal soil types may not under a compressive load. Then, for a simple and
properly characterize the ground conditions encoun- familiar application of LRFD, the SLS design can be
tered in the field, such as glacial deposits. Glacial represented in terms of the factored allowable load
deposits, commonly found across northern regions ( Q∗a ) instead of the displacement. Similar to the ULS,
of North America, are one of most challenging geo- Q∗a for the SLS is produced from the pile nominal
materials with their unsorted grain distributions and allowable load, Qan , factored by a resistance factor,
inconsistent material properties (Legget 1965; Mil- 𝜓SLS , and Q∗a should not be exceeded by the summa-
ligan 1976; Barnett 1992; Clarke 2017). To make tion of factored applied loads:
things even more challenging, there is only a limited ∑
number of site investigation techniques available for Q∗a = 𝜓SLS Qan ≥ 𝛾i Qni (1)
this kind of soil because of its dense or stiff nature
and high gravel or boulder bearing potentials. The where Qni is the ith nominal applied load and 𝛾i is
standard penetration test (SPT) is the most popular the corresponding load factor. A failure of the SLS
and, in many cases, may be the only available site is defined as when the displacement from an applied
investigation technique for such a ground. load, Q , is greater than sa , which corresponds to Qa .
The goal of this paper is to statistically assess the The probability of failure, pf   , should meet a tar-
model factors of RBD at the SLS for driven steel piles get probability of failure, pfT  , which depends on the
in glacial deposits with SPT blow counts (N-values). structural importance and/or intent. The pf is then

Vol:. (1234567890)
13
Geotech Geol Eng

expressed with the loads in the following perfor- In practice, the Qum may be replaced by the pre-
mance function (Reddy and Stuedlein 2017; Tang and dicted capacity,  Qup , since the measured value may
Phoon 2018a; Tang et al. 2019): not be available if site-specific load tests are not con-
ducted. The uncertainty of predicting the capacity is
(2)
( )
pf = Pr Qa − Q ≤ 0 ≤ pfT
generally characterized by the capacity bias, Mu  ,
The symbol Pr () represents the probability of the which is a ratio of Qum to Qup:
event. To determine Q∗a , a reliability analysis was con- Mu = Qum ∕Qup (7)
ducted to calibrate 𝜓SLS for pfT  , and the methodology
in this study was similar to those applied by others
(Misra and Roberts 2009; Uzielli and Mayne 2011; Qum is evaluated by a failure criterion on pile load–
Huffman et  al. 2015, 2016; Reddy and Stuedlein displacement responses, while Qup is calculated by a
2017; Tang and Phoon 2018a, b, c; Tang et al. 2019). design method. Thus, Qa is expressed after consider-
In order to assess Eq. 2, a simple hyperbolic rela- ing Mu:
tionship was adopted to represent the load–displace-
ment response of a pile: Qa = Ms Mu Qup (8)
Q∕Qum = s∕(a + bs) (3)
For the AASHTO (2014) Strength limit I, Q is a
where Qum is the ultimate capacity identified from combination of dead ( QDL ) and live loads ( QLL):
the load–displacement response of the pile, s is the Q = 𝜆DL QDL + 𝜆LL QLL (9)
displacement (in mm) at the pile head; and a and b
are the hyperbolic parameters. In this formula, Q is where 𝜆DL and 𝜆LL are the bias factors for QDL and
normalized by Qum . Equation  3 has provided a good QLL , respectively. As suggested by AASHTO (2014),
representation of the load–displacement response for 𝜆DL and 𝜆LL were assumed for this analysis to have a
many foundation elements, such as driven piles (Dith- mean of 1.05 and 1.15, respectively, and the COV of
inde et  al.2011; Tang and Phoon 2018a, b), bored 𝜆DL and 𝜆LL was 10% and 20%, respectively. Equa-
piles (Dithinde et al. 2011; Tang et al. 2019), helical tion 1 is rearranged after letting the predicted nominal
piles (Tang and Phoon 2018c), augered cast-in-place capacity be equal to Qan:
piles (Reddy and Stuedlein 2017), cement-fly ash- ( )
gravel piles (Wu and Xin 2019), and shallow footings Qup = 𝛾DL QDL + 𝛾LL QLL ∕𝜓SLS (10)
(Uzielli and Mayne 2011; Huffman et al. 2015, 2016).
This relationship is preferred due to its simplicity. In where 𝛾DL and 𝛾LL are respectively the dead load fac-
addition, the hyperbolic parameters ( a and b ) have tor and live load factor, which can be equal to unity
physical meanings, where the initial stiffness, or for SLS.
slope, of the pile response is represented by a and the The SLS limit state function, g , is specified:
asymptote of the normalized load is represented by b.
(11)
( )
g = Qa − Q = Qa − 𝜆DL QDL + 𝜆LL QLL
The Qa can be approximated by the hyperbolic
model for a given sa:
After combining Eqs. 8 and 10 into Eq. 11, the modi-
(4)
( )
Qa = sa ∕ a + bsa Qum
fied g is provided (Tang et al. 2019):
The SLS model factor, Ms 
, is defined by the ( ) ( )
following: g=
𝛾DL QDL + 𝛾LL QLL

sa

𝜓SLS a + bsa (12)
(5)
( )
Ms = sa ∕ a + bsa ( )
Mu − 𝜆DL QDL + 𝜆LL QLL
Then, Qa is simplified as shown below.
Finally, the applied loads are replaced by the QDL
Qa = Ms Qum (6) to QLL ratio, 𝜂 , and the following g is used to calculate
pf for the SLS (Tang et al. 2019).

Vol.: (0123456789)
13
Geotech Geol Eng

Fine-grained tills were likely formed by the addi-


( ) ( )
𝛾DL + 𝛾LL ∕𝜂 sa ( )
g= ⋅ Mu − 𝜆DL + 𝜆LL ∕𝜂
tion of glaciolacustrine deposits and contain clay to

𝜓SLS a + bsa
(13) a maximum of approximately 55% (Barnett 1992).
If g is assumed to be normally distributed, the reli- The different origins and geological processes created
ability index, 𝛽 ′ , can be determined from pf by apply- glacial tills in Ontario with a variety of soil charac-
ing the inverse standard normal cumulative function, teristics and contents. For example, Site 5 near Mara-
𝜙−1 (Phoon et al. 2003; Reddy and Stuedlein 2017): thon contained varved silty clays and clayey silts,
while Site 33 near Buttonville had intermittent layers
𝛽 � = −𝜙−1 pf (14) of clayey silts and silty sands. In all, a variety of soil
( )

conditions were encountered in the studied sites.

3.2 Selection of Test Data for the Analysis


3 Studied pile load test database from MTO
According to Abu-Hejleh et  al. (2015), the load test
3.1 Pile Load Test Database database should contain information on the (1) sub-
surface ground conditions at the test site; (2) pile
The pile load tests were conducted for the Ministry of information; and (3) load–displacement curves. The
Transportation of Ontario (MTO) in various regions MTO load test data were further divided into two
of Ontario, Canada. For this study, the focus was on categories: Group 1 considers the load–displacement
driven steel piles, particularly pipe and H piles, sub- response and Group 2 has sufficient measurements for
jected to axial compression loads. The testing pro- determining the pile capacity.
cedures were similar to the ASTM (2013) D1143/ For Group 1, the piles were selected based on the
D1143M standard for a standard (slow) static-main- following conditions: plunging failure, where the
tained load test, and loads were applied to the top capacity can be identified without extrapolation; non-
of the piles by hydraulic jacks that acted against an organic ground; and steel driven piles. In the end,
anchored reaction frame, weighted box, or weighted a total of 42 piles (22 pipe piles; 20 H piles) were
platform. selected.
The pile test results were accompanied with geo- In addition to meeting the requirements for Group
technical borehole logs that provided the soil clas- 1, a total of 36 piles (19 pipe piles; 17 H piles) were
sifications and strength measurement information, selected for Group 2, where sufficient soil measure-
such as SPT N-values. Glacial deposits, also referred ments, in particular SPT N-values and the corre-
to as tills, are routinely encountered in the Province sponding soil type, along the pile length were known
of Ontario. Due to their formation process, tills are from borehole logs.
vertically and horizontally heterogeneous in situ, and The details of the pile geometry, site conditions,
glacial deposits usually contain an inconsistent mix- and load test data are shown in Table  1. The pipe
ture of particle sizes, mineral types, and weathered piles were closed-ended and filled with concrete,
rock types (Legget 1965; Milligan 1976; Barnett except for two that were open-ended. The outside
1992; Clarke 2017). The tills formed from igneous diameter of the pipe piles was 305  mm to 324  mm,
and metamorphic rocks can contain over 70% sand and the H piles were either 310 × 79 or 310 × 110
and less than 5% clay (Barnett 1992). These tills designations (mm × kg/m). The embedment lengths
likely range from non-plastic to slightly plastic due to ( L ) ranged from 3 to 35 m, and the slenderness ratio
the low clay content (Milligan 1976). However, tills ( L∕D ), where D is the pile width or diameter, varied
from carbonate rocks, limestones, and dolostones can from approximately 9–110. With the recommenda-
have a content of 40–50% silt and 15–30% clay (Leg- tions from Canadian Geotechnical Society (CGS)
get 1965; Barnett 1992). Well-graded sediments from (2006), the field SPT N-values were corrected to N60
carbonate rocks lead to compact to very compact tills.

