See

discussions, stats, and author profiles for this publication at: http://www.researchgate.net/publication/235742744

Reliability of rigid piles subjected to lateral

loads
ARTICLE in ARCHIVES OF CIVIL AND MECHANICAL ENGINEERING MAY 2012
Impact Factor: 1.79 DOI: 10.1016/j.acme.2012.04.007

CITATIONS

READS

468

2 AUTHORS:
Wojciech Pua

Adrian Rozanski

Wroclaw University of Technology

Wroclaw University of Technology

21 PUBLICATIONS 52 CITATIONS

25 PUBLICATIONS 17 CITATIONS

SEE PROFILE

All in-text references underlined in blue are linked to publications on ResearchGate,

letting you access and read them immediately.

SEE PROFILE

Available from: Adrian Rozanski

Retrieved on: 05 November 2015

archives of civil and mechanical engineering 12 (2012) 205218

Available online at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/acme

Reliability of rigid piles subjected to lateral loads

W. Puan, A. Rozanski
Wrocaw University of Technology, Faculty of Civil Engineering, Institute of Geotechnics and Hydrotechnics,
Wybrzeze Stanisawa Wyspianskiego 27, 50-370 Wrocaw, Poland

art i cle i nfo

ab st rac t

Article history:

In this paper the complete solution to the problem of random lateral bearing capacity of rigid

Received 23 April 2012

piles has been presented. The suggested solution is based on the limit states theory approach

Accepted 26 April 2012

proposed by Brinch Hansen [2]. A revised approach utilising the response surface method is

Available online 4 May 2012

proposed and compared with the solution presented in earlier papers. Both cases of non-

Keywords:

cohesive and cohesive soils are studied. In addition, the influence of spatial averaging is also

Reliability index

analysed. Numerical algorithms, for both cases of cohesive and cohesionless soils, have been

Response surface

developed in order to evaluate probability of lateral bearing capacity exceeding. As it has been

Soil strength parameters

demonstrated random fluctuations of soil properties can cause significant changes in the
value of ultimate lateral loading determined according to the Brinch Hansen method. Series of
numerical examples under consideration give some important conclusions concerning an
influence of soil properties randomness on safety evaluations.
& 2012 Politechnika Wrocawska. Published by Elsevier Urban & Partner Sp. z.o.o. All rights
reserved.

1.

Introduction

Methods elaborated in this paper have been inspired by many

computations carried out when the modernisation of main
railway tracks in Poland has been started. In almost all cases
there was a need to replace existing foundations of overhead electrical transmission lines supports by new foundations. One of possible solutions was to apply single pre-cast
concrete piles as a foundations of electrical line supports (as
demonstrated in Fig. 1). In many cases, due to lengths of piles
and soil conditions, piles had to be considered as rigid ones.
Therefore the lateral ultimate soil resistance has been considered. The ultimate lateral resistance of the soil in the
vicinity of a pile is not very often treated in the engineering
practice. This is due to fact that majority of piles used in
foundation engineering are not rigid [1]. Therefore the probabilistic approach, in this area, is almost forgotten in the
literature. Among deterministic methods the solution proposed by Brinch Hansen [2] is considered as a one of the most
precise, however simplified approaches elaborated by Broms

[3,4] are more commonly used. The vital point in the Brinch
Hansesn method is the position of the rotation centre zr of
the pile under consideration (see Fig. 2). It is possible to prove
that the ultimate horizontal loading Hu is highly sensitive to
the location of the centre of rotation zr. The rotation centre
itself is subjected to random fluctuations due to inherent
uncertainty of soil properties (mostly the strength parameters) as well as uncertainty in geotechnical recognition.
Therefore a probabilistic approach to this problem seems to
be quite adequate. On the other hand solutions of deterministic problem could not be written in a closed mathematical
form. Due to this difficulty standard probabilistic approaches
could not be applied straightforward. A solution of the
reliability problem associated with laterally loaded piles is
the objective of this paper.
The problem of the reliability of rigid piles subjected to lateral
loads in the context of piles capacity has been discussed in
earlier works [5,6]. In the papers an algorithm for evaluating
reliability indices (and corresponding failure probabilities),
based on the SORM method [7], was introduced. The algorithm

Corresponding author. Tel.: 48 502209006; fax: 48 713284814.

E-mail address: Wojciech.Pula@pwr.wroc.pl (W. Pua).

1644-9665/$ - see front matter & 2012 Politechnika Wrocawska. Published by Elsevier Urban & Partner Sp. z.o.o. All rights reserved.
http://dx.doi.org/10.1016/j.acme.2012.04.007

206

archives of civil and mechanical engineering 12 (2012) 205218

Fig. 2 A scheme of rigid pile embedded in a soil and

subjected to lateral load.

Fig. 1 An example of overhead electrical transmission line

support founded on a single pile (railway line near Wrocaw,
Poland).
was also supported by symbolic computations utilising Mathematica [8] software as well as some power series expansions.
This solution in the next parts of the present paper will be, for
simplicity, called the symbolic algorithm. The application of the
symbolic algorithm in many computational examples shows,
however some important inconveniences that are described in
further sections of the present paper. Moreover the symbolic
algorithm has given a solution solely for cohesionless soils.
The present paper is a comprehensive study on the reliability
problem mentioned above.
The paper is organised as follows. First, for better understanding the Brinch Hansen method, it is shortly described. In
next Section the probabilistic approach is formulated. Next, the
static equations corresponding to the problem are solved. Then,
for comparison with further results, the symbolic algorithm
is shortly presented. Finally the new approach based on the
response surface method [9,10] is demonstrated. The new
approach covers both cases, namely a cohesionless and a
cohesive soil and allows to overcome numerical difficulties that
appeared in the symbolic algorithm. Moreover the effect of
spatial averaging of soil properties has been also discussed.

2.

Formulation of the problem

2.1.

Deterministic approach

Consider a rigid pile subjected to a lateral load as presented

in Fig. 2. The mechanism associated with the failure assumes
a rotation of the pile around centre O due to external load as
well as reaction of surrounding soil. Assume that Hu and Mu,
denote the ultimate lateral load and ultimate value of the
moment, respectively, inducing an ultimate ground resistance pu(z) on the depth z.
Treating the pile as a strip of the width D (or diameter D in
the case of a pile of circular cross-section) and the length L
the following equilibrium equations can be written:
Z zr
Z L
pu zD dz
pu zD dz
1
Hu
0

zr

M u Hu e 

Z
0

zr

pu zDz dz

L
zr

pu zDz dz

where zr is the ordinate of the rotation centre as indicated in

Fig. 2. The above equations have to be solved with respect to
unknown ordinate zr of the rotation centre and the ultimate
value of the lateral load Hu.
It is evident that the solution of Eqs. (1) and (2) depends
on the resultant (passive minus active) ultimate lateral soil
resistance pu(z) distributed along the piles length. In the
simplest case, i.e. under assumption that the resistance pu(z)
does not depend on z, pu(z) pu, the solution of Eqs. (1) and (2)
takes the following form:


1 Hu
L
3
zr
2 pu D
Hu

pu DL

s




2e 2
2e
:
1
1 1
L
L

archives of civil and mechanical engineering 12 (2012) 205218

It is evident, however, that the simplest approach does not

reflect any real situation in the natural soil. Therefore an
application of some sophisticated method is necessary.

