Mylonakis, G. (2001). GeÂotechnique 51, No.

5, 455±461

Winkler modulus for axially loaded piles

G . M Y L O NA K I S 

The problem of elastic soil±pile interaction and its modelling Nous reprenons le probleÁme de l'interaction eÂlastique entre
using the concept of a Winkler support are revisited. It is sol et pile et sa repreÂsentation en maquette par le concept du
shown that dividing the vertical soil shear tractions and the support de Winkler. Nous montrons qu'en divisant des trac-
corresponding settlements along the pile generates depth- tions de cisaillement de sol verticales et les tassements corre-
dependent Winkler springs that accurately describe the spondants le long de la pile, on engendre des ressorts de
pile±soil interaction, contrary to the widespread belief that Winkler qui deÂpendent de la profondeur, ressorts qui deÂcri-
the Winkler representation is always approximate. A simpli- vent avec exactitude l'interaction pile-sol, contrairement aÁ
®ed theoretical model is then derived for analysing the l'opinion qui veut que la repreÂsentation de Winkler soit
response of an end-bearing cylindrical pile in a homoge- toujours approximative. Nous deÂrivons ensuite une maquette
neous soil stratum. Explicit solutions are obtained for: (a) theÂorique simpli®eÂe pour analyser la reÂaction d'une pile
pile settlement; (b) depth-dependent Winkler modulus; (c) cylindrique porteuse en bout dans une couche de sol homo-
average (depth-independent) Winkler modulus to match pile geÁne. Nous obtenons des solutions explicites pour : (a) le
head settlement. Both in®nitely long piles and piles of ®nite tassement de la pile ; (b) le module de Winkler deÂpendant de
length are examined. The approximate analytical solution la profondeur ; le module moyen de Winkler (ne deÂpendant
compares favourably with ®nite-element and boundary-ele- pas de la profondeur) pour preÂdire le tassement des teÃtes de
ment solutions. A simple regression formula for the average pile. Nous eÂtudions aÁ la fois les piles de longueur in®nie et les
Winkler modulus is developed. piles de longueur ®nie. La solution analytique approximative
montre une bonne correÂlation avec les solutions aÁ eÂleÂment
KEYWORDS: dynamics; elasticity; piles; settlement; stiffness; soil/ ®nis et les meÂthodes marginales. Nous deÂveloppons une sim-
structure interaction. ple formule de reÂgression pour le module de Winkler moyen.