Vol:. (1234567890)
13
Geotech Geol Eng

Table 1  Summary of studied piles


Case no. Site no. Pile no. Pile ­typea Lengthb (m) Embedded soil t­ypec Qum(kN) Qs(kN) Qp(kN) Notes

1 2 5 305 OD Pipe 5.85 Sand 812 Open-ended


2 4 2 324 OD Pipe 35.94 Silt and Clayey Silt 553 343 mm ∅ shoe
3 5 43 HP 310 × 79 20.73 Varved Silty Clay 217
4 7 2 HP 310 × 79 21.70 Clay and Silty Sand 827
5 11 1 HP 310 × 79 26.82 Sand to Silt 550
6 14 2 324 OD Pipe 18.29 Silty Clay 298 343 mm ∅ shoe
7 17 2 HP 310 × 110 26.47 Clayey Silt to Sand 2482
8 22 3 324 OD Pipe 15.30 Clayey Silt 163 148 15 343 mm ∅ shoe
9 22 4 324 OD Pipe 30.15 Clayey Silt 937 343 mm ∅ shoe
10 22 5 324 OD Pipe 15.28 Clayey Silt 233 324 mm ∅ shoe
11 23 2 324 OD Pipe 3.02 Silty Clay 442 217 225 343 mm ∅ shoe
12 23 3 HP 310 × 110 3.05 Silty Clay 425 255 170
13 24 2 324 OD Pipe 15.39 Sand 608 402 206 343 mm ∅ shoe
14 24 3 324 OD Pipe 22.40 Sand 667 443 224 343 mm ∅ shoe
15 24 4 HP 310 × 79 22.40 Sand 1371 420 951
16 24 5 HP 310 × 79 15.39 Sand 702 275 427
17 25 1 324 OD Pipe 5.64 Silty Clay 343 343 mm ∅ shoe
18 25 4 HP 310 × 79 18.44 Silty Clay 873
19 25 5 324 OD Pipe 18.35 Silty Clay 667 343 mm ∅ shoe
20 25 6 324 OD Pipe 9.27 Silty Clay 472 343 mm ∅ shoe
21 25 9 HP 310 × 79 9.39 Silty Clay 489
22 28 1 HP 310 × 79 6.10 Clayey Silt 502
23 28 2 HP 310 × 79 18.29 Clayey Silt 471 331 141
24 28 3 HP 310 × 79 12.19 Clayey Silt 563
25 28 7 324 OD Pipe 6.10 Clayey Silt 658 569 89 343 mm ∅ shoe
26 28 8 324 OD Pipe 18.29 Clayey Silt 659 442 217 343 mm ∅ shoe
27 28 9 324 OD Pipe 12.19 Clayey Silt 567 343 mm ∅ shoe
28 33 2 324 OD Pipe 32.95 Clayey Silt and Silty Sand 2095 342 mm ∅ shoe
29 35 1 HP 310 × 110 14.69 Layered Clayey Silt and 1592 523 1069
Silty Sand
30 35 4 324 OD Pipe 14.69 Layered Clayey Silt and 1507 759 748 343 mm ∅ shoe
Silty Sand
31 35 5 HP 310 × 110 27.58 Layered Clayey Silt and 2714 1524 1190
Silty Sand
32 35 6 324 OD Pipe 28.01 Layered Clayey Silt and 2396 343 mm ∅ shoe
Silty Sand
33 37 3 HP 310 × 79 14.48 Sand to Silty Sand 1042 345 700
34 37 5 HP 310 × 79 31.24 Sand to Sandy Silt 1609 444 1165
35 37 6 HP 310 × 110 14.48 Sand to Silty Sand 717 413 304
36 37 8 HP 310 × 110 30.92 Sand to Silty Sand 1566 699 867
37 38 4 324 OD Pipe 11.30 Silty Clay and Silt to Silty 842 Open-ended
Sand
38 38 5 324 OD Pipe 15.50 Silty Clay and Silt to Silty 2007 Open-ended
Sand
39 39 2 HP 310 × 110 25.50 Silty Sand; Layered Clay 1279 733 546
and Silt

Vol.: (0123456789)
13
Geotech Geol Eng

Table 1  (continued)
Case no. Site no. Pile no. Pile ­typea Lengthb (m) Embedded soil t­ypec Qum(kN) Qs(kN) Qp(kN) Notes

40 39 3 324 OD Pipe 25.40 Silty Sand; Layered Clay 1152 525 627 343 mm ∅ shoe
and Silt
41 40 2 HP 310 × 110 24.50 Layered Sand and Silty Clay 1205 625 580
42 40 3 324 OD Pipe 17.20 Sandy Silt to Sand 1128 544 584 343 mm ∅ shoe
a
 Steel H pile designations are size (mm) by weight (kg/m). Steel pipe piles were filled with concrete before testing, and OD is the
outside diameter (mm)
b
 Embedment Length
c
 The dominating soil type, and classifications are according to MTO standards

for 60% hammer efficiency in cohesive soils and to two pile types. In general, the tested piles have a
(N 1 )60 with an additional overburden pressure cor- diverse set of lengths and soil strengths.
rection in cohesionless soils. A few assumptions
were taken for all cases to correct the SPT N-value,
including a donut hammer type with an estimated rod 4 Collection of Model Factors
energy ratio of 45%; a standard sampler; and a bore-
hole diameter of approximately 100 mm. The average 4.1 Introduction to Model Factors
corrected SPT N-value ( Navg ), using N60 for cohesive
soils and (N 1 )60 for cohesionless soils, along the piles For 𝜓SLS to be calibrated, Eq.  13 requires several
ranged from 5 to 55, and the corrected N-value ( Ntip ) model factors, namely Mu and hyperbolic parameters
at the base of the piles ranged from 5 to 68. Figure 1 a and b . For a specific pfT  , 𝜓SLS was determined in
shows histograms of Navg , Ntip , and L∕D among the this study with MCS using the statistical properties

Fig. 1  Histograms of pile
slenderness ratio, average
corrected N-value along
piles, and corrected N-value
at base of piles

