You are on page 1of 22

Energy Systems (2021) 12:841–861

https://doi.org/10.1007/s12667-019-00362-y

ORIGINAL PAPER

Simulation of random fields of soil properties by the local


average subdivision method and engineering applications

Nikolaos Alamanis1 · Panos Dakoulas2

Received: 6 March 2019 / Accepted: 2 October 2019 / Published online: 7 November 2019
© Springer-Verlag GmbH Germany, part of Springer Nature 2019

Abstract
The seismic performance of oil and natural gas pipelines founded or embedded in
earth slopes encompasses great uncertainty related both to the earthquake shaking
characteristics and to the natural heterogeneity of geomaterials. Regarding the latter,
the parameters of shear strength, stiffness, density, etc., may vary indeed from point
to point even within the same soil layer as a result of the natural formation process.
Apart from their cross-correlation, such random variables exhibit autocorrelation,
in which the soil properties at a given point appear to be spatially correlated with
the properties of neighbouring points. Therefore, there is a need to use stochastic
methods in the safety evaluation of such systems. Aiming at the reliability assess-
ment of such soil-pipeline systems under seismic shaking, this paper introduces
an automated methodology for generating random fields using the Local Average
Subdivision (LAS) method by Fenton and Vanmarcke (J Eng Mech 116(8):1733–
1749, 1990). Subsequently, it performs rigorous nonlinear dynamic analysis of a
given slope using the finite difference method. The automated procedure is used in
a Monte-Carlo simulation scheme for computing the probability of exceeding differ-
ent levels of anticipated permanent slope movement for different levels of shaking
intensity. The results demonstrate that the effect of the spatial variability of the soil
properties on the permanent displacements of natural slopes is important, leading to
a range of variation of about ± 60%.

Keywords Numerical simulation · Stochastic methods · Spatial variability · Random


fields · Seismic performance

* Nikolaos Alamanis
alam@teilar.gr
Panos Dakoulas
dakoulas@uth.gr
1
General Department, University of Thessaly, Vólos, Greece
2
Department of Civil Engineering, University of Thessaly, Vólos, Greece

13
Vol.:(0123456789)

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


842 N. Alamanis, P. Dakoulas

1 Introduction

The construction of vast energy networks of buried oil and natural pipelines through
geohazard areas all over the world imposes major challenges for their structural integ-
rity and safety, as well as for environmental safety. More specifically, oil and natural
gas pipelines crossing seismic regions may be subjected to large ground movements
caused by seismic slope instability, ground settlements and soil liquefaction phenom-
ena [2, 3]. Such severe external loadings may induce significant strains in the pipe-
line wall, leading to local buckling and/or rupture and, possibly, loss of containment.
In these cases, pipeline design should account for the ground-induced strains due to
the differential ground movement, in addition to the strains developed from operational
loads. The intensity and the spatial distribution of the imposed ground movement may
depend in a significant way on the spatial variability of the soil properties in a zone
along the path of the pipeline system.
The uncertainties associated with both the spatial variability of the soil properties of
natural slopes and the characteristics the seismic excitation, require the use of probabil-
istic methods [4–6]. For engineering design, probabilistic methods allow for the selec-
tion of an acceptable risk level, based on the specifications of each project. To this end,
the presented research aims at accounting in a rational way for the effects of the spatial
variability of the soil properties and may be used for the evaluation of the seismic per-
formance of buried oil and natural gas pipelines. The paper introduces an automated
methodology based on Mathematica [7] which generates random fields and subse-
quently performs nonlinear dynamic analysis of a given slope using the finite differ-
ence method [8] in a Monte-Carlo simulation scheme. Statistical analysis of the results
allows computation of the probability of exceeding different levels of anticipated per-
manent slope movement.

2 Literature review

Random fields of spatially variable properties are motivated by the widespread accept-
ance of reliability methods in engineering design. For example, to estimate the failure
probability of a natural slope subjected to earthquake shaking, one may represent both
soil properties and input earthquake ground motion using independent random fields,
with the first varying in space and the second in time. Subsequent dynamic analyses of
such systems using Monte Carlo simulations allow an estimation of the probability of
failure under seismic loading.
Gaussian random fields are completely defined by the mean value 𝜇(x), the variance
𝜎 2 (x) and the autocorrelation function 𝜌(x, x� ). A widely used exponential auto-correla-
tion function is expressed as
( )
|x − x� | |y − y� |
𝜌(x, y) = Exp − − (1)
lx ly