2.2.

The Brinch Hansen method

As it has been already mentioned as one of the best closedform approaches the Brinch Hansen solution [2] is considered.
Due to fact that the probabilistic solution proposed in this paper
is utilising the Brinch Hansen approach its short description is
presented below. Following the Brinch Hansen proposal the
ultimate lateral soil resistance pu(z) (understood as passive
minus active pressure), on a certain depth z, is given by
pu z qzKq z cKc z

where q(z) is the effective vertical overburden pressure at the

depth z:
0

qz p gzd g zs ;

z zd zs




z
p 1
cK0c 1 2 sin
f
D
4 2

In the equations above g, g stands for the unit weights

above and below water table, c is the cohesion and Kq(z), Kc(z)
denote the pressure coefficients. They are the functions of
friction angle f, the ordinate z as well as the diameter of a
pile D. In order to evaluate values of Kq(z) and Kc(z) Brinch
Hansen proposed certain non-linear interpolation formulae
combining solutions for the level z0 and solutions for great
depth (z N). For the ground level (z 0) K0q and K0c are
accepted as the difference of passive and active pressure
coefficients for a rough wall subjected to lateral displacements [11], namely:




p f
p f
e1=2pftgf cos ftg 
K0q e1=2pftgf cos ftg
4 2
4 2
7


 
p f
1 ctgf
K0c e1=2pftgf cos ftg
4 2

For a great depth slip lines associated with failure do not

arrive to the ground level. This case can be treated similarly
to strip deep foundation [12]. It can be shown [2] that:
1
1
1
p1
u z c qzK0 tgfNc dc qzKq cKc

where
4
d1
c 1:58 4:09tg f

Nc is the bearing capacity coefficient determined as



 
p f

1 cot f
Nc ep tan f tan2
4 2

10

Kq z-K0q

K0 1sin f:

Kq z-K1
q

if

z-0

15

and Kc z-K1
c

if

z-1:

16

Moreover, for intermediate depths the pressures should be

approximately equal to that resulting from (14). Coefficients in
(14) satisfy the condition (15) but do not satisfy the condition
(16). Therefore Brinch Hansen has suggested the following
interpolating formulae [2]:
Kq

K0q K1
q aq z=D
1 aq z=D

aq

K0q
1
Kq K0q

Kc

K0c K1
c ac z=D
1 ac z=D

ac

K0c
0
K1
c Kc

K0 sin f
sinp=4 f=2



p f
2 sin
4 2

17

18

19

20

The equations above satisfy the conditions (15) and (16) and
give similar pressure values in comparison with the values
obtained by Eq. (14) for intermediate depth z. In this way it is
possible to determine the pressure pu(z) along the piles length.
Nevertheless, it is necessary to remember that computer
facilities were not sufficiently well spread among engineers
when the Brinch Hansen method was presented. Therefore,
due to complex dependences in equations, Brinch Hansen has
suggested an iterative solution of the equilibrium Eqs. (1) and
(2) as well as he elaborated specially dedicated charts in order
to evaluate coefficients Kq and Kc [2]. Nowadays there is no
need to apply them as the solution of (1) and (2) basing on
Brinch Hansen approach can be easily solved by applying any
of computer solvers.
According to Brinch Hansen proposal a safety of the pile
with respect to lateral capacity is evaluated by the global
safety factor F defined as

11

12

and Kc z-K0c

and

and K0 is a pressure coefficient at rest, given by the following

equation:

14

The Brinch Hansen idea was to obtain such expressions for

depth depending pressure coefficients Kq(z) and Kc(z) that
satisfy the following limiting conditions:

6
0

207

Hu
Ha

21

where Hu is the ultimate lateral force (resulting from Eqs. (1)

and (2)) and Ha is an externally applied lateral force.

3.

Probabilistic formulation

Eq. (9) shows directly that

1
K1
c Nc dc

1
and K1
q Nc dc K0 tgf:

13

In the case of moderate depths Brinch Hansen method

assumes passive Rankine state. By projection of all acting forces
on the plane inclined by angle f to the slip plane the resultant
pressure (passive minus active) on the depth z is obtained as


z
K0 sin f
0
pm
u z qzKq 1
D sinp=4 1=2f

Preliminary numerical studies have evidently showed high

sensitivity of solutions of equilibrium Eqs. (1) and (2), namely
the rotation centre zr and the ultimate loading force Hu, to a
value of the friction angle f. Some results are demonstrated
in Table 1.
On the other hand, it is well known that random fluctuations of soil properties in natural deposits are usually significant and very important in the context of engineering

208

archives of civil and mechanical engineering 12 (2012) 205218

Table 1 Location of the rotation centre zr and values of

the ultimate fore Hu versus variations of the internal
friction angle /.

All integrals in Eq. (24) can be evaluated in a closed form

and consequently Eq. (24) can be reduced to the following
one:

Angle of internal
friction [deg.]

Location of the
rotation centre zr [m]

Ultimate lateral
force Hu [kN]

a0 a1 zr a2 z2r a3 z3r b1 lnD aq zr

cb2 lnD ac zr 0

26.35
27.52
28.63
29.78
30.93
32.10
33.22

2.106
2.110
2.113
2.117
2.121
2.125
2.129

13.30
14.61
16.07
17.67
19.45
21.43
23.62

where c is the soil cohesion, D is the pile diameter and the

coefficients aq and ac are given by Eqs. (18) and (20). The
coefficients a0, a1, a2, a3 are expressed by relations:
a3

In the above relationship Ha and Hu are random variables.

However random variability of Hu is implied by random character
of soil properties as well as possible uncertainty in geometrical
features. An equivalent reliability measure, which is usually
more relevant to engineering design, is the reliability index b.
One-to-one correspondence between probability of failure and
the reliability index is given by the following equation:
23

where F0 is the standard normal cumulative distribution function. The b index will be utilised in further considerations within
this paper. In our case, however, evaluation of reliability measures, pF and b, is not an easy task. This is due to fact that there
is no explicit relation between ultimate resistance Hu and soil
properties treated as random variables.