Vol:. (1234567890)
13
Geotech Geol Eng

of the model factors. MCS assesses the pf by gen- different methods on the 42 pile load tests are pre-
erating many sets of input or model parameters that sented by boxplots in Fig. 3, where a line dividing a
formulate the limit state function (Eq.  13). Each set box is the median, a black diamond is the mean load
of parameters is a simulation, and the parameters are by a criterion, and edges of a box are the first and
randomly sampled from their fitted probability distri- third quartiles. The differences between the median
butions. MCS was applied because it is not limited to and mean or quartiles in Fig. 3a indicate the capaci-
a single type of probability distribution, such as nor- ties are likely asymmetrically or non-normally dis-
mal, lognormal, or Weibull, and is applicable for non- tributed. The figure also shows minimum and maxi-
linear limit state functions (Allen et al. 2005; Uzielli mum failure loads by the whiskers. The ratios of Qum
and Mayne 2011; Reddy and Stuedlein 2017). to the average capacity, ( Qum  ), by each criterion are
shown in Fig.  3b. Based on the analysis, it is found
4.2 Measured Axial Capacity ( Qum) that most methods provide a similar range of vari-
ability. The main difference between the failure meth-
The capacity is a key component in the hyper- ods is the average location on the load–displacement
bolic model, but according to Hirany and Kulhawy curve. Thus, the magnitude of the identified capacity
(1989), at least 42 different methods exist to iden- differs depending on the selected criterion, which will
tify the capacity from measured load–displacement influence the results of the remaining analysis.
responses. A total of 11 failure identification meth- Although the Davisson Offset Method is one of
ods were evaluated for their performances in this the most popular methods, it is an empirical method
study, including Butler and Hoy (1977); Chin (1970); and may inaccurately estimate the yielding point
Davisson (1972); De Beer (1967, 1968); Decourt (Fellenius 1980). The yielding loads for the remain-
(1999); Fuller and Hoy (1970); Hansen 80% (Hansen ing investigation were evaluated with the De Beer
1963); Hansen 90% (Fellenius 1980); both L1 and method, which identifies the pile capacity as the load
L2 from the L1-L2 method (Hirany and Kulhawy at the intersection of two slopes from the load–dis-
1989); and Van Der Veen (1953). An example of the placement curve plotted in a double-log scale (De
results is given on the same load–displacement curve Beer 1967, 1968). This method was selected because
by a 324 mm-diameter pipe pile with an embedment it is one of the methods with the lowest interquartile
length of 30  m. Figure  2a shows the site conditions range. In addition, it provides a relatively conserva-
of the tested pile, while Fig. 2b shows the variations tive estimation of the pile capacity by roughly corre-
of different failure methods. The performances of 11 sponding to the point of maximum curvature on the

Fig. 2  Example of ground
profiles and load–displace-
ment curve for pile 4 at
site 22

Vol.: (0123456789)
13
Geotech Geol Eng

Fig. 3  Variability of a capacity and b normalized capacity by failure criteria

load–displacement curve and does not provide an area of the pile base. For SPT measurements, Qup can
excessively large failure load, like the Chin, Decourt, be calculated with direct or indirect design methods.
or Hansen 80% criteria shown in Fig. 2b. Direct design methods are empirical correlations
linking the pile resistances directly with SPT N-val-
4.3 Predicted Axial Capacity ( Qup) ues. Many linear correlations have been proposed for
a variety of soil types (Thorburn and MacVicar 1971;
The capacity of a pile subjected to a compressive load Aoki and Velloso 1975; Meyerhof 1956, 1976; Shioi
is composed of the side resistance ( Qs ) and tip resist- and Fukui 1982; Martin et  al. 1987; Decourt 1982,
ance ( Qp): 1995; Brown 2001; Shariatmadari et  al. 2008), but
they can be limited to the regional soil conditions. For
Qup = Qs + Qp = Σqs As + qp Ap (15) indirect methods, SPT N-values first derive the soil
where qs is the unit side resistance, qp is the unit tip strength parameters, such as the undrained shear
resistance, As is the side area of the pile, and Ap is the strength ( Cu ) and the friction angle ( 𝜙′ ), and the pile

Vol:. (1234567890)
13
Geotech Geol Eng

resistances are then predicted with the strength cohesive soils and the 𝛽 method by Berezantzev et al.
parameters based on classical soil mechanics theory. (1961) for cohesionless soils. These approaches are
Common indirect design methods can generally be simple and are examples of the traditional 𝛼 and 𝛽
categorized as 𝛼 methods (Tomlinson 1957; API methods. As recommended by Tomlinson and Wood-
2000) for cohesive soils and 𝛽 methods (Berezantzev ward (2008), the Berezantzev et al. method applied a
et  al. 1961; Nordlund 1963, 1979) for cohesionless friction angle of the soil-pile interface, δ, that was
soils or the long-term conditions of cohesive soils. 65% of 𝜙 and an installation coefficient, Kf  , that was

In this study, Qup was calculated with both direct 0.75 for H Piles and 1.0 for pipe piles. The second
and indirect design methods that can account for het- method was based on recommendations from Federal
erogeneous soil profiles and different pile geometries, Highway Administration (FHWA) (Hannigan et  al.
namely pipe and H piles. The three selected methods 2016) for layered soil profiles and included the 𝛼
are summarized in Table 2 for the side resistance and method adopted by API (2000) for cohesive soils and
Table  3 for the tip resistance. The first method is a the 𝛽 method by Nordlund (1963, 1979) for cohesion-
combination of the 𝛼 method by Tomlinson (1957) for less soils. According to a survey conducted by

Table 2  Summary of selected design methods for the pile side resistance


Method Soil type Equation for qs Remarks

Tomlinson (1957) + Berezantzev Cohesive qs = 𝛼Cu 𝛼 is an adhesion factor


et al. (1961) Cohesionless qs = Kf Ko 𝜎 � tan 𝛿 Kf is installation coefficient, Ko is the lateral earth
coefficient at rest, σ′ is the effective stress, and δ is
the friction angle of the soil-pile interface
API (2000) + Nordlund (1963, Cohesive qs = 𝛼Cu ( )−0.5
𝛼 = 0.5 Cu
1979) ( 𝜎 � )−0.25for 𝜎 ′ ≤ 1
Cu

𝛼 = 0.5 𝜎 �Cu
for Cu
𝜎′
>1
Cohesionless qs = K𝛿 CF 𝜎 � sin 𝛿 K𝛿 is the coefficient of lateral earth pressure, and CF
is an empirical correction factor for when 𝛿 is not
equal to 𝜙′
Jesswein and Liu (2021) Cohesive and qs = AN1cor 𝜎 � A is an empirical coefficient. For pipe piles, A ranges
cohesionless from 0.132 for desiccated clays to 0.01 for gravels
or gravelly soils (content > 20 %). For H piles, A
ranges from 0.125 for desiccated clays to 0.0038 for
gravels or gravelly soils. Ncor is N60 for cohesion-
less soils and (­ N1)60 for cohesive soils.

Table 3  Summary of Selected Design Methods for the Pile Tip Resistance


Method Soil Type Equation for qp Remarks

Tomlinson (1957) + Berezantzev Cohesive qp = Nc Cu Nc is an end bearing factor and is equal to 9 (Meyerhof


et al. (1961) 1976)
Cohesionless qp = Nq 𝜎t� Nq is an end bearing factor, and 𝜎t′ is the effective stress
at the pile tip
API (2000) + Nordlund (1963, Cohesive qp = Nc Cu Nc is equal to 9 (Meyerhof 1976)
1979) Cohesionless qp = 𝛼t Nq 𝜎t� 𝛼t is an end bearing correction factor for consideration of
the pile slenderness ratio
Jesswein and Liu (2021) Cohesive and qp = 1.3Ntip 𝜎t� For cohesionless soils, Ntip is (N1 )60 , for cohesive soils,
cohesionless (N1 )60

