You are on page 1of 27

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/317589800

Numerical and Probabilistic Analyses of Deep


Excavations in Soft Soils

Article in Electronic Journal of Geotechnical Engineering · June 2017

CITATIONS READS

0 18

2 authors:

Carlos Javier Sainea Vargas Mario Camilo Torres Suarez


National University of Colombia National University of Colombia
6 PUBLICATIONS 0 CITATIONS 6 PUBLICATIONS 0 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Geotechnical behavior of mudrocks View project

Pile groups under seismic loading View project

All content following this page was uploaded by Carlos Javier Sainea Vargas on 14 June 2017.

The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
Numerical and Probabilistic
Analyses of Deep Excavations in Soft
Soils

Carlos J. Sainea-Vargas* and Mario C. Torres-Suárez


Universidad Nacional de Colombia, Bogotá
*e-mail: cjsaineav@unal.edu.co

ABSTRACT
In densely populated cities deep excavations are commonly required for different infrastructure works.
Often those urban areas are located in soft soils, materials of unfavorable mechanical behavior given
its low strength and high compressibility. In the analysis of deep excavations in soft soils numerical
techniques as finite element method are usually required with appropriate constitutive models. But soft
soil deposits as those present in Bogotá are also inherently variable, so the analyses should be
performed using probabilistic approaches. Numerical analysis allows to predict ground movements
around the excavation and probabilistic approaches are useful to assess the safety conditions of the
works for a given limit state. In this paper serviceability limit states were considered on an idealized
three-dimensional finite element model, in terms of potential damages generated in adjacent structures.
In numerical modelling two different constitutive models were used to represent soil behavior:
Hardening Soil (HS) and Hardening Soil Small Strain (HSsmall). Also, three different probabilistic
methods were considered to assess safety conditions of the excavations: Point Estimate Method
(PEM), Monte Carlo Simulation with a Latin Hypercube Sampling scheme (MCS-LHS) and Response
Surface Method (RSM). In numerical and probabilistic analyses, soil constitutive parameters were
treated as random variables and their selection is based on multivariate and sensitivity analyses. Finally
the results obtained are compared and discussed.
KEYWORDS: Deep excavation; probabilistic analysis; serviceability limit states; soft soils;
three-dimensional finite element analysis

INTRODUCTION
In densely populated cities it has been sought to take advantage of subsoil with several types of
infrastructure works in which deep excavations are needed. Main aspects about analysis and design of
deep excavation in soft soils are explained in detail by different authors [1,2,3]. Often soft soils are
found in urban areas where excavations have to be made, and their presence could lead to substantial
difficulties since they are highly compressible normally or lightly consolidated materials with very
soft to soft consistency and frequently are found in a saturated or nearly fully saturated state [1].
Prediction of ground movements around deep excavations in soft soils is a complex problem in which
the use of numerical techniques is often needed and among those techniques finite element method
has become predominant [4]. With the aim of obtaining good predictions about the performance of
excavations, appropriate soil constitutive models must be used according to the problem under
consideration and soils found in study area; also, reasonable estimates of constitutive parameters

- 3899 -
Vol. 22 [2017], Bund. 10 3900

values should be made. In this paper Hardening Soil [5,6] and Hardening Soil Small Strain [6,7]
constitutive models were considered.
Soils as those present in Bogotá are naturally variable materials, so deterministic approaches
would not be applicable but rather those based on uncertainty in which model parameters such as
geometry, loads or constitutive parameters of materials are considered as random variables [8]. The
system response can also be considered as a random variable and the performance of geotechnical
structures can be evaluated in terms of probabilities of failure or undesired performance in reliability
based analysis [9,10]. In this paper constitutive parameters of Hardening Soil (HS) and Hardening
Soil Small Strain (HSsmall) models are considered as random variables. But these constitutive
parameters are numerous, and if all were taken as random variables, the computational effort required
in probabilistic simulations would be increased to the point of discouraging the use of an uncertainty
based approach. So, before performing numerical and probabilistic analyses a parameter selection
were made based on multivariate and sensitivity analyses.
In this paper three different probabilistic methods were used to assess the performance of
excavations: Point Estimate Method (PEM) [9,11,12], Monte Carlo Simulation with a Latin
Hypercube Sampling scheme (MCS-LHS) [9,10,13] and Response Surface Method (RSM)
[12,14,15,17]. For the first two methods, constitutive parameters selection as random variables was
made considering total variability in parameters and using multivariate techniques and sensitivity
analysis [16,17,18,19]. For the third method, selection of random variables was made after removing
trends with depth and combining different sources of uncertainty by means of a simple second-
moment probabilistic approach [20,21]. Details of parameter selection are not part of the scope of this
paper but further explanations can be found in previous documents [22]. In each case different series
of three-dimensional finite element models were elaborated and analyzed. Numerical simulations
results of ground movements in different points were obtained and used to estimate possible damages
in buildings adjacent to excavations using the Damage Potential Index (DPI) [23,24]. Results
obtained by the three approaches are compared and discussed.

METHODS
Numerical methods
Base information
Base information was taken from a deep excavation project in lacustrine soft soils of Bogotá [25].
At the site three borings with sampling were made between 30 and 70 m depth, and also field tests
were performed: 8 SCPTu between 31 and 46 m depth, 1 Marchetti Dilatometer test (DMT) of 41 m
depth and 1 pressuremeter test (PMT) register up to 30m depth. Samples obtained were tested for
measurement of index properties, physical characteristics and strength and compressibility properties.
Water level in the ground was found at 1.5 to 2 m depth. In surface it was found a fill layer of 1.5 to 2
m of thickness, 38 to 42 m of silty-clay soils with soft consistency classified as CH, MH and OH,
followed by intercalations of more competent materials such as firm clays and sands, and others less
competent as clayey silts and with sand lenses up to about 60 m, from where gravels of sandstone
were found. Soils of interest in this paper are those present from 2 to 42 m depth and results of
SCPTu tests indicate that correspond to clays and silty clays with presence of organic soils, mainly
located in normalized behavior types 3 and 4 [26].
Soil constitutive models and parameters
To simulate the behavior of normal and slightly overconsolidated clayey soils in excavations, the
use of HS and HSsmall models is a good choice [27,28,29,30]. HSsmall model is an extension of HS
Vol. 22 [2017], Bund. 10 3901