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Simulation of random fields of soil properties by the local… 843

where lx and ly are the characteristic autocorrelation lengths in the horizontal and
vertical directions, respectively. In the usual case of fields of random variables rep-
resenting several material properties, the cross-correlation coefficient 𝜌ij between
properties i and j are also required.
Various algorithms have been proposed to produce random fields corresponding
to different levels of accuracy, efficiency and ease of use. The most important among
them are (a) Moving Average (MA) method, (b) the Fast Fourier Transform (FFT)
method (c) the Turning-Bands method (TBM) and (d) the Local Average Subdivi-
sion (LAS) method [1, 4]. Non-Gaussian fields may also be obtained through non-
linear transformations of Gaussian fields. The FFT, TBM and LAS methods are typ-
ically much more efficient than other methods.
The Moving Average method expresses the process as an average of an underly-
ing white noise process [4]. In two or three dimensions, the calculation of required
weighting functions is quite complex, which makes the method difficult to use, espe-
cially in the case of arbitrary covariance functions. In most cases, the MA method is
found to be at least ten times slower than the FFT method discussed below.
The Fast Fourier Transform method is based on spectral representation of con-
tinuous random fields, provided that the space and frequency domains can be dis-
cretized into a series of equally spaced points [4]. For N points in a 1D problem,
the FFT method requires N ­Log2N computations [9], whereas the Discrete Fourier
Transform method would require N ­ 2 computations, leading to much higher compu-
tational time. One disadvantage of the method is the covariance function is always
symmetric with respect to the midpoint of the field, but this is overcome by doubling
the size of the field and utilizing only a segment of it. A second shortcoming is the
relationship between space and frequency discretization, which, if it is not properly
considered, may lead to erroneous results.
The Turning-Bands method generates random fields in two and three dimen-
sions, by using a series of one-dimensional processes along lines passing through
the field [10, 11]. If a sufficiently large number of such lines are used, the method
may yield very accurate variance and covariance results, at the expense of decreas-
ing efficiency.
Fenton and Vanmarcke proposed the efficient and accurate Local Average Subdi-
vision (LAS) method for generating realizations of a homogeneous Gaussian scalar
random process in one, two and three dimensions [1, 4]. The obtained random pro-
cess, achieved by successive subdivision of the field, is defined by its mean, variance
and covariance or autocorrelation function. The ability of the method to increase its
resolution is important for finite element analysis applications of spatially randomly
varied systems. A more detailed description of the LAS method is given below.
Griffiths and Fenton [12] applied the LAS method to investigate slope stability
under static conditions using the finite element method. The analysis used the elas-
toplastic Mohr–Coulomb soil model in Monte Carlo simulations to investigate the
effect of spatial correlation and local averaging on the probability of failure. The
results using the LAS method led to higher probabilities of slope failure compared
to those from simpler analyses. Moreover, Fendon et al. [13] presented the results
of a random finite element method analysis of a slope in which the soil is repre-
sented by a spatially random field and slope stability is evaluated using the finite

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
844 N. Alamanis, P. Dakoulas

element method. The paper investigates the resulting failure probability, as a func-
tion of the soil statistical parameters, and proposes a simplified harmonic averaging
approach to estimating the failure probability that avoids the necessity for Monte
Carlo simulation.
Griffiths et al. [14] investigated the probability of slope failure under static con-
ditions using a simplified method and an advanced random finite element method.
The latter method was combined with Monte Carlo simulation, in which the spatial
variation is considered. It was shown that simplified probabilistic analyses in which
spatial variability of soil properties is not properly accounted for, can lead to uncon-
servative estimates of the probability of failure if the coefficient of variation of the
shear strength parameters exceeds a critical value.
Further research on the slope stability under static conditions that accounts for
the uncertainties and spatial variation of the soil parameters using random fields has
been presented in several publications [15, 16]. It is noted that the research on slope
stability using probabilistic methods in references [12–16] focuses entirely on the
static behavior, as opposed to the seismic performance investigated in Sect. 6 this
paper.
In the following a brief description of the LAS method is presented.

3 The Local Average Subdivision method

In order to take into account soil uncertainty and spatial correlation in slope stabil-
ity analysis, Fenton and Vanmarcke introduced the Random Finite Element Method
(RFEM) using a local average subdivision method [1, 17]. This method combines
finite element analysis [18] and random field theory [1, 4]. For the simpler one-
dimensional case, the evolution of a local average subdivision process using the
LAS method for a normal distribution is depicted schematically in Fig. 1.
This is a repetitive process that starts from the top and moves downwards (Fig. 1).
In step 0 a total average is generated for process Z01. In step 1, the item is subdivided
into two equal areas whose average local values are equal to the average in step 0.
The next steps are produced by dividing the previous cell into two new cells, keep-
ing the average of the previous “parent” cells. Thus, the overall mean value remains
constant.
More specifically, the algorithm for the one-dimensional case is as follows:

Fig. 1  Local average subdivision process [1]

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Simulation of random fields of soil properties by the local… 845