4.
Solution of equilibrium equations provided
by Brinch Hansen approach
In order to evaluate reliability measures (assuming Brinch
Hansen approach to lateral capacity) it is necessary to solve
equilibrium Eqs. (1) and (2) provided that the ultimate lateral
soil resistance pu(z) is determined by evaluation specified in
Section 2.2. Multiplying both sides of Eq. (1) by e and adding
side to side to Eq. (2) one obtains:
Z L
Z zr
pu ze zD dz
pu ze zD dz 0
24
0

zr

Eq. (24) has to be solved with respect to unknown ordinate

of the rotation centre zr. In the case of Brinch Hansen
approach the soil resistance pu(z) is evaluated as
pu z p gzKq z cKc z

25

In the equation above the pressure coefficients Kq(z), Kc(z)

are evaluated according to Eqs. (17) and (18). Additionally, if
the effect of ground water is neglected then Eq. (6) reduces to
qz p gz

28

1
a2 eg pK1
q cKc

computations. If some soil properties are a subject of random

variability then a natural question arises how reliable the total
safety factor (21) is. Then, an appropriate reliability problem can
be formulated as follows: what is the probability (pF) that the
applied loading Ha exceeds the ultimate lateral resistance Hu?


Hu
o1 PfFo1g
22
pF PfHa 4Hu g P
Ha

pF F0 b

2
1
gK1
q cKc
3

26

27

gD 0 1
gcD 0 1
K K
Kc Kc
aq q q
ac



2D 0 1
gD
1
a1 2epK1
Kq Kq eg p
q cKc
aq
aq


2cD 0 1
gD
Kc Kc eg p

ac
ac

29

30



1
eg p 1
gD 0 1
1 3
Kq cK1
K K
a0  gK1
q cKc L
c
3
2
2aq q q


  
D 0 1
gD
Dg 0 1 2
1
c
Kc Kc L  epK1

cK

K
K

eg

p
q
c
2ac
aq q q
aq

 
cD
gD
L
K0c K1
c eg p
ac
ac
(
)
!
D 0 1
gDe pD gD2

Kq Kq ep

2 lnD2 aq LD
aq
aq
aq aq
!
(
)
D 0 1
gDe pD gD2
c
Kc Kc ep

2 lnD2 ac LD
31
ac
ac
ac
ac
Additionally the coefficients accompanying logarithmic
terms take the form:
!
2D 0 1
Dge Dp gD2
Kq Kq ep

2
32
b1
aq
aq
aq aq

b2

2D 0 1
Dge Dp gD2
Kc Kc ep

2
ac
ac
ac
ac

!
33

Note that Eq. (27) cannot be solved in a closed form and

therefore a numerical procedure is necessary in order to
evaluate the rotation centre zr. Assuming that zr has been
previously evaluated one finds the ultimate lateral force in
the following form:
000

000

000

Hu a00o ao c a001 a1 czr a002 a2 cz2r

000

b001 lnD aq zr b1 c lnD ac zr

34

where
a002 DgK1
q
000

35

a2 DgK1
c

36



Dg o
1

K
K

a001 2D pK1
q
aq q q

37



Dg o 1
000
a1 2D pK1
Kc Kc
c
ac

38


 
Dg o
Dg 1 2
K L
Kq K1
a00o  DL pK1
q
q
aq
2 q
)


D2
Dg
Koq K1
lnD2 aq LD
q p
aq
aq

39

archives of civil and mechanical engineering 12 (2012) 205218


 
Dg o 1
Dg 1 2
000
K L
ao  DL pK1
Kc Kc
c
ac
2 c
)


D2
Dg
2
Koc K1

p

a
LD
lnD
c
c
ac
ac
b001
000

b1

40



2D2 o
Dg
Kq K1
q p
aq
aq

41



2D2 o 1
Dg
Kc Kc p
:
ac
ac

42

5.
Solution for cohesionless soil by symbolic
algorithm
5.1.

General solution

The solution presented in this section has been proposed by

first of the authors in earlier papers [5,6] and could be applied
solely for the case of cohesionless soils. In the case of
cohesionless soils Eq. (27) reduces to:
0

a0 a1 zr a2 z2r a3 z3r b1 lnD aq zr 0

43

It is evident that all coefficients in (43) are obtained by means

of Eqs. (28)(33) by substituting zero for the cohesion parameter (c0) in each equation.
As mentioned above it is not possible to obtain the solution
of Eq. (43), with respect to zr, in a closed form. In purpose to
get an approximate equation for zr (in a closed form), and
then for Hu, the logarithmic term in Eq. (43) can be expanded
into a Taylors polynomial of the third degree. In order to
improve the accuracy an optimising parameter y is introduced in the following way:
D aq zr D yaq yaq aq zr o k

44

where o Dyaq and k yaqaqzr. Then, the expansion is

performed around the point o and the increment k, namely
lnD aq zr lno k  lnD yUaq

aq y zr aq 2 y zr 2
aq 3 y zr 3


2
D yUaq
2D yUaq
3D yUaq 3

45

Substituting the right-hand side of Eq. (45) to Eq. (43)

enables to find zr in a closed form (polynomial equation of
the third degree). It is shown in Fig. 3 that the approximation
point y depends on the value of angle f. It has been pointed
out in earlier paper [5,6] that above approximation can be

209

effective provided that one approximation point y is suitable

for relatively wide range of f values. Two branches of the
function y(f)are graphically presented in Fig. 3. The point that
had been specified for further computations is indicated in
the central part of the lower curve (see Fig. 3).
The obtained formula for zr is substituted to Eq. (34). Then
the ultimate lateral force Hu is determined by reducing
Eq. (34), i.e. by inserting c 0 in (34) one gets:
Hu a00o a00 zr a00 z2r b001 lnD aq zr

46

Because the resulting expression is complicated all transformations have to be carried out by a certain computer system
which enables symbolic computations, including Mathematica.
Then the resulting equation can be transferred to a software
executing reliability measures calculations. In the present study
the FORM and SORM methods [13,7] have been implemented
for evaluating the values of b and pF.
It should be mentioned, however, that in the case of cohesive
soil Eq. (27), which allows determination of the rotation centre
zr, contains two logarithmic terms and therefore it is more
complex as in the case of cohesionless soil. In this case finding
y suitable for wide interval of f appeared to be impossible.
Evaluation of the reliability index that corresponds to a
specified value of total safety factor F (Eq. (21)) can be performed
by applying the following procedure:
1. Assume a value of F.
2. Replace all random variables by constants equal to their
expectations and use them to evaluate the ultimate
resistance Hu given by Eq. (46).
3. For given total safety factor F (step 1) and obtained value of
Hu (step 2) calculate the value of applied load Ha utilising
Eq. (21).
4. Treat Ha and f as random variables. Assign to Ha the expectation equal to the value obtained in step 3. Other probabilistic
characteristics have to be additionally assumed. An example
is given in Table 2.
5. Evaluate b and pF by making use of Eq. (22).
6. All steps above has to be repeated for several different
values of factor F.