Vol.: (0123456789)
13
Geotech Geol Eng

AbdelSalam et  al. (2012) in America, the Nordlund and 1.16 for H piles. For the two H and pipe pile
method is one of the most popular approaches for a types, the variability is generally high as the COV of
static analysis. For qp by the Nordlund method, the Mu is 50.9% for the API and Nordlund approaches
effective vertical stress at the pile base, 𝜎t′ , was lim- and 47.4% for the Tomlinson and Berezantzev et  al.
ited to 150  kPa as suggested by Hannigan et  al. approaches. Most of the variabilities by these indirect
(2016). The third method is the direct approach pro- design methods are due to the difficulties in accurately
posed by Jesswein and Liu (2021) based on an earlier characterizing the shear strength parameters, namely
study with the same pile load database. It further clas- Cu and 𝜙′ . Many of the cohesionless soils were classi-
sified cohesive soils into more sub-types based on silt fied as dense ( 30 ≤ field N-value ≤ 50 ) to very dense
contents and moisture changes and considers the (field N-value > 50 ). In addition, the gravel content in
gravel content in cohesionless soils. the ground can lead to excessively high N-values. For
For indirect methods with cohesionless soils, 𝜙′ example, field N-values ranged from 38 to 154 at Site
was determined by the correlation from Wolff (1989): 38 as the gravel content varied between 10 and 50%.
Although corrected N-values were capped to 50, high
𝜙� = 27.1 + 0.3N60 − 0.00054N60 (16)
2
field N-values do not proportionally reflect the shear
strength of the soil with the correlations for Cu and
For cohesive soils, Cu was determined based on
𝜙′ and would result in a greater strength than reality.
the empirical relationships by Sivrikaya and Toğrol
The direct design method by Jesswein and Liu (2021)
(2006). Cu was equal to 4 N60 for clays, 3.8N60 for
provides unbiased predictions with an average Mu of
silty clays, 3N60 for clayey silts, and 9N60 for desic-
0.97 for pipe piles and 1.00 for H piles. The method
cated (hard) clays.
also has a smaller COV of 44.4% for pipe piles and
The load capacity calculations were conducted
25.4% for H piles. Some of the reasons are due to the
with all 36 piles in Group 2 from the database. The
consideration of silts in cohesive soils and the gravel
changing soil profiles and SPT-N-values were consid-
content in cohesionless soils. The main reason is the
ered by dividing the pile length into small segments.
method was developed with the same database, and
The corrected N-values were capped to a maximum
the effectiveness of this method requires further veri-
value of 50. Both H piles and open-ended pipe piles
fications with pile tests in glacial deposits from other
were assumed to be fully plugged as suggested by
sources. In general, variabilities are also found due
Hannigan et al. (2016).
to the inconsistencies in the SPT measurements and
factors related to the pile length, such as the residual
4.4 Comparison between Measured and Predicted
loads and load transfer along piles of different lengths
Capacities
(Nordlund 1963; Semple and Rigden 1986; Fellenius
2019).
The predictions were compared to the measured
results, as shown in Fig. 4, for both H and pipe piles.
4.5 Hyperbolic Parameters
A perfect prediction exists if a datapoint is located on
the solid black 1:1 line. The lower and upper dashed
The load–displacement responses of pipe piles
lines indicate an underprediction or overprediction of
are shown in Fig.  5a with the normalized curves in
the measured capacity by 30%, respectively. For each
Fig.  5b. For H piles, the load–displacement curves
design method, Table  4 summarizes the descriptive
and their normalized ones are shown in Fig. 5c and d,
statistics of Mu , which shows the uncertainties of the
respectively. In Fig. 5b and d, the elastic and plung-
predictions. The API and Nordlund approaches tend
ing slopes on the normalized curves vary due to the
to overpredict the pile capacity by a factor of 1.28
different failure mechanisms, pile geometries, and
on average. This is also indicated by a Mu lower than
soil conditions. The hyperbolic parameters a and b in
one, such as the average Mu of 0.90 for H piles. The
Eq. 3 were collected for each normalized curve with
Tomlinson and Berezantzev et al. approaches slightly
nonlinear least squares regression.
underpredict with an average Mu of 1.10 for pipe piles

Vol:. (1234567890)
13
Geotech Geol Eng

Fig. 4  Comparison of predicted and measured capacities: (a and b) API and Nordlund, (c and d) Tomlinson and Berezantzev et al.,
and (e and f) Jesswein and Liu

Vol.: (0123456789)
13
Geotech Geol Eng

Table 4  Descriptive Statistics of Mu for Driven Piles 5 Statistical Analysis of Model Factors


Pile type n Design method Mean COV (%)
During the MCS, the model factors ( a , b , and Mu )
Pipe piles 19 Tomlinson + Berezantzev 1.10 54.6 are simulated by their source probability distributions
et al
and should be considered as random variables. The
API + Nordlund 1.00 57.3
statistical properties of the model factors were inves-
Jesswein and Liu 0.97 44.4
tigated for all 42 pile load tests. The statistical prop-
H Piles 17 Tomlinson + Berezantzev 1.16 38.6
et al
erties were evaluated by verifying the randomness;
API + Nordlund 0.90 39.4
collecting descriptive statistics, such as the mean and
Jesswein & Liu 1.00 25.4
COV; and fitting probability distributions.

Fig. 5  Measured load–displacement responses from load tests

Vol:. (1234567890)
13
Geotech Geol Eng

5.1 Verification of Randomness be due to the different pile geometries, soil condi-


tions, and failure criteria applied within the studies.
Before assuming the model parameters are ran- Depending on the pile type in this study, a was found
dom variables, the parameters need to be checked to have a fairly strong positive correlation with L∕D ,
for potential biases from the soil or pile properties Ntip , and/or Navg . This dependency can be expected
(Uzielli and Mayne 2011; Reddy and Stuedlein 2017; due to the rigidity of the pile material and soil condi-
Tang and Phoon 2018a; Tang et al. 2019). Potentially tions. The load-transfer effect within a pile is related
influential variables include Navg , Ntip , and L∕D . If to L∕D , and a short pile will likely mobilize both the
any correlation exists, the model factor is transformed side and tip resistance simultaneously and have less
with a function to remove the statistical dependency, elastic displacement than a long pile. A greater Ntip
and the COV of the transformed model factor can be indicates a denser and less compressible soil, and the
reduced compared to the non-transformed model fac- pile deformation may rely more on the tip resistance.
tor (Tang and Phoon 2018a). For pipe piles, b had a negative correlation with L∕D
Correlations between the model factors and L∕D , and Navg , while b solely had a negative correlation
Ntip , and Navg can be assessed with statistical tests, with L∕D for H piles. Overall, the correlations in this
such as the Pearson Product-Moment correlation test study indicate long piles, especially in strong soils,
or Spearman rank-correlation test (Howell 2002). will likely have a more gradual failure response than
The Spearman rank-correlation was used since it can short piles.
assess monotonic relationships, including both lin- Since dependencies were found among the model
ear and non-constant trends, between two variables parameters ( a and b ), the correlations were removed
(Howell 2002). If the p-value is less than 0.05, the by empirically transforming a and b to their inde-
null hypothesis can be rejected within the 95% con- pendent counterparts, at or bt . For pipe piles, the
fidence interval, and the correlation indicated by the following transformation functions were obtained
coefficient, r , is significantly different from zero, indi- by regressing L∕D , Navg , and/or Ntip to the original
cating a dependence between the variables (Howell model parameter:
2002). If the p-value is greater than 0.05 or r is less 0.60
than 0.40, a weak correlation likely exists, and it can a = 0.291 ⋅ exp (0.014 ⋅ L∕D) ⋅ Ntip ⋅ at (17)
be ignored.
For the hyperbolic model factors, Table  5 shows ( ( ) )
b = (−0.060 ⋅ ln(L∕D) + 0.96) −0.11 ⋅ ln Navg + 1.29 ⋅ bt
the results of the Spearman rank-correlation test and (18)
the number of analyzed samples, n . For deep foun-
For H piles, the following expressions were
dations, there are discrepancies among the findings:
obtained:
some researchers noticed a and b were biased by L∕D
(Reddy and Stuedlein 2017; Tang and Phoon 2018b), a = 0.99 ⋅ exp(0.018 ⋅ L∕D) ⋅ at (19)
while others (Dithinde et  al. 2011; Tang and Phoon
2018a, c; Tang et  al. 2019) did not find any biases b = (−0.0028 ⋅ ln (L∕D) + 0.94) ⋅ bt (20)
within their model factors. These discrepancies may

Table 5  Results of Pile type Variable n a b


spearman rank-correlation
for hyperbolic parameters r p-value r p-value

Pipe piles L∕D 22 0.60 0.00  − 0.55 0.01


Navg 19 0.55 0.01  − 0.65 0.00
Ntip 19 0.59 0.01  − 0.27 0.26
H piles L∕D 22 0.65 0.00  − 0.65 0.00
Bold values indicate a Navg 19 0.33 0.18  − 0.12 0.63
potential correlation with Ntip 19 0.61 0.01  − 0.45 0.06
the model factor

Vol.: (0123456789)
13
Geotech Geol Eng

Table 6  Results of Pile type Variable n API + Nordlund Tomlin- Jesswein and Liu


spearman rank-correlation son + Berezantzev
for Mu by design method et al
r p-value r p-value r p-value