1. A normally distributed total mean Z01 is generated.


2. This cell is divided into two equal parts.
3. Two normally distributed Z11 and Z12 values are generated whose average value and
variance meet the following three criteria:
• they display correct variance based on the local average subdivision theory.
• they are correctly correlated.
• they have the same average as the original cell ( Z11 + Z12)/2 = Z01.
4. Each Stage 1 cell is subdivided into two equal parts.
5. Two normally distributed Z21 and Z22 values are generated, whose average value
and variance meet the criteria:
• they display correct variance based on the local average subdivision theory.
• they are appropriately interrelated.
• they have the same average as its original element ( Z21 + Z22)/2 = Z11.
• they are appropriately associated with elements Z23 and Z24.
And so on for the following steps. The process is terminated when the number
of produced cells N = 2m after m successive divisions is equal (or larger) than
the desired resolution (i.e. number of elements) for the finite element discretiza-
tion. For a two-dimensional problem, each parent cell is divided in four equally
sized cells following a similar procedure.
For example, for the case of soil cohesion, the random fields of the desired
cohesion values are evaluated based on the relationship:
( )
ci = 𝜇c + 𝜎c Gc xi (2)
where Gc is the zero-mean random field, xi is a vector containing the coordinates of
the centre of element i and ci is the value of cohesion in element i. The same relation
is also applied to the friction angle.
It is noted that the auto-correlation function given by Eq. (1) allows different
autocorrelation lengths in the horizontal and vertical directions. Indeed, the verti-
cal autocorrelation length may be typically ten times smaller than the horizontal
one, due to the natural soil formation process. If the autocorrelation length tends
to infinity the material the material properties are constant (homogeneous layer),
whereas if it tends to zero the material properties vary randomly without any rela-
tionship to the neighbouring points. The cross-correlation between two soil prop-
erties is calculated by applying the Cholesky decomposition method [19]. The
correlation matrix is decomposed to give the lower-triangular matrix L. Applying
this to a vector of uncorrelated samples u produces a sample vector Lu with the
covariance properties of the system being modelled.
The LAS method is used to create soil random fields with predetermined val-
ues for the mean value 𝜇 , standard deviation 𝜎 , cross-correlation coefficients 𝜌ij
between the properties i and j, and autocorrelation lengths lx and ly in the horizon-
tal and vertical directions, respectively. The typical values of 𝜇 , 𝜎 , 𝜌ij , lx and ly for
different soil properties, which are used in the analyses that follow, are based on
available information from the literature [20–25].

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
846 N. Alamanis, P. Dakoulas

For each analysis, Mathematica initially creates a random four-digit number that
acts as a seed number for applying of the LAS method. The required data for LAS
analysis is:

1. The number of columns nx and rows ny of material zones, each having dimensions
dx in the horizontal direction and dy in the vertical direction.
2. The mean μ, standard deviation σ, and the type of distribution (normal, logarith-
mic, uniform) for each soil property, i.e. cohesion c, friction angle φ, the dilation
angle ψ, modulus of elasticity e, density ρ and Poisson ratio ν.
3. The correlation coefficient matrix ρij (6 × 6) for the above properties.
4. Autocorrelation lengths lx and ly in the horizontal and vertical directions.

The results of the analysis with the fields of random properties are graphically
represented automatically in the Mathematica environment.

4 Confirmation of the LAS methodology

Figure 2 shows an example of a single realization of random fields of the cohesion


c and the friction angle φ in an area with grid dimensions 128 × 64 and zone dimen-
sions dx = dy = 1 m. Similarly, Fig. 3 represents corresponding random fields of the
density 𝜌 and Young’s modulus E. The targeted mean values and standard deviations
of the soil properties are given in Table 1, whereas the correlation coefficients are
given in Table 2. The autocorrelation lengths are lx = 20 m in the horizontal direction
and ly = 2 m in the vertical direction.
Along with the exact (targeted) average values and standard deviations (Table 1)
and correlation coefficients (Table 2), the corresponding values obtained numeri-
cally through simulation are also given. The comparison of targeted and attained
values is very satisfactory for the average values and the correlation coefficients,
and quite satisfactory for the standard deviations. By considering the average value
from a sufficiently large number of realizations (e.g. 200), it has been shown that the
agreement between exact and computed covariance by the LAS method are in very
good agreement [1, 4]. Consequently, the LAS methodology achieves the goal of
generating random property fields with desirable quantitative characteristics.

