3rd International Symposium on Cone Penetration Testing, Las Vegas, Nevada, USA - 2014

Numerically derived CPT based p-y curves for a soft clay modeled as an
elastic perfectly plastic material
P. Truong and B.M. Lehane
The University of Western Australia, Perth, WA, Australia

ABSTRACT: This paper presents the results from numerical analyses which explore the possibility of deriving p-y springs for laterally loaded piles in clay using direct correlations with the CPT qnet value. Three dimensional finite element analyses were performed using a linear elastic perfectly plastic soil model to predict
the response of single piles in clay subjected to lateral loads. The corresponding CPT qnet profile for each lateral load test simulation was determined via an expression that was derived using the same soil model in large
displacement finite element analyses. A wide range of three dimensional finite element computations of lateral pile response was performed and used to derive CPT based p-y springs for piles in linear elastic perfectly
plastic soil. The new p-y formulation is shown to be capable of matching the response measured in lateral
load tests on centrifuge scale piles in soft clay.
1 INTRODUCTION
The industry standard method for estimation of the response to lateral load of piles in clay is to use the
load transfer method with the soil p-y (non-linear) springs recommended by American Petroleum Institute
(API, 2011), which were developed by Matlock (1970). The API p-y curves are a function of the clays
undrained shear strength (su) and rigidity index (Ir), and also depend on the relative depth (z) of any given
soil layer. The selection of a representative su profile is difficult given the wide scatter in measured su values observed in practice due, for example, to sampling disturbance effects (especially in soft clays). A direct CPT-based p-y formulation is therefore clearly preferable to an indirect approach involving su.
This paper presents a n umerical derivation for p-y curves applicable to clay assuming a l inear elastic
perfectly plastic (LEPP) soil model. The p-y curves are deduced from three dimensional (3D) finite element analyses (FEA) for 6 different cases and corresponding numerically derived CPT qnet data are determined using an existing FE solution for cone penetration in a similar LEPP soil. The p-y and qnet data are
compared and used to develop a CPT based p-y formulation as a function of qnet and Ir.
A parallel centrifuge investigation of laterally loaded pile behavior in soft clay was also performed at
the University of Western Australian (UWA) and is presented in a companion paper by Guo et al. (2014).
The formulations derived here are compared with the lateral pile response of the centrifuge piles to assess
their potential for application to laterally loaded pile analysis.
2 NUMERICAL ANALYSIS
The numerical analyses for this study were conducted using the 3D FEA program Plaxis 3D (Plaxis 3D
2011). Six models were analyzed with each having the same pile length to diameter ratio (L/D) of 11.8
and with a lateral load applied to the pile at 1.5 pile diameters above the ground surface. The pile com975

prised volumetric elastic elements (with a Youngs modulus of 30 GPa) and therefore the predicted lateral
pile displacements did not include any component due to the pile itself yielding. The pile-soil interface
was assumed to be fully rough for the present study; other analyses conducted by the authors show that
having an interface roughness halfway between being fully rough and smooth reduces the ultimate lateral
stresses developed on piles by about 10%. A linear elastic beam element with negligible stiffness (E = 1011
GPa) was embedded at the center of the volumetric pile to allow for ease of extraction of the pile bending moments and deflections.
The soil was represented as a LEPP isotropic material with a linear shear modulus (G) profile, linear su
profile, Poissons ratio () of 0.5 and bulk unit weight () of 15.6 kN/m3. Clays with strengths corresponding to overconsolidation ratios (OCRs) of 1 and 2 were examined and the analyses permitted suction values of up to 100 kPa. The full range of parameters investigated is presented in Table 1. It is seen that the
pile diameters and rigidity indices (Ir = G/su) considered were varied by about one order of magnitude. A
typical deformed mesh and displacement vector plot from the Plaxis 3D analyses is presented in Figure 1,
showing the soil block, pile mesh and soil mechanism at failure.
Table 1. Input variables for PLAXIS 3D analyses
Model

Pile diameter,

FE1
FE2
FE3
FE4
FE5
FE6

D
m
0.88
0.88
0.88
0.88
0.44
3.00

Net cone
resistance,
qnet
MPa
13.7 z(m)
23.9 z(m)
11.0 z(m)
15.8 z(m)
13.7 z(m)
13.7 z(m)

Soil stiffness
ratio,
E/qnet
17.5
17.5
5.4
38.0
17.5
17.5

Over consolidation ratio,

OCR
1
2
1
1
1
1

Undrained
shear strength,
su
kPa
1.2 z(m)
2.1 z(m)
1.2 z(m)
1.2 z(m)
1.2 z(m)
1.2 z(m)

Rigidity
index,
Ir = G/su
67
67
17
200
67
67

The following relationship between the net CPT qnet and the clay su was adopted. This was developed
by Lu et al. (2004) based on numerical analyses using the re-meshing and interpolation FE technique in a
LEPP soil i.e. the same soil model used for the laterally loaded pile analysis in this paper.

