You are on page 1of 28

The International Journal of Geomechanics

Volume 2, Number 4, 419–446 (2002)


Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

Kinematic Hardening Model for


Pipeline-Soil Interaction under
Various Loading Conditions
J. Zhang, D.P. Stewart, and M.F. Randolph

Advanced Geomechanics, 4 Leura Street,


Nedlands, WA 6009, Australia
Tel: +61 8 9389 5033; Fax: +61 8 9389 5066
E-mail: georgez@ag.com.au

Golder Associates, 182 Lord Street,


Perth, WA 6000, Australia
Tel: +61 8 9328 7677; Fax: +61 8 9328 8433
E-mail: dstewart@golder.com.au

Centre for Offshore Foundation Systems, The University


of Western Australia Nedlands, WA 6907, Australia
Tel: +61 8 9380 3075; Fax: +61 8 9380 1044
E-mail: randolph@civil.uwa.edu.au

ABSTRACT. A kinematic hardening two-surface model of soil-footing interaction for pipelines under
various loading conditions is presented. The model requires 13 parameters. Most of these parameters can be
derived directly from a set of model tests or from conventional soil mechanics theory. Model performance is
demonstrated through modeling the drained behavior of pipeline-soil interaction under typical loading
conditions. Predictions from the model compare well with experimental data obtained in a series of loading
tests on shallowly embedded pipelines. The model provides an efficient and direct way to evaluate the
behavior of a pipeline under various loading conditions.

I. INTRODUCTION

Offshore pipelines provide a crucial link between offshore oil and gas fields and the
hydrocarbon product users onshore. The serviceability of a pipeline depends to a great extent on

Key Words and Phrases. kinematic hardening, pipeline-soil interaction, offshore pipelines, sand.

© 2003 ASCE DOI: 10.1061/(ASCE)1532-3641(2002)2:4(419)


ISSN 1532-3641

Int. J. Geomech., 2002, 2(4): 419-446


420 J. Zhang et al.

its geotechnical stability as soils provide direct support to the pipeline. Hence a basic consider-
ation in offshore pipeline design is whether the soil resistance is high enough to balance the
hydrodynamic loads if the pipeline is placed directly on the seabed. If the soil resistance is less
than that required for pipeline stability, additional engineering measurements, such as trenching
or anchoring, have to be employed. Thus the evaluation of soil resistance to an untrenched
pipeline is a fundamental consideration in offshore pipeline design.
A considerable amount of research has been carried out with a view of achieving a more
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

reliable prediction of soil restraint for untrenched offshore pipelines. Empirical models were
proposed on the basis of full-scale model test data [1–3]. A general form of these models is

H = H f + Hr (1)

where Hf is a frictional resistance that depends on the net vertical load and Hr represents
additional resistance due to pipeline embedment. It should be noted that this expression predicts
a certain soil resistance, H = Hr , even though the net vertical load is zero, as observed in the
corresponding experimental work.
However, because these models were developed entirely on an empirical basis, it should not
be viewed as complete verification of the pipe-soil interaction model [4]. In an attempt to achieve
better understanding of the physics of pipe-soil interaction of untrenched offshore pipelines, a
new approach has been adopted recently [5,6]. Instead of proposing separate equations for the
evaluation of soil resistance and pipeline settlement, the pipe-soil system is considered as a
macro-element in the new approach — a plasticity model. This approach has been widely adopted
in modeling soil-footing interaction of shallow foundations [7–11]. This type of model provides
a direct link between the incremental loads acting on a pipeline and the induced incremental
displacements. Hence, the soil resistance and the movement of a pipeline can be evaluated
directly from these models.
In order to construct a plasticity hardening model, an initial elastic response, a yield function,
hardening law, and flow rule are required. Within a conventional elasto-plastic model [5], the
yield function defines a boundary of elastic states. Pipeline response under unloading and
reloading within this boundary is fully elastic. Nonrecoverable displacements occur only when
the loading point is located on the yield surface and causes the surface to expand. Thus the
conventional model gives an abrupt change in the horizontal stiffness when the load reaches the
elastic boundary and cannot predict cumulated pipeline settlement when the load is cycled within
this boundary.
To overcome these drawbacks of the conventional model, a bounding surface model has
been developed [6]. Model performance is greatly improved with a gradual transition of the
horizontal response and cumulated pipe settlement under cyclic loading. However, the bounding
surface model always predicts plastic displacement even though the amplitude of the cyclic
loading is very small compared to the size of the bounding surface, since the elastic range at the
start of loading or reverse loading has vanished in this model.
The concept of the kinematic hardening model or the so-called “bubble” or “two-surface”
model has been introduced in stress space to describe the monotonic and cyclic behavior of soils
[12,13]. As illustrated in p-q stress space in Figure 1, in a kinematic hardening model, the pre-
loading history is memorized by a bounding surface, F = 0, the evolution of which during
reloading is determined from the isotropic hardening law linking the size of the bounding surface
to the cumulated volumetric strain. An elastic range is enclosed by a yield surface, f = 0, that takes
similar form to the bounding surface but is much smaller. The yield surface may be dragged along

Int. J. Geomech., 2002, 2(4): 419-446


A Kinematic Hardening Model for Pipeline-Soil Interaction 421
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

FIGURE 1. Illustration of a two-surface soil model in p-q space.

with the loading point and moves around within the bounding surface like a “bubble” following
the kinematic hardening law. The area between the bubble (yield) surface and the bounding
surface defines a gradual transition from elastic to fully plastic states. As such, the model allows
plastic strain to occur within the bounding surface and makes it possible to simulate the
accumulation of plastic strain for overconsolidated soils under cyclic loading but avoids strain
accumulation under small amplitude cyclic loading.
A kinematic hardening model incorporating these features is proposed to simulate the
behavior of offshore pipelines in this article. The model is developed on the basis of experimental
data that were mainly obtained from pipe-soil interaction tests using pipeline segments shallowly
embedded in calcareous sand [14]. The formulation of the model is described in the next section.
Following that, single surface models are discussed in comparison with the current model. The
performance of the model is evaluated through modeling the drained behavior of soil-pipeline
interaction. Model predictions are compared with experimental data obtained from centrifuge
tests.