5 Representative results of random fields

Figure 4a shows an example of a random field of cohesion c, following a normal dis-


tribution with an average value μ = 30 kPa, standard deviation σ = 9 kPa, and auto-
correlation lengths lx= 40 m and ly = 2 m in the horizontal and vertical directions.
Figure 4b shows a similar distribution of cohesion c, except that the autocorrelation
lengths are equal to lx= ly = 2 m in both directions. Comparison of the distribution of
cohesion values in the two figures demonstrates the effect of autocorrelation length
lx . Indeed, in Fig. 4a (lx= 40 m) longitudinal horizontal layers with relatively similar

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Simulation of random fields of soil properties by the local… 847

Fig. 2  Creation of random fields for a cohesion c and b internal friction angle φ with targeted p­ roperty
distribution characteristics given in Tables 1 and 2

cohesion values are created, whereas in Fig. 4b (lx = 2 m) there is no difference in


the distribution of cohesion in the two directions.
Figure 5a shows a similar random field of cohesion values with the same average
μ = 30 kPa, standard deviation σ = 9 kPa and autocorrelation length lx = 40 m in the
horizontal direction, but with autocorrelation length ly = 5 m in the vertical direc-
tion. This distribution compared with that in Fig. 5b corresponding to lx = ly = 5 m
leads to similar conclusions as those in Fig. 4. Moreover, comparison of the random
fields corresponding to lx=ly = 2 m (Fig. 4b) and lx=ly = 5 m (Fig. 5b) shows that the
smaller autocorrelation lengths lead to a more intense local variability, approaching
a state of spatially uncorrelated cohesion values.

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
848 N. Alamanis, P. Dakoulas

Fig. 3  Creation of random fields for a Young’s elastic modulus E and b density ρ with targeted property
distribution characteristics given in Tables 1 and 2

Figure 6 shows random fields of cohesion with an average value μ = 30 kPa


and auto-correlation lengths lx= 40 m in the horizontal direction and ly = 2 m in
the vertical direction. The standard deviation ratio σ/μ assumes values of 0.1, 0.2,
0.3 and 0.4 in Fig. 6a–d, respectively. As expected, the range of cohesion values
in Fig. 6 widens significantly as the standard deviation ratio 𝜎∕𝜇 increases.
Finally, the effect of the correlation coefficient 𝜌c𝜙 between cohesion c and
friction angle 𝜙 is demonstrated in Fig. 7, obtained using the LAS method for the
soil properties given in Table 3. Indeed, for a value of 𝜌c𝜙 = − 0.75 a strong nega-
tive correlation appears between cohesion c and friction angle 𝜙 . By contrast,

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Simulation of random fields of soil properties by the local… 849

Table 1  Comparison of exact and attained soil properties using the LAS method
Exact average Attained average Exact standard Attained
value, μ value, μ deviation, σ standard devia-
tion, σ

Cohesion c (kPa) 30 30.0746 9 7.579


Friction angle φ° 30° 30.0096° 6° 5.111°
Dilation angle ψ° 0° 0° 0° 0°
Density, ρ (kg/m3) 2000 2.0019 200 172
Young’s modulus E (kPa) 60,000 59,687 12,000 10,628
Poisson’s ratio v 0.3 0.3 0 0

Table 2  Comparison of exact Exact value Attained value


and attained correlation
coefficients 𝜌ij 𝜌c𝜙 − 0.5 − 0.5075
𝜌c𝜌 0.5 0.5228
𝜌𝜙𝜌 0.5 0.4920
𝜌cE 0.2 0.1946
𝜌𝜙E 0.2 0.2097

for 𝜌c𝜙 = 0 (not shown here) no correlation appears between cohesion and friction
angle values.

6 Probability of permanent movement of slope due to earthquake

The automated procedure involves development of appropriate random fields of


material properties and numerical analysis of an earth slope using the finite differ-
ence program FLAC [8]. After the initial static analysis, a simulation of the seis-
mic behavior of the slope is performed for the evaluation of the permanent hori-
zontal and vertical displacements at the end of shaking. The numerical analysis is
employed in a Monte Carlo simulation procedure in order to compute the probability
of exceeding different levels of anticipated permanent slope movement at different
levels of peak ground acceleration. The Monte Carlo simulations are repeated for
five different seismic excitations and for six different levels of peak ground excita-
tion ranging from 0.05 to 0.5 g.