INTRODUCTION subgrade reaction and is measured in units of force per length

A versatile way of modelling soil±pile interaction is through a squared.y It is well known that k(z) is not a property of the soil
series of independent Winkler springs distributed along the pile alone, but depends on the characteristics of both pile and soil,
shaft. Although approximate, Winkler models are widely ac- and varies with depth even in a homogeneous layer.
cepted in the analysis of both axially and laterally loaded piles Under the assumption of negligible Poisson effects in the
subjected to static or dynamic loads (Terzaghi, 1955; Coyle & pile,{ introducing the function k(z) reduces the continuum
Reese, 1966; Novak, 1974; Poulos & Davis, 1980; Randolph & problem to the one-dimensional equation (Scott, 1981)
Wroth, 1978; Scott, 1981; Fleming et al., 1992; Gazetas et al.,
d2 W (z)
1992). Their popularity stems primarily from their ability to Ep Ap ‡ k(z)W (z) ˆ 0 (2)
dz 2
(a) yield predictions that are in satisfactory agreement with which is amenable to analytical treatment, often leading to
more rigorous solutions closed-form solutions (Novak, 1974; Scott, 1981; Guo, 2000).
(b) incorporate variation of soil properties with load amplitude Note that using the exact ratio k(z) ˆ f (z)=W (z) in equation (2)
(non-linearity) and depth (inhomogeneity) reproduces the pile response accurately, contrary to the wide-
(c) be extended to dynamic loads by adding pertinent
spread view that the Winkler representation is always approx-
distributed dampers to the spring bed imate.
(d ) incorporate group effects through pertinent pile-to-pile Current methods for determining k(z) can be classi®ed into
interaction models three main groups:
(e) require smaller computational effort than rigorous ®nite-
element or boundary-element formulations. (a) experimental methods
(b) calibration with rigorous numerical solutions
The key problem in the implementation of Winkler models lies (c) simpli®ed theoretical models.
in the assessment of the modulus of the Winkler springs. In the
case of axially loaded piles, the springs are de®ned as In the experimental methods (group a), k(z) is obtained directly
f (z) from equation (1) by measuring the longitudinal strains along
k(z)  (1) an axially loaded pile and computing analytically the corre-
W (z) sponding distributions of vertical soil reaction and pile settle-
ment (Coyle & Reese, 1966). With the methods of group b,
where f (z) denotes the vertical soil reaction per unit pile length average k values along the pile can be determined by matching
and W (z) the corresponding pile settlement, at depth (z). a key response parameter (e.g. pile head settlement) with results
Following early applications of the model to settlement analysis from Winkler models, as for instance done by Thomas (1980)
of surface footings, k(z) is often referred to as the modulus of and Sanchez-Salinero (1982) using ®nite-element formulations.
Group c consists of approximate analytical methods, notably the
plane-strain model (Novak, 1974; Randolph & Wroth, 1978),
Manuscript received 16 November 2000; revised manuscript accepted 16 which introduce simpli®cations to derive simple theoretical
March 2001. estimates of k.
Discussion on this paper closes 2 November 2001, for further details Notwithstanding the signi®cance of the above methods in
see the inside back cover.
 Civil Engineering Department, City University of New York. engineering research and practice, they all can be criticised for
{ Note that k differs from the so-called coef®cient of subgrade reaction certain drawbacks. For instance, experimentally determined k
(denoted by k s ), which is de®ned as pressure over settlement, and which values (expressed through the well-known `p± y' or `t±z' curves)
thereby has units of force per length cubed. have been developed primarily for inelastic conditions, and do
{ The effect of radial displacements on pile response is usually minor not properly account for the low-strain stiffness of the soil
as, for instance, shown by Mattes (1969) and Pak & Ji (1993). (Reese & Wang, 1996). On the other hand, calibrations with
455
456 MYLONAKIS
rigorous numerical solutions in group b may encounter numer- @(ô rz r) @ó z
ÿ rˆ0 (3)
ical dif®culties in certain parameter ranges, as, for example, in @r @z
the case of long compressible piles (El-Sharnouby & Novak,
1990). Also, these approaches are often limited by the analytical where ó z is the vertical normal stress and ô rz is the associated
and computational complexities associated with the underlying shear stress.
numerical procedures, which can make them unappealing to Fundamental to the approximate analysis presented is the
geotechnical engineers. Finally, the plane-strain theories of assumption that the normal stress, ó z , and shear stresses, ô rz , in
group c involve empirical parameters that need to be calibrated the vertically loaded medium are controlled exclusively by the
with other methods (Randolph & Wroth, 1978), and do not vertical displacement, uz ; the in¯uence of radial displacement,
account for important factors such as the continuity of the ur , on these two stresses is considered to be negligibly small.
medium in the vertical direction and the stiffness mismatch Based on this physically motivated simpli®cation, the stress±
between pile and soil. displacement relations for ó z and ô rz are respectively
With reference to the methods in group c, it seems that a
simple rational model capable of providing improved estimates @uz
of k(z) to be used in engineering applications would be ó z ' ÿM (4)
@z
desirable. In the framework of linear elasticity, an approximate
analytical solution is presented in this paper for an axially @uz
ô rz ' Gs (5)
loaded pile in a homogeneous soil stratum. While maintaining @r
analytical simplicity, the proposed solution has distinct advan-
tages over the existing models in group c. Speci®cally where M is a pertinent constant to be discussed later on.
Equations (4) and (5) were apparently ®rst employed by Nogami
(a) it accounts for the continuity of the medium in both the & Novak (1976) for analysing the corresponding dynamic
horizontal and vertical directions problem. In that work, however, the radial displacement of the
(b) it accounts for pile±soil stiffness ratio, pile length to medium was assumed to be zero. In the present study the
diameter ratio, and compressibility of the soil material assumption would be less restrictive: ur has negligible in¯uence
(c) it does not involve empirical constants. on ó z and ô rz, but is not zero. The importance of this modi®ca-
tion is discussed later on.
Apart from its intrinsic theoretical interest, the proposed approx- From equations (3), (4) and (5), the equation of vertical
imate solution may be used to provide a more rational basis for equilibrium of the soil medium is expressed as
assessing and improving other related methods. @