II. MODEL FORMULATION

The development of this model is based on the following assumptions, as illustrated in


Figures 2 and 3.

1. The pipeline is rigid and is placed on a homogeneous soil that has a flat surface. As such,
the soil-pipeline interaction considered is a plane-strain problem. For a pipeline under
inclined loading, this problem can be fully described in vertical and horizontal load space.
Moment loading is ignored as the position of H varies within a limited range and thus
results in only a small variation in the amount of moment loading.
2. The loading history of a pipeline is represented by the bounding surface, F = 0. The size
of the bounding surface is defined by its positive intercept on the vertical load axis and
expands or contracts isotropically without change in aspect ratio.

Int. J. Geomech., 2002, 2(4): 419-446


422 J. Zhang et al.
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

FIGURE 2. A pipeline under V-H loading.

FIGURE 3. Two-surface model for a pipeline-soil system in V-H load space.

Int. J. Geomech., 2002, 2(4): 419-446


A Kinematic Hardening Model for Pipeline-Soil Interaction 423

3. A yield surface fb = 0 encloses the elastic domain in vertical and horizontal load space. It
translates, expands, or contracts within the bounding surface.
4. The plastic hardening modulus is interpolated according to the distance, δ, between the load
point (A on the yield surface) and the conjugate point (C on the bounding surface), which
is characterized by the same direction of the exterior normal (see Figure 3). As such the
modulus varies along the loading path from a very large value on the yield surface to a
small value on the bounding surface. It defines the gradual transition from elastic to plastic
state.
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

5. A nonassociated flow rule governs the plastic displacement vector. The isotropic hardening
rule describes the expansion or contraction of the yield and bounding surfaces. The
kinematic hardening law defines the translation of the yield surface within the bounding
surface.

A. Yield Functions
Two similar functions for the yield surface and the bounding surface are proposed. They both
take the form of a parabola, with the center of the yield surface located at N (VN, HN) and the center
of the bounding surface located at M (VM, HM) where VM = (1–β) Vmax /2 and HM = 0, as shown
in Figure 3. The size of the yield surface is r times that of the bounding surface. The two surfaces
are expressed as follows.
The bounding surface [6] is:

 V 
F = H − µ + β (Vmax − V ) = 0 (2a)
 Vmax 

or by reference to the center, VM = (1–β)Vmax /2 and HM = 0:

 V − VM 1 + β   1 + β 
F = H − H M − µ +  Vmax − (V − VM ) = 0 (2b)
 Vmax 2  2 

This expression gives a positive intercept on the horizontal load axis and a negative intercept on
the vertical load axis. While the positive intercept on the horizontal load axis provides the
pipeline a horizontal capacity of µβVmax for V = 0, the negative intercept on the vertical load axis
implies that the pipeline has an uplift capacity which may not be strictly true in practice for slow
loading.
The yield or bubble surface is:

 V − VN 1 + β   1 + β 
fb = H − H N − µ  +  rVmax − (V − VN ) = 0 (3)
 rVmax 2  2 

where r is a parameter defining the size of the bubble or the elastic domain. As the pipeline has
been assumed rigid, the behavior of soil-footing interaction is solely controlled by the soil part,
which implies the size of the elastic domain is determined from soil properties. Values for r of
0.05 and 0.1 have been adopted previously [15] in modeling the behavior of normally consoli-
dated and overconsolidated clayey soils, respectively, implying that r probably increases with the
overconsolidation ratio. In the current model, to account for the overconsolidated state of the soil,

Int. J. Geomech., 2002, 2(4): 419-446


424 J. Zhang et al.

the size of the yield surface is linked to the overloading ratio, R = Vmax/V0, of the pipeline at the
start of loading. The relationship of r = r0 R is proposed here to include the effect of loading
history, where a small value is used for r0.

B. Flow Rule
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

As indicated by previous experimental work [5,10,14,16], a nonassociated flow rule should


be adopted in describing soil-footing interaction in V-H space. Hence a bullet-shaped plastic
potential surface is proposed, as given below.

m
 V 
G = H − µt  + β (Vmax − V ) − C = 0 (4)
 Vmax 

where C is a constant adjusting the position of the plastic potential surface to allow the surface
to pass through the conjugate point on the bounding surface. Since C is a constant, this expression
gives a plastic displacement vector depending only on the loading state, thus providing a
convenient way to the derivation of the vector and being straightforward to use. One of the
surfaces, with C = 0.02Vmax, together with the bounding surface, is illustrated in Figure 4 in
normalized form, where µ = 0.5, β = 0.05, µt = 0.6, m = 0.2.

C. Hardening Functions
In this model, the size of the bounding surface changes (expands or contracts) with the plastic
vertical displacement. The size of the yield surface is directly scaled from the bounding surface
with its center moving inside the bounding surface.

FIGURE 4. Normalized bounding surface and plastic potential surface.