5.2.

Numerical example

The aim of the example is to check the effectiveness of the

presented above procedure in the case of cohesionless soil.

Fig. 3 Approximation point h versus friction angle /. The selected value is h 1.97842.

210

archives of civil and mechanical engineering 12 (2012) 205218

Table 2 Parameters of the problem.

Parameter

Probability distribution

Expected value

C.O.V. (%)a

Friction angle f
Applied load Ha
Surface overburden p
Eccentricity e
Unit weight g
Pile diameter D
Pile length L

Lognormal
Normal
Normal
Normal
Constant (non-random)
Constant (non-random)
Constant (non-random)

33.61
825 kN
8.8 kNm-2
8.64 m
20.15 kNm-3
0.36 m
2.9 m

10; 15
10; 15; 20
5
5

C.O.V.coefficient of variation (standard deviation/expected value)n100%.

Fig. 4 Reliability index b versus total safety factor F.

Table 2 presents input data for the example. Utilising the

procedure formulated in the previous section the reliability
index versus the total safety factor F has been computed.
The SORM method serves to evaluate reliability indices and
corresponding failure probabilities.
Results of computations are presented in Fig. 4. One can
observe that for the 15% friction angle coefficient of variation
the total safety factor F should be assumed about 3 in order to
assure the reliability index b on the level 3. If random
variation of friction angle is smaller (10%) then the range
2.1-2.4 of F is sufficient to obtain b on the level about 3.
Generally the results demonstrate that rather high values of
total safety factor F should be used if the Brinch Hansen
method is applied in engineering practice.
Additionally sensitivity analysis has been carried out in
order to evaluate impact of all random variables on the value
of reliability index b. It has been found that the most
important, for the problem under consideration, are random
fluctuations of friction angle f (the sensitivity coefficients
af 0.88, provided that c.o.v.{f} 10% and c.o.v.{Ha} 10%).
Random variability of the applied load Ha is another important factor (aH 0.37 provided that c.o.v.{f} 10% and
c.o.v.{Ha} 10%).

6.

Solutions by the response surface

As it has been already mentioned in Section 5.1 the symbolic

algorithm has solved the problem in the case of cohesionless
soils only. In the case of cohesive soil the angle of internal
friction is usually smaller than in the case of cohesionless one.
This creates difficulties in mentioned above expansion, which
appeared not accurate enough. On the other hand, taking into
account more terms of the power series, disables getting
solution of equilibrium equations in a closed analytical form.
Consequently some new investigations were carried out in
order to find a solution utilising another approach, namely a
solution based on the response surface method.
First examples utilising response surface methodology in
the cohesionless soils have been presented by authors in
the conference proceedings of 28th Winter School on Rocks
Mechanics and Geoengineering 2005 [14]. Next considerations
presented in this paper were completed later and have not
been yet published.
The response surface method (see for example [15,10])
approximates an unknown function by the known function
chosen appropriately.

211

archives of civil and mechanical engineering 12 (2012) 205218

If numerical calculations are performed a functional dependence between the model parameters x1, x2,yxn, treated as the
input data (e.g. random parameters) and the values received as
the output data yf(x1, x2,yxn), has to be established. Here
the resulting variable is y Hu f(x1, x2,yxn). The function f is
obtained by utilising a non-linear regression, based on the
Marquardt algorithm [16]. Then, if the f function is evaluated
the SORM method is easily applicable.
Let us now come back to the example demonstrated in
Section 5.2. We assume the values of coefficients of variation
for friction angle f and external load Ha as c.o.v.{f}c.o.v.{Ha}
15%. The surface overburden and eccentricity are now assumed
to be non-random. Other assumptions are the same as in the
Table 2. A set of evaluations of the force Hu, using Eqs. (43) and
(46) with different values of f have been carried out. In these
evaluations Eq. (43) has been solved numerically. This way a
response surface y Hu(f) has been established.
Next several forms of the relationship fHu have been
considered. However, usually accepted polynomials of second
degree have given not satisfactory results. Therefore several

other non-linear functions have been fitted to the numerical

results. The criterion of fitting was the sum of least square
differences. Finally, the following relation has been accepted:
p
Hu a b fc d

47

where a, b, c, d are functional constants (without any physical

interpretation). They have been established by non-linear
regression algorithm based on the Marquardt compromise
[16] as equal to the following values: a 1.113, b 0.1123,
c4.208, d 0.5829. The above curve fits very well fHu
dependence as it can be seen in Fig. 5.
As a next step, by means of Eq. (47) evaluations of reliability
measures have been carried out by applying two different ways,
namely the symbolic algorithm and the response surface. In
both cases, in the final stage, the SORM method has been used
to determine reliability indices b. Table 3 contains values of b
indices obtained in the computations.
It is evident that reliability indices, that correspond to the
same value of total safety factor F, but resulting from two
different approaches (symbolic algorithm and response
surface), coincide to each other. Therefore the symbolic
algorithm can be successfully replaced by the response
surface approach. Results demonstrated in Tables 3 and 4
are associated with the length of the pile equal to L 2.9 m
and L3.4 m, respectively. Additionally analogical computations (by means of the Eq. (47)) have been repeated for other
pile lengths, namely L2.4 m, L 2.8 m, L 3.0 m, L3.2 m. In
each case a different set of coefficients a, b, c and d has been
received. Table 4 contains results for the pile length L 3.4 m. In
all cases, with accuracy to one digit after point, the reliability
indices computed for the same value of safety factor coincide
for symbolic algorithm and for response surface method.
As a next step an idea of building one response surface, for
different lengths of piles, has been tested. In this case three
random variables have been considered, namely the angle of
internal friction f, the applied load Ha and the length of the
pile L. A new response surface, as a function of variables f
and L, is given by the following equation:
Hu aLb cfd

Fig. 5 The response surface; Pile length L 2.9 [m].