3.4 + 1.6 ln 1.9 + 1.3 with =

(1a)

4.7 + 1.6 ln

with = 0 and = 1

(1b)

The value in this equation is referred to as the anisotropy factor, which for the current analysis is taken
as zero as the in-situ earth pressure coefficient (K0) adopted in the pile analyses was unity. A cone roughness factor (c) of unity is assumed when applying equation (1) as this leads to predicted Nkt values comparable to those indicated in practice (i.e. between 11 and 14 for Ir between 50 and 300); adoption of this
c is consistent with the adoption of a fully rough interface between the pile and soil. The qnet profiles
adopted implicitly in the FE analyses are provided in Table 1. These vary linearly with depth, matching
the linear variation of su with depth. The values of the corresponding soil stiffness ratio (E/qnet) are also
given in Table 1 and the relationship between this ratio and the rigidity index is given in Equation 2. The
p-y formulation employed by API (2011) uses the strain at a half of the maximum stress recorded in an
unconfined compressive strength test (c) and the relationship between this parameter and Ir is also given
in Equation 2.
1
1
=
=
3
3

(2)

976

Figure 1. Plaxis 3D deformed mesh and displacement vector plot of a laterally loaded pile

3 DERIVATION OF P-Y CURVES

The lateral pile displacements were taken directly from the FE output at each level of applied load while
the net soil resistance per unit length of pile (p) was determined by double differentiation of the pile bending moment profiles for each load. The MATLAB (MATLAB 2013) smoothing spline tool was used to assist differentiation and obtain p at normalized depths (z/D) between 0.5 and 7; the net pressure (P) is equal
to p/D. The soil movement patterns tended to have a larger vertical component at z/D values greater than 7
(as they are close to the point of fixity) and therefore p-y data at these depths were not developed.
The maximum ultimate pressures (Pu) required normalized displacements (y/D) of up to 2 to develop.
This pressure, which is obtained by normalizing the ultimate p value by the pile diameter, is related in the
following equation, to the qnet value via a bearing factor, Npq:
=

=
+ [1.5 0.14 ln ] tanh 0.65

4.7 + 1.6

(3)

The calculated Npq factors for each of the six FE models are presented in Figure 2, where they are seen to
vary with depth at z/D ratios less than about 3. In addition to this depth dependence, computed Npq values
were also found to vary with Ir. The Npq predictions do not exhibit a dependence on the strength profile or
on the pile diameter (e.g. compare predictions of FE1 with FE2, FE5 and FE6), and regression analyses of
the computed data (z/D 1) showed that the following expression provided a best fit to the set of six FE
cases (with a coefficient of determination (r2) of 0.97). Additional FE analyses on a range of Ir values
would help to refine this expression.
(4)

The initial term in Equation (4), which essentially gives the value of Npq at the ground surface, is based
on the research of Broms (1964) and leads to what appears from Figure 2 to be a conservative ground surface Npq value (of ~0.25). It is noted that the estimation of Npq at z/D < 1 was highly sensitive to small
numerical errors and therefore this (conservative) assumption was adopted.
The Npq values predicted using Equation 4 are compared on Figure 3 with those given by the API
(2011) recommendations for soft clays. It is evident that API (2011) recommends a lateral resistance that
977

is typically about 20% less than that predicted. The deduced Npq value at z/D > 3 for Ir = 200 (of about
unity) is comparable to the fully rough ultimate factor assessed by Randolph & Houlsby (1984) for the
flow around mechanism.
The relationship between y/D with the net pressures normalized by the computed ultimate pressures
(P/Pu) is presented on Figure 4 for the three rigidity indices considered. It is evident that contrary to API
(2011) recommendations, these curves are not unique for all z/D values. There is a clear tendency for the
initial slope of these curves to reduce as z/D increases with curves at z/D = 3 (where failure occurs by soil
flowing around the pile) being essentially independent of z/D; curves at lower z/D ratios reflect patterns
associated with wedge and cavity expansion type mechanisms. The rigidity index also has a clear effect on
the form of the normalized curves but the similarity of FE1, FE2, FE5 and FE6 (all with Ir = 67) indicates
the normalized curves are not a function of the pile diameter or the soil strength profile. Regression analyses of all the FE predictions led to the development of two equations, depending on the relative depth

0.85 0.5
= (0.26 + 3.98)

0.85
= (0.15 + 2.3)

0<

<3

(5)

(5)

Equations 5a and 5b provide a fit to the FE data with respective r2 values of 0.98 and 0.99. Further
analyses will help to refine these equations. The equations are compared on Figure 5 with the API (2011)
recommendations for soft clay with rigidity indices of 200 and 50 (used for illustrative purposes). It is evident that the API curve (applicable for all z/D values) is stiffer at low y/D values. Overall, however,
Equation 5 predicts P/Pu values at any given y/D value that are on average within 15% of the API recommendation- despite the fact that these equations were derived for a LEPP soil.