model to include soil behavior at small deformations. These are two constitutive models of isotropic
hardening and two yielding surfaces, one of them with non-associated flow and the other with
associated flow. The ultimate state is defined by Mohr-Coulomb failure criterion. In deviatoric
envelope plastic potential is defined to ensure a hyperbolic stress-strain response in triaxial
compression and in volumetric limit surface the hardening parameter corresponds to preconsolidation
pressure. These models allow the representation of aspects of nonlinear behavior of soil including
stress-dependent stiffness, effects of stress history with evolution of preconsolidation pressure and
different stiffness in loading and unloading. HSsmall model also allows to consider stiffness decrease
with increasing shear strain amplitude from small to larger strains [28,29].
The parameters in common for both models are cohesion c’, tensile strength ft, friction
angle φ' and dilatancy angle ψ to stablish Mohr-Coulomb envelope, parameter m to define
stress dependent stiffness according to a potential law, secant stiffness at 50% of ultimate
deviator stress E50ref, unloading and reloading stiffness at engineering strains Eurref, tangent
stiffness for primary oedometric load Eoedref, reference stresses for stiffness σref and σoedref,
Poisson ratio for unloading-reloading cycles υur, deviator stress ratio at failure Rf,
overconsolidation ratio OCR, coefficient for lateral pressure at rest k0, maximum void ratio in
the ultimate state emax, a parameter to control the law of dilatancy in the contractancy domain
D, a parameter to define the shape of the elliptical cap yield surface M and a parameter to
define the rate of plastic volumetric strain and the preconsolidation pressure H. HSsmall
model has two additional parameters which are initial tangent modulus E0ref and
characteristic shear strain level for which initial shear stiffness of soil degrades to 70% of its
initial value γ0.7.
For parameters ft, ψ, σref, σoedref, υur, Rf, k0, emax, D, M and H default values were used as suggested
in technical manuals [27,29]. Remaining parameters were obtained using existing equations based on
available information. Parameters m and γ0.7 were found from consistency index test results, and the
others mainly using correlations with results of SCPTu tests [26,29,31]. Numerical simulations of
oedometric tests were also performed using ZSoil® software to fit mean values of parameters m,
Eoedref, E50ref, Eurref, and E0ref.

Finite Element model


Several series of finite element models as the one shown in Figure 1 were elaborated and
analyzed with ZSoil® software [32]. The excavation is located in a corner and is surrounded by 10
storey buildings modeled as plates with distributed loads of 100kPa.
It is an idealized excavation of 40x64m and 12m depth, supported by a retaining system
composed of 30m depth diaphragm walls and concrete beams with f'c = 28MPa. 18 m length and 0.60
m diameter piles spaced every 4m were used in the bottom of the excavation to prevent heave. The
extension of the model correspond to a 160x184x60m block whose dimensions were defined
following recommendations found in technical literature [33,34]. Construction stages are presented in
Table 1 and illustrated in Figure 2.
Vol. 22 [2017], Bund. 10 3902

(a) (b)

Figure 1: Deep excavation in soft soils model geometry considered in the analyses (a)
Excavation and distribution of adjacent plates and loads (b) Sections for DPI determination

Table 1: Construction sequence considered in numerical analyses


Stage Construction activities
1 Pile installation
1.5 Retaining wall construction
2 First level beams and plate (0.0m)
3 Excavation 1 (0.0-3.0m)
4 Second level beams and plate (-3.0m)
5 Excavation 2 (-3.0-6.0m)
6 Third level beams and plate (-6.0m)
7 Excavation 3 (-6.0-9.0m)
8 Fourth level beams and plate (-9.0m)
9 Excavation 4 (-9.0-12.0m)
10 Bottom slab
11 Framed building construction

(a) (b)
Vol. 22 [2017], Bund. 10 3903

(c) (d)
Figure 2: Construction sequence considered in the analyses (a) Initial state (b) Pile
installation and wall construction (c) Excavation, beam and slabs construction (d) Finished
excavation and building
Damage Potential Index (DPI)
To define the system response a serviceability limit state is considered in this paper, in this case
possible damage in adjacent buildings to excavations. As a response variable it was chosen the
damage potential index (DPI) [35]. DPI is based on the maximum principal stress strain εp originated
by excavation in adjacent buildings which combines angular distortion β, lateral deformation εl and
the direction of crack formation measured from vertical plane θmax as defined as shown in Figure 3
[36]. Expressions for εl, εp, θmax, β and DPI are presented in equations (1),(2),(3),(4) and (5)
respectively. In this paper critical values were taken as DPIcrit=25 for sagging and DPIcrit=20 for
hogging [35].
δ H 2 − δ H1
εl = (1)
L
δ −δ
β = V 2 V1 (2)
L

θ max = 1/ 2 tan −1  β ε  (3)


 l
ε p = ε l ( cos θ max ) + β ( sin θ max )( cos θ max )
2
(4)
εp
DPI = ×100
( )
1
200
(5)
Vol. 22 [2017], Bund. 10 3904

(a) (b)

Figure 3: (a) Linear and (b) angular distortion in buildings

Probabilistic Methods
Point Estimate Method (PEM)
In this method a continuous random variable is replaced by a discrete random variable whose
probability mass function has the same moments and for reliability analyses the moments of
performance function are obtained by evaluating this function in a selected set of discrete points
[9,11]. This method does not require knowing the probability distribution function of individual
variables or joint probability distribution. PEM seeks to replace a continuous probability distribution
function with a discrete function that has the same first three central moments.
For this method three cases have been established: 1) When Y is a function of a variable X whose
first three moments are known, 2) When Y is a function of a variable whose distribution is symmetric
and approximately Gaussian, 3) When Y is a function of variables X1,X2,…,Xn whose distributions are
symmetric and may be correlated. In Case 3, the most used, starting from a function of random
variables Y=f(X1,X2,…,Xn) and known values of their means and standard deviations 2n points of
interest are defined. Points are defined such that the value of each variable is an standard deviation
above (si=+1) or below (si=-1) its mean. If variables are not correlated, function Y is evaluated in
each point and weights are equal to P=1/2n, but if they are correlated the weights are given by
equation (6).

1  n−1 n 
P ( s1 , s2 ,..., sn ) = n 
1 + ∑ ∑ ( si ) ( s j )( ρij ) (6)
2  =i 1 =j i +1 
The expected value of Y raised to m is given by equation (7).

E Y m  ≈ ∑ Pi ( yi )
m
(7)

yi is the value of Y evaluated in Xi, i is a proper combination of signs + y – indicating the location
of xi in which the hypercube corners are located. A disadvantage of this method is that requires the
evaluation of performance function 2n times, a large number when the number of uncertain
parameters grow. PEM also has limitations to find moments of Y over the second.
Vol. 22 [2017], Bund. 10 3905

Monte Carlo Simulation -Latin Hypercube Sampling (MCS-LHS)


It is a procedure in which a stochastic process is simulated by random selection of input values in
an analysis model in proportion to their joint probability density function [37]. A general procedure
for applying Monte Carlo Simulation (MCS) method has the following steps [10]:
• Define a function Y=f(X1,X2,…,Xn) which describes the problem in terms of all random variables
involved.
• Determine the probability distribution and their parameters for each random variable Xi.
• Generate random values for each Xi based on their probability distribution.
• Evaluate function Y deterministically using realizations of each random variable a sufficiently
large number of times.
• Extract statistical information of results.
• Find efficiency and precision of performed simulation.
To assess failure probability the limit state function g must be evaluated in each iteration. If N
simulations are performed and in NF of them the outcome is system faulure, the failure probability can
be approximated as:

N F ( g ( xˆ ) ≤ 0)
pf ≈ (8)
N
This technique is applicable to linear or non-linear problems, but may require a very large number
of simulations to yield a reliable distribution of response. To reduce the number of simulations and
computational effort there are different variance reduction techniques and one of them is Latin
Hypercube Sampling (LHS) [9,10]. In LHS, if the objective function is defined in terms of k variables
X1,…,Xk, range of each variable is divided in m intervals of equal probability and for each one a value
is selected as representative. This representative value can be the mean of the interval and then
representative values are combined such each one is considered only once in all simulation process.
Response Surface Method (RSM)
Considering the response of a variable of interest y as dependent on a set of predictor variables
{x1,…,xk}, if a functional relation y=f(x1,…,xk) is unknown as usually happens in different
experimental fields, it can be approximated using an empirical model and including error in the
system as y=g(x1,…,xk)+ε [15,17]. In case of numerical computations x1,…,xn are the input data and
corresponding y values are given by output data for each model analyzed. In the particular problem
considered in this paper, x1,…,xn are soil constitutive parameters selected as random variables and
values of y are the system response in terms of DPI values as explained in previous section. In the
application of RSM usually the function g is a low-order polynomial of first or second order. A
polynomial of second order is shown in equation (9).
k k
= ( x1 , x2 ,..., xk ) β0 + ∑ βi xi + ∑ βii xi2 + ∑ βij xi x j + ε
y g= (9)
=i 1 =i 1 i< j

Least squares method is used to estimate the parameters of polynomial equations and response
surface analysis is performed on fitted surface. For the estimation of model parameters a suitable
experimental design should be chosen. In this paper, a second order polynomial approximate function
is used without cross terms as shown in equation (10).
Vol. 22 [2017], Bund. 10 3906

k k
= ( x1 , x2 ,..., xk ) β0 + ∑ βi xi + ∑ βii xi2
y g= (10)
=i 1 =i 1

Performance function is evaluated around a center point {xc1,xc2,…,xck} and other 2k points around
it: {xc1+/-mσx1, xc2…,xck},{xc1,xc2+/-mσx2,…,xck},…,{xc1, xc2,…,xck+/-mσxk}, where m is a parameter
determining the relative distance of points and σxi is the standard deviation of predictor variable xi.
Coefficients can be found equating values of performance function and values of polynomial equation
evaluated at selected points. Response surface polynomial equation can be used to estimate statistical
moments of distribution by means of Monte Carlo Simulation [12] or to evaluate the design point
using first order reliability method FORM [9,10,37,38,39], and then the reliability index and failure
probabilty. Different applications of RSM in geotechnical engineering are presented in literature
[12,38,39,40].

Performance function and probability of undesired performance


To determine the probability of undesired performance a limit state function must be defined. For
serviceability limit states related with possible cracking in adjacent buildings to excavation, the limit
state function is given as a margin of safety in equation (11), where R is the capacity or resistance of
the system and Q the loads or solicitations imposed. For reliability analysis, DPIcrit values were given
in previous section, while DPIcalc values can be found from numerical simulation for sagging and
hogging deformation patterns and for each section shown in Figure 1.

= − Q DPI crit − DPI calc


M R= (11)
If M>0 the excavation is stable for a serviceability limit state defined, and if M≤0 there is a
condition of undesired performance implying cracking in adjacent buildings. System resistance and
loads imposed are random variables, so failure probability is obtained integrating the joint probability
density function over the failure region. If R and S are independent variables, the probability of
undesired performance is then given in equation (12).

= [ M ≤ 0]
Pf P= ∫∫ f ( r ) f ( s ) drds
R S (12)
M ≤0

RESULTS
Model parameters
For the application of PEM and MCS-LHS, a selection of constitutive parameters as random
variables was made considering total variability in parameters and using multivariate techniques and
sensitivity analysis. Soil profile in models analyzed for PEM and MCS-LHS are composed of three
layers: from 0 to -2m a fill layer, from -2m to -6m a lightly overconsolidated soft clay layer and from
-6m to -60m a normally consolidated soft clay layer. For the fill layer Mohr-Coulomb M-C
constitutive model were used with deterministic parameters, and for the soft clay layers 1 and 2
constitutive models HS and HSsmall were considered with deterministic parameter values given as
mean values found in statistical analyses, except for variables selected as random. A summary of
model parameters for PEM and MCS-LHS series of experiments is presented in Table 2.
Vol. 22 [2017], Bund. 10 3907

Table 2: Soil constitutive parameters values used in PEM and MCS-LHS


Fill Soft clay layers (HS, HSsmall)
Parameter
(M-C) Layer 1 Layer 2
ft (kPa) - 0 0
σref (kPa) - 100 100
σoedref (kPa) - 218.83 184.61
υur 0.30 0.20 0.20
Rf - 0.90 0.90
k0 - 0.4561 0.5415
emax - 2.06 3.48
D - 0.25 0.25
M - 1.1978 0.9927
H - 4420.46 1141.81
γ (kN/m3) 18.00 15.41 15.58
c’ (kPa) 0.00 1.47 1.62
φ (°)b,c 36.00 32.95 27.29
ψ (°) 12.00 4.71 0
m - 0.35 0.45
E50ref (MPa)b,c 100 9.00 4.00
Eurref (MPa) - 27.00 12.00
Eoedref (MPa) - 5.20 1.40
OCR - 2.61 1.10
E0 ref (MPa)a,c - 90.00 44.00
γ0.7a - 0.00048 0.00048
a
Parameters only for HSsmall b Random variables HS c Random variables HSsmall

Inferior and superior limits for random variables assumed in PEM were obtained from descriptive
statistical analysis and are shown in Table 3. For HS model the random variables considered are φ and
E50ref, while for HSsmall model the random variables are E50ref and E0ref.
Table 3: Limits for selected random parameters in PEM
Soft clay layers (HS, HSsmall)
Parameter Layer 1 Layer 2
Inf. Sup. Inf. Sup:
φ (°) 28.69 37.21 24.97 29.61
E50ref (MPa) 3.60 14.40 1.94 6.06
E0 ref (MPa) 75.96 104.04 38.72 49.29
For MSC-LHS series of experiments the probability distributions from Pearson system and their
parameters were obtained after performing inferential analyses and are shown in Table 4.
Table 4: Probability distributions used in MCS-LHS
Soft clay layers (HS, HSsmall)
Parameter Layer 1
pdf Par1 Par2 Loc. Scale p-value
φ (°) I-Beta 2.330 0.577 16.027 21.114 0.733
E50ref (MPa) VI-Beta prime 0.998 7.504 4375.424 30126.54 0.733
E0 ref (MPa) I-Beta 0.334 0.511 74571.29 39007.03 0.707
Layer 2
Parameter pdf Par1 Par2 Loc. Scale p-value
φ (°) I-Beta 2.474 0.941 19.394 10.900 0.914
E50ref (MPa) I-beta 0.1596 0.808 2714.079 7797.139 0.130
E0 ref (MPa) IV-Cauchy 3.624 0.127 44262.59 10883.52 0.995
Vol. 22 [2017], Bund. 10 3908

