Article

Development of unified p–y curve model for clays using

finite element analysis of laterally loaded piles
Ahmad Souria , Mohsen Amirmojahedi b a
, and George Z. Voyiadjisb
Can. Geotech. J. Downloaded from cdnsciencepub.com by Indian Institute of Technology Madras (IITM) on 10/27/24

, Murad Abu-Farsakh
a
Louisiana Transportation Research Center, Louisiana State University, Baton Rouge, LA 70808; b Department of Civil and
Environmental Engineering, Louisiana State University, Baton Rouge, LA 70803

Corresponding author: Murad Abu-Farsakh (email: [email protected])

Abstract
In this study, a unified p–y curve model is developed for clays in undrained conditions using the results of 3D finite element
(FE) modeling. The p–y curves are primarily used in the analysis of laterally loaded piles. Formulas for the ultimate lateral
bearing capacity factor (Np ), and the reference deflection in the p–y curve (y50 ), were obtained using the results of parametric
studies and regression analysis. The tangent hyperbolic function was used to model the p–y curve shape. In the FE models,
the clay soil material was modeled as elastic-perfectly plastic material using Mohr–Coulomb criteria, and the pile material was
modeled as elastic only. The parametric study results show that Np varies nonlinearly with depth. The proposed model for Np is
composed of two regions: a nonlinear zone and a linear zone. The model for Np used the undrained shear strength (su ), effective
unit weight of soil (γ  ), and pile width (D) as parameters. Further, the proposed model for y50 was found to be dependent on
For personal use only.

the soil stiffness (Es ), D, and Np . The dependency of y50 on Np was often overlooked in previous studies. Finally, the proposed
model was imported in LPILE program and the results from previous case studies were compared with the proposed model
predictions.
Key words: p–y curve, lateral resistance, single pile, static loading, finite element, parametric study

1. Introduction where Ep and Ip are the pile’s elastic modulus and second mo-
The p–y curve is a subgrade reaction technique that de- ment of area, respectively. The value of Pz is referred to as the
scribes the nonlinear relationship between the mobilized axial force and z is the depth.
soil resistance, p, and the lateral deflection of the pile, y, The differential equation of the pile–soil system can be
in the behavior of laterally loaded piles. The p–y curves solved analytically if Epy is assumed to be constant with
were introduced as an improvement to the Winkler method, known variation with depth as in the solutions presented
which assumes constant spring stiffness. In the laterally by Hetenyi (1946) and Reese and Matlock (1956). However,
loaded pile problem, the pile is idealized as a beam sup- the major weakness in these solutions is the elastic soil as-
ported by a series of springs over its length. The reaction sumption, which ignores the nonlinearity of soil reaction as
from each soil spring represents the net soil resistance devel- revealed in later experiments. The p–y curves concept, first
oped around the pile that is working against the movement introduced by McClelland and Focht (1958), addressed the
direction. nonlinearity of the soil behavior. In the p–y curves, the mod-
Following the beam on elastic foundation representation ulus of subgrade reaction Epy varies as the pile deflection
by Winkler (1867), the stiffness of the springs known as the progresses. It starts with the highest value and gradually ap-
modulus of subgrade reaction is defined as proaches to a zero value at the ultimate soil resistance value.
Reese et al. (1974) argued that although Epy is related to the
p soil stiffness (Es ), its variation is a result of the soil–structure
(1) Epy = interaction process.
y
The p–y curve function is controlled by three main ele-
ments: initial slope (Epy ), ultimate soil resistance (pu ), and the
where p is the force per unit length of the pile and y is the characteristic curve shape. The initial slope is related to the
pile deflection. The pile–soil system behavior is governed by coefficient of horizontal subgrade modulus and has different
the following differential equation: definitions for cohesive and cohesionless soils. The ultimate
soil resistance represents the maximum resistance that can
d4 y d2 y be mobilized in the p–y curve, which is controlled by two fail-
(2) Ep Ip 4
+ Pz 2 + Epy y = 0
dz dz ure modes: wedge failure and flow-around failure (Reese and

Can. Geotech. J. 60: 1055–1072 (2023) | dx.doi.org/10.1139/cgj-2021-0014 1055

 Canadian Science Publishing

Van Impe 2010). For the curve shape, various shapes were 2. p–y curve model development for
used in the literature, such as hyperbolic and power func-
tions. Different forms of p–y curves were developed for cohe-
clays under undrained conditions
sive soils by different researchers (e.g., Matlock 1970; Reese As discussed earlier, there are three elements to construct
and Welch 1975; Reese et al. 1975; Sullivan 1977; Dunnavant the p–y curve: the ultimate lateral soil resistance, initial slope,
and O’Neill 1989; Wu et al. 1998; Georgiadis and Georgiadis and characteristic shape function. In clay soils, the ultimate
2010; Hong et al. 2017; Jeanjean et al. 2017; Zhang and An- lateral soil resistance is a function of the undrained shear
dersen 2019). A summary of the p–y curve models from these strength (su ), pile width (D), and the lateral bearing capacity
studies is shown in Table 1. factor (Np ). The initial slope is controlled by the parameter y50,
More recently, a new framework for calculation of p–y which is a function of the soil strain level corresponding to
Can. Geotech. J. Downloaded from cdnsciencepub.com by Indian Institute of Technology Madras (IITM) on 10/27/24

curves in clays was developed (Jeanjean et al. 2017; Zhang and 50% of su in the triaxial test (ε50 ), D, and Np as will be shown
Andersen 2017; Zhang and Andersen 2019). Zhang and An- later. A power function inside the hyperbolic tangent func-
dersen (2017) developed a procedure to obtain the p–y curve tion is used in the study to describe the characteristic shape
from the direct simple shear (DSS) test results, in which nor- of the p–y curve as in eq. 3:
malized p–y curves were “scaled” from the DSS stress–strain
   
curve using curve fitting and correlations between the lateral y n
displacement and the shear strain from the DSS test results. (3) p = Apu tanh B
y50
Jeanjean et al. (2017) proposed an alternate scaling procedure
using correlations obtained from finite element simulations where A, B, and n are model constants, and pu is the ultimate
for DSS tests and statistical analyses. They used the tangent soil resistance.
hyperbolic function to model the p–y curve shape. Zhang and
Andersen (2019) proposed an improved model based on their
2.1 Ultimate lateral soil resistance
previous work in Zhang and Andersen (2017). Their model
The ultimate lateral soil resistance (pu ) is related to the soil
used two distinct p–y curve models: one for the wedge failure
strength and pile width as follows:
zone and one for the flow-around failure zone. They also in-
cluded the later soil resistance contribution at the pile tip. In
(4) pu = Np su D
For personal use only.