Pipe Piles L∕D 19  − 0.62 0.00  − 0.39 0.10  − 0.39 0.10


Navg 19  − 0.09 0.70  − 0.13 0.60  − 0.16 0.23
Ntip 19  − 0.01 0.96  − 0.05 0.85 0.18 0.45
Qum 19 0.25 0.30 0.39 0.10 0.40 0.09
H Piles L∕D 17  − 0.72 0.00  − 0.34 0.18  − 0.15 0.57
Navg 17  − 0.21 0.42  − 0.30 0.25 0.23 0.38
Bold values indicate a Ntip 17  − 0.31 0.23  − 0.26 0.31 0.26 0.31
potential correlation with Qum 17  − 0.20 0.44  − 0.05 0.85 0.39 0.13
the model factor

Reddy and Stuedlein (2017) and Tang and Phoon 5.2 Fitting of Probability Distributions
(2018b) offered different functions because the corre-
lations vary with the capacity failure criterion, data- In order to conduct MCS, the statistical distributions
base, and foundation type. of the model factors need to be identified. Distribu-
For the three design methods, Table  6 shows the tions were considered if they assumed the model fac-
potential correlations between Mu and L∕D , Ntip , Navg , tors could only possess values greater than zero (posi-
and Qum . Bathurst and Bozorgzadeh (2019) suggested tive values), and the goodness-of-fit was conducted
small or large Qum values may influence Mu , but none with the Anderson–Darling test (Anderson and Dar-
of the design methods or pile types indicated a signif- ling 1952) at the 95% confidence interval. Mu and Mut
icant relationship. Notably, predictions for the capac- were fitted to lognormal distributions, while at and
ity with the API and Nordlund methods were biased bt were fitted to Weibull distributions with a best-fit
by L∕D , and Mu was transformed to Mut with the scale 𝛼w and shape 𝛽w parameter obtained from maxi-
function below for pipe piles: mum likelihood estimation. Figures 6 and 7 show the
fitted cumulative distribution functions offer a reason-
Mu = (−0.62 ⋅ ln (L∕D) + 3.41) ⋅ Mut (21) able representation of the model factors.

5.3 Simulation of Hyperbolic Parameters


while H piles had transformation function below.
Figure 8 shows that the hyperbolic parameters a and
Mu = 1.72 ⋅ exp (−0.011 ⋅ L∕D) ⋅ Mut (22) b are weakly and inversely correlated, and this is con-
firmed with a Pearson Product-Moment correlation
Many of the studied piles in cohesive soils have
coefficient, 𝜌 , of -0.64 for pipe piles and -0.48 for
high Cu values at shallow depths. Yet, for high
H piles. The figure also shows the Kendall tau coef-
strength ratios ( Cu∕𝜎 � ), the API 𝛼 method uses a low
ficient, 𝜌𝜏 , between a and b . However, for the trans-
adhesion factor and underestimates the side resistance
formed parameters ( at and bt ), the relationships are
of short piles as a result. Longer piles generally have
significantly weaker with a 𝜌 of − 0.45 for pipe piles
low Cu∕𝜎 � ratios as 𝜎 ′ increases, but the API method
and − 0.085 for H piles. This change in dependency
applies a higher adhesion factor, which overestimates
may indicate that a and b are likely indirectly corre-
the side resistance. Overpredictions may also occur
lated by the underlying soil and pile properties.
with cohesionless soils due to the difficulty to prop-
The relationship between at and bt for both pipe
erly determine the shear strength from high N-values.
and H piles must be considered in the analysis; oth-
The Mu from the Jesswein and Liu and Tomlinson
erwise, biased results may develop during the reli-
and Berezantzev et al. methods did not share signifi-
ability analysis if these two parameters were assumed
cant correlations with the variables.
to be independent. Copulas represent the correlation

Vol:. (1234567890)
13
Geotech Geol Eng

Fig. 6  Cumulative distribution for at and bt

Fig. 7  Cumulative distribution for Mu or Mut for a Pipe piles and b H piles

structure between two or more variables by math- piles, similar to several other studies (Uzielli and
ematically joining their cumulative probabilities Mayne 2011; Reddy and Stuedlein 2017; Tang et al.
(Nelsen 2007). Copula theory was used in this study 2019). For the Gaussian, Clayton, Frank, Plackett,
to simulate the load–displacement response of the Gumbel, and Joe copulas, the Multivariate Copula

Vol.: (0123456789)
13
Geotech Geol Eng

Fig. 8  Observed and simulated correlation of hyperbolic parameters

Analysis Toolbox (MvCAT) by Sadegh et  al. (2017) the mode and maximum likelihood estimate (MLE)
in Matlab (Mathworks 2017) was used to estimate of the parameter. For the pipe piles, the 𝜃 by MLE
the copula parameter 𝜃 , which indicates the strength was − 0.27, 1.80, 0.45, 1.95, and 2.69 for the Gauss-
of the correlation between at and bt . With the cumu- ian, Clayton, Plackett, Gumbel, and Joe copulas,
lative probabilities of at and bt from the previously respectively. These values were similar to the results
fitted Weibull distributions, MvCAT finds 𝜃 with from local optimization, which provided a 𝜃 of − 0.49,
two approaches: a local optimization algorithm and 1.12, 0.45, 1.54, and 2.69 for the Gaussian, Clayton,
Bayesian framework (Sadegh et  al. 2017). Sadegh Plackett, Gumbel, and Joe copulas, respectively. For
et  al. (2017) suggests local optimization approaches the Frank copula (Fig.  9c), the local optimization
may lead to biased values for 𝜃 as an algorithm may approach likely became trapped, and the estimated 𝜃
become trapped on a local minimum. On the other significantly differed from the Bayesian analysis. For
hand, a Bayesian approach can be more robust to H piles, 𝜃 mostly coincided between the local optimi-
locate the global optimum and can describe the zation and Bayesian approaches. For both pile types,
underlying uncertainty to estimate 𝜃 . It should be as indicated by the estimated 𝜃 with the Gumbel and
noted that the Clayton, Gumbel, and Joe copulas were Clayton copulas, the Bayesian analysis suggests the
intended for positively correlated data. While deter- correlations between at and bt are slightly stronger
mining 𝜃 with these copulas, the observed negative compared to the local optimization, but the correla-
correlation between at and bt was rotated 90 degrees tions are slightly weaker with the Gaussian and Frank
by replacing the cumulative probability ( u ) of at with copulas by the Bayesian analysis.
1-u (Reddy and Stuedlein 2017). As shown in Table  7, the copulas were assessed
Figures  9 and 10 show the results from MvCAT with the Bayesian Information Criterion (BIC). Based
for pipe piles and H piles, respectively. Both opti- on that a lower BIC suggests a better fit, a Clayton-
mization approaches were applied to estimate 𝜃 . type copula was selected with a 𝜃 of 1.80 for pipe
The red star is the result from local optimization, piles, and a Gumbel-type copula was chosen with a
while a histogram shows the posterior distribution 𝜃 of 1.36 for H piles. To verify the fit, these copu-
of 𝜃 from the Bayesian analysis. A black cross shows las were used to simulate 5000 pairs of hyperbolic

Vol:. (1234567890)
13
Geotech Geol Eng

Fig. 9  Posterior distribution of the copula parameter for pipe piles with a Gaussian, b Clayton, c Frank, d Plackett, e Gumbel, and f
Joe Copula

Vol.: (0123456789)
13
Geotech Geol Eng

Fig. 10  Posterior distribution of the Copula Parameter for H Piles with a Gaussian, b Clayton, c Frank, d Plackett, e Gumbel, and f
Joe Copula