@uz

@ 2 uz
r ‡ç 2 r ˆ0 (6a)
@r @r @z

PROBLEM DEFINITION AND MODEL DEVELOPMENT where ç is a dimensionless parameter given by

The system considered in this study is depicted in Fig. 1: a M
solid cylindrical pile embedded in a homogeneous soil layer ç2 ˆ (7)
over a rigid base, subjected to an axial head load, P. Both soil Gs
and pile are assumed to be homogeneous, isotropic and linearly Note that if the variation with depth of the vertical normal
elastic. The pile is described by its length, L, diameter, d, and stress ó z is neglected (i.e. @ó z =@z ˆ 0), equation (6) simpli®es
Young's modulus, Ep , and the soil by its Young's modulus, Es , to
and Poisson's ratio, í. Stresses and displacements are assumed
to be uniformly distributed within the pile cross-section. Perfect
 
@ @uz
contact (i.e. no gap or slippage) is considered at the pile±soil r ˆ0 (6b)
@r @r
interface.
With reference to the cylindrical coordinate system of Fig. 1,
which is the governing equation of the plane-strain model. The
the vertical equilibrium of the soil medium in the axisymmetric
solution to this equation is
state of deformation is written as
uz ˆ c1 ln r ‡ c2 (8)
which clearly diverges with increasing radial distance from the
pile. To overcome the problem, Randolph & Wroth (1978) and
Baguelin & Frank (1979) consider an empirically determined
P
`magical radius' around the pile beyond which soil displacement
is assumed to be zero. As will be shown below, the solution to
equation (6a) is free of this problem.
Introducing separation of variables and accounting for the
r
boundary conditions of zero normal tractions at the soil surface
z and bounded displacements at large radial distances from the
origin, equation (6a) yields the solution
σθ
uz (r, z) ˆ BK 0 (áçr) cos áz (9)
τrz
σr where K 0 ( ) denotes the modi®ed Bessel function of zero order
L and ®rst kind, and á is a positive variable. B is a constant to be
uθ = 0 determined from the boundary conditions.
dθ
Because of the approximate nature of the analysis employed,
ut the equilibrium of forces in the horizontal direction is not
σz uz
satis®ed in this approach, nor is the boundary condition of
vanishing shear stresses at the soil surface. Nevertheless, as
d demonstrated in studies of several related probems (Tajimi,
1969; Nogami & Novak, 1976; Veletsos & Younan, 1994), these
violations have typically only a minor in¯uence on the solution.
This will be demonstrated further in this paper through com-
Fig. 1. System considered parisons with results from pertinent numerical studies.
 WINKLER MODULUS FOR AXIALLY LOADED PILES 457
In®nitely long pile shear modulus of the soil material. A problem arising from the
For an in®nitely long pile, the displacements and shear use of this equation is that the solution will exhibit a high
stresses in the medium are obtained by integrating equations (5) sensitivity to Poisson's ratio (recall that the constrained modulus
and (9) over the positive variable á: tends to in®nity as í approaches 0´5), a behaviour that has not
1 been observed in rigorous numerical solutions of such problems
uz (r, z) BK 0 (áçr) cos áz dá (10) (see, for instance, Butter®eld & Banerjee, 1971; Selvadurai &
0 Rajapake 1985). As an alternative, one may assume that the two
1
horizontal normal stresses, ó r and ó è, in the vertically loaded
ô rz (r, z) ˆ ÿGs ç áBK 1 (áçr) cos áz dá (11) medium are zero, which is analogous to the assumption used by
0
Veletsos & Younan (1994) for the laterally-loaded problem. In
With reference to the pile, the differential equation of vertical such a case, equation (18) should be replaced by
equilibrium is
ç2 ˆ 2(1 ‡ í) (19)
2
@ W (z)
ÿEp Ap ˆ ð dô rz (d=2, z) ‡ F(z) (12) Perhaps a better choice for the problem at hand is to consider
@z 2 ó r ˆ 0 and åè ˆ 0, which captures better (though only approxi-
where F(z) represents body forces distributed along the pile mately) the condition of zero tangential displacement in the
axis. For the problem at hand, F(z) is determined by resolving domain. With the latter assumption,
the force at the pile head into equivalent body forces through 2
the Cosine transformation ç2 ˆ (20)
1ÿí
1
2P Results obtained from equations (18)±(20) are compared
F(z) ˆ cos áz dá (13)
0 ð graphically in Fig. 2. It can be seen that the predictions of
equations (19) and (20) remain close over the entire range of í
From equations (10)±(13), and considering perfect bonding values, whereas equation (18) exhibits a singular behaviour as í
at the pile±soil interface [i.e. W (z) ˆ uz (d=2, z)], an explicit approaches 0´5 and ceases to be acceptable. Except where
solution is obtained for the pile settlement W : speci®cally otherwise indicated, the solutions presented herein
1
are based on equation (20).
2P K 0 (çá d=2) cos áz
W (z) ˆ  dá
ð dçGs