Int. J. Geomech., 2002, 2(4): 419-446


A Kinematic Hardening Model for Pipeline-Soil Interaction 425

The change in the size of these two surfaces is related to the plastic vertical displacement [14]
by the following equation,

kve kvp
Vmax = zp (5)
kve − kvp

where kve and kvp are elastic and plastic stiffnesses of the monotonic vertical loading, and
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

kve = λ1kvp, where λ1 is a model parameter. Because soil around a pipeline will be densified under
cyclic horizontal loading, an additional increase in Vmax is expected after cycling. The variation
of Vmax is related to the cumulated plastic deformation, and is given by

′ = cVmax
Vmax (6a)

c = cmax − (cmax − 1)e


(
− ζz pc B ) (6b)

where zpc is the cumulated settlement induced from cyclic loading, B is the pipe diameter, c is
a multiplier, and ζ is a parameter that must be derived from test data.
The centers of the bounding and yield surfaces will move as they change in size. For the
bounding surface, the center moves along the vertical load axis only, therefore

 δVM  1 − β δVmax 


M:  = 2  (7)
δH M   0 

The center of the yield surface moves partly because of the size change and partly because
of the kinematic feature. For the load point located on the bounding surface, or, in other words,
when the two surfaces are in contact, the conjugate point C (VC, HC) on the bounding surface is
identical to the load point A (V, H) on the yield surface, and the following expression holds:

 V − VM  1  V − VN 
 =   (8)
H − HM  r H − HN 

Hence the center of the bubble surface can be given as

 VN   V  1 − β V 
  = (1 − r )  + r  2 max  (9)
HN   H   0 

For a load point A (V, H) located on the yield surface, there is a conjugate point C (VC , HC)
on the bounding surface F = 0, where the gradients of yield and bounding surfaces are identical,
hence

 VC  1  V − VN   VM  1 V − VN + 1 − β rVmax 
 =  + =  2  (10)
 HC  r  H − H N   H M  r  H − HN 

Int. J. Geomech., 2002, 2(4): 419-446


426 J. Zhang et al.

The vector connecting points A and C can be expressed as

 VC − V  1  V − VN   V  (1 − β) V 
 =   +   2 max 
(11)
 HC − H  r  H − H N   H  0 
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

It is assumed that the relative motion of loading point A, with respect to its conjugate point
C, occurs along AC, Figure 3, thus

δV   δVC   VC − V 
 − = S (12)
δH  δHC   HC − H 

where S is a scalar factor. The evolution of the bounding and the yield surfaces are described by

δV   δVN   V − VN  δVmax


 = +  (13a)
δH  δH N   H − H N  Vmax

 δVC   δVM   VC − VM  δVmax


 = +  (13b)
δHC  δH M   HC − H M  Vmax

from which the evolution of the bubble center can be derived as follows

 δVN   VC − V   δVM   (VC − VM ) − (V − VN )  δVmax


 = S +  + 
δH M  ( HC − H M ) − ( H − H N ) Vmax
(14a)
δH N   HC − H 

Equation (14a) can be rewritten by substituting equations (7) and (10), to give

1 − β 
 δVN   p δVmax  + 1 − r  V − VN  δVmax
  =  S +  2    (14b)
δH N   q  r  H − H N  Vmax
 0 

where

1 1− β 1
p = (VC − V ) = ( V − VN ) − V + Vmax , q = ( HC − H ) = ( H − H N ) − H (14c)
r 2 r

From the consistency condition, δfb = 0, by using equations (3) and (14b), the following
expression is obtained

1    1 − r V − VN 1 − β  1 − r H − H N  
S=  aδV + δH +  c − a + −  δVmax  (15)
ap + q    r Vmax 2  r Vmax  

Int. J. Geomech., 2002, 2(4): 419-446


A Kinematic Hardening Model for Pipeline-Soil Interaction 427

where

∂fb ∂f 2µ(V − VN )
a= =− b = (15a)
∂V ∂VN rVmax

∂fb  1  V − V  2 r(1 + β)2 


c= = −µ  N
+ 
∂Vmax  r  Vmax  (15b)
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

 4 

D. Plastic Hardening Modulus


For any loading path, the plastic hardening modulus, K, varies from its initial value of Kmax,
to the value of KC when the yield surface contacts the bounding surface. The following expression
can be adopted to perform this transition

ρ
 δ 
K = KC + ( K max − KC )  (16)
 δ max 

where ρ is a model parameter, δmax = the maximum value of δ, and δ is the distance between the
load point (A) and the conjugate point (C) as shown in Figure 3, given by

δ= p2 + q 2 (17)

where p and q are defined by equation (14c). The reference distance δmax is obtained when the
bubble contacts the bounding surface at a point diametrically opposite to the conjugate point C,
Figure 3,

2(1 − r )
δ max =
r
(V − VN )2 + ( H − H N )2 (18)

The plastic stiffness interpolation parameter ρ changes the rate of stiffness variation with
load change. Appropriate values for this parameter have to be determined from curve-fitting of
sideswipe or probe test results under overloaded (R = Vmax/V0 > 1) conditions. Because the
transformed distance δ/δmax < 1, higher values of ρ will give weaker response.
It is convenient to relate the elastic modulus, Kmax, to the plastic stiffness, kvp, by Kmax = λ3kvp,
where λ3 is a model parameter. The plastic hardening modulus reduces to KC when the two
surfaces are in contact at the current loading point. The expression for KC can be derived by using
the following governing equations:

 F=0

 δF = 0 (19)
δz p = λn
 V

Int. J. Geomech., 2002, 2(4): 419-446


428 J. Zhang et al.

where λ is the plastic multiplier and nV the normalized vertical displacement at the contacting
point. Noting that ∂F/∂H = 1, λ is given by

1 1
λ=
KC f
(a1δV + δH ) (20)
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

The nonassociated flow rule gives

nV = b1 g (21)

where f and g are unit vectors, f = (a12 +1)0.5, g = (b12 +1)0.5, a1 = ∂F/∂V, and b1 = ∂G/∂V.
From the consistency condition, δF = 0, defining c1 = ∂F/∂Vmax, one may write

δH + a1δV = − c1δVmax (22)

Using equations (19), (20), (21), and (22), the plastic modulus at the conjugate point can be
derived,

kve kvp 1
KC = − c1b1 (23)
kve − kvp fg

E. Elastic Deformation
For the plane strain problem described here, the elastic component may be expressed as

δV  e  δz  kve 0  δz 


δH 
[ ]
 = D  =
δx   0
 
khe  δx  (24)

where [De] is the elastic matrix, V and H are the vertical and horizontal loads per meter length
of the pipeline, and kve and khe are the elastic stiffness for vertical and horizontal directions,
respectively (with no cross-coupling between the two). In this model, the horizontal stiffness khe
is related to kvp by khe = λ2kvp, where λ2 is a model parameter.

F. Incremental Relations
The elasto-plastic stiffness matrix may be written as

T
 ∂G  ∂F 
[ ]
D    D e
e
 ∂ p  ∂ p 
[ ]
[ ] [ ]
Dep = De − T
 ∂F   
e  ∂G 
(25)
 ∂ p 
[ ]
  D   + Kfg
 ∂ p 

Int. J. Geomech., 2002, 2(4): 419-446


A Kinematic Hardening Model for Pipeline-Soil Interaction 429

to give

δV  ep δz   a11 a12  δz 


δH 
[ ]
 = D  =
δx  a21
 
a22  δx  (26)

where
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

2
a1b1kve b1kve khe
a11 = kve − ; a12 = −
Kfg + a1b1kve + khe Kfg + a1b1kve + khe

a1kve khe 2
khe (27)
a21 = − ; a22 = khe −
Kfg + a1b1kve + khe Kfg + a1b1kve + khe

The corresponding flexibility matrix may be obtained by inverting equation (26) to give:

δz  −1 δV  b11 b12  δV 


 = D
δ
 
x
ep
[ ]  =
δH  b21
 
b22  δH  (28)

where

a1b1 1 b a 1 1
b11 = + ; b12 = 1 ; b21 = 1 ; b22 = + (29)
Kfg kve Kfg Kfg Kfg khe

III. SINGLE SURFACE MODELS:


COMPARISON WITH THE BUBBLE MODEL

Two single surface models, a conventional elastic-plastic model [5] and a bounding surface
model [6], have been proposed previously. These two models are briefly described below and
compared with the kinematic hardening model.

A. The Conventional Elastic-Plastic Model


In the conventional elastic-plastic model, the yield surface, plastic potential, hardening law,
and elastic response are defined by equations (2a), (4), (5), and (24). Fully elastic response is
assumed inside the yield surface. The elasto-plastic stiffness matrix Dep can be derived using
standard theory to yield

 a1b1kve
2
b1kve khe 
[ D ] =  0
kve 0 1
ep
 −   (30)
khe  K + a1b1kve khe a1kve khe
2
khe 

Int. J. Geomech., 2002, 2(4): 419-446


430 J. Zhang et al.

where

∂F kve kvp ∂F ∂G ∂F
K = − b1 p =− b1c1 , a1 = , b1 = , and c1 =
∂z kve − kvp ∂V ∂V ∂Vmax

It can be seen that this expression is identical to that given by equation (25) in the kinematic
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

hardening model when the bubble and bounding surfaces are in contact. Thus for normally loaded
(R = Vmax/V0 = 1) conditions, the two models give identical results. Therefore the conventional
elasto-plastic model may be viewed as a special example of the two-surface model, where the
yield surface coincides with the bounding surface, r = 1, VN = (1–β) Vmax/2, and HN = 0.

B. The Bounding Surface Model


In the bounding surface model, the bounding surface, plastic potential, hardening law, and
elastic response are defined by equations (2a), (4), (5), and (24). Using the consistency condition
on the bounding surface (δFb = 0), the hardening equation, and nonassociated flow rule, the
plastic modulus at the image point, Kb, can be obtained.

kve kvp b1c1


Kb = (31)
kve − kvp fg

This equation gives results identical to those in equation (23) for the same point on
the bounding surface. For simplicity, a radial mapping rule is employed to associate load point
A (V, H) and image point B (VB, HB), as shown in Figure 5. The plastic modulus at the load point
(V, H) has been expressed as [6]:

FIGURE 5. Illustration of the bounding surface model.