Vol:. (1234567890)
13
Geotech Geol Eng

Table 7  Goodness-of-fit Test for the Selected Copula of parameters at and bt . Equations  17–20 transform
Hyperbolic Parameters at and bt to a and b to simulate the load–displace-
Pile type n Copula Copula BIC ment responses. For the simulations, L∕D , Navg , and
Parameter,𝜃 Ntip were assumed to range from 10 to 110, 5 to 50,
Pipe Piles 22 Gaussiana  − 0.27  − 134.5
and 5 to 50, respectively, with a uniform distribu-
Claytona,c 1.80  − 152.3
tion for both pile types. These ranges were based on
Franka  − 1.61  − 136.1
the collected database, and corrected N-values may
Placketta 0.45  − 136.2
be limited to 50 to prevent overestimations in prac-
Gumbel1,3 1.95  − 137.5
tice (Brown  2001). The range for L∕D roughly cor-
Joea,c 2.69  − 127.0
responds to a pile length of 3  m to 35  m. Then, for
H Piles 20 Gaussianb  − 0.10  − 125.0
both pipe piles and H piles, Eq.  3 allowed Q to be
Claytona,c 0.67  − 130.3
determined for each set of L∕D , Navg , Ntip , at , and bt .
Franka  − 0.92  − 124.6
If a simulation provided b with a value greater than
Placketta 1.16  − 124.6
one, the simulation was removed. Logically, b should
Gumbela,c 1.36  − 139.2
be less than one since the identified failure load by
Joea,c 1.62  − 135.9
the De Beer criterion is normally conservative. This
situation occurred with less than 1.5% of the total
Bold text indicates the selected copula samples. The results of the simulations are shown in
a
 𝜃 is maximum likelihood estimate Fig. 8 for the two hyperbolic parameters and Fig. 11
b
 𝜃 is from localoptimization for the load–displacement curves. The figures help
c
 copula was rotated 90 degrees verify the process to create the simulations and show
the simulated load–displacement responses are within
a similar range as the observed responses; thus, the

Fig. 11  Observed and simulated load–displacement responses

Vol.: (0123456789)
13
Geotech Geol Eng

results indicate the copulas can satisfactorily repre- increases with a greater 𝜂 , but the difference is almost
sent the pile load–displacement responses. negligible between a 𝜂 of 3 to 10. With 𝜂 equal to 3,
Table  8 also summarizes the calibrated 𝜓SLS values
for an sa of 13  mm (approximately 0.5 inches) and
6 Calibration of Resistance Factors for SLS 25 mm (1 inch).
Among the three design approaches, the API and
For a sa that ranged from 5 to 25  mm and a 𝜂 that Nordlund method has the lowest 𝜓SLS because it tends
ranged from 1 to 10, MCS calibrated 𝜓SLS with the to overpredict the capacity the most with the high-
following steps suggested by Abu-Farsakh et  al. est variability, which increases the chances of fail-
(2009, 2010, 2013) and Tang et al. (2019): ure. Most of the uncertainty is due to the bias with
L∕D and the Qs predictions with H piles. Particularly
1. Chose a starting value for 𝜓SLS. for cohesive soils, it is difficult to accurately deter-
2. For a given sa and 𝜂 , generated values for L∕D , mine the settlement as the design method relies on
Navg , Ntip , 𝜆DL , 𝜆LL , and Mu or Mut by indepen- empirical relationships with SPT N-values to deter-
dently sampling them from their source distri- mine Cu . In addition, the API method tends to give
butions. Lognormal distributions were assumed higher predictions for the capacity than the Tomlin-
to represent 𝜆DL and 𝜆LL . Model factors at and bt son method. The Tomlinson and Berezantzev et  al.
were first randomly sampled using a Gaussian methods are more robust than the API and Nordlund
copula with their probability distributions. Equa- methods and have higher 𝜓SLS values as a result. For
tions  17–20 transformed the simulated at and instance, 𝛿 and Kf for the Berezantzev et al. 𝛽 method
bt to a and b to consider the dependencies with can be empirically adjusted for various pile geom-
L∕D , Navg , and/or Ntip . For the API and Nordlund etries and soil conditions, but the equivalent param-
design methods, Eqs. 21 and 22 transformed Mut eters depend on the displaced soil volume by the pile
to Mu. and soil shear strength with the Nordlund 𝛽 method.
3. Counted the number of failures, where the result The Tomlinson and Berezantzev et  al. methods are
from the limit state function (Eq. 13) is equal to slightly conservative compared to the Jesswein and
or less than zero. Calculated pf as the number of Liu direct method, and these approaches provide a
failures divided by the total number of samples similar range of 𝜓SLS for pipe piles. For H piles, the
and 𝛽 ′ with Eq. 14. The total number of simula- Jesswein and Liu method offers the lowest variabil-
tions was slightly less than five million since the ity and provided approximately 15% higher resistance
simulations were removed when b was greater factors. Although the accuracy of the Jesswein and
than one. However, less than 2% and 0.02% of Liu method requires further confirmation with more
the total simulations were affected for H piles and tests on driven piles, these results indicate the impor-
pipe piles, respectively. tance of applying robust methods that consider the
4. Repeated Steps 1 to 3 until the absolute differ- soil gradation, in particular the gravel content, with
ence between calculated ( 𝛽 ′ ) and target reliabil- SPT N-values.
ity index ( 𝛽T  ) was less than the desired tolerance,
which was set as 0.01 in this study. 𝛽T was 2.33,
3, and 3.5 that respectively corresponds to redun- 7 Conclusions and Discussions
dant piles, non-redundant piles (typical geotech-
nical designs), and bridge designs (AASHTO With a database of 42 pile load tests, reliability-based
2014). design methods were calibrated for the serviceability
limit state (SLS) of driven piles in glacial deposits.
Figure  12 shows the resulting 𝜓SLS for the three The SLS was simulated by adopting a hyperbolic rela-
design methods for pipe piles and H piles subjected tionship for the load–displacement responses. First,
to compression loads. In general, the resistance factor the model parameters ( Mu , a , and b ) were collected
increases as the sa increases, and the most significant from the pile load–displacement responses. Then, a
change is between 5 and 10 mm. With less variability statistical analysis was conducted to characterize the
from the live load, the resistance factor also slightly dependency of the model factors and to fit them to

Vol:. (1234567890)
13
Geotech Geol Eng

Fig. 12  Variation of 𝜓SLS with sa and 𝛽T for the design methods

Vol.: (0123456789)
13
Geotech Geol Eng

Table 8  Variation of Pile type sa (mm) Design method 𝜓SLS


𝜓SLS for an 𝜂 of three and
different values of sa and 𝛽T 𝛽T = 2.33 𝛽T = 3.00 𝛽T = 3.50

Pipe piles 13 Tomlinson + Berezantzev et al 0.23 0.16 0.12


API + Nordlund 0.16 0.11 0.08
Jesswein & Liu 0.22 0.16 0.12
25 Tomlinson + Berezantzev et al 0.30 0.21 0.16
API + Nordlund 0.22 0.16 0.12
Jesswein & Liu 0.29 0.21 0.16
H Piles 13 Tomlinson + Berezantzev et al 0.34 0.25 0.20
API + Nordlund 0.25 0.19 0.16
Jesswein & Liu 0.39 0.30 0.25
25 Tomlinson + Berezantzev et al 0.42 0.31 0.26
API + Nordlund 0.35 0.27 0.23
Jesswein and Liu 0.49 0.40 0.34