For the application of RSM, random variables selection was made after removing trends with
depth and combining different sources of uncertainty by means of a simple second-moment
probabilistic approach. Soil profile here is composed by sixteen layers: from 0 to -2m a fill layer,
from -2m to -60m a lightly overconsolidated to normally consolidated soft clay divided in one layer
of 1m (-2 to -3m), nine layers of 3m (-3 to -30m) and five layers of 6m (-30 to -60m). A summary of
model parameters for RSM application is presented in Table 5.
Table 5: Soil constitutive parameters values used in RSM
Soft clay layers (HS, HSsmall)
Fill
Parameter L1 L2 L3 L4 L5 L6 L7 L8 L9 L10 L11 L12 L13-
(MC)
L15
ft (kPa) - 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
σref (kPa) - 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0
σoedref (kPa) - 254.2 240.5 225.9 216.4 210.6 207.2 205.7 205.3 205.5 205.6 204.5 197.7 191.2
υur 0.30 0.20 0.20 0.20 0.20 0.20 0.20 0.20 0.20 0.20 0.20 0.20 0.20 0.20
Rf - 0.90 0.90 0.90 0.90 0.90 0.90 0.90 0.90 0.90 0.90 0.90 0.90 0.90
k0 - 0.39 0.42 0.44 0.46 0.48 0.48 0.49 0.49 0.49 0.49 0.49 0.51 0.52
emax - 2.08 2.04 5.52 4.02 4.44 3.19 4.16 3.95 3.42 5.41 3.30 3.30 1.23
D - 0.25 0.25 0.25 0.25 0.25 0.25 0.25 0.25 0.25 0.25 0.25 0.25 0.25
M - 1.28 1.24 1.18 1.15 1.08 1.01 1.10 1.10 1.10 1.10 1.10 1.07 1.04
H - 2328 1944 1477 1153 959 889 899 961 1039 1097 615 1570 1570
γ (kN/m3) 18.00 13.76 13.62 13.51 13.48 13.54 13.65 13.81 14.00 14.19 14.38 14.61 14.73 14.68
c’ (kPa) 0.00 3.08 2.76 2.50 2.49 2.67 3.01 3.45 3.95 4.47 4.97 5.57 5.86 5.69
φ (°)b 36.00 37.35 35.75 33.87 32.55 31.67 31.16 30.92 30.86 30.88 30.90 30.73 29.62 28.49
ψ (°) 12.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
m - 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50
E50ref (MPa)b,c 100 12.88 10.30 7.52 5.70 4.68 4.26 4.29 4.57 4.93 5.20 5.04 2.84 1.32
Eurref (MPa) - 48.70 39.25 28.65 21.73 17.82 16.24 16.33 17.40 18.79 19.82 19.20 10.82 3.97
Eoedref (MPa) - 3.80 3.06 2.24 1.70 1.39 1.27 1.27 1.36 1.47 1.55 1.50 0.84 1.32
OCR - 4.23 3.40 2.47 1.86 1.52 1.38 1.38 1.47 1.58 1.66 1.56 1.00 1.00
E0 ref (MPa)a,c - 84.87 82.17 80.43 81.05 83.59 87.59 92.61 98.21 103.92 109.30 115.76 119.03 117.36
γ0.7a - 4.8E-4 4.8E-4 4.8E-4 4.8E-4 4.8E-4 4.8E-4 4.8E-4 4.8E-4 4.8E-4 4.8E-4 4.8E-4 4.8E-4 4.8E-4
a
Parameters only for HSsmall b Random variables HS c Random variables HSsmall

Inferior and superior limits for random variables in each layer assumed in RSM are shown in
Table 6. Again, for HS model, random variables considered are φ and E50ref, while for HSsmall model
the random variables are E50ref and E0ref.
Table 6: Limits for selected random parameters in RSM
Soft clay layers (HS, HSsmall)
Parameter
Lim. L1 L2 L3 L4 L5 L6 L7 L8 L9 L10 L11 L12 L13-L15
φ (°) Inf. 33.31 30.96 27.81 25.35 23.75 23.00 23.10 23.70 24.55 25.37 25.63 25.56 25.75
Sup. 41.38 40.54 39.94 39.74 39.59 39.32 38.74 38.02 37.21 36.44 35.83 33.68 31.22
E50ref Inf. 6.08 4.43 4.77 1.86 1.43 1.28 1.35 1.55 1.82 2.10 1.95 1.25 0.79
(MPa) Sup. 19.48 16.17 12.27 9.54 7.92 7.25 7.22 7.58 8.04 8.30 8.13 4.43 1.86
E0 ref Inf. 56.27 56.70 65.95 71.00 73.89 76.73 80.85 86.22 92.38 97.28 105.0 110.2 109.8
(MPa) Sup. 113.5 107.6 94.90 91.10 93.28 98.45 104.4 110.2 115.5 121.3 126.5 127.8 124.9

Numerical analyses
Before performing the probabilistic analyses using PEM, MCS-LHS and RSM, deterministic
numerical analyses were made using mean values in all random variables for each soil layer as shown
in Table 2 and Table 5. Graphs of settlement troughs and wall deflection profiles were obtained to
check trends of behavior and to identify zones of sagging and hogging patterns of deformation. Also,
Vol. 22 [2017], Bund. 10 3909

initial values of DPI were assessed to check if selected retaining system and construction sequence
were in a condition of undesired behavior according to critical values given before. Construction
sequence and retaining system chosen for subsequent analyses were defined based on DPI values
obtained in models analyzed with HSsmall, trying not to exceed critical values.
Wall deflection profiles for 3-layer models

(a) (b)

c) d)
Figure 4: Wall deflection profiles obtained in numerical analyses for 3-layer models in cross
sections (a) LE-C (HS) (b) LE-C(HSsmall) (c) LW-C(HS) (d) LW-C(HSsmall)

Wall deflection profiles found in deterministic numerical analyses of 3-layer models using mean
values of random variables are presented in Figure 4. General patterns of wall deflection are similar
for LW-C section, with rotation points located at -12 to -20m, -25m and -28m approximately, but
deflections are lesser in model analyzed using HSsmall. For final construction stage the maximum
wall deflection on LW-C section is of 7.4cm in the model analyzed using HS and of 2 cm in the one
analyzed with HSsmall. For LE-C section, similar patterns of wall deflection are also found but in this
case wall base moves 9cm in the model analyzed using HS and 2cm in the one using HSsmall.
Maximum wall deflections in LW-C sections are of 12cm and 5cm respectively.

