You are on page 1of 4

Comparison of Three Covariance Functions on Stochastic Settlement

Analysis of Shallow Foundation in Anisotropy Field


Ryan Kurnia

Universiti Teknologi Petronas, Malaysia

Indra Sati Hamonangan Harahap

Universiti Teknologi Petronas, Malaysia

ABSTRACT: In recent decades, there has been a remarkable increase application of probabilistic methods to
more traditional areas of geotechnical engineering (i.e. random field theory). Random field is fully characterized by its mean and correlation or covariance function. Hence, it is very important carefully investigate the
type of covariance function used to generate the random field specific to the task. In this research random finite element method will be used to quantify the reliability of shallow foundation settlement. Three types of
covariance functions are used to generate the random field their effectiveness are assessed and analyzed on
this specific problem. The covariance function type II seem more consistent to this task, due to it produce the
lowest variance and the settlement distribution is reasonably fit the lognormal distribution. However, maybe
another covariance function is probably more suitable for another problem. Therefore, it is necessary to check
the effectiveness of covariance function in small scale before a move to a larger scale in every specific task.
1 INTRODUCTIONS

The safety factor is frequently used in conventional geotechnical practices when dealing with uncertainties. However, it is not very logical strategy
since the same safety factor is used for different type
of application without regard to the degree of uncertainty involved in the calculation. Meanwhile, the
probabilistic method allows the geotechnical engineer to evaluate the combining effect of uncertainties (Duncan 2000), in others word the degree of uncertainties are taken into account. Hence, using the
probabilistic method seems more logical. The good
and comprehensive example of the use of probability
in geotechnical engineering have been presented by
Christian et al. (Christian et al. 1994), Tang et al.
(Tang et al. 1999). Furthermore, Duncan (Duncan
2000) had presented the probability theory can be
applied in simple way without more additional data
and time.
One of a rigorous method to apply the probability
theory in geotechnical engineering can be achieved
by representing each point of the site using random
variable (i.e. Random processes or field) (Fenton &
Griffiths 2008) (Vanmarcke 2010). Numerous studies have been conducted in recent years to develop
the application of random field theory in the geotechnical engineering field. For example, reliability
analysis using random field theory combined with

Monte-Carlo simulation (Cho 2007; El-Ramly et


al. 2002), using random finite element method
(RFEM) (Griffiths & Fenton 2004; Huang et al.
2010), and reliability analysis using RFEM combined with Monte-Carlo simulation and first-order
reliability method (FORM) analysis (Griffiths et al.
2009).
To characterize the random field we need to
know the field mean, field variance and how rapidly
the field varies in space (Assessment et al. 2008).
Covariance function is the convenient measure in
order to get some idea how the field varies in space.
Several covariance functions are available in literature i.e. triangular covariance function, polynomial
decaying covariance function, Markov covariance
function, and Gaussian covariance function. The
Markov and Gaussian covariance function are used
this study due to its simplicity and it only require the
scale of fluctuation or correlation length () and variance ( ) to compile the covariance structure of the
field. Random finite element method (RFEM) is
used extensively in this study. This method has been
successfully used by severer researcher (Huang et al.
2010; Vessia et al. 2009; Paiboon et al. 2013)
Random field is fully characterized by its mean and
covariance function (Santoso et al. 2011)

(Assessment et al. 2008). Furthermore, the nature of


soil and rock are varied through space. Therefore,
careful assessment covariance function must be conducted in each specific geotechnical task. The purpose of this study is to assess several types of covariance function in reliability assessment of settlement
of single shallow foundation. The complete explanation of the covariance functions will be fully explained in the methodology section.
2 METHODOLOGY