probability distributions. With the source distribu- It is recommended to use the calibrated 𝜓SLS with
tions of the model factors, Monte Carlo simulations the De Beer method for the pile capacity as the find-
were used to calibrate the resistance factors, as shown ings will likely differ on the selected failure criterion.
in Fig.  12 and Table  8, for three SPT-based design Among the studied piles, the average capacity was
methods: (1) Tomlinson (1957) and Berezantzev et al. 891 kN by the De Beer method, which would be 46%
(1961); (2) API (2000) and Nordlund (1963, 1979); lower by the L1 method and 140% more by the Chin
and (3) Jesswein and Liu (2021). method. The failure criterion also controls the magni-
The calibrated resistance factors ranged from 0.16 tude and variability of the model factors ( Mu , a , and
to 0.40 for a sa of 25 mm for the conditions of 𝛽T of 3 b ). However, the calibrated 𝜓SLS mainly depends on
and 𝜂 of 3, and these values mainly depended on the the ability to predict the capacity since many of the
selected design method. Among all these piles, the design methods had the same statistical characteris-
combination of Tomlinson (1957) and Berezantzev tics with a and b for a pile type.
et al. (1961) methods had a Mu with a mean of 1.10 Due to the number of available pile load tests, the
and COV of 47.4%. These approaches had less vari- presented analysis mainly considers the difference
ability and were less likely to overpredict compared between pipe piles and H piles and does not differen-
to the combination of API and Nordlund methods, tiate 𝜓SLS between embedded soil types. The biases,
which had a mean of 0.96 and COV of 50.9% for correlations, and load–displacement responses would
Mu . In particular, the Berezantzev et  al. method for likely vary by the soil type. Although SPT N-values
cohesionless soils can be robust as its design vari- may be the only available soil testing technique, piles
ables, namely 𝛿 and Kf  , can be empirically adjusted in cohesive soils may be influenced by the soil plas-
based on the pile type and local soil conditions. Due ticity, sensitivity, and stress history. A larger database
to the reduced variability with the Jesswein and Liu would help to confirm the findings.
method, which had a COV of 36.4%, resistance fac-
tors were par to 45% higher than those from the other Acknowledgements  The presented research was made pos-
sible with funding from the Ministry of Transportation of
two approaches. This heightened performance is a Ontario (MTO) and support through a graduate scholarship
result of the method being developed with the same from the National Sciences and Engineering Research Coun-
database, and its accuracy requires further confirma- cil of Canada. In addition, the authors would like to thank Mr.
tion with pile tests from other sources. In general, the David Staseff and Ms. Minkyung Kwak from MTO for sharing
the data of pile load tests.
analysis demonstrates the importance of using locally
developed and robust design methods for the capacity.

Vol:. (1234567890)
13
Geotech Geol Eng

Funding  This research was funded by Ministry of Trans- the Fifth Pan-American Conference on Soil Mechanics
portation Ontario Highway Infrastructure Innovations Fund- and Foundation Engineering, Buenos Aires, Argentina,
ing Program grant entitled "Improving Pile Design in Ontario 367–376
Soils." Barnett PJ (1992) Chapter 21: quaternary geology of Ontario.
In Thurston PC, Williams HR, Sutcliffe RH, Scott GM
Data availability  Some or all data, models, or code gener- (Eds.) Geology of Ontario: Ontario Geological Survey
ated or used during the study are proprietary or confidential in (Special Vol 4, Part 2, pp 1011–1088). Sudbury, ON:
nature and may only be provided with restrictions. Ontario Ministry of Northern Development and Mines
Bathurst RJ, Bozorgzadeh N (2019) A simple probabilistic
Declarations  internal stability analysis and design of reinforced soil
walls. GeoSt.John’s 2019, St. John’s, NL: CGS. Retrieved
Conflict of interest  The authors have not disclosed any com- from, http://​www.​geost​johns​2019.​ca/
peting interests. Becker DE (1996) Eighteenth Canadian Geotechnical Col-
loquium: limit states design for foundations. Part I: an
overview of the foundation design process. Can Geotech
J 33:956–983
References De Beer EE (1967) Proefondervindelijke bijdrage tot de studie
van het grensdraagvermogen van zand onder funderingen
AbdelSalam SS, Ng KW, Sritharan S, Suleiman MT, Roling op staal (Deel 1). Annales des travaux publics de Belgique
M (2012) Development of LRFD procedures for bridge (2nd series), 68(6), 481–504
pile foundations in Iowa. Vol. III: Recommended resist- De Beer EE (1968) Proefondervindelijke bijdrage tot de studie
ance factors with consideration of construction control van het grensdraagvermogen van zand onder funderingen
and setup (Report No. IHRB: Project TR-584). Ames, IA: op staal (Deel 2–3). Annales des travaux publics de Bel-
Iowa Department of Transportation gique (2nd series), 69(1), 44–88; 69(4), 321–360
Abu-Farsakh MY, Yoon SM, Tsai C (2009) Calibration of Berezantzev, V.G., Khristoforov, V.S., & Golubkov, V. N.
resistance factors needed in the LRFD design of driven (1961). Load bearing capacity and deformation of piled
piles (Report FHWA-LA-09–449). Baton Rouge, LA: foundations. In: Proceedings of the 5th International Con-
Louisiana Transportation Research Center ference on Soil Mechanics and Foundation Engineering,
Abu-Farsakh MY, Yu XB, Yoon SM, Tsai C (2010) Calibra- 2, 11–15
tion of resistance factors needed in the LRFD design of Brown RP (2001) Predicting the ultimate axial resistance of
drilled shafts (Report FHWA-LA-10-470). Baton Rouge, single driven piles (Doctoral dissertation). University of
LA: Louisiana Transportation Research Center Texas, Austin, TX, Department of Civil Engineering
Abu-Farsakh MY, Chen QM, Haque MN (2013) Calibration Butler HD, Hoy HE (1977) Users manual for the Texas quick-
of resistance factors for drilled shafts for the new FHWA load method for foundation load testing (Report IP-77-8).
design method (Report FHWA-LA-12–495). Baton FHWA, Washington, DC
Rouge, LA: Louisiana Transportation Research Center Canadian Geotechnical Society (CGS) (2006) Canadian foun-
Abu-Hejleh NM, Abu-Farsakh M, Suleiman MT, Tsai C (2015) dation engineering manual, 4th edn. CGS, Richmond, BC
Development and use of high-quality databases of deep Canadian Standards Association (CSA) (2019) Canadian High-
foundation load tests. Transp Res Rec 2511:27–36. https://​ way Bridge Design Code. Mississauga, ON: CSA
doi.​org/​10.​3141/​2511-​04 Chin FK (1970) Estimation of the ultimate load of piles from
Allen TM, Nowak AS, Bathurst RJ (2005) Calibration to deter- tests not carried to failure. In: Proceedings of second
mine load and resistance factors for geotechnical and southeast asian conference on soil engineering, Singapore
structural design. Transportation Research Board Circular City, Singapore, 81–92
E-C079. Washington, DC: Transportation Research Board Clarke BG (2017) Engineering of glacial deposits. CRC Press,
American Association of State Highway and Transportation Boca Raton, FL
Officials (AASHTO) (2014) AASHTO LRFD bridge Davisson MT (1972) High capacity piles. In: Proceedings of
design specifications. AASHTO, Washington, DC Lecture Series on Innovations in Foundation Construc-
American Petroleum Institute (API) (2000) Recommended tion, Chicago, IL: ASCE, March 22, 81–112
practice for planning, designing, and constructing fixed Decourt L (1995) Prediction of load-settlement relationships
offshore platforms-working stress design: API Recom- for foundations on the basis of the SPT-T. Ciclo De Con-
mended Practice 2A-WSD (RP 2A-WSD), 21st edn. API, ferencias Internationale, Mexico City, Mexico 1:85–104
Washington, DC Decourt L (1982) Predictions of bearing capacity based exclu-
American Society for Testing and Materials (ASTM) (2013) sively on N values of the SPT. In: Proceedings of the 2nd
Standard test methods for deep foundations under static European Symposium on Penetration Testing, Amster-
axial compressive load (D1143/D1143M-07). West Con- dam, 1, 29–34
shohocken, PA: ASTM Decourt L (1999) Behavior of foundations under working load
Anderson TW, Darling DA (1952) Asymptotic theory of cer- conditions. In: Proceedings of the 11th Pan-American
tain goodness-of-fit criteria based on stochastic process. Conference on Soil Mechanics and Geotechnical Engi-
Ann Math Stat 23(2):193–212 neering, Foz DoIguassu, Brazil, August, 4, 453–488
Aoki N, Velloso DA (1975) An approximate method to esti-
mate the bearing capacity of piles. In: Proceedings of