Wall deflection profiles for 16-layer models


Wall deflection profiles found in deterministic numerical analyses of 16-layer models are
presented in Figure 5. For LW-C section, rotation point is located at depths between -5 and -10m;
wall top moves a maximum of 4.7cm in model analyzed with HS model and less than 1cm in model
analyzed with HSsmall. Wall base moves a maximum of 8.5cm in stage 9 in model analyzed with HS
and 3.15cm in the one analyzed with HSsmall. In LE-C section, wall base moves a maximum of 20cm
Vol. 22 [2017], Bund. 10 3910

in stage 10 in model analyzed with HS, and 1.7cm in model analyzed with HSsmall. In each section
the patterns of deflection are similar when using HS or HSsmall but with smaller movements in the
second case as seen in 3-layer models, although this time the effect is more marked. Deflection
patterns in 3-layer and 16-layer models are different because of the distribution of soil stiffness with
depth, because in the second case in normally to lightly consolidated soil under -6m depth the
stiffness is growing while in the former is constant.

(a) (b)

c) d)
Figure 5: Wall deflection profiles obtained in numerical analyses for 16-layer models in
cross sections (a) LE-C (HS) (b) LE-C(HSsmall) (c) LW-C(HS) (d) LW-C(HSsmall)

Settlement troughs for 3-layer models


Settlement troughs found in deterministic numerical analyses using mean values of random
variables for 3-layer models are presented in Figure 6. Graphs shown in each case served to define the
limits of sagging and hogging patterns of deformation to perform the DPI assessment in subsequent
analyses. Settlement troughs found in LE-C section show a sagging pattern 20m to approximately
44m away from the face of the excavation, and hogging pattern from this point to 60m. In LE-C
section, the nearest building is located from 20m to 60m, while from 0 to 20m there is a road and
sagging patterns in such infrastructure are not considered here. Maximum settlement in LE-C sections
is about 30 cm in model analyzed using HS and about 9 cm in the one using HSsmall. In LW-C
sections a sagging pattern appears from 0 to approximately 34m and a hogging pattern from there to
60m away from the face of the excavation. Maximum settlement in LW-C sections is about 44 cm in
model analyzed using HS and about 20 cm in the one using HSsmall. In each case, considering LE-C
or LW-C sections, general patterns of settlement are similar but again the effects are lesser in models
analyzed using HSsmall constitutive model.
Vol. 22 [2017], Bund. 10 3911

(a) (b)

(c)
(d)
Figure 6: Settlement troughs obtained in numerical analyses for 3-layer models in cross
sections (a) LE-C (HS) (b) LE-C(HSsmall) (c) LW-C(HS) (d) LW-C(HSsmall)

Settlement troughs for 16-layer models


Settlement troughs for 16-layer models are presented in Figure 7. In this case trends in settlement
are different respect to 3-layer models and the sagging and hogging patterns are not as clear as seen
before, and they are defined depending of the upward or downward concavity of settlement graph.
Settlement graphs of LE-C section show a sagging pattern between 20m and 60m away from the
excavation face. In LW-C section there is a sagging pattern from 0m to 40m and a hogging pattern
from 40m to 60m away from the excavation face.

(a) (b)
Vol. 22 [2017], Bund. 10 3912

(c) (d)
Figure 7: Settlement troughs obtained in numerical analyses for 16-layer models in cross
sections (a) LE-C (HS) (b) LE-C(HSsmall) (c) LW-C(HS) (d) LW-C(HSsmall)

Ground near the excavation is heaving, and this is much more marked in section LE-C and
particularly in model analyzed with HS where it reaches 97cm, which certainly would generate
damages in road and service lines. Nearest building in LE-C section is 20m away from the excavation
face where ground heaves 49cm and 37cm at 60m, both values obtained at stage 10. In models
analyzed with HSsmall, heaving near the excavation at stage 10 in LE-C is of 5cm, of 1.2cm at 20m
away from the excavation face and of 1.7cm at 60m away. In LW-C ground next to the excavation
face is heaving at stages 1 and 10, and sinking in the other stages. Maximum settlement in LW-C
sections is of 21cm in model analyzed using HS and about 3cm in the one using HSsmall.

Damage potential index (DPI) for 3-layer models


Figure 8 and Figure 9 show the evolution of DPI values with construction sequence in 3-layer
models, for sagging and hogging patterns of deformation respectively. Construction stages and
features of retaining system were defined taking as reference the models analyzed using HSsmall
constitutive model, so critical values of DPI defined as threshold of acceptable behavior are not being
surpassed when using this model. In sagging patterns the critical stages for sections LE-A, LE-B, LE-
C are 10 and 11, while for LW-A and LW-E are 2, 3 and 9, exceeding critical DPI value for models
analyzed with HS in that stages. In hogging patterns the critical stages for sections LE-A to LE-E and
LW-A to LW-E are 9, 10 and 11, and in models analyzed with HS acceptable threshold are surpassed
from stage 6 to stage 11. In general, DPI values differ in sections analyzed in each model, so it
remarks the importance of considering tridimensional modelling for this kind of geotechnical
projects.
Vol. 22 [2017], Bund. 10 3913

(a) (b)

(c) (d)
Figure 8: DPI values obtained in numerical analyses of 3-layer models for sagging profiles
in sides of excavation (a) LE(HS) (b) LE(HSsmall) (c) LW(HS) (d) LW(HSsmall)

(a) (b)

(c) (d)
Figure 9: DPI values obtained in numerical analyses of 3-layer models for hogging profiles
in sides of excavation (a) LE(HS) (b) LE(HSsmall) (c) LW(HS) (d) LW(HSsmall)
Damage potential index (DPI) for 16-layer models
Vol. 22 [2017], Bund. 10 3914

Figure 10 and Figure 11 show the evolution of DPI values with construction sequence in 16-layer
models, for sagging and hogging patterns of deformation respectively. In this case critical results are
obtained in sagging patterns of deformation on models analyzed with HS, whose DPI values assessed
exceed tolerable limit. In sagging pattern, LE-A and LW-A sections give the greater values of DPI,
which decrease gradually to the lower values found in sections LE-E and LW-E. In hogging pattern in
models analyzed with HS, DPI values are under 15 and are lower than critical value, showing a
markedly different behavior respect to sagging pattern. Models analyzed with HSsmall present small
DPI values, far from the critical value, for sagging and hogging patterns of deformation. There is also
a great contrast between DPI values found in models analyzed with HS and HSsmall constitutive
models.

(a) (b)

(c) (d)
Figure 10: DPI values obtained in numerical analyses of 16-layer models for sagging
profiles in sides of excavation (a) LE(HS) (b) LE(HSsmall) (c) LW(HS) (d) LW(HSsmall)

(a) (b)
Figure 11: DPI values obtained in numerical analyses of 16-layer models for hogging
profiles in sides of excavation (a) LW(HS) (b) LW(HSsmall)
Vol. 22 [2017], Bund. 10 3915

Probabilistic analyses
As explained in previous sections, probabilistic analyses were performed using PEM, MCS-LHS
and RSM, using constitutive parameters from Table 2 and Table 5, probability distributions from
Table 4 and limit values from Table 3 and Table 6. Probabilities of undesired performance were
assessed using these three different methods and results are presented and discussed in this section.
To this end, in each case safety margin values M were calculated using DPIcrit defined and DPIcalc
values obtained from ground displacements in numerical models. Considering the excavation as a
system in series with statistically independent resistances in system elements, total probability of
undesired performance for each construction stage were found using equation 13.