6.1 Numerical model

Figure 8 illustrates the geometry and the numerical discretization of a slope with
inclination equal to 2:1. The constitutive model used for the soil is the classic
elastoplastic Mohr–Coulomb model, which is defined by the cohesion c, fric-
tion angle φ, density ρ, modulus of elasticity Young E, Poisson’s ratio ν, and

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
850 N. Alamanis, P. Dakoulas

Fig. 4  Effect of autocorrelation length in the distribution of cohesion c: a lx = 40 m, ly = 2 m and b


lx = 2 m, ly = 2 m

dilatancy angle ψ. All material properties are considered as randomly varied in


space, except of the angle of dilatancy ψ and Poisson’s ratio ν which are taken
as constant. Each of the above properties is considered as a Gaussian random
field defined by its mean value 𝜇 , variance 𝜎 2 and the autocorrelation function
𝜌(x, y) given by Eq. (1). The mean values, μ, and standard deviations, σ, of soil
properties used in this study are given in Table 4. The correlation coefficients 𝜌ij
of the above soil properties, obtained from available published data, are given in
Table 5 [6]. The values of the autocorrelation lengths are taken lx = 20 m for the
horizontal direction and ly = 2 m for the vertical direction. For the seismic analy-
sis, the Mohr–Coulomb constitutive model is further enhanced with a hysteretic

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Simulation of random fields of soil properties by the local… 851

Fig. 5  Effect of autocorrelation length in the distribution of cohesion c: a lx = 40 m, ly = 5 m and b


lx = 5 m, ly = 5 m

sig3 model [8] with parameters a = 1, b = − 0.55 and xo = − 1.22, allowing a


proper modelling of the hysteresis loops with respect to evolution of shear mod-
ulus degradation and material damping during cyclic loading.

6.2 Earthquake excitations

The numerical excitation consists of a series of acceleration records obtained from


the Kalamata (1986), Lefkada (2003), Kobe (1995), Friuli (1976) and Northridge
(1994) earth-quakes. Table 6 provides the key elements of historical records. These
records have been adjusted to match approximately the Eurocode 8 acceleration

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
852

13
Fig. 6  The effect of standard deviation ratio 𝜎∕𝜇 on the distribution of the cohesion: a 𝜎∕𝜇 = 0.1, b 𝜎∕𝜇 = 0.2, c 𝜎∕𝜇 = 0.3 and d 𝜎∕𝜇 = 0.4 (𝜇 = 30 kPa)
N. Alamanis, P. Dakoulas

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Simulation of random fields of soil properties by the local… 853

Fig. 7  Correlation between cohesion c and friction angle 𝜙 for 𝜌c𝜙 = − 0.75

Table 3  Soil properties for the Cohesion c (kPa) Friction angle (φ°)
analyses in Fig. 7
Average value 30 30°
Standard deviation 9 6°
Autocorrelation length, (m) lx = 40, ly = 2 lx = 40, ly = 2

spectra for hard soil and rock sites, calibrated for a peak ground acceleration of
0.3 g. Figure 9 plots the acceleration response spectra of the five excitations and the
Eurocode 8 design spectra.

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
854 N. Alamanis, P. Dakoulas

Fig. 8  Slope geometry and numerical discretization

Table 4  Soil properties for Average value, μ Standard


seismic analysis of slope devia-
tion, σ

Cohesion c (kPa) 30 9
Friction angle (φ°) 20° 6°
Dilation Angle (ψ°) 0° 0°
Density, ρ (kg/m3) 2000 200
Young’s modulus E (kPa) 60,000 12,000
Poisson’s ratio (v) 0.3 0

Table 5  Historical seismic excitation records


Earthquake Mw R (km) Recording Component PGA (g)

Kalamata (1986) 6.0 12 Prefecture Hor. 0.25


Lefkada (2003) 6.4 10 Lefkada Trans. 0.60
Kobe (1995) 7.2 20 Port Island Horizontal 57
Northridge (1994) 6.7 30 Rinaldi Hor. 318 47
Friuli (1976) 6.5 19 Friuli Hor. 35

6.3 Seismic analysis—permanent displacements

Figure 10 illustrates a typical example from one random field realization of the hori-
zontal permanent displacement ux at the end of the seismic shaking, having a maxi-
mum value of about 2.10 m in the area of the lower half of the slopes. The results
show that a significant mass of soil moves to the left due to the inertial forces that
develop during seismic vibration.
To investigate the effect of the spatial variability of soil properties, it is desir-
able to separate the effect of the characteristics of the seismic excitation used
in the analysis from the influence of spatial variability of material properties.
To this end, the ratio fx = ux ∕̄ux is considered, where ux is the permanent hori-
zontal displacement at the end of the shaking and ū x is the permanent horizontal

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Simulation of random fields of soil properties by the local… 855

Fig. 9  Acceleration spectra of modified seismic excitations and Eurocode 8 spectra for rock sites scaled
at peak ground acceleration equal to ag = 0.3 g