Int. J. Geomech., 2002, 2(4): 419-446


A Kinematic Hardening Model for Pipeline-Soil Interaction 431

ρ
kve kvp  M   Vmax − Vm 
K = Kb +  1+  
η   Vmax − V0 
(32)
kve − kvp 

where Vm is the intercept of the loading surface on the V-axis, as shown in Figure 5. V0 is the initial
vertical load of the current loading event, (Vmax – Vm) is the current reference distance between
the loading surface and the bounding surface and (Vmax – V0) is the maximum reference distance
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

between the initial loading point and the bounding surface. Parameter η is the ratio of H/V and
M is the ratio of H/V where horizontal load peaks; and

 (1 + β)2 
M = 0.5µ  
 (1 − β) 
(33)

Equation (32) guarantees that K has an infinite value at the beginning of unloading when η
is zero (and Vm = V0), to simulate the elastic response, and K = Kb when Vm = Vmax, meaning that
the loading surface and bounding surface are coincident, to simulate the plastic response. In this
equation ρ is a parameter that adjusts the rate of decay of stiffness with displacement and will
generally lie in the range 3 to 5.
However, this equation also gives K = Kb initially in an unloading event starting from the
bounding surface, thus a soft response is predicted in the early stage of unloading, which is not
correct. This situation can be avoided if the following expression is used in describing the
degradation of the plastic modulus.

ρ
 δ 
K = Kb + ( K max − Kb )  (34)
 δ max 

where Kmax is a very large value, δ = H′B – H for loading events, δ = H′B + H for reverse loading
events, and δmax = 2H′B, as illustrated in Figure 5. This equation effectively gives K a large value
at the starting of loading or reverse loading. The value of K decreases gradually and approaching
Kb when the load point reaches the bounding surface. Thus the bounding surface model may be
viewed as another special example of the two-surface model, where the yield surface shrinks to
a point and the elastic response vanishes.

C. Differences between The Three Models


The fundamental difference between the three models is how to describe the load-displace-
ment response inside the bounding surface, i.e., the assumption of elastic range and the transition
from elastic to plastic state. In a conventional elastic-plastic model, the elastic domain expands
to coincide with the bounding surface. When the load point is located inside this surface, fully
elastic response is assumed. The response is transformed to a plastic state abruptly as soon as the
load point is located on the bounding surface. In a bounding surface model, the elastic domain
shrinks to a point. A gradual transition from elastic to plastic state is performed inside the whole
region enclosed by the bounding surface. In between these two extremes, a small elastic range
is assumed in the two-surface model. The elastic region is dragged with the loading point and a

Int. J. Geomech., 2002, 2(4): 419-446


432 J. Zhang et al.

fully elastic response occurs inside the bubble each time the load reverses its direction, regardless
of the amplitude and bias of the loading. A gradual transition from elastic to plastic state is
achieved in the region between the elastic domain and the bounding surface. These differences
are summarized in Table 1.
Because of the differences shown in Table 1, the conventional model could not be used to
simulate pipeline response if the load is cycled inside the bounding surface because it always
gives elastic behavior. The bounding surface model normally can be used under both monotonic
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

and cyclic loading conditions, but may overpredict the cumulated settlement under small ampli-
tude load cycles due to ignoring of the elastic region.
On the other hand, for the improved performance, the number of model parameters must
increase, as can be seen in Table 1. The complexity of the model also increases, not only because
of the selection of an increasing number of parameters but also due to the need for frequent
updating of the size and position of the bubble in the kinematic hardening model.

D. Comparison of the Three Models: Monotonic Loading Events


All three models are capable of simulating monotonic behavior. The differences between
them can be demonstrated more clearly through example calculations, as given below. In the
bubble model, r = 0.01 R and ρ = 2 are adopted. The initial bounding surface model, using
equation (32) to interpolate K with ρ = 5 is used. Other parameter values are listed in Table 2.
A detailed description of model parameters is given in Table 3. Methods as to how to derive these
parameters have been discussed previously [6].

TABLE 1
Differences of the Three Models

Number of parameters
Transition from
Size of the elastic to Monotonic Cyclic Loading
Model elastic region plastic loading loading conditions

Conventional Identical to the Abrupt 8 Nil Monotonic


model bounding
surface
Bounding A point Gradual 9 10 Monotonic
surface and cyclic
model
Two-surface Enclosed by Gradual 11 13 Monotonic
(bubble) the bubble and cyclic
model

Int. J. Geomech., 2002, 2(4): 419-446


A Kinematic Hardening Model for Pipeline-Soil Interaction 433

TABLE 2
Parameters Used in Example Calculations

Parameter Value Unit Parameter Value Unit

kvp 400 kPa µ0 0.4 —


λ1 20 — κ 0.65 —
λ2 µt
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

20 — 0.6 —
λ3 200 — m 0.18 —
β 0.06 —

TABLE 3
Model Parameters

Parameter Description Value

kvp Plastic stiffness of vertical loading 350 kPa


λ1 Ratio of kve to kvp 20
λ2 Ratio of khe to kvp 20
β Intersect of normalized bounding surface on vertical load axis 0.06
µ0 Shape parameter of bounding surface for surface footing 0.4
κ Gradient of parameter µ increase with embedment 0.65
µt Shape parameter in plastic potential equation 0.6
m Exponent in plastic potential equation 0.18
ρ Exponent in the interpolation of plastic modulus 2
ζ A parameter linking c with cumulated settlement 20
cmax The maximum value of c 1.5
λ3 Ratio of Kmax to kvp 200
r0 The scale factor for normally loaded pipelines 0.01

1. Monotonic Horizontal Loading at Fixed Vertical Position: Sideswipe Events


In a sideswipe event, the horizontal load increases initially and decreases after the peak while
the vertical load decreases monotonously until the residual state is reached. This loading event
will approximately follow the bounding surface if the elastic vertical stiffness, kve, is very large
compared to the plastic stiffness kvp [17].
Figure 6 shows the response of a pipeline in three sideswipe events at different overloading
ratios. It shows that the horizontal load increases initially to a peak and then drops to the residual
load. The vertical load decreases monotonically and remains constant when the residual horizon-
tal load is mobilized. An initial elastic response is predicted by the model for the overloaded
cases.
For comparison, pipeline response predicted by the conventional elastic-plastic model and
the bounding surface model is shown in Figure 7. It can be seen that the bounding surface model
gives a gradual transition from elastic to plastic states, resulting in certain reductions in maximum

Int. J. Geomech., 2002, 2(4): 419-446


434 J. Zhang et al.
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

FIGURE 6. Sideswipe events of different R at Vmax = 80 kN/m: (a) horizontal response; (b) loading traces.

horizontal resistance for overloaded pipelines. Lower ρ would lead to softer response and
significant decrease in the peak horizontal load at R > 1.