2.1 Text and indenting

The simulation is started to input the data, setting


the finite element properties, and choose the covariance function for simulation. Started with covariance
function type I. Then generate the random field realization up to 3000 realization. The realization then
mapped into finite element mesh, and finally the settlement of foundation can be evaluated. The simulations are conducted using the random finite element
method (RFEM) software. Which are the combination of random field theory, the classical finite element method, and Monte Carlo simulations developed by Fenton and Griffiths (Assessment et al.
2008). After all the process complete, then the same
process is conducted with covariance function type
II and III.
The model is illustrated in Fig. 2 to Fig.4 the soil
mass is discretized into 60 and 20 four node quadrilateral elements vertical and horizontal respectively.
The size of an element is 0.05 meters for vertical and
horizontal. Therefore, the real size of soil deposit is
3x1 meters. 60 x 20 discretization was considered
adequate by the previous researcher (Fenton &
Griffiths 2002). The foundation is assumed to be rigid, possess the rough interface with underlying soil
and do not undergo any rotation. As a consequence,
the finite element model is constrained against horizontal movement but not vertically. The nodes in the
bottom boundary are fixed. A fixed load P=1 N/m is
applied to the footing, 1 m also for footing width.
Because of a linear function of load, the results are
easily scaled to different values of P and footing
width.
The settlement is predicted using elastic theory.
Therefore, the elastic modulus (E) and the Poisson
ratio (v) are used in this study. Only elastic modulus
is considered random and assumed to follow the
lognormal distribution with mean 1 and standard deviation 0.3. The lognormal distribution is chosen due
to the elastic modulus value is strictly nonnegative
and it properties still possesses a simple relationship

with the normal distribution. The poison ratio (v) is


fixed with 0.25 for the entire soil mass in the simulations. The random field is anisotropy with correlation length () 2 meters in the horizontal direction by
1 meter in the vertical direction.
The covariance functions used in this study are
Markovian covariance function and Gaussian covariance function. For Markovian covariance function
there are two ways to determine the covariance between two points; based on absolute distance, later it
called covariance function type I and according to
the product of the 2 directional 1-D Markovian covariance functions, later it called covariance function
type II. For the Gaussian counterpart which later
called covariance function type III. Thus, there is
three type of covariance functions are considered in
this study.
For calculation of covariance structure for type I
can be calculated using Equation.1. for covariance
function type II can be calculated using Equation.2,
and for covariance function type III can be calculated using Equation.3.
( , )=
( , )=
( , )=

(1)

+
| |

| |

(2)
(3)

Where: ( , )= Covariance between two points;


= Horizontal distance between two points (m);
= Vertical distance between two points (m);
=
Horizontal correlation distance (m);
=
vertical
correlation distance (m)
3 RESULT AND DISCUSSIONS

Figure 2 to Figure 3 shows one of a possible realization of the elastic modulus field along with the
mesh for covariance function type I, type II, and
type III respectively. A gray-scale represents the
value of elastic modulus. Lighter areas denote
smaller values of E. While the mesh representing the
deformation. Although, this is just one possible realization the next realization could be different, or
even opposite. The elastic modulus fields, as we can
see in Fig. 2 to Fig 4 indicate that the change of elastic modulus seems smoother for Gaussian covariance function compared to its counterpart the Markovian covariance functions.

cepted while the Gaussian is rejected with p-value


almost zero. Interestingly, the covariance functions
type II is the most fitted to the lognormal distribution, with p-value 0.524573, the highest one.

Type I

Type II

Probability

0.09
0.08
0.07
0.06
0.05
0.04
0.03
0.02
0.01
0

Type III

Settlement

Figure 1. RFEM model for type II covariance function.

In Fig.1, three typical histograms of the settlement under the foundation are generated using three
different covariance functions. The histograms are
normalized. The settlement distributions are predicted using 3000 realizations. The shape of all three
histogram appears to resemble the lognormal distribution. However, when chi-square the goddess of fit
is used to evaluate the goodness of the fit the data to
a lognormal distribution, the hypothesis are then are
accepted for Markovian covariance functions and
rejected for Gaussian covariance function with the
P-value just around 8.04E-12 (see Table 1). It is very
small since large p-values up to 1.0 support the
lognormal hypothesis. However, the chi-square test
is quite sensitive to the smoothness of histogram,
particularly when the number of samples is large.
Since the histogram for settlement distribution that
generated from Gaussian covariance function is
strongly resembles the lognormal distribution, the
rejection may come from the roughness of the histogram. Fenton and Griffiths (Fenton & Griffiths
2002) suggested that still reasonable to assume that
the distribution fit the lognormal distribution.

Figure 3. RFEM model for type II covariance function.