Fig. 10  Permanent horizontal displacement at the end of shaking (Input excitation: Lefkada 2003 record,
scaled to a ground acceleration of ag = 0.3 g)

displacement of a homogeneous slope with property values equal to the average


values of those of the heterogeneous slope.
Figure 11a shows the probability density of the results of permanent horizontal
displacement ratio fx = ux ∕̄ux at the top of the slope (point A in Fig. 8) from all ran-
dom filed realizations and the five earthquake excitations for a peak ground accel-
eration of 0.3 g. Similarly, Fig. 11b plots the probability density of the permanent
vertical displacement ratio fy = uy ∕̄uy at the top of the slope. As shown in Fig. 11a,
the average permanent horizontal displacement of slope A with spatial variability of
properties is about 19% greater than the horizontal displacement of a homogeneous
slope with the same geometry subjected to the same excitation. The maximum theo-
retical deviation of the displacement corresponding to μ +3σ is equal to 1.89, i.e.
89% greater than the horizontal displacement value of the homogeneous slope. This
means that about 78% of cases of heterogeneous slopes will undergo permanent dis-
placements that will be larger than those predicted for homogeneous slopes.
Moreover, the results of all analyses show that, in addition to the spatial vari-
ability of soil properties, the distribution of the permanent horizontal and vertical
displacements (ux and uy ) depends substantially on the frequency characteristics

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
856 N. Alamanis, P. Dakoulas

Fig. 11  Effect of spatial variability of soil properties on the permanent displacement at the end of shak-
ing a horizontal displacement ratio fx = ux ∕̄ux and b vertical displacement ratio fy = uy ∕̄uy, in which ū x
and ū y are the corresponding displacements for a homogeneous embankment having the same mean value
of properties (Peak ground excitation ag = 0.3 g)

of the seismic excitation. Figure 12a, b plot the distribution of the permanent
horizontal displacement ux and (b) vertical displacement uy , from all numerical
simulations using the geometry in Fig. 8 subjected to peak ground acceleration
of 0.3 g. The range of variation for ux is from 0.5 to 2.5 m and for uy from 0.4 to
1.6 m. The numerical results in Fig. 12 are best fitted with a Weibull distribution
having a probability density function given by
( )𝛼 ( )𝛼−1
𝛼 − 𝜒−𝜇 𝜒 −𝜇
f (𝛼, 𝛽, 𝜇, 𝜒) = e 𝛽 𝜒 ≥𝜇 (3)
𝛽 𝛽

in which 𝛼 = 2.2334, 𝛽 = 0.8632 and 𝜇 = 0.421 for the horizontal displacement and
𝛼 = 3.1081, 𝛽 = 0.688 and 𝜇 = 0.3529 for the vertical displacement.
In addition, the effect of the intensity of the seismic excitation was investigated
by considering peak ground accelerations values ag equal to 0.05 g, 0.1 g, 0.2 g,
0.3 g, 0.4 g and 0.5. The results from the analyses for each peak ground accelera-
tion value are used to produce the corresponding parameters of the fitting Weibull

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Simulation of random fields of soil properties by the local… 857

Fig. 12  Distribution of permanent displacement at the end of shaking a horizontal displacement ux and b
vertical displacement ratio uy for all earthquake excitations scaled at a peak ground excitation of ag = 0.3 g

Table 6  Correlation coefficient Properties Correlation coefficient 𝜌ij


matrix 𝜌ij for seismic analysis
of slope c φ ψ ρ E ν

c 1 − 0.5 0 0.5 0.2 0


φ − 0.5 1 0 0.5 0.2 0
ψ 0 0 1 0 0 0
ρ 0.5 0.5 0 1 0 0
E 0.2 0.2 0 0 1 0
ν 0 0 0 0 0 1

distributions for the horizontal and vertical displacements, which are summarized in
Table 7.

6.4 Slope fragility curves for earthquake loading

The results of the analyses in described in 6.3 allows a comprehensive statisti-


cal description of the permanent displacements after shaking for all random field

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
858 N. Alamanis, P. Dakoulas

Table 7  Probability density function coefficients of Weibull distribution for the horizontal and vertical
displacements at different levels of peak ground acceleration
Acceleration, (g) Horizontal displacement Vertical displacement
α β μ α β μ

0.05 2.2324 0.0587 0.0286 3.1081 0.2809 0.1441


0.10 2.2324 0.1661 0.0810 3.1081 0.3972 0.2037
0.20 2.2324 0.4699 0.2291 3.1081 0.5618 0.2881
0.30 2.2324 0.8632 0.4210 3.1081 0.6880 0.3529
0.40 2.2324 1.3290 0.6481 3.1081 0.7945 0.4075
0.50 2.2324 1.8574 0.9058 3.1081 0.8882 0.4556

realizations, the five earthquake excitations and the peak ground accelerations rang-
ing from 0.05 to 0.5 g. By combining the effects of the spatial variability of soil
properties, the frequency characteristics of excitation and the nonlinearity due to
intensity, the probability of exceedance of a given value of displacement for differ-
ent levels of peak ground acceleration is evaluated. Figure 13a, b plot the probability
of exceedance of various levels of permanent displacement in the horizontal and
vertical direction respectively, for different levels of peak ground accelerations. For
example, for a peak ground acceleration of 0.3 g, the probability of exceeding a per-
manent horizontal movement of 1 m at the top of the slope is given from Fig. 13a
equal to about 67%.