the wedge failure zone, they used a power function for the p–y
curve with y50 as the parameter controlling the curve slope. In
the flow-around failure zone, they used the p–y curve model The bearing capacity factor Np varies with depth due to the
from the scaling procedure previously developed in Zhang two failure mechanisms in the soil: wedge failure and flow-
and Andersen (2017). These p–y curve models based on the around failure. The upper-bound values for Np (or Npu ) for the
new framework are advantageous compared with the early flow-around failure mechanism were estimated in the ana-
p–y curve models since it allows the use of DSS test results to lytical solution by Randolph and Houlsby (1984) and later re-
calculate the p–y curve, is applicable for all clays (soft to hard), vised by Martin and Randolph (2006). Npu values falls in the
and incorporates the effect of pile roughness coefficient (α) range of 9.2–11.94 and are dependent on the pile adhesion
and soil gap. The drawbacks, however, are that they require factor (α), which can have values in the range of 0 ≤ α ≤ 1.0.
more effort for p–y curve calculations, and they require hav- Np varies with depth such that it increases from the ground
ing DSS test results, which may not be available. surface up to the critical depth after which it reaches the
A limitation in the early studies from 1960s, 1970s, and upper-bound limit Npu . The actual variation of Np is debatable
1980s is that the p–y curve models were based on limited as different models in the literature exist for soft or stiff clay
number of pile load tests. Additionally, field load tests can and above or below water table. In the following section, the
mobilize the ultimate soil resistance only within limited finite element modeling was used to study the factors affect-
depth near the ground surface. This prevents mobilization of ing Np variation with depth, and then a generalized model
the ultimate lateral soil resistance over the whole pile length, for Np in clays in undrained conditions was developed.
and, in consequence, the bearing capacity factor (Np ) would
not be calculated at the ultimate lateral soil resistance. In this
2.1.1 Finite element model for studying the
study, the 3D finite element (FE) modeling was used to sim-
bearing capacity factor
ulate the laterally loaded pile and mobilize the ultimate soil
resistance over the whole pile length. A parametric study us- To estimate Np , the results of pu are needed first, which
ing multiple FE simulations was performed to evaluate the in- will be normalized to obtain Np using the su and D. Figure
fluence of several parameters on the ultimate soil resistance 1 shows the finite element model used for studying Np . The
and the initial slope in the p–y curve. Using the FE results, a FE model consists of a cylindrical pile surrounded by a soil
new model for Np variation with depth in clay is developed. domain. Only half of the pile and soil were modeled due to
Also, a model is proposed for the reference deflection (y50 ), symmetry. The pile and soil were modeled using solid contin-
which controls the initial slope in the p–y curve. Finally, the uum elements (C3D8R) and the FE mesh density was varied
proposed p–y curve model is imported into the LPILE software to have increased mesh density near the pile. The undrained
and the model predictions are compared with the experimen- clay constitutive behavior was modeled as linear elastoplastic
tal results and existing p–y curve models’ predictions from material with Mohr–Coulomb yield criteria (c = su , φ = 0) and
two previous case studies. Poisson’s ratio, ν = 0.49. The total depth of the soil domain

1056 Can. Geotech. J. 60: 1055–1072 (2023) | dx.doi.org/10.1139/cgj-2021-0014

 Can. Geotech. J. 60: 1055–1072 (2023) | dx.doi.org/10.1139/cgj-2021-0014

Table 1. Summary of p–y curve models for clay.

Can. Geotech. J. Downloaded from cdnsciencepub.com by Indian Institute of Technology Madras (IITM) on 10/27/24

Reference Soil type Ultimate soil resistance (pu ) Initial slope Characteristic shape function
 1
3
Matlock (1970) Soft clay below pu = Np su D – p = 0.5pu y y ≤ pu
 50
water Np = 3.0 + γsuz + J Dz ≤ 9.0 y50 = 2.5ε 50 D
J is constant = 0.5
 1
4
Reese and Stiff clay above pu = Np su D – p = 0.5pu y y ≤ pu
50
Welch (1975) water γ z
Np = 3.0 + su + J Dz ≤ 9.0 y50 = 2.5ε 50 D
 1
2
Reese et al. Stiff clay below pu = Np su D Ei = k s z p = 0.5pu y y ≤ pu
50
(1975) water γ z
Np = 2.0 + su + 2.83 Dz ≤ 11 y50 = ε 50 D
 1
3
p = 0.5pu y y , 0 < y < 8y50
50

Sullivan (1977) Unified clay pu = Np su D Emax = kx pR

= 1, for z > 12D, y > 8y50
⎧  pu
⎪
⎪ 2.0 + γsuz + 0.833 Dz 0 < z < 3D
⎪ pR
= F + (1 − F ) 12D ,
z
⎨ pu for z ≤ 12D, y > 8y50
Np = 3.0 + 0.5 Dz 3D < z < 12D y50 = Aε 50 D
⎪
⎪
⎪
⎩
9.0 z > 12D
  0.7 
pu = 1.02pu tanh 0.537 y y
For personal use only.

Dunnavant and Stiff clay below pu = Np su D –

50
O’Neill (1989) water γ z
Np = 2.0 + su + 0.4 Dz ≤ 9.0 y50 = 0.0063ε50 DKR−0.875
Ep Ip
KR = Es L4p
p y/y50
Wu et al. (1998) Unified clay pu = Np su D – = β 
pu + β −2 y/y50
Np(NC) = 2.5 + 1.5 Dz ≤ 10.0 β  −1 β  −1

Np(OC) = 0.77Np(NC) y50 = Np ε 50 D, β  = 8 − 12

 4
 1
E50 D 12 y
Georgiadis and Unified clay pu = Np su D   Ki = 3E50 Ep Ip p= 1 y
Ki + pu
Georgiadis −λ(z/D)
Np = Npu − Npu − Npo cos θ exp 1+tan θ
(2010)
Npo = 2 + 1.5α, λ = 0.55 − 0.15α
   
(y/D)p 0.5
tanh A (y/D)
p pf
Jeanjean et al. Unified clay pu = Np su D – pu = tanh(A)
γz
(2017) Np = Np0 + s +s ≤ Npd - with soil gap A = 1.33 + 0.45α
u0 u1 z
Np = 2Np0 ≤ Npd - no soil gap (y/D)pf = γpf [2.5 − 1.2ln (α)]
 1.35
z 0.6
Np0 = N1 − (1 − α) − (N1 − N2 ) 1 − dD
N1 = 12; N2 = 3.22
Npd = 9 + 3α
  n 
yp
−

Canadian Science Publishing

p
pu = (Np0 su + γ  z)D - for wedge failure
p p y50
Zhang and Unified clay y50 = ye50 + y50 pu=1−2 - for wedge failure
Andersen (2019) pu = Npd su D - for flow-around failure; y50/D (Gmax /su ) = 
e
n = 0.55 − 0.05 (z/D) ≥ 0.325
  0.5 Np0 / kpy /Gmax p τ
pu = su (DSS scaling) - for flow-around failure
z 0.6 1.35
Np0 = N1 − (1 − α) − (N1 − N2 ) 1 − 14.5D p   p
y50 z
N1 = 11.94; N2 = 3.22 D = a + b D γf
Npd = 9.14 + 2.8α
Note: Np is ultimate lateral bearing capacity factor, su is undrained shear strength, γ  is effective unit weight, z is depth from mudline, D is pile width, y is lateral pile displacement, ε50 is strain corresponding to 50% of
su , E50 is soil stiffness corresponding to ε50 , k is modulus of horizontal subgrade reaction, θ is the angle of slope in front of pile, Ep Ip is pile flexural rigidity, Lp is pile length, and α is pile roughness coefficient.
1057
 Canadian Science Publishing

Fig. 1. Geometry and mesh of the FE model used to study the bearing capacity factor Np .
Can. Geotech. J. Downloaded from cdnsciencepub.com by Indian Institute of Technology Madras (IITM) on 10/27/24

was 15D, which was found sufficient for studying Np variation 0, while for rough piles (α = 1), the tangential stress limit is
with depth. τ i_max = su .
The soil side boundary was placed far from the pile (at The simulation was performed in two steps: (i) applying the
40D) to eliminate the effects from boundary conditions. The geostatic stress in the soil domain and (ii) pushing the whole
side boundary was only constrained in the x- and y-directions, pile laterally a distance of 0.15D, which was found sufficient
For personal use only.