Monotonic Horizontal Loading at a Constant Vertical Load: Probe Events


For normally loaded or slightly overloaded pipelines (R < 10), the horizontal load and the
vertical displacement increase monotonically in a probe event, as shown in Figure 8. When the
overloading ratio R is larger than about 10, the horizontal load reaches a peak followed by post-
peak strain-softening and accompanied by upward movement. The initial gradient of pipe
movement decreases with the increase of overloading ratio R.
For comparison, pipeline response calculated from the conventional model and the bounding
surface model is shown in Figure 9. It can be seen that the bounding surface model effectively

Int. J. Geomech., 2002, 2(4): 419-446


A Kinematic Hardening Model for Pipeline-Soil Interaction 435
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

FIGURE 7. Simulated sideswipe events at different R with Vmax = 80 kN/m, using elastic-plastic (thinner lines) and
bounding surface (thicker lines) models: (a) horizontal response; (b) loading traces.

gives a gradual transition in horizontal stiffness for all values of R, while the conventional model
gives an abrupt change for R > 1. Pipeline movements calculated from both models are very
similar.

IV. COMPARISON WITH EXPERIMENTAL DATA

A. Experimental Set-Up
Full details of the experimental program have been presented previously [14], but a brief
description of the experimental set-up is given below.

Int. J. Geomech., 2002, 2(4): 419-446


436 J. Zhang et al.
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

FIGURE 8. Probe events of different R at V0 = 10 kN/m: (a) lateral response; (b) pipe movement.

The tests were conducted on the beam centrifuge at The University of Western Australia. The
soil used was calcareous sand recovered from the North West Shelf of Australia. Particle size
distribution for the material, after removing the small quantity of debris, is shown in Figure 10.
The samples had an average saturated unit weight of 18.5 kN/m3, and average void ratio of about
0.95.
The tests involved application of vertical and horizontal loads to a model pipeline segment,
with instrumentation to measure loads and displacements. The general arrangement of the test
apparatus is shown in Figure 11. A 20-mm-diameter model pipeline segment of 160-mm length
and 3-mm wall thickness was fabricated from aluminium. At a scale of 1:50 this segment
represents a prototype pipe segment of 1.0-m diameter and 8.0-m length.

Int. J. Geomech., 2002, 2(4): 419-446


A Kinematic Hardening Model for Pipeline-Soil Interaction 437
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

FIGURE 9. Simulated probe events at different R with V0 = 10 kN/m, using elastic-plastic (thinner lines) and bounding
surface (thicker lines) models: (a) horizontal response; (b) pipe movement.

A number of different tests were undertaken as follows:

• Monotonic vertical penetration to an embedment of about two diameters, without lateral


loading.
• Sideswipe tests, where after penetration to a particular depth, the vertical position of the
pipe was held constant while it was displaced laterally.
• Probe tests, where after penetration to a particular depth, the vertical load was held constant
while the pipe was displaced laterally.
• Vertical cyclic load tests, where the vertical load was cycled between fixed load limits.

Int. J. Geomech., 2002, 2(4): 419-446


438 J. Zhang et al.
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

FIGURE 10. Particle size distribution for the soil used in the experiments.

FIGURE 11. General arrangement of the test package.

• Horizontal cyclic load tests, where after penetration under a particular vertical load, the
vertical load was held constant while the horizontal load was cycled between fixed limits
of equal positive and negative value.

In the probe and sideswipe tests, the pipe was initially penetrated to a particular depth before
holding either the vertical load or position constant, respectively. Where the value of the load or
position being held was equal to the maximum experienced during penetration to that depth, the
tests are referred to as “normally loaded”. Where the vertical load was first reduced prior to
holding load or displacement, the tests are referred to as “overloaded”. The terminology is
intended to be a direct transposition of the normal- and overconsolidation concept, and the use
of an “overloading ratio”, R, is similar to an overconsolidation ratio, but in respect of the loading
history of the pipeline.

Int. J. Geomech., 2002, 2(4): 419-446


A Kinematic Hardening Model for Pipeline-Soil Interaction 439

B. Model Parameters
The bubble model requires 13 independent parameters. Parameter descriptions and values
adopted in the simulations are given in Table 3. All these parameters can be evaluated from back
analysis of a series of loading tests at model or full scale. The background as to how this may
be done and how these parameters can be estimated in the absence of model or field test data has
been given elsewhere [6].
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

C. Key Performance Ratios


Key ratios of pipeline-soil interaction derived from the model and obtained from centrifuge
tests are summarized in Table 4. Because the horizontal resistance peaked at a very small
displacement and the recorded data fluctuated significantly, the initial horizontal stiffness for
sideswipe test was calculated using peak resistance and the corresponding displacement.
It can be seen from this table that the key performance ratios from the theoretical model and
from experiments are reasonably close except for the final penetration in probe events. The main
reason for the discrepancy in ∆z is that a mound was formed in front of the pipeline during probe
test. The lateral resistance increased fairly quickly and approached the ultimate value at a
displacement much smaller than that calculated.