6.5 Seismic performance of soil‑pipeline systems

The above methodology may also be applied to natural slopes with overlying or
embedded oil and natural gas pipeline systems in order to evaluate the probability
of global or local buckling of the pipeline. To this end, a simplified modelling of the
pipe—soil system would allow the use of Monte Carlo simulation to account for the
uncertainties related to the spatial variability of soil properties, the frequency con-
tent, and the intensity of the earthquake excitation.

7 Conclusions

This paper investigated the effect of uncertainty related to both the natural heteroge-
neity of geomaterials and the earthquake shaking characteristics on the seismic per-
formance of slopes. An automated methodology for generating random fields of tar-
geted statistical characteristics was developed using the Local Average Subdivision
method and combined with the finite difference program FLAC for seismic analysis.
The methodology was used in a Monte-Carlo simulation scheme for computing the
probability of exceeding different levels of anticipated permanent slope movement
damage.
The main conclusions of this study are the following:

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Simulation of random fields of soil properties by the local… 859

Fig. 13  Probability of permanent displacement u exceeding a certain value u* versus peak ground accel-
eration a horizontal displacement and b vertical displacement

1. The automated Local Average Subdivision methodology for the development of


cross-correlated and autocorrelated random fields of material properties allowed
an efficient realization of random fields that satisfy successfully the target statisti-
cal parameters. Such random fields can be used for numerical analysis in Monte
Carlo simulations to investigate the effects of spatial variability of soil properties
on the seismic performance of various geotechnical systems.
2. Investigation of the seismic behavior of the examined slope using random fields
shows that the effect of the spatial variability of the soil properties on the per-
manent displacements is very important. The range of variation of the perma-
nent horizontal and vertical displacement is about ± 60% from its mean value.
Moreover, it is found that a slope with spatial variability of soil properties has
a 78% probability to experience larger horizontal displacements than a similar
homogeneous slope.

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
860 N. Alamanis, P. Dakoulas

3. The effect of the uncertainty of the frequency content of the earthquake excita-
tion is also important. For horizontal displacements it leads to a variation of
about ± 40% and for vertical displacements to a variation of ± 20%.
4. The statistical analysis of the results of the extensive parametric investigation
leads to the development of slope fragility curves, which provide the probability
of exceedance of various levels of permanent displacement in the horizontal and
vertical direction for different levels of peak ground accelerations. Such curves
may be useful in the design of pipeline systems embedded in or founded on such
slopes.

References
1. Fenton, A.G., Vanmarcke, E.H.: Simulation of random fields via local average subdivision. J. Eng.
Mech. 116(8), 1733–1749 (1990)
2. Vazouras, P., Dakoulas, P., Karamanos, S.A.: Soil–structure interaction effects of steel pipelines
crossing active seismic faults. J. Soil Dyn. Earthq. Eng. 72, 45–65 (2015)
3. Sarvanis, G., Karamanos, S.A., Vazouras, P., Mecozzi, E., Lucci, A., Dakoulas, P.: Permanent
ground-induced actions in buried pipelines: numerical modeling and experimental verification. J.
Earthq. Eng. Struct. Dyn. 47(4), 966–987 (2017)
4. Fenton, A.G., Griffiths, D.V.: Risk Assessment in Geotechnical Engineering. Wiley, Hoboken
(2008)
5. Griffiths, D.V., Fenton, G.A.: Probabilistic Methods in Geotechnical Engineering. International
Centre for Mechanical Sciences. Springer Wien, New York (2007)
6. Alamanis, N.O.: Effect of the spatial variability of soil properties on the permanent seismic dis-
placements of road slopes. Ph.D. thesis, University of Thessaly, Department of Civil Engineering,
Volos, Greece (2017)
7. Wolfram Research, Inc.: Mathematica, version 11. Champaign, IL (2018). www.wolfr​am.com
8. Itasca Consulting Group: Fast Langrangian Analysis of Continua, FLAC v.7. User Manual, Minne-
apolis, MN (2011)
9. Cooley, J.W., Tukey, J.W.: An algorithm for the machine calculation of complex Fourier series.
Math. Comput. 19(90), 297–301 (1965)
10. Matheron, G.: The intrinsic random functions and their applications. Adv. Appl. Probab. 5, 439–468
(1973)
11. Mantoglou, A., Wilson, J.L.: Simulation of Random Fields with the Turning Bands Method. Mas-
sachusetts Institute of Technology, Cambridge (1981). (Report 264)
12. Griffiths, D.V., Fenton, G.A.: Probabilistic slope stability analysis by finite elements. J. Geotech.
Geoenviron. Eng. 130(5), 507–518 (2004)
13. Fenton, G.A., Griffiths, D.V., Urquhart, A.: A slope stability model for spatially random soils. In:
Kiureghian, A., et al. (eds.) Proc. 9th Int. Conf. Applications of Statistics and Probability in Civil
Engineering (ICASP9), pp. 1263–1269. Millpress, San Francisco, CA (2003)
14. Griffiths, D.V., Huang, J., Fenton, G.A.: Influence of spatial variability on slope reliability using
2-D random fields. J. Geotech. Geoenviron. Eng. 135, 1367–1375 (2009)
15. Griffiths, D.V., Huang, J., Fenton, G.A.: Probabilistic infinite slope analysis. Comput. Geotech. 38,
577–584 (2011)
16. Fortsakis, P., Stylianidi, E., Kavadas, M.: Slope Stability Analysis using stochastic methods. In: 6th
Panhellenic Conference of Geotechnical and Geoenvironmental Engineering, TEE, pp. 1–8. (2010)
17. Vanmarcke, E.H.: Probabilistic modeling of soil profiles. J. Geotech. Eng. 103(11), 1227–1246
(1977)
18. Smith, I.M., Griffiths, D.V.: Programming the Finite Element Method, 4th edn. Wiley, New York
(2004)