Ep Ap ð 0
á2 K0 (çá d=2) ‡ K 1 (çá d=2)
Ep Ap á MODEL VALIDATION
(14) Figure 3 compares results for the stiffness of end-bearing
piles computed with the proposed approximate model and with
four ®nite-element and boundary-element solutions by Poulos &
Davis (1980), Blaney et al. (1975), Sanchez-Salinero (1982),
End-bearing pile and El-Sharnouby & Novak (1990). It can be seen that with
For a pile of ®nite length, one should consider the condition small Ep =Es ratios (Fig. 3(a)) the numerical results are sensitive
of vanishing soil displacement at the base of the soil layer. to the discretisation of the pile. For instance, when a small
Imposing this requirement on equation (9) yields number of elements is used, an increase in stiffness with
ð increasing pile length is observed in some of the solutions for
á ˆ ám ˆ (2m ‡ 1), m ˆ 0, 1 . . . (15) L=d . 50Ðan obviously erroneous trend for end-bearing piles.
2L
El-Sharnouby & Novak (1990) report that a dense discretisation
which corresponds to the solution of the eigenvalue problem (of the order of 50 pile elements) is generally needed to remove
cos(áL) ˆ 0. In addition, in the same spirit as with the in®nitely this anomaly. In contrast, the present solution exhibits a stable
long pile, the pile-head force, P, can be expanded in Cosine behaviour and agrees well with the most rigorous results by El-
series as Sharnouby & Novak. Similar good agreement is observed with
1 large Ep =Es ratios in Fig. 3(b).
X 2P
F(z) ˆ cos á m z (16) With reference to a long hollow pile in homogeneous half-
mˆ0
L space, Table 1 compares results for pile stiffness obtained with
the proposed model and with a rigorous elasto-static solution by
The solution to equation (12) is obtained by replacing the Pak & Ji (1993). Although the two solutions are not strictly
integrals in equations (10) and (11) with corresponding in®nite comparable (hollow against solid pile), the agreement between
sums involving á m : the predictions is very satisfactory, with the average difference
1
2P X K 0 (çá m d=2) cos á m z
W (z) ˆ 4
ð d çGs
 