− ∏ (1 − Pfi ) 1 − (1 − Pf 1 )(1 − Pf 2 ) ... (1 − Pfk )


k
=Pf 1= (13)
i =1

Where Pf is the total probability of undesired performance for each stage and Pfi are each one of
probabilities of undesired performance assessed for each side of the excavation.

Assessment of undesired performance using PEM


Having two layers of soft soils with two random parameters each one (φ’ and E50ref), 17 numerical
models were analyzed with HS constitutive model. In models analyzed with HSsmall, two layers of
soft soil were considered with three random parameters each one (φ’, E50ref and E0ref) for a 65 model
series. In each case there were 2n models from combinations of lower and upper limits of random
parameters and one with mean values of them. Minimum M values obtained in each stage for defined
cross sections in sagging and hogging profiles were used to estimate its mean and standard deviation
as shown in Table 7. Assuming a normal distribution, partial values of undesired performance
probabilities Pfi were assessed and are summarized in Figure 12 and Figure 13 for models analyzed
using HS and HSsmall respectively.

Table 7: Mean and standard deviation found using PEM and HS, HSsmall
LE-HS LW-HS BN-HS BS-HS LE-HSsmall LW-HSsmall BN-HSsmall BS-HSsmall
Stage SAG HOG SAG HOG SAG HOG SAG HOG SAG HOG SAG HOG SAG HOG SAG HOG
µ 14.45 15.16 22.23 19.02 18.35 15.88 20.11 19.29 24.61 19.80 24.68 19.92 24.72 19.84 24.54 19.93
1
σ 1.48 0.69 0.42 0.10 1.08 0.60 0.74 0.08 0.22 0.12 0.18 0.05 0.17 0.10 0.25 0.05
µ 15.53 6.90 1.08 6.23 16.63 8.77 -14.99 8.02 21.30 17.80 14.60 16.56 21.31 17.93 7.02 16.98
2
σ 1.93 1.22 0.49 1.04 0.44 0.90 0.77 0.80 0.74 0.48 2.57 0.69 0.77 0.46 4.33 0.59
µ 15.96 4.84 -2.02 4.01 15.17 7.38 -20.00 5.97 20.62 16.46 12.58 14.60 20.61 16.84 3.58 15.70
3
σ 1.22 1.45 0.53 1.30 0.43 1.03 1.02 1.01 0.68 1.29 2.69 1.14 0.97 1.14 4.55 0.97
µ 15.42 3.89 5.47 3.74 15.99 6.88 -8.17 5.62 20.35 15.69 15.04 14.44 20.82 16.44 7.71 15.59
4
σ 1.68 1.59 0.65 1.49 0.41 1.16 1.48 1.13 0.55 1.58 2.31 1.18 0.98 1.34 3.88 0.93
µ 15.56 3.80 4.46 3.43 15.67 6.75 -10.04 5.37 20.25 15.58 14.69 14.25 20.65 16.32 6.95 15.51
5
σ 1.48 1.59 0.66 1.50 0.41 1.15 1.49 1.13 0.51 1.59 2.34 1.22 0.99 1.36 3.95 0.95
µ 12.03 -2.02 4.35 -2.84 10.17 1.41 -10.97 -1.10 17.69 14.03 14.65 12.35 18.14 14.85 6.19 13.76
6
σ 0.34 2.14 0.90 1.78 0.49 1.56 1.54 1.31 0.54 1.84 2.08 1.71 1.25 1.67 3.86 1.36
µ 11.72 -2.12 3.09 -3.17 9.77 1.22 -13.29 -1.35 17.62 13.93 14.13 12.06 17.97 14.66 5.22 13.58
7
σ 0.33 2.14 0.90 1.78 0.48 1.57 1.54 1.32 0.54 1.88 2.14 1.80 1.27 1.73 3.97 1.44
µ 4.54 -11.83 1.10 -14.10 1.13 -8.03 -17.53 -12.32 14.29 11.73 12.95 8.82 14.81 12.67 3.11 10.71
8
σ 0.37 2.54 1.21 1.30 0.53 1.93 2.03 1.08 1.04 2.55 2.50 2.88 2.14 2.36 4.62 2.48
µ 4.13 -12.01 -0.66 -14.80 0.59 -8.38 -20.89 -12.83 13.95 11.33 12.04 7.64 14.38 12.22 1.59 9.78
9
σ 0.37 2.54 1.23 1.29 0.53 1.93 2.06 1.07 1.26 2.88 2.63 3.37 2.35 2.68 4.83 3.00
µ -0.93 -21.87 2.76 -21.42 -3.24 -15.48 -2.72 -21.88 8.71 6.51 19.65 3.89 11.62 9.26 19.03 5.97
10
σ 0.95 2.81 0.65 1.13 0.42 2.15 1.47 1.12 2.46 4.99 1.72 4.52 3.65 4.25 1.68 4.37
µ -3.58 -24.23 10.04 -26.91 -7.13 -18.52 11.12 -26.96 7.32 5.68 18.16 2.32 10.05 8.09 12.63 4.60
11
σ 0.88 2.92 0.67 1.38 0.44 2.24 1.67 1.20 3.03 5.36 1.24 5.15 4.14 4.77 2.63 5.10
Vol. 22 [2017], Bund. 10 3916

Figure 12: Pfi values found using PEM in models analyzed with HS

For models analyzed with HS, there are not a preferential stage or deformation pattern for the
occurrence or undesired behavior, because partial values of Pf reach values over 70% for sagging and
hogging in each face of the excavation at different stages. In models analyzed with HSsmall, the
highest partial Pf values were found in LW side in hogging pattern (33%) at stage 11, and in BS side
in sagging pattern (37%) at stage 9.

Figure 13: Pfi values found using PEM in models analyzed with HSsmall

Assessment of undesired performance using MCS-LHS


In this case there were two layers of soft soils with two random parameters each one (φ’ and
E50ref), 100 numerical models were analyzed with HS constitutive model. In models analyzed with
Vol. 22 [2017], Bund. 10 3917

HSsmall, two layers of soft soil were considered with three random parameters each one (φ’, E50ref and
E0ref) and 100 numerical models were analyzed. Sampling space of each variable was divided in 100
parts and sampling points were chosen following a Latin Hypercube Scheme. Again, as been done in
PEM, minimum M values obtained in each stage for defined cross sections in sagging and hogging
profiles were used to estimate its mean and standard deviation. A normal distribution was assumed
for M to assess statistical moments and Pfi as shown in Table 8, in Figure 14 and Figure 15 for models
analyzed using HS and HSsmall.