D. Ultimate Soil Resistance


For a normally loaded pipeline, it can be shown that the following equation applies in
predicting the ultimate lateral resistance in probe and sideswipe events [18],

mµ(1 + β)  kvp 
2
H
=   (35)
γ ′z 2 (1 + m)2  γ ′z 

TABLE 4
Key Feature Ratios from Model and Test (z0 = 0 to 0.3 m)

Sideswipe event Probe event

Ratio Model Test Ratio Model Test

Hmax Hmax
Vmax 0.115 – 0.173 0.116 – 0.179 V0 0.571 – 0.857 0.7 – 1.1

V H = H max ∆z
0.48 0.47 z0 8.84 1.5 – 2.5
Vmax
Hres  dz 
Vmax 0.058 – 0.086 0.054 – 0.112  dx  initial 0.59 0.4 – 0.7

 dH   dH 
 dx  initial 1490 – 2020 Around 500  dx  initial 90 – 140 10 – 100

Int. J. Geomech., 2002, 2(4): 419-446


440 J. Zhang et al.

where H = Hres for sideswipe tests and H = Hmax for probe tests. This expression is shown for
comparison with the experimental data in Figure 12 using γ′= 8 kN/m3. The test data are obtained
from both previous [19] and current centrifuge modeling conducted in calcareous sand, and full-
scale model tests performed in silica sand [3]. Several data points obtained from sideswipe test
give lower values for normalized load, probably due to errors in the measurement of small
embedments. Otherwise the expression fits the test data well in a large range of embedment.
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

E. Pipelines under Monotonic Loading


Figures 13 and 14 show comparisons between experimental data and model simulations for
sideswipe and probe events. Good agreement has been achieved for pipe sideswipe events, while
reasonably good agreements are obtained for probe events. In the latter case, the values of peak
soil resistance from tests and model simulations are quite close. The computed displacement
gradients are very similar, although the predicted initial embedment is somewhat different from
that measured.

F. Pipelines under Cyclic Loading


Calculated and measured responses of a pipeline under cyclic loading are compared in
Figures 15 and 16. For a normally loaded pipeline, good agreement was achieved regarding

FIGURE 12. Comparison of ultimate lateral resistance.

Int. J. Geomech., 2002, 2(4): 419-446


A Kinematic Hardening Model for Pipeline-Soil Interaction 441
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

FIGURE 13. Comparison between sideswipe test results [14] and model simulations.

pipeline settlement with number of cycles. Both experimental data and model simulation show
similar trend to the stiffness variation with increasing number of cycles although the model gives
a larger horizontal displacement in the initial stage of loading. The calculated and observed load-
displacement response agrees reasonably well for a probe event carried out after cyclic loading.

V. CONCLUSIONS

A kinematic hardening two-surface model has been developed for pipe-soil interaction of
untrenched offshore pipelines. The hardening behavior is linked to the plastic vertical displace-

Int. J. Geomech., 2002, 2(4): 419-446


442 J. Zhang et al.
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

FIGURE 14. Comparison between probe test results [14] and model simulations.

ment in a linear relationship. The bounding surface takes the shape of a parabola and gives an
intercept on the horizontal load axis. A nonassociated flow rule has been adopted through a
bullet-shaped plastic potential surface. A yield surface similar but smaller than the bounding
surface was proposed to define the elastic region. Interpolation has been made for the plastic
modulus regarding the transformed distances when the load point is located in the area between
the yield and the bounding surfaces.
The model provides a suitable basis for predicting the load-displacement response of
pipelines of various embedments under different loading conditions. It reproduces the general
behavior of pipeline-soil interaction observed from experimental work.

Int. J. Geomech., 2002, 2(4): 419-446


A Kinematic Hardening Model for Pipeline-Soil Interaction 443
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

FIGURE 15. Horizontal response under cyclic loading: (a) test results [14]; (b) model simulations.

FIGURE 16. Pipeline settlement during cyclic loading.

Int. J. Geomech., 2002, 2(4): 419-446


444 J. Zhang et al.

At this stage, only drained behavior has been addressed in this article. However, the model
is being developed further, to simulate the undrained or partially drained behavior of pipelines
under both monotonic and cyclic loading conditions. Although attention has been restricted
mainly to pipelines, or strip foundations, the model also has potential for application to other
shapes of foundation, such as skirted or caisson foundations and flat footings, at least in the V-
H loading plane.
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

ACKNOWLEDGMENTS

The article was prepared when the first author was a research student and the second author
was a Lecturer at The University of Western Australia (UWA). The work described here is a part
of a continuing research program on offshore foundation systems at (UWA). Funding from the
Australian Research Council is gratefully acknowledged. The first author was supported by a
University Postgraduate Award and an Ernest and Evelyn Havill Shacklock Scholarship in Civil
Engineering at The University of Western Australia.