Ep Ap L mˆ0 2
á m K 0 (çá m d=2) ‡ K 1 (çá m d=2)
Ep Ap ám
(17) Constrained medium:
3 equation (18)
where á m is given by equation (15). The above equation can be
obtained directly from equation (14) by replacing the integral Partially constrained
with an in®nite sum, and the factor ð outside the integral with medium: equation (20)
L, to account for the differences in the forcing functions F(z). η 2 2
Unconstrained medium:
equation (19) 1·73
√2
Determination of coef®cient ç
Mention has been already made of the Nogami & Novak 1
(1976) dynamic solution based on the assumption of vanishing
radial displacement, ur . The static part of that solution can be
deduced from equation (17) by assigning factor ç the value
0
2 2(1 ÿ í) 0·1 0·2 0·3 0·4 0·5
ç ˆ (18) Poisson's ratio, ν
1 ÿ 2í
which expresses the ratio of the constrained modulus to the Fig. 2. Sensitivity of compressibility coef®cient, ç, to Poisson's ratio
458 MYLONAKIS
6 1 
K1 (çá d=2) cos áz dá
k(z) ˆ ðd çGs
Poulos 0 ð dçGs
(10 elements) á[K 0 (çá d=2) ‡ K 1 (çá d=2)]
Ep Ap á
1
Blaney K0 (çá d=2) cos áz dá
(20 elements)
(21)
5 0 ð dçGs
El-Sharnouby á2 [K 0 (çá d=2) ‡ K 1 (çá d=2)]
Ep Ap á
(50 elements)
For piles of ®nite length, the corresponding solution is
1
Normalised pile stiffness Pd/2W(0)EsAp


Salinero X K 1 (çá m d=2) cos á m z
(>20 elements) k(z) ˆ ð dçGs
ð dçGs
4 mˆ0 á m [K 0 (çá m d=2) ‡ K 1 (çá m d=2)]
Ep Ap á m
1
X K 0 (çá m d=2) cos á m z
(a) ð dçGs
25 mˆ0 á2 [K 0 (çá m d=2) ‡ K 1 (çá m d=2)]
m
Ep Ap á m
Unconstrained medium: equations (17) and (19)
Partially constrained medium: (22)
equations (17) and (20)
With reference to an ini®nitely long pile, the variation of
20 k(z) with depth is presented in Fig. 4 as a function of pile±soil
stiffness ratio, Ep =Es . A decreasing trend with depth is ob-
served in all curves. For points located between about 3 and 20
pile diameters from the surface, k(z) varies between approxi-
mately one and two times Gs , which is in agreement with
values reported in the literature (Thomas, 1980; Sanchez-
15
Salinero, 1982; Fleming et al., 1992). With small Ep =Es ratios,
Blaney k(z) tends to increase close to the surface but decreases more
Poulos
El-Sharnouby rapidly with depth. The singularity observed at z ˆ 0 is
Salinero analogous to that encountered in elastic analyses of surface
25 50 75 100 125
Dimensionless pile length, L/d 0
(b)
500 Ep/Es = 100
Fig. 3. Normalised stiffness of end-bearing piles in a homogeneous
103
soil stratum over rigid bedrock; comparison of the proposed 5
approximate model with results from four numerical solutions. (a)
Soft piles, Ep =Es 100; (b) stiff piles, Ep =Es 1000. Modi®ed from
104
El-Sharnouby & Novak (1990); í 0:5
Depth, z/d

10

being about 5%. The minor effect of the pile Poisson's ratio on 15
the solution is evident.

20
EVALUATION OF WINKLER MODULUS 0·5 1·0 1·5 2·0 2·5 3·0 3·5
For an in®nitely long pile, the Winkler modulus, k(z), is Normalised Winkler modulus, k/Gs
obtained by dividing the side shear traction, f [ˆ ð dô rz (d=2)]
(equation (11)), by the corresponding pile settlement, W (equa- Fig. 4. Variation with depth of Winkler modulus for an in®nitely
tion (14)). Accordingly, long pile in homogeneous halfspace; í 0:5

Table 1. Normalised stiffness of a long hollow pile of wall thickness t and Poisson's ratio íp
in a homogeneous halfspace. Comparison of the proposed approximate model with a
rigorous elasto-static solution by Pak & Ji (1993); í 0:25, t=d 0:05
íp log(Gp =Gs ) Ep  =Es { Normalised pile Pd Difference
stiffness: 2W (0)Es Ap (b) ÿ (a)
:%
Pak & Ji (1993) Proposed model (a)
(a) (b)
0 1´0 1´52 0´94 0´98 4´2
1´5 4´81 1´33 1´44 8´3
2´0 15´2 2´00 2´21 5´5
2´5 48´1 3´23 3´48 7´7
0´25 1´0 1´90 1´04 1´05 1´0
1´5 6´01 1´47 1´56 6´1
2´0 19´0 2´31 2´41 4´3
2´5 60´1 3´64 3´81 4´7
{ Ep  ˆ Ep [1 ÿ (1 ÿ 2t=d)2 ] denotes the Young's modulus of an `equivalent' solid pile having the
same axial rigidity as the hollow pile.
 WINKLER MODULUS FOR AXIALLY LOADED PILES 459
9ÿ2
footings, and has been reported in the literature (Pak & Ji,
8
ð2 > 1 K0 (çá d=2) dá >
1993). k ˆ Ep Ap < =
ð