Table 8: Mean and standard deviation found using MCS-LHS and HS, HSsmall
LE-HS LW-HS BN-HS BS-HS LE-HSsmall LW-HSsmall BN-HSsmall BS-HSsmall
Stage SAG HOG SAG HOG SAG HOG SAG HOG SAG HOG SAG HOG SAG HOG SAG HOG
µ 14.56 15.11 22.14 19.02 18.45 15.83 19.96 19.30 24.59 19.80 24.68 19.92 24.70 19.84 24.53 19.93
1
σ 1.80 0.67 0.40 0.12 1.34 0.61 0.69 0.10 0.16 0.10 0.13 0.04 0.13 0.08 0.20 0.04
µ 15.33 6.67 0.93 6.15 16.49 8.60 -15.31 7.93 21.27 17.81 14.45 16.58 21.26 17.93 6.72 16.99
2
σ 2.01 1.98 1.20 2.00 0.89 1.42 2.26 1.75 0.60 0.38 2.10 0.57 0.63 0.37 3.52 0.47
µ 15.75 4.60 -2.20 3.93 15.00 7.19 -20.34 5.88 20.41 16.46 12.40 14.57 20.56 16.84 3.26 15.68
3
σ 1.94 2.25 1.39 2.19 0.97 1.60 2.71 1.86 0.55 1.22 2.16 0.99 0.81 1.11 3.63 0.85
µ 15.07 3.64 5.26 3.63 15.79 6.69 -8.69 5.51 20.10 15.68 14.83 14.40 20.77 16.44 7.26 15.57
4
σ 2.36 2.47 1.66 2.61 1.24 1.74 3.21 2.07 0.54 1.47 1.81 1.04 0.82 1.31 3.02 0.82
µ 15.25 3.54 4.25 3.33 15.47 6.56 -10.57 5.27 20.02 15.55 14.48 14.21 20.59 16.29 6.50 15.49
5
σ 2.31 2.47 1.69 2.62 1.20 1.74 3.24 2.08 0.53 1.49 1.84 1.08 0.83 1.33 3.10 0.84
µ 11.62 -2.30 4.04 -2.95 9.93 1.19 -11.54 -1.21 17.41 13.90 14.43 12.24 17.77 14.66 5.74 13.74
6
σ 3.12 2.95 1.96 2.94 1.56 2.13 3.60 2.32 0.72 1.69 1.67 1.50 1.06 1.66 3.09 1.18
µ 11.32 -2.40 2.78 -3.28 9.54 1.00 -13.87 -1.46 17.36 13.77 13.92 11.94 17.56 14.40 4.78 13.56
7
σ 3.11 2.95 1.97 2.93 1.52 2.13 3.63 2.31 0.71 1.74 1.74 1.57 1.07 1.71 3.18 1.27
µ 3.95 -12.18 0.72 -14.21 0.83 -8.29 -18.23 -12.47 13.96 11.44 12.75 8.68 14.34 12.27 2.63 10.66
8
σ 4.61 3.39 2.59 2.81 2.44 2.57 4.67 2.29 1.18 2.50 2.07 2.58 1.86 2.42 3.75 2.20
µ 3.55 -12.36 -1.03 -14.91 0.30 -8.64 -21.59 -12.97 13.70 11.05 11.81 7.55 13.97 11.83 1.12 9.68
9
σ 4.62 3.39 2.65 2.81 2.44 2.56 4.77 2.28 1.29 2.85 2.17 2.98 2.00 2.72 3.90 2.59
µ -1.73 -22.31 2.72 -21.60 -3.65 -15.82 -2.76 -22.08 7.97 6.07 19.54 3.45 11.09 8.81 18.90 5.71
10
σ 7.23 4.09 1.31 2.96 3.53 3.12 1.60 2.50 2.39 4.70 1.60 4.03 2.90 4.05 1.76 3.78
µ -4.37 -24.68 9.79 -27.08 -7.49 -18.86 10.65 -27.16 5.67 4.69 17.82 1.42 9.16 7.54 12.13 3.63
11
σ 7.38 4.25 2.04 3.17 3.74 3.27 5.08 2.65 7.90 5.65 2.85 4.61 3.30 4.51 2.71 5.70

Figure 14: Pfi values found using MCS-LHS in models analyzed with HS
Vol. 22 [2017], Bund. 10 3918

Pfi values reach values over 60% for sagging and hogging in each face of the excavation at
different stages. In models analyzed with HSsmall, highest partial Pf values were found in LW side in
hogging pattern (38%) at stage 11, and in BS side in sagging pattern (39%) at stage 9.

Figure 15: Pfi values found using MCS-LHS in models analyzed with HSsmall

Assessment of undesired performance using RSM


In this case sixteen layers of soft soils were considered, with two random parameters each one: φ’,
E50ref in models analyzed with HS, and E50ref, E0ref in models analyzed with HSsmall. In each case a
total of 61 models were analyzed, which corresponds to 2n+1 models from combinations of lower
and upper limits of random parameters and one with mean values. Tesults obtained in this section
were found using m=1. 61 coefficients of the polynomial equations of grade two without cross-terms
were found by regression, for sagging and hogging patterns of deformation and for each construction
stage considered. All values of coefficient of determination R2 found where of 1. In each case 105
Monte Carlo Simulations were performed using polynomial equations to find the first two statistical
moments of M and to assess Pfi and Pf as done previously. Table 9, Figure 16 and Figure 17 present
Pfi values found for models analyzed using HS and HSsmall respectively.
Table 9: Mean and standard deviation found using RS and HS, HSsmall
LE-HS LW-HS BN-HS BS-HS LE-HSsmall LW-HSsmall BN-HSsmall BS-HSsmall
Stage SAG HOG SAG HOG SAG HOG SAG HOG SAG HOG SAG HOG SAG HOG SAG HOG
µ 11.05 0.00 32.03 4.32 13.91 0.00 23.33 2.62 24.21 0.00 24.15 19.78 24.19 0.00 24.32 19.88
1
σ 0.67 0.00 0.85 0.00 0.69 0.00 0.78 0.00 3.03 0.00 7.38 0.52 1.97 0.00 7.52 17.73
µ 11.26 0.00 32.27 5.99 14.33 0.00 23.50 4.38 23.93 0.00 20.57 18.85 23.99 0.00 20.94 19.05
2
σ 0.53 0.00 0.68 0.01 0.54 0.00 0.50 0.00 9.86 0.00 1.66 0.78 2.43 0.00 2.07 18.27
µ 11.28 0.00 32.31 6.01 14.71 0.00 23.47 4.44 23.73 0.00 20.22 18.71 23.88 0.00 20.50 19.07
3
σ 0.54 0.00 0.66 0.01 0.53 0.00 0.47 0.00 4.11 0.00 1.53 0.83 2.66 0.00 2.37 21.62
µ 11.51 0.00 32.90 4.52 14.41 0.00 24.67 2.76 24.09 0.00 22.13 18.78 24.44 0.00 22.04 19.22
4
σ 0.75 0.00 0.82 0.00 0.69 0.00 0.77 0.00 5.42 0.00 6.08 0.83 5.41 0.00 1.23 19.47
µ 11.46 0.00 32.98 4.61 14.48 0.00 24.14 2.60 24.15 0.00 21.89 18.75 24.43 0.00 21.83 19.19
5
σ 0.74 0.00 0.81 0.00 0.67 0.00 0.75 0.00 7.88 0.00 5.02 0.85 5.26 0.00 1.88 20.58
µ 11.9 0.00 33.2 4.5 14.4 0.00 24.5 4.8 23.88 0.00 21.42 18.48 24.03 0.00 21.17 18.83
6
σ 0.7 0.00 0.8 0.0 0.7 0.00 0.7 0.0 10.10 0.00 2.81 1.22 4.67 0.00 3.79 23.95
µ 11.82 0.00 33.22 6.55 14.44 0.00 24.46 4.64 23.82 0.00 21.22 18.44 23.96 0.00 21.02 18.88
7
σ 0.71 0.00 0.79 0.01 0.64 0.00 0.70 0.00 11.88 0.00 1.92 1.39 4.51 0.00 3.79 24.65
8 µ 12.32 0.00 33.66 6.90 14.44 0.00 25.12 4.52 23.23 0.00 20.73 18.08 23.50 0.00 20.33 18.64
Vol. 22 [2017], Bund. 10 3919