48

By applying a non-linear regression algorithm, parameters

of the above function have been found and the function

Table 3 Reliability indices b corresponding to specified values of the total safety factor F. Pile length L 2.9 [m]; response
surface coefficients: a 1.113; b 0.1123; c 4.208; d 0.5829.
Total safety factor F

Expected value of the

lateral force Ha [kN]

Reliability index b by
symbolic algorithm

Reliability index b by
response surface

1.2
1.4
1.6
1.8
2.0
2.2
2.4
2.6
2.8
3.0
3.2

21.92
18.79
16.44
14.61
13.15
11.95
10.96
10.12
9.39
8.77
8.22

0.37
0.74
1.08
1.39
1.69
1.96
2.21
2.45
2.68
2.89
3.10

0.38
0.75
1.09
1.41
1.70
1.97
2.22
2.46
2.69
2.91
3.12

212

archives of civil and mechanical engineering 12 (2012) 205218

Table 4 Reliability indices b corresponding to specified values of the total safety factor F. Pile length L 3.4 [m]; response
surface coefficients: a0.9485, b 0.1002, c 3.721, d 0.02616.
Total safety factor F

Expected value of the

lateral force Ha [kN]

Reliability index b
by symbolic algorithm

Reliability index b by
response surface

1.2
1.4
1.6
1.8
2.0
2.2
2.4
2.6
2.8
3.0

34.45
29.53
25.84
22.97
20.67
18.79
17.23
15.90
14.76
13.78

0.36
0.72
1.06
1.36
1.65
1.92
2.17
2.40
2.63
2.84

0.37
0.74
1.07
1.38
1.67
1.93
2.18
2.42
2.65
2.86

shallow foundations reliability analysis the application of

spatial averaging (of ground properties random fields) results
in more reasonable values of reliability measures. Thus, in
this work, we have decided to include into reliability calculations the approach based on spatial averaging.
The procedure of spatial averaging was first introduced by
Vanmarcke [17]. In the present study we have applied a onedimensional averaging of the soil mechanical property random field, namely the angle of internal friction. As a consequence, we get a new random variable fL expressed as
Z
1
Lfzdz
50
fL
L

Fig. 6 The response surface obtained in the two-dimensional

case.
obtains the following form:
Hu 0:17U1020 L2:868 141:6f9:411

49

Response surface, namely the function (49), is demonstrated in Fig. 6. Results of reliability indices evaluations are
demonstrated in Table 5. Computations have been carried out
with the piles length coefficient of variation equal to 2%, in
case of symbolic algorithm, and for two different coefficients of variation, namely 2% and 5% in the case of response
surface method. As before coincidence of two methods
(symbolic algorithm and response surface algorithm) is
very good. This shows that the response surface method
can successfully replace the symbolic algorithm.

7.

Application of spatial averaging

Earlier studies have not included any spatial averaging in the

reliability analysis of rigid piles lateral capacity. However, it
has been shown in the literature (e.g. [6]) that in the case of

In the equation above f(z) is the random function describing random fluctuations of f, occurring on the depth z, while
L is the length of the pile under consideration; it should be
emphasised that in the case where L is a random variable
then L is the expectation of L. In what follows the aforementioned function f(z) is assumed to be stationary. It possesses
constant both mean value, say mf as well as the variance, s2f .
Another one parameter characterising a correlation structure
is the fluctuation scale d (see e.g. [17]). This parameter is
sometimes called the correlation length or correlation radius.
The value of fL variance can be obtained by making use of
the following expression, i.e.:
VARfL  s2L gLs2f

51

Eq. (51) defines the so called variance function g(L). This

function can be evaluated when the autocorrelation function
of the field is given [18]. It can be found in the literature [17]
that, for the problem of soil properties random variability, the
Gaussian autocorrelation function is the most suitable one.
Therefore, in this work, this function has been chosen for
further considerations. It is given by the following equation [17]:


p
52
rDz s2 exp  2 Dz2
d
where Dzz1z2 is the distance between two points of the field
domain z1 and z2. According to [6] it can be shown that
choosing the autocorrelation function (52) implies the variance
function in the following form:
gL

p
2
p=dLerf p=dL1 expp=d2 L
2

p=d2 L

53

213

archives of civil and mechanical engineering 12 (2012) 205218

Table 5 Reliability indices received from analysis with three random variables (/, L, Ha).
Total safety
factor F

Reliability index b by
symbolic algorithm
c.o.v.{L} 0.02

Reliability index b by
response surface
c.o.v.{L} 0.02

Reliability index b by
response surface
c.o.v.{L} 0.05

1.2
1.4
1.6
1.8
2.0
2.2
2.4
2.6
2.8
3.0

0.36
0.73
1.07
1.38
1.66
1.93
2.18
2.42
2.64
2.87

0.37
0.73
1.07
1.38
1.67
1.94
2.19
2.44
2.67
2.90

0.35
0.70
1.01
1.30
1.56
1.81
2.04
2.26
2.46
2.66

Table 6 Values of the variance function for 3 different

values of fluctuation scale.
Averaging length

Values of the fluctuation scale

L [m]
d 0.6 [m]

d 0.8 [m]

d 1.0 [m]

Values of the variance function

2.9

0.1933

0.2516

0.3070

In the equation above d is the so-called fluctuation scale. The

parameter d can be found from laboratory testing. According to
the literature [19], in the case of non-cohesive soils in vertical
direction, it takes values about 1 [m] or less. Here the computations were carried out for three different values of d. The
resulting values of the variance function are given in Table 6.
Next the reliability computations (with the same assumptions as in the last example considered) incorporating the
results of spatial averaging have been carried out. Only the
response surface method has been applied. The results, namely
the values of reliability indices are summarised in Table 7. For
the transparency of the presentation the reliability indices with
no spatial averaging applied are also provided (the 2nd column
of Table 7). In addition, the results are graphically presented in
Fig. 7. One can simply observe a strong dependence between
spatial averaging and the values of reliability indices. As it can
be seen in Fig. 7, when spatial averaging is applied, relatively
large values of b, greater than 3.4, are obtained (regardless of
the value of d) for the total safety factor F equal to or greater
than 2.4. On the contrary we also see a high influence of the
fluctuation scale d on reliability measuresthe lower value of d
is assumed then the greater value of b is obtained. Therefore, it
has to be strongly emphasised that the value of the fluctuation
scale should be chosen very carefully. Moreover, if possible, it
should be also supported by lab investigations.

8.

Solutions for cohesive soils

8.1.