4 0 2
dçG s
The effect of pile length on k(z) for end-bearing piles is : á K 0 (çá d=2) ‡ E A á K 1 (çá d=2) >
> ;
examined in Fig. 5. It can be seen that in a short pile k(z) is p p
always larger than in a more slender pile having the same (26)
Ep =Es ratio. For instance, with L=d ˆ 15, k(z) can exceed the
value 2Gs over the entire pile length, which is more than twice
that of the corresponding in®nitely long pile. The decreasing Results obtained from the above expressions are plotted in
trend with depth is analogous to that observed in Fig. 4. Fig. 6. The following points are worthy of note. For the pile
lengths of the most practical interest (say 15 , L=d , 50), k
varies between about 2´7 Gs and 1´8 Gs , and tends to decrease
with increasing Ep =Es and L=d. In the limiting cases of
AVERAGE (DEPTH-INDEPENDENT) WINKLER MODULUS Ep =Es ! 1 and Ep =Es ! 0, it can be shown from equations
A common approximation in Winkler analyses is that the (25) and (26) that k tends to zero and in®nity respectively. For
ratio k(z)=Gs is constant along the pile length. While this slenderness ratios less than about 50, k is practically indepen-
introduces some error in the solution, it usually simpli®es the dent of pile±soil stiffness ratio. In addition, the effect of
analysis by allowing equation (2) to be solved in closed form Poisson's ratio on the solution was found to be of secondary
within a homogeneous soil layer. Corresponding average Wink- importance, as shown in Fig. 7.
ler moduli can be derived by matching a key response para- In Fig. 8, results from the model are compared graphically
meter (e.g. pile head settlement) with results from Winkler against four empirical expressions from the literature. These
formulations. For instance, assuming k=Gs to be constant within expressions represent average Winkler moduli obtained by curve
a homogeneous layer over rigid rock, the solution to equation ®tting based on the numerical solutions by Banerjee, Blaney,
(2) is (Scott, 1981) and Poulos (see list of references). In general, the predictions of
the model lie close to the average of the empirical values.
P Nevertheless, it is evident that these formulae do not satisfy the
W (z) ˆ (cosh ëz tanh ëL ÿ sinh ëz) (23)
Ep Ap ë limiting behaviour of the solution for small and large values of
Ep =Es and L=d, as discussed above. An improved regression
where ë is a parameter (units of 1=length) given by formula is presented below.
s
k
ëˆ (24)
Ep Ap
Enforcing the settlements at the pile head in equations (17) and 3
(23) to be equal, the following implicit solution for k is L/d = 15
obtained:
Winkler modulus, k/Gs

25
s
k 2
50
tanh L
Ep Ap
100
s
k ∞
L
Ep Ap 1
1
2 X K 0 (çá m d=2)
ˆ 2 (25)
ð dçGs
 
L mˆ0 2
á m K 0 (çá m d=2) ‡ K 1 (çá m d=2)
Ep Ap á m 0
102 103 104
which can be easily solved iteratively once the value of the Pile–soil relative stiffness, Ep/Es
right-hand side has been determined.
For an in®nitely long pile, setting tanh(hL) ˆ 1 in equation Fig. 6. Depth-independent (`average') Winkler modulus for end-
(23) and using equation (14) leads to the explicit solution bearing piles in a homogeneous soil layer over rigid bedrock; í 0:5