Vol.: (0123456789)
13
Geotech Geol Eng

Dithinde M, Phoon K-K, De Wet M, Retief JV (2011) Char- Misra A, Roberts LA (2009) Service limit state resistance fac-
acterization of model uncertainty in the static pile design tors for drilled shafts. Geotechnique 59(1):53–61
formula. J Geotech Geoenviron Eng ASCE 137(1):70–85 National Research Council Canada (NRCC) (2015) National
Duncan JM (2000) Factors of safety and reliability in geo- Building Code of Canada. Ottawa, ON: NRCC​
technical engineering. J Geotech Geoenviron Eng ASCE Nelsen RB (2007) An introduction to copulas (2nd ed.). New
126(4):307–316 York, NY: Springer. Retrieved from, https://​ebook​centr​
Fellenius BH (1980) The analysis of results from routine pile al-​proqu​est-​com.​ezpro​xy.​lib.​ryers​on.​ca/​lib/​ryers​on/​reader.​
load tests. Ground Eng 13(6):19–31 action?​docID=​371373#
Fellenius BH (2019) Basics of foundation design. Pile Buck Nordlund RL (1963) Bearing capacity of piles in cohesionless
International Inc, Vero Beach, FL soils. J Soil Mech Found Div ASCE 89(3):1–36
Fenton GA, Naghibi F, Griffiths DV (2016) On a unified the- Nordlund RL (1979) Point bearing and shaft friction of piles
ory for the reliability-based geotechnical design. Comput in sand. In: Missouri-Rolla 5­ th Annual short course on the
Geotech 78:110–122 fundamental of deep foundation design, St. Louis, MI.
Fuller FM, Hoy HE (1970) Pile load tests including quick load Paikowsky SG (2004) Load and resistance factor design
test method, conventional methods, and interpretations (LRFD) for deep foundations, vol 507. Transportation
(Record 333). Highway Research Board, Washington, DC Research Board, Washington, DC
Hannigan PJ, Rausche F, Likins GE, Robinson BR, Becker ML Phoon K-K, Kulhawy FH (2008) Serviceability limit state
(2016) Geotechnical Engineering Circular No. 12 – Vol- reliability-based design. In: Phoon K-K (ed) Reliability-
ume I: Design and construction of driven pile foundations based design in geotechnical engineering: computations
(Publication No. FHWA-NHI-16-009). Washington, DC: and applications. Taylor & Francis, New York, NY, pp
Federal Highway Administration (FHWA). 344–384
Hansen JB (1963) Discussion on hyperbolic stress-strain Phoon K-K, Kulhawy FH, Grigoriu MD (2003) Development
response of cohesive soils. J Soil Mech Found Eng of a reliability-based design framework for transmission
89(4):241–242 line structure foundations. J Geotech Geoenviron Eng
Hirany A, Kulhawy FH (1989) Interpretation of load tests on ASCE 129(9):798–806
drilled shafts, part 1: axial compression. In: Kulhawy FH Reddy SC, Stuedlein AW (2017) Serviceability limit state
(ed) Foundation engineering: current principles and prac- reliability-based design of augered cast-in-place piles in
tices, vol 2. ASCE, New York, NY, pp 1132–1149 granular soils. Can Geotech J 54(12):1704–1715
Howell DC (2002) Statistical methods for psychology, 5th edn. Sadegh M, Ragno E, AghaKouchak A (2017) Multivariate
Duxbury, Pacific Grove, CA copula analysis toolbox (MvCAT): describing dependence
Huffman JC, Strahler AW, Stuedlein AW (2015) Reliability- and underlying uncertainty using a Bayesian framework.
based serviceability limit state design for immediate set- Water Resour Res 53:5166–5183. https://​doi.​org/​10.​1002/​
tlement of spread footings on clay. Soils Found 55(4):798– 2016W​R0202​42
812. https://​doi.​org/​10.​1016/j.​sandf.​2015.​06.​012 Salgado S, Woo SI, Kim D (2011) Development of load and
Huffman JC, Martin JP, Stuedlein AW (2016) Calibration and resistance factor design for ultimate and serviceability
assessment of reliability-based serviceability limit state limit states of transportation structure foundations (Report
procedures for foundation engineering. Georisk Assess FHWA/IN/JTRP-2011/03). Indianapolis, IN: Indiana
Manag Risk Eng Syst Geohazards 10(4):280–293. https://​ Department of Transportation
doi.​org/​10.​1080/​17499​518.​2016.​11837​97 Semple RM, Rigden WJ (1986) Shaft capacity of driven pipe
Jesswein, M., & Liu, J. (2021). A new SPT-based method for piles in clay. Ground Engineering, 11–19
estimating axial capacity of driven piles in glacial depos- Shariatmadari N, Eslami A, Karimpour-Fard M (2008) Bearing
its. Geotechnical and Geological Engineering, 40, 1043– capacity of driven piles in sands from SPT-applied to 60
1060.https://​doi.​org/​10.​1007/​s10706-​021-​01941-6  case histories. Iranian J Sci Tech 32(B2):125–140
Legget RF (ed) (1965) Soils in Canada: Geological, Pedologi- Shioi Y, Fukui J (1982) Application of N-value to design
cal, and Engineering Studies (Special Publications No. 3). foundations in Japan. In: Proceedings of the 2­nd Euro-
The Royal Society of Canada, Toronto, ON pean Symposium on Penetration Testing, Amsterdam, 1,
Martin RE, Seli JJ, Powell GW, Bertoulin M (1987) Concrete 159–164.
pile design in Tidewater, Virginia. J Geotech Eng ASCE Sivrikaya O, Toğrol E (2006) Determination of undrained
113(6):568–585 shear strength of fine-grained soils by means of SPT and
Mathworks (2017) Matlab [Computer software]. Natick, MA: its applications in Turkey. Eng Geol 86(1):52–69
Matworks Tang C, Phoon K-K (2018a) Statistics of model factors in reli-
Meyerhof GG (1956) Penetration tests and bearing capac- ability-based design of axially loaded driven piles in sand.
ity of cohesionless soils. J Soil Mech Found Div ASCE Can Geotech J 55:1592–1610
82(1):1–19 Tang C, Phoon K-K (2018b) Evaluation of model uncertainties
Meyerhof GG (1976) Bearing capacity and settlement of pile in reliability-based design of steel H-piles in axial com-
foundations. J Geotech Eng Div ASCE 102(3):197–228 pression. Can Geotech J 55:1513–1532
Milligan V (1976) Geotechnical aspects of glacial tills. In R.F. Tang C, Phoon K-K (2018c) Statistics of model factors and
Legget (Ed.), Glacial Till: An Inter-Disciplinary Study consideration in reliability-based design of axially loaded
(Special Publication No. 12) (pp. 269–291). Ottawa, ON: helical piles. J Geotech Geoenviron Eng ASCE. https://​
The Royal Society of Canada doi.​org/​10.​1061/​(ASCE)​GT.​1943-​5606.​00018​94

Vol:. (1234567890)
13
Geotech Geol Eng

Tang C, Phoon KK, Chen YJ (2019) Statistical analysis of Van Der Veen C (1953) The bearing capacity of a pile. In: Pro-
model factors in reliability-based limit-state design of ceedings of 3rd ICSMFE. Zurich, Switzerland
drilled shafts under axial loading. J Geotech Geoenvi- Wolff TF (1989) Pile capacity prediction using parameter func-
ron Eng. https://​doi.​org/​10.​1061/​(ASCE)​GT.​1943-​5606.​ tions. In: Predicted and observed axial behavior of piles:
00020​87 results of a pile prediction symposium (ASCE Geotechni-
Thorburn S, MacVicar SL (1971) Pile load tests to failure in cal Special Publication No. 23), 96–106
the Clyde alluvium. Behav Piles Lond UK Inst Civil Eng Wu XZ, Xin J-X (2019) Probabilistic analysis of site-specific
1(7):53–54 load-displacement behavior of cement-fly ash-gravel
Tomlinson MJ, Woodward J (2008) Pile design and construc- piles. Soils Found 59(5):1613–1630. https://​doi.​org/​10.​
tion practice, 5th edn. Taylor and Francis, New York, NY 1016/j.​sandf.​2019.​07.​003
Tomlinson, M. J. (1957). The adhesion of piles driven in clay
soils. In: Proceedings of the 4th International Conference Publisher’s Note  Springer Nature remains neutral with regard
of Soil Mechanics, 2, 66–71 to jurisdictional claims in published maps and institutional
Uzielli M, Mayne PW (2011) Serviceability limit state CPT- affiliations.
based design for vertically loaded shallow footings on
sand. Geomech Geoeng 6(2):91–97

Vol.: (0123456789)
13

You might also like