General algorithm

As it was already mentioned one of the reasons of applying

response surface methodology was a creation of a solution

useful for both cohesionless and cohesive soils. In what

follows we consider a rigid pile embedded in a cohesive soil
of strength parameters c and f. Three following random
variables are now considered: f, c and Ha.
In order to reduce a computational effort in creating a
response surface for ultimate horizontal force Hu an iterative
algorithm has been created. The algorithm is a modified
version of the response surface algorithm proposed by Bauer
and Pua [15] for the purpose of probabilistic settlement
analysis. The main steps are as follows:
1. Given the expected values of variables f and c, one may
assume that an approximate interval of their variation is a
double standard deviation of length with the expected
value in the centre, for each variable.
2. If nine pairs of values of f and c are chosen from the
interval of their variation (three different of each), a set of
values for ultimate force Hu could be obtained by numerical solution of equilibrium Eqs. (1) and (2).
3. A response surface for Hu can be determined by non-linear
regression in the following form, that has been obtained
by steps analogical to the ones described for Eq. (47):
Hu fc

A1 c
A4

A2 A3 f f

54

4. Utilising the surface Hu(f, c) the design point on the limit

state surface is found using the FORM method.
5. The above steps are repeated in the neighbourhood
of a design point. Intervals of variability are reduced, for
example, to approximately half a standard deviation from
the values determining a design point.
6. For the last design point, by applying the SORM method,
the reliability index b is found.

8.2.

Numerical analyses

In order to demonstrate the efficiency of the above algorithm

consider now an example based on the data presented in
Table 8. Satisfactory results have been obtained after three
iterations of the algorithm that has been outlined in previous
section. The coefficients in Eq. (54) took the following values:
A1 8.104, A2 2.223, A3 0.06714, A4 20.10.
The convergence of the reliability index as well as coordinates
of the design point (in the FORM method) are demonstrated in

214

archives of civil and mechanical engineering 12 (2012) 205218

Table 7 The effect of spatial averaging on reliability indices (c.o.v.{L} 0.02).

Total
safety
factor F

Reliability index b
without spatial
averaging

Reliability index b with

spatial averaging (d 0.6 m)

Reliability index b with

spatial averaging (d 0.8 m)

Reliability index b with

spatial averaging (d 1.0 m)

1.2
1.4
1.6
1.8
2.0
2.2
2.4
2.6
2.8
3.0

0.37
0.73
1.07
1.38
1.67
1.94
2.19
2.44
2.67
2.90

0.76
1.42
2.01
2.56
3.06
3.53
3.96
4.37
4.77
5.13

0.70
1.30
1.85
2.35
2.82
3.25
3.66
4.04
4.40
4.75

0.65
1.22
1.73
2.20
2.64
3.05
3.43
3.79
4.13
4.46

Table 8 Parameter characteristics involved into the

problem considered.
Parameter

Probability
distribution

Expected
value

Standard
deviation

Friction
angle f
Cohesion c
External
lateral
load Ha
Unit weight g
Surface
overburden p
Eccentricity e
Pile
diameter D
Pile length L

Lognormal

151

2,51

Lognormal
Lognormal

25 kPa
14.5 kN

3 kPa
2.175 kN

Non-random
Non-random

19.62 kNm3
8.8 kNm2

Non-random
Non-random

8.64 m
0.36 m

Non-random

2.9 m

Fig. 7 The influence of spatial averaging on the value of

reliability indices b.

Table 9. One can observe quite satisfactory behaviour of the

reliability index and the design point coordinates. The results
were compared to others obtained when sixteen pairs of f and
c were selected in the beginning (see step 2 of the algorithm).
The final results were the same with the accuracy of two digit
after the point. Then nine pairs of f and c seems to be sufficient
in this case.
Given the Eq. (54) one can also investigate an effect of
mutual correlation between strength parameters on reliability measures (see Table 10). It is evident that the strong
negative correlation results in higher values of the reliability
index.

8.3.

Application of spatial averaging

Consider, once again, a rigid pile embedded in a cohesive soil

of parameters c and f. Essential data are the same as in
previous case, see Table 8. In what follows, we assume that
there is no mutual correlation between f and c. In this
section one-dimensional averaging of both the internal friction angle and cohesion fields has been applied. These
random fields vary with depth z, parallel to the pile shaft.

Table 9 Reliability indices obtained after three iterations of the algorithm.

Iteration
number

1
2
3

Reliability
index b

2.638
2.618
2.617

Design point
coordinates
f [deg.]

c [kPa]

11.657
11.717
11.703

21.540
21.553
21.554

Table 10 The effect of mutual correlation between / and

c on reliability measures.
Correlation
coefficient r (f,c)

Reliability
index b

Probability
of failure pF

0.0
0.2
0.4
0.6
0.8

2.653
2.801
2.975
3.186
3.459

0.00399
0.00255
0.00147
0.00072
0.00027

215

archives of civil and mechanical engineering 12 (2012) 205218

Averaging length L is equal to the pile embedding length,

namely 2.9 [m]. On the contrary to the case of cohesionless
soils, now we have two random variables, i.e. fL given by
relation (50) and cL which is defined by the following
analogical relation:
cL

1
L

czdz

55

where c(z) is the random function describing the variability of

cohesion c with the depth z. As it was in the case of friction
angle, we assume that the function c(z) is stationary with
constant mean value mc and variance s2c . Furthermore, the
variance of cL is expressed as
VARcL  gLs2c

56

As before the Gaussian correlation function is accepted,

which implies the variance function in the form (53). The
value of the fluctuation scale is assumed to be the same for
both friction angle and cohesion fields. The calculations have
been performed for four different values of fluctuation scale
d. The values of variance function corresponding to different
values of d are summarised in Table 11.
Furthermore, the values of reliability indices b are given in
Table 12. Note that for each value of d the reliability index b
has been determined by making use of procedure formulated
in Section 8.1. Observing the results one can simply notice
that as it was in the case of cohesionless soils there is a
high influence of spatial averaging on reliability measures.
We see that reliability indices determined with spatial averaging are greater when compared to the ones obtained without averaging. Obtained result is strongly affected by the
assumed value of fluctuation scale d. Therefore it should be
strongly emphasised that this parameter has to be chosen
very carefully selection procedure should be supported by
laboratory investigations.

8.4.