REFERENCES
[1] D.A. Wagner, J.D. Murff, H. Brennodden, and O. Sveggen, Pipe-soil interaction model, OTC 5504, Proc.
19th OTC, 181–190, 1987.
[2] H. Brennodden, J.T. Lieng, T. Sotberg, and R.L.P. Verley, An energy-based pipe-soil interaction model,
OTC 6057, Proc. 21st OTC, 147–158, 1989.
[3] R.L.P. Verley and T. Sotberg, A soil resistance model for pipelines placed on sandy soils, J. OMAE, 116, 145–
153, 1994.
[4] J.R. Hale, W.F. Lammert, and D.W. Allen, Pipeline on-bottom stability calculations: Comparison of two
state-of-the-art methods and pipe-soil model verification, OTC 6761, Proc. 23rd Annual OTC, 567–582, 1991.
[5] J. Zhang, M.F. Randolph, and D.P. Stewart, An elasto-plastic model for pipe-soil interaction of offshore
pipelines in sand, Proc. ISOPE’99, Brest, France, Vol. II, 185–192 (1999).
[6] J. Zhang, D.P. Stewart, and M.F. Randolph, Soil restraint of shallowly embedded offshore pipelines in
calcareous sand, ASCE J. Geotech. Geoenviron. Eng., submitted 2001b.
[7] G.J.M. Schotman, The effects of displacements on the stability of jack-up spud can foundations, OTC 6026,
Proc. 21st Annual OTC, 515–524, 1989.
[8] R. Nova and L. Montrasio, Settlements of shallow foundations on sand, Geotechnique 41, No. 2, 243–256,
1991.
[9] C.M. Martin, Physical and Numerical Modeling of Offshore Foundations under Combined Loads. Ph.D.
Thesis, Oxford University, 1994.
[10] L. Montrasio and R. Nova, Settlements of shallow foundations on sand: geometrical effects, Geotechnique 47,
No. 1, 49–60, 1997.
[11] B.W. Byrne and G.T. Houlsby, Drained behaviour of suction caisson foundations on very dense sand, OTC
10994, Houston, Texas, 1999.
[12] Z. Mroz, V.A. Norris, and O.C. Zienkiewicz, Application of an anisotropic hardening model in the analysis
of elasto-plastic deformation of soils, Geotechnique 29, No. 1, 1–34, 1979.
[13] M. Rouainia and D. Muir Wood, A kinematic hardening constitutive model for natural clays with loss of
structure, Geotechnique 50, No. 2, 153–164, 2000.

Int. J. Geomech., 2002, 2(4): 419-446


A Kinematic Hardening Model for Pipeline-Soil Interaction 445

[14] J. Zhang, D.P. Stewart, and M.F. Randolph, Centrifuge modeling of pipe-soil interaction for offshore
pipelines, I. J. Physical Modelling in Geotechnics, 1, No. 1, 25–39 (2001a).
[15] Z. Mroz, V.A. Norris, and O.C. Zienkiewicz, An anisotropic, critical state model for soils subject to cyclic
loading, Geotechnique 31, No. 4, 451–469, 1981.
[16] G. Gottardi, G.T. Houlsby, and R. Butterfield, Plastic response of circular footings on sand under general
planar loading, Geotechnique 49, No. 4, 453–469, 1999.
[17] F.S. Tan, Centrifuge and Theoretical Modelling of Conical Footings on Sand, Ph.D. Thesis, Cambridge
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

University, 1990.
[18] J. Zhang, Geotechnical Stability of Offshore Pipelines in Calcareous Sand, Ph.D. Thesis, University of Western
Australia, 2001.
[19] L.T.I. Wallace, Pipeline Performance in Calcareous Soil, Honours Thesis, University of Western Australia,
1995.

NOTATIONS

The following symbols are used in this article:


aij, bij = elements in the elastic-plastic matrix;
B = pipe diameter;
c = multiplier;
De, Dep = elastic, elastic-plastic matrixes;
F = bounding surface function;
G = plastic potential function;
f, g = magnitude of the unit vectors on the bounding and plastic potential surfaces;
H = horizontal load component;
HC = Horizontal load at conjugate points;
HM, HN = position of the centers of bounding and yield surfaces on H direction;
K, KC = plastic modulus at load and conjugate points;
Kmax = elastic modulus on the yield surface;
khe = elastic horizontal stiffness;
kve, kvp = elastic, plastic vertical stiffnesses;
m = exponent in plastic potential function;
p = difference in vertical load coordinates of load and conjugate points;
q = difference in horizontal load coordinates of load and conjugate points;
r = proportion between the sizes of yield and bounding surfaces;
r0 = r for normally-loaded pipeline (R = 1);
R = Vmax/V0, overloading ratio;
S = scalar factor;

Int. J. Geomech., 2002, 2(4): 419-446


446 J. Zhang et al.

V = vertical load component;


V0 = initial vertical load during horizontal loading;
VC = vertical load at conjugate points;
VM, VN = center positions of the bounding and yield surfaces on V direction;
Vmax = maximum vertical load experienced by the pipeline;
Downloaded from ascelibrary.org by Cambridge University on 09/21/22. Copyright ASCE. For personal use only; all rights reserved.

x = horizontal displacement;
z = vertical displacement;
z0 = initial embedment;
zp = plastic vertical displacement;
zpc = cumulated plastic vertical displacement induced in cyclic loading;
β = parameter in bounding surface function;
δ = distance between load and its conjugate points;
δmax = the maximum value of δ;
κ = gradient of parameter µ increasing with embedment z0;
λ = plastic multiplier;
λ1 = ratio of kve to kvp;
λ2 = ratio of khe to kvp;
λ3 = ratio of Kmax to kvp;
µ = shape parameter in bounding surface function;
µ0 = µ in bounding surface function for surface footing;
µt = shape parameter in plastic potential function;
ρ = exponent for the interpolation of plastic modulus at load point;
ζ = model parameter linking multiplier c to cumulated settlement during cyclic loading.

Int. J. Geomech., 2002, 2(4): 419-446

You might also like