1.6
1.4
1.2
1

Npq

0.8
0.6
0.4

FE1

FE2

FE3

0.2

FE4

FE5

FE6

z/D

Figure 2. Ultimate soil resistance factor, Npq, derived in finite element analyses.

978

1.6
1.4

Ir = 17

Ir=17

1.2

Ir = 67
Ir = 200

Npq

Ir = 17 (API, 2011)

0.8

Ir = 67 (API, 2011)

0.6

Ir = 200 (API, 2011)

Ir=67

0.4

Ir=200

0.2
0

z/D

Figure 3. Comparison of Npq given by Equation 4 with API (2011) recommendations.

1.2

1.2

Ir = 67

Ir = 17
1

0.8

z/D = 1.0

P/Pu0.6

z/D > 3.0

Eqn. 5 for z/D = 0.5

0.4

Eqn. 5 for z/D = 1.0

0.2

0.4

0.6

Eqn. 5 for z/D = 2.0

Eqn. 5 for z/D > 3

Eqn. 5 for z/D > 3

0.2

Eqn. 5 for z/D = 1.0

0.2

Eqn. 5 for z/D = 2.0

z/D = 0.5
z/D = 1.0
z/D = 2
z/D > 3.0
Eqn. 5 for z/D = 0.5

P/Pu0.6

z/D = 2

0.4

0.8

z/D = 0.5

0.8

y/D

Ir = 200
1
z/D = 0.5
z/D = 1.0

P/Pu 0.6

z/D = 2
z/D > 3.0

0.4

Eqn.5 for z/D = 0.5

Eqn. 5 for z/D = 1.0

0.2
0

Eqn. 5 for z/D = 2.0

Eqn. 5 for z/D > 3
0

0.02

0.1

y/D

1.2

0.8

0.05

0.04

0.06

y/D

Figure 4. P/Pu vs. y/D curves derived from finite element analyses.
979

0.15

0.2

1.2

1.2

Ir = 50

Ir = 200

0.8

0.8

P/Pu

P/Pu
0.6

0.6

0.4

0.4

Eqn. 5 for z/D = 1.0

Eqn. 5 for z/D >3

0.2

Eqn. 5 for z/D = 1.0

Eqn. 5 for z/D > 3

0.2

API (2011)
0

0.1

0.2

API (2011)
0

0.3

y/D

0.02

0.04

0.06

y/D

Figure 5. Comparison of API (2011) recommendations with normalized p-y curves given by Equation 5.

4 VERIFICATION OF P-Y CURVES

The accuracy of Equations 4 and 5 was checked by using these equations directly in a standard load transfer program and then comparing the predicted lateral load displacement response with that predicted directly in the FE analyses. The program Oasys ALP (Oasys 2013) was employed, which (as with many
commercially available laterally loaded pile programs) models the pile as a series of beam elements and
the soil as a series of non-interacting p-y (non-linear) springs located at nodes between the beam elements.
The p-y springs were derived using Equations 4 and 5 and input into ALP at a spacing of D/2 along each
pile.
The lateral load (H) with pile head displacement predictions obtained for the 6 FE predictions (see Table 1) are compared on Figure 6 with the corresponding ALP predictions. Evidently, very good agreement
is observed. Such agreement verifies the numerical procedures used for the derivation of Equations 4 and
5 and also demonstrates that the load-transfer (beam-spring) approach is capable of reproducing results of
the full 3D FE models, even for diameters as large as 3m.
Figure 6 also shows corresponding predictions obtained using the API (2011) recommendations for soft
clay (noting that the rigidity index is related to the API c parameter via Equation 2). The API predictions
provide a reasonable match to the other sets of predictions up to about half of the ultimate capacities, but
then deviate considerably because of the lower API Npq values (see Figure 3).
Direct application of Equations 4 and 5 is problematic in practice as the rigidity index (Ir) is difficult to
quantify. Backanalyses of existing lateral load test case history data using these equations is the best way
of assessing a representative Ir value (which is analogous to the selection of c when employing API
2011). One example of such a backanalysis is provided on Figure 7, which plots the lateral loaddisplacement response of two centrifuge scale piles reported by Guo et al. (2014). These piles were jacked
into kaolin at OCRs of 1 and 2 and load tested after full equalization. The predicted responses of these
piles obtained using Oasys ALP and Equations 4 and 5 with Ir = 150 are seen on Figure 7 to provide a
good match to the measured responses and are a clear improvement on the predictions obtained using API
(2011) with the same Ir value. Future investigations will examine how a suitable Ir value can be derived
from shear wave velocities and qnet measurements obtained directly in seismic cone penetration tests.