while the bottom boundary was constrained in the z-direction to mobilize the ultimate soil resistance.
only. For the symmetry plane, only the y-direction was con-
strained.
2.1.2 Estimation of soil resistance and bearing
The pile length was varied in this study and its length was
capacity factor
chosen to extend above and below the soil boundaries by
1D (Fig. 1). This is because the loading scheme was to push To estimate Np with depth, the soil resistance profile was
the whole pile laterally so that the ultimate soil resistance obtained first. The soil resistance (p) represents the net soil
is mobilized over the whole depth. This approach is more resistance acting against the lateral pile movement at certain
time efficient compared with loading the pile at the top only depth (i.e., the net normal stress and surface friction around
and allows estimating the ultimate soil resistance within the the pile perimeter). The common approach to obtain the soil
whole depth in a single run. The pile material constitutive resistance profile is by differentiating the bending moment
behavior was modeled as linear elastic with concrete  prop- (BM) or the shear force functions. The soil resistance (p) is
erties: fc = 38 MPa, Young’s modulus (Ep = 57 000 fc ), and related to BM and shear force as follows:
Poisson’s ratio (ν = 0.2).
Figure 2 illustrates the pile loading scheme as well as the
d2 M (z) dV (z)
development of two failure mechanisms (wedge and flow- (5) −p (z) = 2
=
dz dz
around) within the soil. The wedge failure mechanism is char-
acterized by soil yielding in front of pile and formation of a
gap on the backside, whereas the flow-around failure is char- where M (z) is the BM function, V (z) is the shear force func-
acterized by soil yielding around the pile with no gap. tion, z is the depth.
The effect of pile adhesion is simulated using interface The BM or shear force measurements are obtained at
model, which was incorporated using the contact modeling several points over the pile length, and then a high order
approach in ABAQUS. The interface has zero thickness and polynomial or cubic spline fit is used to obtain a differen-
transfers normal and tangential stresses between the pile tiable function. Georgiadis et al. (2011) obtained the soil
and soil and allows controlling the pile adhesion factor (α). resistance profile by differentiating the shear force profile,
The interface normal stress is governed by the “Hard con- while Nip and Ng (2005) obtained the resistance profile
tact” rule, which (i) transfers the normal stress (in bearing by double differentiating the BM profile. Haiderali and
mode) only when the surfaces are in contact, (ii) does not Mudabhushi (2016) studied and compared the accuracy of
transfer tensile normal stress, and (iii) allows pile–soil sepa- high-order polynomials and cubic splines for constructing
ration (i.e., gap). The tangential stress is transferred when the the BM profile from instrumented piles and then obtaining
surfaces are in contact, and is limited by the maximum shear the soil resistance profiles by differentiation. They concluded
stress value, which is controlled by α and su (τ i_max = αsu ). For that cubic splines are more accurate than polynomial fits.
smooth piles (α = 0), the tangential stress limit is τ i_max = Therefore, in this study cubic spline fits were used to create

1058 Can. Geotech. J. 60: 1055–1072 (2023) | dx.doi.org/10.1139/cgj-2021-0014

 Canadian Science Publishing

Fig. 2. Illustration of the loading scheme used in the FE model and soil plastic zones around the pile.
Can. Geotech. J. Downloaded from cdnsciencepub.com by Indian Institute of Technology Madras (IITM) on 10/27/24

the shear force profile and then perform differentiation to stress) in stiffer clays. The observation is in agreement with
obtain the soil resistance profile. Np models proposed by Matlock (1970), Reese et al. (1975),
In the FE model, the shear force was obtained at equidis- and Sullivan (1977). Near the ground surface, the results show
For personal use only.

tant points (1D spacing) for a total of 15 points starting from that Np values were around 2.0, which is similar to Reese et
the ground surface. Then, the shear force profile was ob- al. value but lower than Matlock’s model value at 3.0. Assess-
tained using a cubic spline fit. The soil resistance profile (p) ing the value of Np near the ground surface is difficult in real
was obtained from the slope (first derivative) of the cubic experiments due to the singularity at this location. In this
spline function. Finally, the bearing capacity factor (Np ) was study, pile adhesion was considered as an input, and the ef-
estimated by normalizing the soil resistance profile using the fect of singularity was minimized by adding artificial points
undrained shear strength and pile width (Np = supD ), as shown to the shear force profile above the ground surface with con-
in Fig. 3. The soil resistance profile in the figure shows that stant force magnitude. The artificial points restrain the cubic
Np is nonlinearly increasing and reaches a maximum value at spline fit for the shear force curve. The value of Np at the
the depth where the flow-around failure mechanism governs. ground surface was between 1.8 and 2.2.

2.1.3 Parametric study for Np variation with depth Effect of soil effective unit weight
The effect of several factors on the ultimate soil resistance The results for the effect of γ  on Np variation are shown
and the variation of Np with depth was investigated in the in Fig. 4b. The rate of Np increase with depth decreases when
form of parametric study. For undrained clay conditions, the γ  decreases. This is because the overburden stress is directly
factors known to influence Np are su , D, α, and the effective related to γ  . Also, similar to su effect, the depth at which
unit weight of soil (γ  ) (see Table 1). In this parametric study, Npu is reached becomes deeper with smaller γ  . However, the
su was varied from 8.14 to 95.76 kPa, γ  was varied from 7.85 impact of γ  is less than su , which is due to the fact that γ 
to 18.85 kN/m3 , D was varied from 0.61 to 1.22 m, and α was values are bounded in nature (11–21 kN/m3 ) and it is linearly
varied from 0 to 1.0. The undrained soil behavior was en- related to Np as will be shown later.
forced by using a Poisson’s ratio of ν = 0.49.
Effect of pile adhesion
Effect of undrained shear strength The effect of α on Np variation is depicted in Fig. 5. The re-
The results for Np variation at different su are shown in sults are presented for three pile cases with α = 0.0 (smooth),
Fig. 4a. The selected values for su cover the range from soft 0.5 (intermediate), and 1.0 (rough) at different γ  (18.85, 14.14,
(<11.97 kPa) to very stiff (>95.76 kPa) clays. It can be seen that and 7.85 kN/m3 ). The effect of pile adhesion is known to in-
su has strong influence on Np . The results show that the rate crease the Np value, which is due to the additional resistance
of Np increase with depth decreased when su increased, and from the pile skin friction. Following the analytical solution
consequently the depth for the Np upper-bound limit (Npu ) be- by Randolph and Houlsby (1984) and Martin and Randolph
came deeper. This indicates that the conditions for the flow- (2006), the Npu values in the flow-around failure mechanism
around failure require greater confinement (or overburden at α = 0.0, 0.5, and 1.0 are 9.2, 10.8, and 11.94, respectively.

Can. Geotech. J. 60: 1055–1072 (2023) | dx.doi.org/10.1139/cgj-2021-0014 1059

 Canadian Science Publishing

Fig. 3. Example showing the steps to obtain the soil resistance profile and the bearing capacity factor Np .
Can. Geotech. J. Downloaded from cdnsciencepub.com by Indian Institute of Technology Madras (IITM) on 10/27/24

Fig. 4. Effect of su (a) and γ  (b) on Np variation.

For personal use only.

Fig. 5. Effect of pile adhesion (α) on Np variation.

1060 Can. Geotech. J. 60: 1055–1072 (2023) | dx.doi.org/10.1139/cgj-2021-0014

 Canadian Science Publishing

Fig. 6. Effect of pile width (D) on Np variation.

Can. Geotech. J. Downloaded from cdnsciencepub.com by Indian Institute of Technology Madras (IITM) on 10/27/24

Fig. 7. General characteristics of Np variation used for model 2.1.4 Model development for Np variation with
development. depth
Using the results of parametric study, a unified model for
Np variation with depth is developed. There are several mod-
els in the literature for predicting Np variation. However,
For personal use only.

these models were developed for specific clay type (soft or

stiff clay) and others for cases with soil above water table
or below water table (e.g., Matlock 1970; Reese et al. 1975;
Sullivan 1977). Other prediction models for Np in clays (Murff
and Hamilton 1993; Georgiadis and Georgiadis 2010) were
not included in the discussion since they do not use su , γ  , D
as the main model parameters. The developed model in this
study is applicable for all clay types (soft or stiff) and different
soil conditions (above or below water).
The characteristics of the proposed Np model were selected
following the observations from the parametric study. The
characteristics of the model can be described as follows: (i)
within the shallow depth range 0 < z < 2D, Np increased non-
linearly with depth starting at a value of 2.0 for smooth piles
(α = 0), (ii) within the depth range 2D < z < Zc , Np increased lin-
It can be noticed in Fig. 5 that Np increases at slightly higher early with depth until it reached Npu , (iii) for nonsmooth piles
rate initially (up to z/D = 2) when α increased and then the (α > 0), Np increased at higher rate within the shallow depth
rate reaches a constant value until the depth of Npu is reached. (0 < z < 2D), and (iv) for depths 2D < z < Zc , Np increased at
Also, Np value at the ground surface increased slightly when the same rate regardless of α value. The depth at which Npu is
α increased (+0.3 at α = 0.5, +0.55 at α = 1.0 compared with reached increased slightly with α due to the higher Npu value.
cases with α = 0.0). Figure 7 summarizes the characteristics of Np considered in
the model development.