0
2

10 L/d = 15

25
Winkler modulus, k/Gs
Depth, z/d

20
50
1 1/2
30

ν = 1/3
∞
40
1/4

50
1·0 1·5 2·0 2·5 3·0 3·5 0
Normalised Winkler modulus, k/Gs 102 103 104
Pile–soil relative stiffness, Ep/Es
Fig. 5. Variation with depth of Winkler modulus for end-bearing
piles in a homogeneous soil layer over rigid bedrock; Ep =Es 1000, Fig. 7. Effect of soil Poisson's ratio on average Winkler modulus for
í 0:5 an in®nitely long pile in homogeneous halfspace
460 MYLONAKIS
3·5 2´7 Gs and 1´8 Gs , and tends to decrease with increasing
Ep =Es and L=d. In the limiting cases of very stiff
(Ep =Es ! 0) and very soft piles (Ep =Es ! 1), k tends
to zero and in®nity respectively. On the other hand, k
3·0 remains ®nite with in®nitely long piles, while it tends to
in®nity as L=d ! 0. k tends to increase with increasing soil
Sanchez-Salinero (1982): Poisson's ratio, but this increase is of secondary importance.
7·5(Ep/Es)–0·2
Normalised Winkler modulus, δ

2·5
Thomas (1980):
2·55(Ep/Es)–0·033 This paper NOTATION
(Eqn 27) (Eqn 25) Ap pile cross-sectional area
2·0 B integration constant
d pile diameter
Gp , Ep pile shear modulus, Young's modulus
Gs , E s soil shear modulus, Young's modulus
f (z) vertical soil reaction per unit pile length
1·5 F body force along pile axis
Modified k, k(z) Winkler modulus of subgrade reaction
Randolph & Wroth (1978): ks coef®cient of subgrade reaction
<2π/ln(δL/d) L pile length (= soil thickness)
O'Rourke & Dobry (1978): M soil compression modulus
1·0
1 + 5·5(L/d)–0·74 + 0·6(L/d)1·3(Ep/Es)–1·04 P pile head load
r radial coordinate
t wall thickness of hollow pile
uq , ur , uz soil displacement components
0·5 W (z) pile displacement (settlement)
100 1000 10000 z vertical coordinate
Pile–soil relative stiffness, Ep/Es
á, ám positive variable
Fig. 8. Depth-independent Winkler modulus for end-bearing piles in ç compressibility coef®cient
a homogeneous soil layer over rigid bedrock. Comparison of results ë Winkler parameter
from the proposed model with four empirical expressions from the í soil Poisson's ratio
literature: L=d 50; í 0:5 íp pile Poisson's ratio
ôrz soil shear stress
ó r, ó q, ó z soil normal stresses