980

200

300

FE1 - D=0.88m - OCR=1 - Ir=67

FE2 - D=0.88m - OCR=2 - Ir=67

150
200

H (kN)
100

H (kN)

50

100

Plaxis 3D

Plaxis 3D

New formula
0

New formula

API (2011)
0.0

0.1

0.2

0.3

0.4

0.5

API (2011)
0.0

0.1

y (m)
200

0.2

0.3

0.4

0.5

y (m)
200

FE3 - D=0.88m - OCR=1 - Ir=17

150

FE4 - D=0.88m - OCR=1 - Ir=200

150

H (kN)

H (kN)

100

100

50

50

Plaxis 3D

Plaxis 3D

New formula
0

New formula

API (2011)
0.0

0.5

1.0

1.5

0
0.00

2.0

API (2011)
0.05

y (m)
25

0.10

0.15

0.20

y (m)
8000

FE5 - D=0.44m - OCR=1 - Ir=67

20

FE6 - D=3.0m - OCR=1 - Ir=67

6000

15

H (kN)
4000

H (kN)
10

2000

Plaxis 3D

Plaxis 3D

New formula

New formula

API (2011)
0.0

0.1

0.2

0.3

y (m)

API (2011)
0.0

0.5

1.0

1.5

2.0

y (m)

Figure 6. Comparison of load displacement results between finite element analyses, derived p-y curves, and the API
(2011) recommendations.

981

250

50

D=0.88m - OCR=1 - Ir=150

D=0.44m - OCR=2 - Ir=150

200

40

150

30

H (kN)

H (kN)

100

20
Guo et al. (2014)

50
0

New formula
API (2011)
0.0

0.2

0.4

Guo et al. (2014)

10
0

0.6

yh (m)

New formula
API (2011)
0.0

0.1

0.2

0.3

yh (m)

Figure 7. Backanalysis of Guo et al (2014) centrifuge results (prototype units) with derived p-y relationships.

5 CONCLUSION
This paper presents a numerical derivation of p-y curves that are suited to the prediction of the response to
lateral load of piles in clays that can be represented by a linear elastic modulus and shear stress strength,
su. The p-y curves can be derived directly from CPT qnet profiles but require an assumption to be made regarding the value of the Ir of the soil. The formulations for the p-y curves when implemented in a standard
load transfer (beam-spring) computer model are shown to lead to virtually the same pile response predictions as full 3D Finite Element analyses. Preliminary backanalyses of lateral load test data measured in
kaolin at OCR = 1 and 2 show that Equations 4 and 5 lead to good predictions of the observed response
assuming Ir = 150.
REFERENCES
American Petroleum Institute 2011, Geotechnical and foundation design considerations, ANSI/API Recommended
Practice 2 GEO First Edition, April 2011.
Broms, B.B 1964, Lateral resistance of piles in cohesive soils, American Society of Civil Engineers Journal of Soil
Mechanics, 90(SM2): 27-63.
Guo, F. & Lehane, B.M. and Ju, J. 2014, Experimentally derived CPT-based p-y curves for soft clay, Proceedings
of the International Symposium on Cone Penetration Testing, Las Vegas.
Lu, Q., Randolph, M.F., Hu, Y. & Bugarski, I.C. 2004, A numerical study of cone penetration in clay, Gotechnique, 54(4): 257-267.
MATLAB 2013, software, version 2013a, The MathWorks Inc.
Matlock, H. 1970, Correlation for design of laterally loaded piles in soft clay, Proceedings of the 2nd Offshore Technology Conference, Houston, 1(Paper OTC 1204): 577-594.
Oasys ALP 2013, software, version 19.2, Oasys Limited.
Plaxis 3D 2011, software, version 2011.1, Plaxis bv.
Randolph, M.F. & Houlsby, G.T. 1984, The limiting pressure on a circular pile loaded laterally in cohesive soil,
Gotechnique, 34(4): 613-623.

982