Nonlinear zone (0 < z < 2D)

Effect of pile width
In this zone, it is assumed that Np variation follows a power
The effect of pile width (D) is presented in Fig. 6, in which
function starting from Npo at the ground surface as follows:
three pile widths were studied (0.61, 0.91, and 1.22 m). The
results show that a larger pile width results in increasing the z  z n
(6) Np = Npo + k (0 < z < 2D)
rate of Np increase with depth. This is attributed to the fact D D
that larger pile width mobilizes greater soil resistance at the
same relative depth (z/D). At the ground surface, Np values where k and n are model parameters, and z is the depth from
were not affected by the change in the pile width. the ground surface. The model function was initially devel-

Can. Geotech. J. 60: 1055–1072 (2023) | dx.doi.org/10.1139/cgj-2021-0014 1061

 Canadian Science Publishing

Fig. 8. Np slope in the linear zone is directly related to the Fig. 9. Results for critical depth Zc versus γ  D/su .
term γ  D/su .
Can. Geotech. J. Downloaded from cdnsciencepub.com by Indian Institute of Technology Madras (IITM) on 10/27/24

where Np@z/D = 2 is the Np value at depth z/D = 2 obtained from

oped using the results for smooth piles (α = 0), and then ad- eq. 6, and Npu is the upper-bound limit for Np from the solu-
justed for nonsmooth piles (α > 0). Np results for smooth piles tion by Martin and Randolph (2006).
from the parametric study were used to evaluate the parame- The Np value continues to increase with depth until it
ters k and n in eq. 6. The function was fitted to Np results using reaches the upper-bound limit Npu at the critical depth (Zc ).
an iterative scheme, and k and n were found to correlate with The results for Zc (only for smooth piles, α = 0) were found to
the ratio γ  D/su . A linear regression was used to obtain a re- correlate with γ  D/su as seen in Fig. 9, and were fitted by the
lation for k and n as a function of γ  D/su , and the regression following equation:
results are as follows: Zc 5.36
(12) = 

γ D D γ D/su
For personal use only.

(7) k = 0.82 + 1.95

su It is not necessary to estimate the critical depth (Zc ) to es-
γ D tablish the Np variation with depth. This is because Zc is indi-
(8) n = 0.13 + 0.6 rectly determined by enforcing the upper-bound limit Npu . It
su
should be noted here that the best-fit equation for Zc is based
Eq. 7 and eq. 8 make the Np model in the shallow zone on Np results for homogeneous soil properties and, therefore,
dependent on γ  D/su , which is consistent with the previous is not applicable for nonhomogeneous layered soils. Equation
observations from the FE results and the subsequent part of 12 was found to be closely similar to the critical depth for-
the model in the linear zone. mula proposed by Matlock (1970) (i.e., Zc = γ  D/s
6D
u +J
), in which
For nonsmooth piles (α > 0), the values of Npo and pa- the differences were in the constant (6.0 vs. 5.36) and the in-
rameter k increased when α increased, while parameter clusion of the empirical factor “J”.
n remained fairly unchanged. The adjustment for Npo as a
function of α was obtained using linear regression as follows:
2.1.5 Comparison with Np models from the
(9) Npo = 2.0 + 0.6α literature
Similarly, the adjustment for k as a function of α was ob- The developed Np model is compared with other models
tained using an exponential function as follows: from literature presented in Table 1. The Np variation is eval-
uated using each model for several cases by varying su , γ  ,
(10) k(α) = k(0) + (1 − exp [−0.71α]) and D. Figure 10 presents the comparison for D = 0.91 m,
γ  = 18.85 kN/m3 , and three su values of 11.97 (soft clay),
where k(0) is the value of parameter k at α = 0.0 using eq. 7.
23.94 (medium stiff clay), and 47.88 kPa (stiff clay). Matlock
model and Sullivan model produced fairly similar Np varia-
Linear zone (2D < z < Zc ) tions, while Dunnavant and O’Neill model had greater Zc /D
In this zone, the results of parametric study showed that due to the smaller constant in the third term (0.4 z/D). The
Np increased linearly with depth (Fig. 7). The slope of Np vari- results for Reese et al. model show a slight change in Np vari-
ation in the linear zone for smooth and nonsmooth piles was ation with different su , which indicates that the model is less
obtained and plotted against γ  D/su . The results were found sensitive to su . The issue in Reese et al. model is that the con-
to closely align along the 1:1 line leading to the conclusion stant in the third term (2.83 z/D) is relatively large, which
that the slope equals to the term γ  D/su (Fig. 8). dominates the Np slope value compared with the second term
Following this observation, the model proposed for the lin- (γ  D/su ).
ear zone (2D < z < Zc ) is For the proposed model, Np results are higher in the shal-
z γ D  z  low zone (0 < z/D < 2) when compared with Matlock and
(11) Np = Np @z/D=2 + − 2 ≤ Npu (z > 2D) Sullivan models (+1 at z/D = 2), and notably higher than
D su D

1062 Can. Geotech. J. 60: 1055–1072 (2023) | dx.doi.org/10.1139/cgj-2021-0014

 Canadian Science Publishing

Fig. 10. Comparison of Np results for the proposed model and literature models at different su values.
Can. Geotech. J. Downloaded from cdnsciencepub.com by Indian Institute of Technology Madras (IITM) on 10/27/24
For personal use only.

Dunnavant and O’Neill model (+1.8 at z/D = 2). The critical Dunnavant and O’Neill models. All models predicted greater
depth in soft clay for the proposed model is similar to Matlock Zc /D when the pile width decreased. The results from the
and Sullivan models (z/D = 3.3). For medium stiff clay, Zc /D for proposed model are closely similar to Matlock and Sullivan
the proposed model is at z/D = 6.3, which is similar to Dun- models at D = 0.91 m only, while it is closer to the result of
navant and O’Neill model. For stiff clay, Zc is notably greater Dunnavant and O’Neill model at D = 0.61 m. For D = 0.31 m,
in the proposed model (z/D = 12) compared with Dunnavant the proposed model has the largest Zc /D at 9.3, which was +2
and O’Neill model (z/D = 9). One should recall that Matlock higher than Matlock and Sullivan models and +1.3 higher
and Sullivan models were developed for soft clay, while Dun- than Dunnavant and O’Neill model. Reese et al. model pre-
navant’s model was developed for stiff clay. dictions are not sensitive to the change in pile width, which
The main difference between the proposed model and the is again due to the large constant “R” in the third term.
other models is that the Np slope is controlled by single term The models were also compared for different effective soil
(γ  D/su ), whereas in the other models, it is controlled by two unit weight (γ  ), and the results show that the Np values for

terms ( γsuD + R Dz ), where R is a constant. The proposed model all models decrease with lower γ  . The Zc /D result from the
eliminates this empirical part in previous models by being proposed model is similar to Matlock and Sullivan models at
solely dependent on the term γ  D/su . Matlock (1970) reasoned γ  = 18.85 kN/m3 , while it is closer to Dunnavant and O’Neill
that the term with the constant “R” was added to account for model at γ  = 14.14 and 9.43 kN/m3 . The results of Reese et
the “geometrically related restraint against the upward flow al. model were insensitive to γ  , similar to the previous ob-
of soil in weightless soil”. servations for the studied cases for su and D.
The proposed model was then compared with other mod-
els from literature for different pile widths (D = 0.3, 0.61, and
2.1.6 Np variation in layered soils
0.91 m), and the results are shown in Fig. 11. Np variations
from the proposed model are slightly higher in the shallow The new Np model was developed for homogeneous soils;
zone (0 < z/D < 2) as compared with Matlock, Sullivan, and however, in practice, piles are driven in multilayered soil. Np