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Simulation of random fields of soil properties by the local… 861

19. Gentle, J.E.: Cholesky Factorization. Numerical Linear Algebra for Applications in Statistics, pp.
93–95. Springer, Berlin (1998)
20. Babu, S.G.L., Mukesh, M.O.: Effect of soil variability on reliability of soil slopes. Geotechnique
54(5), 335–337 (2004)
21. Cho, S.E.: Effects of spatial variability of soil properties on slope stability. Eng. Geol. 92, 97–109
(2007)
22. Cheng, Y.M., Lau, C.K.: Slope Stability Analysis and Stabilization, pp. 138–151. Routledge, New
York (2008)
23. Low, B.K., Lacasse, S., Nadim, F.: Slope reliability analysis accounting for spatial variation.
Georisk Assess. Manag. Risk Eng. Syst. Geohazards 1(4), 177–189 (2007)
24. Wolff, T.F.: Probabilistic slope stability in theory and practice. In: Shackelford, C.D., et al. (eds.)
Uncertainty in the Geological Environment: From Theory to Practice. Geotechnical Special Publi-
cation No. 58, pp. 419–433. ASCE, New York (1996)
25. Wu, X.Z.: Trivariate analysis of soil ranking correlated characteristics and its application to proba-
bilistic stability assessments in geotechnical engineering problems. Soils Found. 53(4), 540–556
(2013)

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published
maps and institutional affiliations.

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center
GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers
and authorised users (“Users”), for small-scale personal, non-commercial use provided that all
copyright, trade and service marks and other proprietary notices are maintained. By accessing,
sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of
use (“Terms”). For these purposes, Springer Nature considers academic use (by researchers and
students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and
conditions, a relevant site licence or a personal subscription. These Terms will prevail over any
conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription (to
the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of
the Creative Commons license used will apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may
also use these personal data internally within ResearchGate and Springer Nature and as agreed share
it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not otherwise
disclose your personal data outside the ResearchGate or the Springer Nature group of companies
unless we have your permission as detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial
use, it is important to note that Users may not:

1. use such content for the purpose of providing other users with access on a regular or large scale
basis or as a means to circumvent access control;
2. use such content where to do so would be considered a criminal or statutory offence in any
jurisdiction, or gives rise to civil liability, or is otherwise unlawful;
3. falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association
unless explicitly agreed to by Springer Nature in writing;
4. use bots or other automated methods to access the content or redirect messages
5. override any security feature or exclusionary protocol; or
6. share the content in order to create substitute for Springer Nature products or services or a
systematic database of Springer Nature journal content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a
product or service that creates revenue, royalties, rent or income from our content or its inclusion as
part of a paid for service or for other commercial gain. Springer Nature journal content cannot be
used for inter-library loans and librarians may not upload Springer Nature journal content on a large
scale into their, or any other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not
obligated to publish any information or content on this website and may remove it or features or
functionality at our sole discretion, at any time with or without notice. Springer Nature may revoke
this licence to you at any time and remove access to any copies of the Springer Nature journal content
which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or
guarantees to Users, either express or implied with respect to the Springer nature journal content and
all parties disclaim and waive any implied warranties or warranties imposed by law, including
merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published
by Springer Nature that may be licensed from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a
regular basis or in any other manner not expressly permitted by these Terms, please contact Springer
Nature at

onlineservice@springernature.com

You might also like