NON-LINEAR REGRESSION ANALYSES

For practical applications, it appears desirable to have a REFERENCES
simpli®ed formula to estimate k. Using non-linear regression Baguelin, F. & Frank, R. (1979). Theoretical studies of piles using the
analyses based on the Levenberg±Marquardt method (Bevington ®nite element method. In Numerical methods in offshore piling, pp.
& Robinson, 1992), the following equation was derived: 83±91. London: Institution of Civil Engineers.
Bevington, P. R. and Robinson, D. K. (1992). Data reduction and error
 ÿ1=40 "  ÿ0:6 # analysis for the physical sciences. New York: McGraw Hill.
k E p L
' 1:3 1‡7 (27) Butter®eld, R. & Banerjee, P. K. (1971). The elastic analysis of
Gs Es d compressible piles and pile groups. GeÂotechnique 21, No. 1, 43±60.
Blaney, G. W., Kausel, E. & Roesett, J. M. (1975). Dynamic stiffness of
which corresponds to the depth-independent modulus in equa- piles. Proc. 2nd Int. Conf. Num. Methods Geomech., Blacksburg 2,
tions (25) and (26). It is evident that this equation satis®es the 1010±1012.
limiting behaviour of the solution, i.e. k ˆ 0 at Ep =Es ! 1; Coyle, H. M. & Reese, L. C. (1966). Load transfer for axially-loaded
k ! 1 at Ep =Es ! 0 or L=d ! 0; k ˆ finite as L=d ! 1. piles in clay. J. Soil Mech. Found. Div., ASCE 92, No. SM2, 1±26.
The correlation coef®cient between the values in Fig. 6 and El-Sharnouby, B & Novak, M. (1990). Stiffness constants and inter-
those computed with equation (27) is 0´973, and the standard action factors for vertical response of pile groups. Can. Geotech. J.
27, 813±822.
error is 0´114. Fleming, W. G. K., Weltman, A. J., Randolph, M. F. & Elson, W. K.
(1992). Piling engineering, 2nd edn. John Wiley & Sons.
Gazetas, G. et al. (1992). Seismic response of soil±pile±foundation±
CONCLUSIONS structure systems: some recent developments. In Piles under dy-
An approximate analytical model was developed for the soil namic loads, Geotech. Special Publ. No. 34, pp. 56±93. American
Society of Civil Engineers.
reaction along an axially loaded pile in a homogeneous elastic
Guo, W. D. (2000). Vertically-loaded pile in Gibson soil. J. Geotech.
soil stratum. The model was shown to be reasonably accurate Engng, ASCE 126, No. 2, 189±193.
and exhibit stable behaviour over a wide range of parameters Mattes, N. S. (1969). The in¯uence of radial displacement compatibility
without developing the numerical instabilities with long piles on pile settlement. GeÂotechnique 19, No. 1, 157±159.
observed in numerical solutions. The main conclusions of the Nogami, T. & Novak, M. (1976). Soil±pile interaction in vertical
study are: vibration. Earthquake Engng & Struct. Dyn 4, 277±293.
Novak, M. (1974). Dynamic stiffness and damping of piles. Can.
(a) k(z) is singular at the pile head, and tends to decrease with Geotech. J. 11, No. 4, 574±598.
depth. With long piles and at depths between 3 and 20 pile O'Rourke, M. J. & Dobry, R. (1978). Spring and dash pot coef®cients
diameters, k(z) varies between approximately one and two for machine foundation on piles, American Concrete Institute,
times the soil shear modulus, Gs . With end-bearing piles, Detroit, SP-10, 177±198.
k(z) increases with decreasing pile length, and may exceed Pak, R. Y. S. and Ji, F. (1993). Rational mechanics of pile±soil
the value of 2 Gs over the entire pile length. interaction. J. Engng Mech., ASCE 119, No. 4, 813±832.
Poulos, H. G. & Davis, E. (1980). Pile foundation analysis and design,
(b) Depth-independent (`average') Winkler moduli can be John Wiley & Sons.
obtained by matching predictions of pile head settlement Randolph, M. F. & Wroth, C. P. (1978). Analysis of deformation of
with results from Winkler formulations. In the parameter vertically loaded piles. J. Geotech. Engng, ASCE 104, No. 12,
range of most practical interest (say 100 , Ep =Es , 10 000 1465±1488.
and 15 , L=d , 50), k ranges between approximately Reese, L. C. & Wang, S. T. (1996). Computer program GROUP:
 WINKLER MODULUS FOR AXIALLY LOADED PILES 461
Analysis of a group of piles subjected to axial and lateral loading. GeÂotechnique 5, No. 4, 297±326.
Austin, Texas: Ensoft, Inc. Tajimi, H. (1969). Dynamic analysis of a structure embedded in an
Sanchez-Salinero, I. (1982). Static and dynamic stiffness of single piles, elastic stratum. Proc. 4th WCEE, Chile.
Geotech. Engng Rep. GR82-31. Austin: University of Texas. Thomas, G. E. (1980). Equivalent spring and damping coef®cient for
Scott, R. F. (1981). Foundation analysis. Prentice Hall. piles subjected to vertical dynamic loads. MS thesis, Rensselaer
Selvadurai, A. P. S. & Rajapakse, R. K. N. D. (1985). On the load- Polytechnic Institute, New York.
transfer from a rigid cylindrical inclusion into an elastic half-space. Veletsos, A. S. & Younan, A. H. (1994). Dynamic soil pressures on
Int. J. Solids Struct. 21, 1213±1229. rigid cylindrical vaults. Earthquake Engng Struct. Dyn. 23, 645±
Terzaghi, K. (1955). Evaluation of coef®cients of subgrade reaction. 669.