Can. Geotech. J. 60: 1055–1072 (2023) | dx.doi.org/10.1139/cgj-2021-0014 1063

 Canadian Science Publishing

Fig. 11. Comparison of Np results for the proposed model and literature models at different pile width (D).
Can. Geotech. J. Downloaded from cdnsciencepub.com by Indian Institute of Technology Madras (IITM) on 10/27/24
For personal use only.

models for homogeneous soils still can be used to establish With heq determined, the Np variation in the second layer
the Np variation in layered soil profile using the equivalent is obtained by locating heq in the Np variation for a homoge-
depth method proposed by Georgiadis (1983). In this method, neous soil established using Layer 2 properties.
Np variation in the first layer is established using the same
procedure in homogeneous soils. Then, Np variation in the
second layer is determined after estimating the equivalent 2.1.7 Np variation in soils with water table
depth at the top of this layer. The equivalent depth (heq ) is
determined by first estimating the force (F1 ) needed to induce In addition to soil layering, the presence of water table is
failure at the bottom of the first layer using the first-layer another factor affecting Np variation. The presence of soil lay-
properties ers below water reduces the effective unit weight (γ  ). In the
proposed model, γ  is one of the main parameters control-
 H1  H1 ling Np variation and the presence of water table is readily
(13) F1 = pu1 dz = Np (z) su D 1 dz
0 0
accounted for by using the effective unit weight (γ  = γ sat −
γ w ), where γ sat is the submerged unit weight and γ w is the
where pu1 is the ultimate soil resistance, H1 is the thickness of unit weight of water. There are two scenarios for soils under
Layer 1, Np (z) 1 is the Np variation for homogeneous soil es- water table: the first is when the soil is completely submerged
tablished using Layer 1 properties, and (su )1 is the undrained from the ground surface, and the second is when the water ta-
shear strength for Layer 1. ble exists below the ground surface. For the first scenario, uti-
Then, F1 is used to back-calculate the equivalent depth (heq ) lizing the effective unit weight for the whole depth accounts
at top of the second layer as follows: for the effect of water table. In the second scenario, the wa-
ter table exists at certain depth below the ground surface and
 heq
in this case the effective unit weight changes abruptly at the
(14) F1 = Np (z) su D 2 dz
0 water table elevation. For this case, the soil profile should be

1064 Can. Geotech. J. 60: 1055–1072 (2023) | dx.doi.org/10.1139/cgj-2021-0014

 Canadian Science Publishing

Fig. 12. Geometry and mesh of the FE model used to study 30D to ensure flexible pile behavior. Similar to the study for
y50 . Np , only half of the pile and soil domain were modeled to
reduce the solution time. The pile and the soil were modeled
using solid continuum elements (C3D8R) and the FE mesh
density was varied so that it is denser near the pile. The
pile constitutive behavior was modeled as linear elastic with

 properties: fc = 33.9 MPa, Young’s modulus Ep =
concrete

5000 fc = 29.1 GPa (ACI 2008), and Poisson’s ratio ν = 0.2.
The clay constitutive behavior was modeled as linear elasto-
plastic material with Mohr–Coulomb yield criteria (c = su , φ
Can. Geotech. J. Downloaded from cdnsciencepub.com by Indian Institute of Technology Madras (IITM) on 10/27/24

= 0) and Poisson’s ratio ν = 0.49 to enforce the undrained

behavior. The effect of pile adhesion was accounted for by
using an interface model in a similar way to the study of Np .
The simulation was performed in two steps. In the first
step, the geostatic stress in the soil was established by apply-
ing gravity load in the global z-direction. In the second step,
the pile top was pushed laterally a distance of 0.3D, which
was chosen to fully develop the p–y curves within the upper
divided into two layers and the Np profile is calculated using
15D depth.
the method proposed by Georgiadis (1983) as discussed in the
previous section.
2.2.2 Obtaining p–y curves and y50 from FE results
2.2 Initial slope and y50
Constructing the p–y curve requires obtaining the soil re-
The second element needed toward the development of the
sistance and pile displacement at several points during the
p–y curve is the initial slope, which can be defined using ei-
lateral pile loading. In the FE model, the soil resistance was
ther the coefficient of horizontal subgrade reaction (k) or the
obtained first by evaluating the shear profile and then the
reference deflection y50 in the p–y curve. y50 is a reference de-
For personal use only.

slope of the shear profile, as presented earlier in Fig. 3. The

flection and refers to the pile deflection at 50% of the ultimate
soil resistance and pile deflection at certain depth are com-
soil resistance (pu ) in the p–y curve. y50 is commonly used in
bined to construct the p–y curve. For y50 , the deflection corre-
the p–y curve models for clays (e.g., Matlock 1970; Reese et
sponding to half of the ultimate soil resistance (pu /2) is found
al. 1975; Sullivan 1977; Dunnavant and O’Neill 1989; Wu et
from the constructed p–y curve. Note that pu (= Np su D) was
al. 1998), and can be estimated using the characteristic strain
estimated using the proposed Np model. Figure 13 illustrates
(ε 50 ) and the pile width (D) as follows:
the construction of the p–y curves from the soil resistance
(15) y50 = Aε50 D and pile deflection profiles.
where A is a constant, and ε50 is known as characteristic
strain and is the axial strain corresponding to 50% of su in 2.2.3 Parametric study for y50
the unconsolidated undrained (UU) compression test.
Different values for A were used in the literature: Matlock The factors affecting y50 in the p–y curve studied were (i)
(1970) assumed A = 2.5, Reese et al. (1975) used A = pile’s material stiffness (Ep ), (ii) pile width (D), (iii) undrained
1.0, Stevens and Audibert (1979) suggested A = 8.9D−0.5 , shear strength (su ), and (iv) soil stiffness (Es ). The parameter
and Dunnavant and O’Neil (1989) found that A depends values used in the study for y50 were Ep = 14.6, 21.8, 29.1,
on the pile–soil relative stiffness and proposed A = 36.4, and 43.7 GPa; D = 0.61, 0.91, and 1.22 m; su = 11.97,
  23.94, 35.91, and 47.88 kPa; and Es = 100, 200, 300, 400, and
E I −0.875
0.0063 EspLp4 , where Ep Ip is pile flexural rigidity, Es is soil
500 su . In all cases, the soil properties (su , γ  ) were constant
stiffness, and L is pile length. Wu et al. (1998) argued that A with depth and γ  = 18.85 kN/m3 was used. A smooth pile
is approximately equal to the bearing capacity factor (Np ) for condition (α = 0) was assumed for the interface.
normally consolidated clays.
In the following section, a parametric study using FE anal-
ysis was conducted to develop a model for the reference de- Effect of soil stiffness
flection y50 . The p–y curves were first obtained from the FE The soil stiffness (Es ) was varied in respect to the undrained
simulations for laterally loaded single pile, and then y50 was shear strength, su . The base value for su was 11.97 kPa. For
determined from the p–y curves. Potential factors affecting each case, y50 was obtained from the p–y curves at several
y50 that were investigated are Es , Ep , su , and D. depths z/D = 1, 2, 3, 4, 5, and 6. Figure 14 presents the results
of y50 with depth, which were normalized by D. It is observed
that y50 decreases as the soil stiffness Es increases at a non-
2.2.1 Finite element model for y50
linear rate. Another observation is that y50 varied with depth
The FE model used for studying y50 is composed of a flex- even though Es was constant for the whole soil profile. The
ible pile with circular cross section (Fig. 12). The pile width latter indicates that y50 is dependent on another variable be-
(D) was varied in the analysis and the length was selected as sides Es , which was found to be Np as shown later.