Figure 4. RFEM model for type III covariance function

Therefore, the covariance function type II is strongly


recommended to generate the random field for this
problem. The previous research (Cheon & Gilbert
2014) has also chosen this type of covariance function to model the spatial variability in the Gulf of
Mexico.
Table 1. Summary of simulation result
Mean

Covariance Function
Type I

1.29

0.28

Accepted

Type III

1.30

0.32

Rejected

No
2

Figure 2. RFEM model for type I covariance function.

The means resulted from three covariance function appears to be consistent, only slightly different
as we can see in Table 1, However, Markovian covariance function (type II). That is, the covariance
between points in the field decays according to the
product of the 2 directional 1-D Markovian covariance functions produced the lowest variance. In term
of the goodness of the fit of the simulation result to
the lognormal distribution, the Markovian are ac-

Type II

1.28

4 CONCLUSSIONS

0.26

Chi-square test
Hypothesis

Accepted

P-value

0.19197

0.52457
8.04E12

The quantification of the uncertainties in the geotechnical field can be achieved through the random
field theory. Because of the role of covariance function is significant to random field analysis, the careful analysis is needed to guide the engineer choose
the suitable covariance for the specific task. As a result, the type II covariance function is more suitable
for this type of problem and the problem similar to
this problem. However, the analysis in this research
only uses the small scale. In addition, it only uses

single footing, thus, the analysis of partial settlement


cannot be incorporated. Using the larger scale analysis with multiple footing is needed to verify the consistency of the result.
5 REFERENCES

Assessment, R. et al., 2008. Risk Assessment in


Geotechnical Engineering,
Cheon, J.Y. & Gilbert, R.B., 2014. Modeling spatial
variability in offshore geotechnical properties
for reliability-based foundation design.
Structural Safety, 49, pp.1826.
Cho, S.E., 2007. Effects of spatial variability of soil
properties on slope stability. Engineering
Geology, 92(3), pp.97109.
Christian, J.T., Ladd, C.C. & Baecher, G.B., 1994.
Reliability applied to slope stability analysis.
Journal of Geotechnical Engineering, 120(12),
pp.21802207.
Duncan, J.M., 2000. Factors of safety and reliability
in geotechnical engineering. Journal of
geotechnical
and
geoenvironmental
engineering, 126(4), pp.307316.
El-Ramly, H., Morgenstern, N.R. & Cruden, D.M.,
2002. Probabilistic slope stability analysis for
practice. Canadian Geotechnical Journal,
39(3), pp.665683.
Fenton, G.A. & Griffiths, D. V, 2002. Probabilistic
foundation settlement on spatially random soil.
Journal of Geotechnical and Geoenvironmental
Engineering, 128(5), pp.381390.
Fenton, G.A. & Griffiths, D.V., 2008. Risk
assessment in geotechnical engineering, John
Wiley & Sons Incorporated.
Griffiths, D. V & Fenton, G.A., 2004. Probabilistic
slope stability analysis by finite elements.
Journal of Geotechnical and Geoenvironmental
Engineering, 130(5), pp.507518.
Griffiths, D. V, Huang, J. & Fenton, G.A., 2009.
Influence of spatial variability on slope
reliability using 2-D random fields. Journal of
Geotechnical
and
Geoenvironmental
Engineering, 135(10), pp.13671378.
Huang, J., Griffiths, D. V & Fenton, G.A., 2010.
System reliability of slopes by RFEM. Soils
and Foundations, 50(3), pp.343353.
Paiboon, J. et al., 2013. International Journal of
Solids and Structures Numerical analysis of
effective elastic properties of geomaterials
containing voids using 3D random fields and
finite elements. International Journal of Solids
and Structures, 50(20-21), pp.32333241.
Available
at:
http://dx.doi.org/10.1016/j.ijsolstr.2013.05.031.
Santoso, A.M., Phoon, K. & Quek, S., 2011. Effects
of soil spatial variability on rainfall-induced
landslides. Computers and Structures, 89(1112),
pp.893900.
Available
at:
http://dx.doi.org/10.1016/j.compstruc.2011.02.0
16.

Tang, W.H., Stark, T.D. & Angulo, M., 1999.


Reliability in back analysis of slope failures.
Soil and foundations, 39(5), pp.7380.
Vanmarcke, E., 2010. Random fields: analysis and
synthesis, World Scientific.
Vessia, G. et al., 2009. Application of random finite
element method to bearing capacity design of
strip footing. Journal of GeoEngineering, 4(3),
pp.103112.