The case of frictionless soils

In the case of the very cohesive or ideal cohesive (frictionless)

soils it is evident that the response surface for Hu given by
relation (54) is not a proper function. Therefore some numerical analysis have been carried out and it has been stated that
if c is a large value, while f tends to zero or even equals zero
(ideal cohesive soil) then the response surface for Hu can be
expressed by the following function:
Hu fc

A1 c
A4 f
A2 A3 f

57

Note that in the expression above, the first component

remained unchanged in relation to the Eq. (54). Only the
second term on the right side of this equation has changed,
i.e. instead of the non-linear component A4f1, which
excludes solving the problem of f 0, we propose to use a
linear component A4f.
In order to verify whether the function given by relation
(57) well describes the variation of the ultimate load Hu we
have solved an example in which cohesion is much greater
than the internal friction angle. As it was in the previous
case, the solution is obtained using an algorithm provided in
Section 8.1. Essential data used within this example are
shown in Table 13. Note that all remaining parameters,
namely overburden pressure p, eccentricity e, pile diameter
D and pile length L are the same as the ones given in Table 8.
Satisfactory results have been obtained after two iterations
of the algorithm. The following set of coefficients has been
derived: A1 1.794, A2 2.268, A3 0.0701, A4 0.4879. The
results, namely reliability index and the coordinates of the
design point are shown in Table 14.
Taking into account the correction resulting from the SORM
method, the final result for this example is: b 3.256
(pF 0.000565).

Table 11 Values of variance function for different

fluctuation scales.

Table 13 Parameter characteristics involved into the

problem considered.

Averaging length

Parameter

Probability
distribution

Expected
value

Standard
deviation

Friction
angle f
Cohesion c
External
lateral
load Ha
Unit weight g

Lognormal

51

11

Lognormal
Lognormal

80 kPa
24 kN

8 kPa
3.6 kN

Non-random

19.62 kNm3

Values of the fluctuation scale

L m
d 0.8
[m]

d 1.0
[m]

d 1.2
[m]

d 1.4
[m]

Values of the variance function

2.9

0.2516

0.3070

0.3593

0.4086

Table 12 The values of reliability indices b evaluated for

different values of fluctuation scale.
Fluctuation
scale d [m]

0.8
1.0
1.2
1.4

Table 14 Reliability indices obtained after two iterations

of algorithm.

Reliability index b
With spatial
averaging

Without spatial
averaging

3.427
3.349
3.285
3.201

2.653

Iteration
number

1
2

Reliability
index b

3.234
3.235

Design point
coordinates
f [deg.]

c [kPa]

4.260
4.261

67.099
67.099

216

8.5.

archives of civil and mechanical engineering 12 (2012) 205218

The case of random pile length

Table 16 Reliability indices obtained after two iterations

of algorithm.

It happens quite often that due to the accidental error the

actual length of the pile may be somewhat different from the
value resulting from the design calculations. Therefore, it is
worth solving the example of the pile embedded in a cohesive
soil with the pile embedding L being random variable. We
consider now analogous example as in Section 8.2 (see
Table 8). We assume, in addition, that the pile embedding L
is a random variable. Then, the example becomes to be a
problem with four random variables: friction angle f, cohesion c, external load Ha and pile embedding L. We assume
that L is a random variable of normal distribution with
expected value 3.1 [m] and standard deviation 0.06 [m]. After
some numerical calculations a new response surface for the
ultimate load Hu has been found in the following form:
Hu fcL

A1 cA2 A3 L A6 A7 A8 A9 A10 A11

2 2 2
f
c
L
c
A4 A5 f
L
f

58
Final results have been obtained after two iterations of
algorithm provided in Section 8.1 and the following set of
coefficients has been derived. The results of performed calculations, are summarised in Table 15.
A1 15:779;

A2 4:705;

A5 0:268;

A6 73:927;

A7 99:133;

A3 2:718;

A8 75:449;

A4 8:298;

A9 349:985;

A10 679:243;

A11 179:392

59

It should be mentioned that application of the SORM

method leads to the final result: b 3.280 (pF 0.00052).
In this section a one-dimensional averaging of both the
internal friction angle and cohesion fields has also been
applied. Averaging length L is assumed to be equal to the
expected value of pile embedding, namely 3.1 [m]. The
calculations have been performed for only one value of the
fluctuation scale, i.e. d 1.2 [m]. In this case the variance
function, given by relation (53), is gL 0:3394. The results
corresponding to the case of spatial averaging are summarised in Table 16.
Taking into consideration the correction resulting from the
SORM method, the final result for this example has been
determined; reliability index is b 4.000 and the corresponding
probability of failure pF 0.0000317. The comparison of the
results obtained for the cases with and without spatial averaging is shown in Table 17. As it has been observed in previous
examples, a greater value of reliability index is always obtained
when the spatial averaging of random fields is involved into the
problem.

Table 15 Reliability indices obtained after two iterations

of algorithm.
Iteration
number

1
2

Reliability
index b

3.295
3.240

Design point coordinates

f [deg.]

c [kPa]

L [m]

10.585
11.238

21.040
20.893

3.062
3.058

Iteration
number

1
2

Reliability
index b

4.073
3.985

Design point coordinates

f [deg.]

c [kPa]

L [m]

12.865
12.784

22.843
22.907

3.039
3.039

Table 17 The values of reliability index b and probability of failure evaluated with and without spatial
averaging (d1.2 [m]).

Reliability index b
Probability of
failure pF

Without spatial
averaging

With spatial
averaging

3.280
0.00052

4.000
0.0000317

The purpose of this section was to show that the response

surface method gives the possibility of solving the problems
with more than two random variables. The addition of each
new variable, however, causes a significant increase in the
number of calculations. Some difficulties are also observed
when the accurate response surface has to be evaluated.
Nevertheless, as mentioned earlier, for the case of the
laterally loaded rigid piles the greatest impact on the value
of the reliability index are due to random fluctuations of both
the friction angle and cohesion. Therefore, it makes sense to
restrict the considerations just to these two random variables, which makes the response surface method a very
efficient tool which can be successfully used for quick
determining the reliability measures for this type of geotechnical problems.

9.

Conclusions

Within this study a complete solution to the problem of

random lateral bearing capacity of rigid piles has been
presented and discussed. The suggested solution is based
on the limit states theory approach proposed by Brinch
Hansen [2], which is a widely spread tool in computations
of piles under transversal loading. Numerical algorithms, for
both cases of cohesive and cohesionless soils, have been
developed in order to evaluate probability of lateral bearing
capacity exceeding. It has been shown that, in the case of
Brinch Hansen approach, random variation of soil mechanical properties can lead to significant fluctuations of the
ultimate resistance Hu. Hence, the reliability investigations
seem to be of paramount importance when the parameters
for the design process are selected. Nowadays the probabilitybased design becomes very important way in the process of
constructing safe and reliable structures [20]. Therefore
proposed algorithms are important both from theoretical
and practical points of view.