Can. Geotech. J. 60: 1055–1072 (2023) | dx.doi.org/10.1139/cgj-2021-0014 1065

 Canadian Science Publishing

Fig. 13. Construction of the p–y curves from the soil resistance and pile deflection profiles.
Can. Geotech. J. Downloaded from cdnsciencepub.com by Indian Institute of Technology Madras (IITM) on 10/27/24
For personal use only.

Fig. 14. Effect of soil stiffness on y50 .

Effect of pile material stiffness can be considered independent of Ep without introducing sig-
The effect of pile material stiffness (Ep ) was investigated nificant error.
by varying the reference value (Ep-ref = 29.1 GPa) by ±25%
and ±50% as discussed earlier. The reference Ep value was Effect of undrained shear strength
estimated
 for a concrete pile with fc = 33.9 MPa and Ep = The effect of su on y50 was investigated. The su values

5000 fc . The results show that y50 slightly increases when Ep were selected for soft (11.97 kPa), medium stiff (23.94 and
increased (Fig. 15). However, the effect is negligible as noticed 35.91 kPa), and stiff clays (47.88 kPa). Figure 16 demonstrates
in the % change in y50 , which was within ±3%. Therefore, y50 that su has notable effect on y50 , in which y50 decreased with

1066 Can. Geotech. J. 60: 1055–1072 (2023) | dx.doi.org/10.1139/cgj-2021-0014

 Canadian Science Publishing

Fig. 15. Effect of pile material stiffness (Ep ) on y50 .

Can. Geotech. J. Downloaded from cdnsciencepub.com by Indian Institute of Technology Madras (IITM) on 10/27/24

Fig. 16. Effect of undrained shear strength (su ) on y50 .

For personal use only.

higher su . Also, increasing su resulted in a different variation Fig. 17. Effect of pile width (D) on y50 .
for y50 with depth. For example, for su = 11.97 kPa, y50 in-
creased up to depth z/D = 4 and then remained fairly constant
below z/D = 4. On the other hand, for su = 23.94, 35.91, and
47.88 kPa, y50 values kept increasing with depth at different
rates. The figure shows that the variation of y50 is closely sim-
ilar to Np variation for each case. Recall that the ultimate soil
resistance (pu ) reaches its limit value at the critical depth (Zc ),
and for the presented cases, Zc was 4D, 7.4D, 11D, and 14.7D
for su = 11.97, 23.94, 33.52, and 47.88 kPa, respectively. For
su = 11.97 kPa, 4D corresponds to the depth after which y50
remained constant. The latter indicates that y50 is dependent
on Np .

Effect of pile width

The effect of pile width (D) on y50 was studied for three
pile sizes: 0.61, 0.91, and 1.22 m. The results in Fig. 17 show case: at z/D = 3, 4, 6 for D = 1.22, 0.91, 0.61 m, respectively.
that y50 values increased with larger pile width. However, These depth values are closely similar to Zc for Np in each
the variation of y50 with depth was not identical in each case (Zc /D = 2.8, 3.7, 5.5 for D = 1.22, 0.91, 0.61 m, respec-
case and has similarities with Np variation. It is noticed here tively). This is another indication that y50 is directly related
that y50 values stopped increasing at a certain depth in each to Np .

Can. Geotech. J. 60: 1055–1072 (2023) | dx.doi.org/10.1139/cgj-2021-0014 1067

 Canadian Science Publishing

2.2.4 Proposed model for y50 Fig. 18. Shape of proposed p–y curve model.
Using the results of FE parametric study results, a model
for y50 is proposed as follows.

(16) y50 = pu /2Epy

It was found that Epy = Es in the FE model. This, however,

does not represent a realistic clay behavior since it is known
that Es varies with strain. Addressing this issue requires the
use of an advanced constitutive model in the FE simulation,
Can. Geotech. J. Downloaded from cdnsciencepub.com by Indian Institute of Technology Madras (IITM) on 10/27/24

which is not within the scope of this study.

Es is often substituted by ε50 (= 0.5su /Es50 ). Substituting the
definitions of pu (eq. 4) and ε 50 in eq. 16, the proposed model
for y50 becomes

(17) y50 = ε50 Np D

It can be noticed from eq. 17 that y50 is dependent on the

depth (z) through Np . This behavior is neglected in most pre- Fig. 19. Comparing the p–y curves from the proposed model
vious studies. and FE model.
When eq. 17 was used with the proposed p–y curve model
(discussed below) to simulate case studies, it was noticed that
y50 values were relatively high for overconsolidated stiff clays.
This resulted in a softer soil resistance at small deflections
and overprediction for lateral pile deflections. Therefore,
eq. 17 was amended by incorporating the effect of OCR on
For personal use only.

y50 . Wu et al. (1998) similarly proposed a reduction to y50 for

overconsolidated clays by 33%. In this study, the well-known
relation between the undrained stiffness of normally consol-
√
idated (NC) and overconsolidated clays (OC), EOC = ENC OCR
(Bowles 2001), was used to modify y50 as follows:

ε50 Np D
(18) y50 = √
2 OCR

In eq. 18, the constant with the OCR term was needed to 3. Comparison with case studies and
ensure that y50 corresponds to 50% of pu in the proposed p–
existing p–y curve models
y curve model. The influence of this constant is negated by
the B and n parameters used within the tangent hyperbolic The proposed p–y curve model predictions are compared
function as presented in the following section. with two experimental studies and existing p–y curve mod-
els by Matlock (1970) and Reese et al. (1975). The pro-
2.3 Proposed p–y curve model posed p–y curve model was imported into the computer
The tangent hyperbolic function presented in eq. 3 was program LPILE v2019 (Ensoft 2022) through the p–y curve
adopted to model the p–y curve shape. The parameters A, B, user-defined option. LPILE v2019 also has built-in p–y curve
and n were determined by trial and error and optimized so models from Matlock (1970) and Reese et al. (1975). The
that the curve passes through y50 at pu /2, and reaches pu at experimental studies were simulated in LPILE and the re-
10y50 . The proposed p–y curve model is as follows: sults for pile’s load–deflection curve and BM profile are
 compared.
 0.65 
y
(19) p/pu = tanh 0.4 3.1 Pile load tests at Lake Austin, Texas
y50
(Matlock 1970)
The general shape of the p–y curve using eq. 19 is shown in Matlock (1970) and his team performed static lateral load
Fig. 18. tests on 324 mm diameter, 12.8 m steel pipe pile. The pile was
Figure 19 presents a comparison between the p–y curves driven near Lake Austin, Texas, in soft clay with average su of
from the FE model and those obtained using the proposed 38 kPa. The site was considered under water as indicated in
model for example case of a pile in clay with su = 14.36 kPa, the study. The lateral load was applied at the mudline using
γ  = 18.85 kN/m3 , D = 0.91 m, Es = 200Su , and α = 0. The a ram-jack with no pile head restraints (i.e., free-head). The
figure shows good agreement between the p–y curves from pile was instrumented with a series of strain gages to allow
proposed model and those from the FE model. determination of the pile’s deflection profile, BM profile, soil

1068 Can. Geotech. J. 60: 1055–1072 (2023) | dx.doi.org/10.1139/cgj-2021-0014

 Canadian Science Publishing

Fig. 20. Comparison of computed load–deflection results us- 3.2 Pile load tests near Manor, Texas (Reese et
ing the proposed model and Matlock (1970) p–y curve model. al. 1975)
Reese et al. (1975) performed static lateral load tests on
641 mm diameter, 15.2 m long steel pipe pile. The pile was
driven at a site near Manor, Texas. The soil strata consisted
of stiff to very stiff overconsolidated clay with su ranging be-
tween 96 and 290 kPa. The piles were driven in a pit that was
inundated for 4 months prior to piles installation and test-
ing. Similar to Matlock (1970), the lateral load was applied
at the mudline using a ram-jack with no pile head restraints
Can. Geotech. J. Downloaded from cdnsciencepub.com by Indian Institute of Technology Madras (IITM) on 10/27/24

(i.e., free-head). The pile was instrumented with a series of

strain gages to allow determination of the pile’s deflection
profile, BM profile, soil resistance profile, and p–y curves. Ex-
perimental results for load–deflection at pile head and BM
profiles were provided in the reference at four lateral loads:
179.7, 317.7, 485.6, and 606.2 kN.
resistance profile, and p–y curves. Only experimental BM re- The 641 mm pile was modeled in LPILE and the lateral load
sults were provided in the reference and at three lateral loads: tests were simulated using the built-in p–y curve model for
32.9, 68.9, and 109.4 kN. stiff clays with free water by Reese et al. (1975) and the pro-
The 324 mm pile was modeled in LPILE and the lateral load posed model through the user-defined option. In the simula-
tests were simulated using the built-in p–y curve model for tion, the soil profile was divided into two layers: Layer 1 ex-
soft clays (Matlock 1970) and the proposed model through tending from the ground surface to 2.74 m, and Layer 2 from
the user-defined option. The parameters used in Matlock’s 2.74 to 15.2 m from the ground surface. su was assumed to be
built-in p–y curve model were ε50 = 0.015, su = 38 kPa, γ  96 kPa for Layer 1 and 215.4 kPa for Layer 2. The parameters
= 10.2 kN/m3 , and J = 0.25. The parameters used in the pro- used in the built-in p–y curve model were ε50 = 0.007 and γ 
posed model were ε 50 = 0.015, su = 38 kPa, γ  = 10.2 kN/m3 , α = 10.2 kN/m3 for Layer 1, and ε50 = 0.005 and γ  = 10.2 kN/m3
For personal use only.

= 0, and OCR = 1. It should be noted that ε50 value here was for Layer 2. The parameters used for the proposed model were
obtained by trial and error using Matlock’s model to achieve ε 50 = 0.007, γ  = 10.2 kN/m3 , α = 1, and OCR = 7 for Layer
the closest predictions to the experimental results for BM pro- 1, and ε 50 = 0.005, γ  = 10.2 kN/m3 , α = 1, and OCR = 5 for
file. Layer 2. The ε50 value here was obtained by trial and error
The results for load–deflection at the pile head are com- using Reese et al.’s model to achieve the closest predictions
pared in Fig. 20. Only computed results from LPILE are com- to the experimental results for pile head deflection.
pared here because the experimental results were not pro- As mentioned in the discussion for y50 , the OCR term in
vided in the reference. The figure shows good agreement be- the proposed model was needed to lower y50 values and in-
tween the proposed model and Matlock’s model at lateral crease the initial slope of the p–y curve. The OCR values were
loads of 68.9 and 109.4 kN, in which the pile head deflec- assumed here for overconsolidated clay based on our judg-
tion predictions for the proposed model were within 5%–10% ment. Reese et al. (1975) addressed the initial slope issue in
from Matlock’s model. At the lowest load (32.9 kN), the pile his model by using a separate linear function in the p–y curve
head deflection results from the proposed model were no- that was independent from y50 (Epyi = kz, where k is the rate of
tably higher than those from Matlock’s model by 40%. This soil stiffness increase with depth, and z is the depth). There-
is attributed to the higher initial p–y curve slope (i.e., stiffer) fore, before using the proposed model, we recommend veri-
at small deflections in Matlock’s model compared with the fying the initial slope and y50 for the model using experimen-
proposed p–y curve model. tal results such as from single pile tests or in situ testing to
The BM results are compared in Fig. 21. The figure shows obtain the initial modulus of subgrade reaction (ki ).
the experimental as well as the computed BM profiles. The The results for load–deflection at the pile head are com-
maximum BM results obtained using the proposed model pared in Fig. 22. The figure shows good agreement between
were notably smaller than the experimental results and the the experimental and computed load–deflection results for
computed results using Matlock’s model. The differences in both the proposed model and Reese’s model. The computed
the maximum BM between the experimental results and pro- deflection results using the proposed model were +5% at
posed model predictions were −48%, −21%, and −18% (from 485.6 and 606.2 kN, and +25% at 179.7 and 317.7 kN from
lowest to highest lateral load). On the other hand, the differ- the experimental results. On the other hand, the computed
ences in the maximum BM between the experimental results deflection results using the Reese et al.’s model were within
and Matlock’s model predictions were −37%, −5%, and −3% −4% to +13% from the experimental results.
(from lowest to highest lateral load). The proposed model had The experimental and computed BM profiles are compared
better prediction for the depth of inflection point (i.e., zero in Fig. 23. Similar to the soft clay case, the maximum BM re-
BM). The predicted inflection point depth using the proposed sults obtained using the proposed model were smaller than
model and Matlock’s was within 1%–6% and 14%–40% of the the experimental results and the computed results using
experimental results, respectively. Reese et al.’s model. The difference in the maximum BM

Can. Geotech. J. 60: 1055–1072 (2023) | dx.doi.org/10.1139/cgj-2021-0014 1069

 Canadian Science Publishing

Fig. 21. Comparison of experimental and computed bending moment results using the proposed model and Matlock (1970) p–y
curve model.
Can. Geotech. J. Downloaded from cdnsciencepub.com by Indian Institute of Technology Madras (IITM) on 10/27/24

Fig. 22. Comparison of experimental and computed load– Fig. 23. Comparison of experimental and computed bending
deflection curves using the proposed model and Reese et al. moment results using the proposed model and Reese et al.
(1975) p–y curve model. (1975) p–y curve model.
For personal use only.

between the experimental and proposed model results was

−24%, −25%, −30%, and −32% (from lowest to highest lat-
eral load). On the other hand, the difference between the ex-
perimental and Reese et al.’s model results was −19%, −19%,
−18%, and −32% (from lowest to highest lateral load). Fur-
ther, the depth of inflection point using the proposed model
was deeper. The predicted inflection point depth using the
proposed model was within +18 to +31% of the experimen-
tal results, while it was within −9% to +3% using Matlock’s
model.
Overall, the proposed model was able to predict the lateral
pile behavior with acceptable accuracy. The load–deflection and pu . This approach is similar to what Reese et al. (1975)
curves were close to the experimental results and the predic- used in his p–y curve model.
tions using the existing p–y curve models. However, the max-
imum BM predictions using the proposed model were lower
than the experimental and the predictions using the existing
4. Conclusions
models. The proposed model predictions can be further im- A unified p–y curve model for undrained clays was devel-
proved by using multipiece function for the p–y curve shape, oped. The model was developed using the results of paramet-
in which the initial slope is controlled independently of y50 ric studies conducted using finite element simulations for