archives of civil and mechanical engineering 12 (2012) 205218

It was mentioned earlier in the paper that, for the problem of

random lateral bearing capacity of rigid piles, classical (i.e.
commonly used) reliability methods can lead to large difficulties in obtaining the solutionreliability measures. This is due to
the complicated form of solution of equilibrium equations. In
order to avoid this inconvenience two separate procedures have
been proposed. The first one is the application of symbolic
calculations associated with classical FORM/SORM methods
(this approach is called, in the paper, the symbolic algorithm).
Another one proposition is to utilise the response surface
procedure in conjunction with FORM/SORM methods. It should
be strongly emphasised that the symbolic algorithm solves
only the case of cohesionless soils. However, the second
approach, i.e. the response surface based one, overcomes
numerical difficulties that appeared in the symbolic algorithm
and allows to solve the problem of both cohesionless and
cohesive soils. Note that, for the case of cohesionless soil, the
results have been compared and it has been found that both
approaches (the symbolic algorithm and the response surface) lead to almost the same results. It is evident, however,
that even though the response surface approach solves the
problem of cohesive soils there exist a significant growth of
numerical effort when the number of random variables considered in the problem is increasing.
Within numerical examples, presented in the paper, the
random fluctuations of soil mechanical properties as well as
the uncertainty in the embedding of the pile have been
widely considered. The vital role of friction angle random
fluctuations in randomness of lateral bearing capacity has
been demonstrated. Nevertheless, it is evident that the
random variability of applied external lateral load is important, provided that its variation coefficient exceeds 10%.
Numerical examples have revealed a strong dependence
between spatial averaging and the values of reliability
indices. Furthermore, it was shown that performing reliability calculations in conjunction with spatial averaging leads to
greater values of reliability indices when compared to the
ones obtained by classical structural reliability approach, i.e.
without spatial averaging. It has been also shown that there
exist a high influence of the fluctuation scale d on reliability
measures. The lower value of fluctuation scale is assumed in
calculations then the greater value of reliability index is
obtained. In the authors opinion the spatial averaging (along
the pile length) should be involved into reliability analysis of
rigid piles, provided a credible estimation of fluctuation scale
is available. Similarly including a negative correlation of
strength parameters into reliability computations results in
growth of reliability indices. Unfortunately the recognition of
the correlation coefficient level of these properties is usually
very poor in engineering practice.
Results of computations clearly present that selecting one
deterministic value of the total safety factor for this problem
seems not to be reasonable. It is evident that possibility of
bearing capacity exceeding is strongly affected by random
character of soil properties, which should be reflected in the
safety evaluation.
Finally it is necessary to mention that the paper deals with
rigid piles embedded into non-stratified homogeneous soil.
Usually rigid piles are not very long then the possibility of
pile contact with several soil layers is practically bounded.

217

Original approach given by Brinch-Hansen [2] is also dedicated for homogeneous subsoil. Due to the complex relations
in Brinch-Hansen approach the extension for layered subsoil
and/or variable soil properties is not straightforward. However, two approaches are possible. The first one is to consider
two (or more) soil layers and evaluate the resistance pu (by
Brinch-Hansen method) in each layer separately, investigate
equilibrium equations numerically and by applying response
surface method carry out probabilistic calculations. As an
alternative approach (instead of Brinch-Hansen) a numerical
solution (e.g. finite element or finite differences methods) in
conjunction with the response surface could be utilised. The
problem of stratified subsoil and variable soil properties is
currently investigated by authors.

Acknowledgements
This paper, as a research project, was supported by Polish
Government funds for science in 20082010.

r e f e r e nc e s

[1] H.G. Poulos, E.H. Davis, Pile Foundation Analysis and Design,
J. Wiley, New York, 1980.
[2] J. Brinch Hansen, The ultimate resistance of rigid piles
against transversal force, The Danish Geotechnical Institute,
Bulletin, No.12, Copenhagen, 1961.
[3] B.B. Broms, Lateral resistance of piles in cohesive soils,
Journal of Soil Mechanics and Foundation Engneering Division 90 (SM2) (1964) 2763.
[4] B.B. Broms, Lateral resistance of piles in cohesionless soils,
Journal of Soil Mech. and Foundation Eng. Division 90 (SM3)
(1964) 123156.
[5] W. Pua, Reliability analysis of rigid piles subjected to lateral
loads, in: S. Geomechanics, Pietruszczak, G.N. Pande (Eds.),
Numerical Models, Balkema, Rotterdam, 1997.
[6] W. Pua, Applications of Structural Reliability Theory to
Foundations Safety Evaluation (in Polish), Wroclaw University of Technology Press, Wrocaw, 2004.
[7] M. Hochenbichler, S. Gollwitzer, W. Kruse, R. Rackwitz, New
light on first and second-order reliability methods, Structural
Safety 4 (1987) 267284.
[8] Wolfram Research Inc., 2003.
[9] G.P. Box, N.R. Draper, Empirical Model-Building and Response
Surface, J. Wiley & Sons, New York, 1996.
[10] L.A. Faravelli, A response surface approach for reliability
analysis, Journal of the Engineering Division, ASCE 115 (12)
(1989) 27632781.
[11] J. Brinch Hansen, J. Lundgren, Hautprobleme der Bodenmechank, Springer Verlag, Berlin, 1960.
[12] J.Brinch Hansen, A general Formula for Bearing Capacity, The
Danish Geotechnical Institute Bulletin, No. 11, Copenhagen,
1961.
[13] O. Ditlevsen, H.O. Madsen, Structural Reliability Methods,
Chichester John Wiley & Sons, 1996.
[14] W. Pua, A. Rozanski, I. Shahrour, Reliability of rigid piles
subjected to lateral loads. A revised approach, in: Geotechnics in Civil and Mining Engineering (Proceeding of 28th
Winter School on Rocks Mechanics and Geoengineering),
Wroclaw University of Technology Press, 2005.
[15] J. Bauer, W. Pua, Reliability with respects to settlement limitstates of shallow foundation linearly deformable subsoil,
Computers & Geotechnics 25 (34) (2000) 281308.

218

archives of civil and mechanical engineering 12 (2012) 205218

[16] D.W. Marquardt, An algorithm for least-squares estimation

of non-linear parameters, Journal of the Society for Industrial and Applied Mathematics 11 (2) (1963).
[17] E.H. Vanmarcke, Probabilistic modeling of soil profiles, Journal of the Geotechnical Engineering Division ASCE 103 (GT11)
(1977) 12271246.
[18] E.H. Vanmarcke, Random Fields-Analysis and Synthesis, MIT
Press, Cambridge, 1983.

[19] C. Cherubini, Data and Consideration on the variability of

geotechnical properties of soils, in: Proceedings of the ESREL
Conference, 1997.
[20] A. Kudzys, R. Kliukas, Probability-based design of spun
concrete beam-columns, Journal of Civil Engineering and
Management 16 (2) (2010) 451461.