1070 Can. Geotech. J. 60: 1055–1072 (2023) | dx.doi.org/10.1139/cgj-2021-0014

 Canadian Science Publishing

single pile. New models for the ultimate soil resistance (pu ) Author contributions
and the reference deflection (y50 ) were proposed. pu is esti- Data curation: AS, MA
mated from the undrained shear strength su , pile width, and Formal analysis: AS, MA, MA-F
the bearing capacity factor Np . The hyperbolic tangent func- Funding acquisition: MA-F
tion was used to model the p–y curve shape. The following is Investigation: AS, MA, MA-F
a summary of the observations from the study: Methodology: AS
Project administration: MA-F
r Previous experimental studies proposed different formulas Supervision: MA-F, GZV
to estimate the Np variation with depth depending on soil Writing – original draft: AS
undrained shear strength and the effect of water table. The Writing – review & editing: MA-F, GZV
Can. Geotech. J. Downloaded from cdnsciencepub.com by Indian Institute of Technology Madras (IITM) on 10/27/24

results from our analysis shows that these variables can be

incorporated into a unified single Np model for all clay con-
Competing interests
There is no financial or nonfinancial interest.
sistencies and for cases with or without water table.
r The results indicate that Np variation is nonlinear within
the zone between the ground surface and a depth of 2D, Funding information
and then it becomes linear below depth 2D up to the depth There is no funding to report for this submission.
where the ultimate value of Np (Npu ) is reached. This was
different than Np formulas from previous studies, which as-
sumed only a linear variation for Np .
References
r The reference deflection (y50 ) in the p–y curve was found ACI. 2008. Building code requirements for structural concrete (ACI 318-
08) and commentary. American Concrete Institute, Farmington Hills,
to be dependent on the Np variation, unlike what was pro-
MI.
posed in most previous studies. Bowles, J.E. 2001. Foundation analysis and design. 5th ed. McGraw-Hill,
r The effect of pile material stiffness (Ep ) was found to be New York.
negligible on y50 , and therefore the p–y curve is considered Dunnavant, T.W., and O’Neill, M.W. 1989. Experimental p-y model for
submerged, stiff clay. Journal of Geotechnical Engineering, 115(1):
independent of Ep . However, some of the previous studies
95–114. doi:10.1061/(ASCE)0733-9410(1989)115:1(95).
suggested that y50 is dependent on Ep . Ensoft. 2022. LPILE v2019: program for analysis of laterally loaded piles.
For personal use only.

r The proposed model was used to simulate case studies Ensoft, Inc., Austin, TX. Available from www.ensoftinc.com.
and the predictions were close to the experimental and Georgiadis, M. 1983. Development of p-y curves for layered soils. ASCE.
pp. 536–545.
the predictions using existing p–y curve models. The load–
Georgiadis, K., and Georgiadis, M. 2010. Undrained lateral pile response
deflection curves were in good agreement with the experi- in sloping ground. Journal of Geotechnical and Geoenvironmental
mental results and the results obtained using existing mod- Engineering, 136(11): 1489–1500. doi:10.1061/(ASCE)GT.1943-5606.
els. However, the maximum BM results were underesti- 0000373.
Georgiadis, M., Anagnostopoulos, C., and Saflekou, S. 2011. Centrifugal
mated by the proposed model and were within −5% to
testing of laterally loaded piles in sand. Canadian Geotechnical Jour-
−16% of the predictions by existing models. nal, 29: 208–216. doi:10.1139/t92-024.29208.
Haiderali, A.E., and Madabhushi, G., 2016. Evaluation of curve fitting
techniques in deriving p–y curves for laterally loaded piles. Geotech-
Article information nical and Geological Engineering, 34(5): 1453–1473. doi:10.1007/
s10706-016-0054-2.
History dates Hetenyi, M. 1946. Beams on elastic foundations. Scientific series. Univer-
sity of Michigan Press, Ann Arbor, MI.
Received: 19 January 2021 Hong, Y., He, B., Wang, L.Z., Wang, Z., Ng, C.W.W., and Mašín, D. 2017.
Accepted: 4 January 2023 Cyclic lateral response and failure mechanisms of semi-rigid pile
Accepted manuscript online: 14 January 2023 in soft clay: centrifuge tests and numerical modelling. Canadian
Version of record online: 17 May 2023 Geotechnical Journal, 54(6): 806–824. doi:10.1139/cgj-2016-0356.
Jeanjean, P., Zhang, Y., Zakeri, A., Andersen, K., Gilbert, R., and
Senanayake, A. 2017. A framework for monotonic p-y curves in clays.
Copyright In Proceedings of the Offshore Site Investigation Geotechnics 8th In-
© 2023 The Author(s). Permission for reuse (free in most ternational Conference, Society for Underwater Technology.pp. 108–
141.
cases) can be obtained from copyright.com.
Martin, C., and Randolph, M. 2006. Upper-bound analysis of lateral pile
capacity in cohesive soil. Géotechnique, 56(2): 141–145. doi:10.1680/
Data availability geot.2006.56.2.141.
Data will be available upon request from the corresponding Matlock, H. 1970. Correlations for design of laterally loaded piles in soft
clay. Offshore Technology in Civil Engineering Hall of Fame Papers
author. from the Early Years. pp. 77–94.
McClelland, B., and Focht, J.A. 1958. Soil modulus for laterally loaded
piles. Transactions of the American Society of Civil Engineers, 123(1):
Author information 1049–1063.
Murff, J.D., and Hamilton, J.M. 1993. P-ultimate for undrained analysis
Author ORCIDs of laterally loaded piles. Journal of Geotechnical Engineering, 119(1):
91–107. doi:10.1061/(ASCE)0733-9410(1993)119:1(91).
Mohsen Amirmojahedi https://orcid.org/
Nip, D., and Ng, C., 2005. Back-analysis of laterally loaded bored piles.
0000-0001-9661-4821 Proceedings of the Institution of Civil Engineers – Geotechnical En-
Murad Abu-Farsakh https://orcid.org/0000-0003-0901-6116 gineering, 158(2): 63–73. doi:10.1680/geng.2005.158.2.63.

Can. Geotech. J. 60: 1055–1072 (2023) | dx.doi.org/10.1139/cgj-2021-0014 1071

 Canadian Science Publishing

Randolph, M.F., and Houlsby, G. 1984. The limiting pressure on a circular Offshore Technology Conference, Houston, Texas, 2(OTC 2312). pp.
pile loaded laterally in cohesive soil. Géotechnique, 34(4): 613–623. 672–690.
doi:10.1680/geot.1984.34.4.613. Stevens, J., and Audibert, J. 1979. Re-examination of p-y curve formula-
Reese, L.C., and Matlock, H. 1956. Non-dimensional solutions for tions. In Offshore Technology Conference, Paper OTC 3402, Houston,
laterally loaded piles with soil modulus assumed proportional Texas. pp. 397–403.
to depth. In Proceedings of the Eighth Texas Conference on Sullivan, W.R. 1977. Development and evaluation of a unified method
Soil Mechanics and Foundation Engineering, The University of for the analysis of laterally loaded piles in clay. Unpublished M.S. the-
Texas. sis, University of Texas at Austin.
Reese, L.C., and Van Impe, W.F. 2010. Single piles and pile groups under Winkler, E. 1867. Die Lehre von Elastizitat und Festigkeit (The theory of
lateral loading. 2nd ed. CRC Press. elasticity and stiffness). H. Domenicus. Prague.
Reese, L.C., and Welch, R.C. 1975. Lateral loading of deep foundations in Wu, D., Broms, B.B., and Choa, V. 1998. Design of laterally loaded piles in
stiff clay, Journal of the Geotechnical Engineering Division, 101(7): cohesive soils using p-y curves. Soils and Foundations, 38(2): 17–26.
633–649. doi:10.1061/AJGEB6.0000177. doi:10.3208/sandf.38.2_17.
Can. Geotech. J. Downloaded from cdnsciencepub.com by Indian Institute of Technology Madras (IITM) on 10/27/24

Reese, L.C., Cox, W.R., and Koop, F.D. 1974. Analysis of later- Zhang, Y., and Andersen, K.H. 2017. Scaling of lateral pile p-y response
ally loaded piles in sand. Offshore Technology in Civil En- in clay from laboratory stress-strain curves. Marine Structures, 53:
gineering Hall of Fame Papers from the Early Years. pp. 124–135. doi:10.1016/j.marstruc.2017.02.002.
95–105. Zhang, Y., and Andersen, K.H. 2019. Soil reaction curves for monopiles
Reese, L.C., Cox, W.R., and Koop, F.D. 1975. Field testing and analysis of in clay. Marine Structures, 65: 94–113. doi:10.1016/j.marstruc.2018.
laterally loaded piles on stiff clay. In Proceedings of the VII Annual 12.009.
For personal use only.

1072 Can. Geotech. J. 60: 1055–1072 (2023) | dx.doi.org/10.1139/cgj-2021-0014