σ 0.69 0.00 0.78 0.02 0.62 0.00 0.67 0.00 11.17 0.00 1.87 1.71 4.37 0.00 4.18 27.51
µ 11.90 0.00 33.90 6.81 14.62 0.00 25.10 4.71 23.22 0.00 20.52 18.06 23.48 0.00 20.14 18.57
9
σ 0.68 0.00 0.76 0.02 0.60 0.00 0.64 0.00 14.81 0.00 1.91 1.76 4.45 0.00 4.24 28.03
µ 13.41 0.00 36.48 8.73 14.25 0.00 30.25 4.17 23.27 0.00 18.03 19.19 23.50 0.00 16.31 18.98
10
σ 0.89 0.00 0.99 0.15 0.88 0.00 0.99 0.00 8.39 0.00 24.80 3.48 7.10 0.00 15.25 67.96
µ 13.70 0.00 39.27 7.94 14.75 0.00 31.10 4.46 23.60 0.00 20.65 18.40 23.49 0.00 20.26 19.09
11
σ 0.71 0.00 0.67 0.20 0.67 0.00 0.51 0.02 4.96 0.00 2.53 2.99 2.77 0.00 2.91 0.98

Figure 16: Pfi values found using RSM in models analyzed with HS

In models analyzed with HS, Pfi values for sagging patterns of deformation are critical as they
reach values over 50% in all excavation sides from stage 2 to stage 11. In models analyzed with
HSsmall, the highest partial Pf values were found in LW side in sagging pattern (23%) at stage 10,
and in BS side in hogging pattern (39%) also at stage 10.

Figure 17: Pfi values found using RSM in models analyzed with HSsmall

Assessment of total undesired performance in 3-layer and 16-layer models


Vol. 22 [2017], Bund. 10 3920

A summary of total Pf values found for 3-layer and 16-layer models is presented in Figure 18. 3-
layer models, those analyzed using HS with methods PEM and MCS-LHS, show the same behavior
reaching 100% at stage 2 and keeping that value till stage 11, while results of 16-layer models
analyzed with RS show a similar behavior but reaching 100% since stage 1.

Figure 18: Pf values assessed for 3-layer (PEM, MCS-LHS) and 16-layer models (RSM)

3-layer models, those analyzed using HSsmall with sampling schemes PEM and MCS-LHS, show
a very similar behavior with peaks on stage 3 of about 20%, on stage 9 of about 40% and at stage 11
reaching the maximum value of 55% (HS) and 74% (HSsmall). On the other hand, 16-layer models or
the RS results show a gradual evolution of Pf with construction stages until a peak of 60% at stage 10
and a sudden decrease to zero in stage 11. There is a marked difference between total Pf results of
models analyzed with HS and HSsmall constitutive models, in the former case reaching a certain
condition of undesired performance almost since the beginning. As explained before, retaining system
features and excavation sequence were defined based on results of a model analyzed with HSsmall
considering mean values of random parameters, trying not to exceed critical DPI values in sagging
and hogging. However in models analyzed with all three methods PEM, MCS-LHS and RS there were
found high total probabilities of undesired performance.

CONCLUSIONS
(1) The numerical techniques like the finite element method considering suitable constitutive
models for soil are useful to simulate the behavior of deep excavations in soft soils in terms of wall
deflections and ground settlements which may cause damages in adjacent buildings. After defining a
suitable performance function for a given limit state condition, the values of wall and soil
displacements found in numerical modelling can serve to evaluate probabilities of undesired
performance of the system using different probabilistic methods. So the combination of numerical
and probabilistic methods end up being advantageous in the analysis of geotechnical complex
problems with uncertainty in soil constitutive parameters and including different construction stages.
(2)When combining probabilistic techniques based on reliability and the results of numerical
finite element analyses, a usual constraint is the number of models to analyze because of the
computational effort needed, especially when the number of random variables considered grows. So,
before starting the analyses a selection of the variables to consider as random must be done, and also
use probabilistic techniques with suitable sampling schemes which allow the reduction in the number
Vol. 22 [2017], Bund. 10 3921

of finite element analyses to perform. In this paper, three of those techniques where used: the point
Estimate Method (PEM), Monte Carlo Simulation using a Latin Hypercube sampling scheme (MCS-
LHS) and Response Surface Method (RSM).
(3) Different behaviors were observed in each excavation side and in each one of the sections
considered for the analyses. This was observed in wall deflection profiles, in settlement troughs, in
DPI values and finally in statistical moments of the performance functions and in assessed
probabilities of undesired performance for the excavations. It remarks the importance of performing
numerical analyses on tridimensional models to include the boundary conditions properly. Also, there
were observed different behaviors in the same aspects respect to the stages in the excavation sequence
which also highlights the importance of modeling the construction process.
(4) A marked difference were observed in results when using HS and HSsmall constitutive
models. In general when using HS model wall deflections were higher, settlement troughs were
deeper, DPI values exceed acceptable limits, Pfi values were much higher and Pf values indicate the
undesired performance of the system as a certain event from the early stages of the construction
process. When using HSsmall model, the behavior in general were better but despite of not exceeding
acceptable DPI values the Pfi and Pf values found indicate high probabilities of undesired
performance specially at the final stages of the construction process. This remarks the need of using
probabilistic approaches to analyze this kind of complex geotechnical problems. Finally, about the
differences in results obtained with HS and HSsmall, it is related with the modelling of soil small
stiffness in the latter case, which is the range in which the soil is in this kind of projects.
(5) 3-layer and 16-layer models present a different behavior because of the differences in the
distribution of soil parameters with depth. The former comes from total variability multivariate
analyses considering an overconsolidated layer from -2 to -6m depth while the latter were obtained
after detrending with depth and combining different sources of uncertainty. The presence of
overconsolidated soils near the surface was considered in the trend and in the variability of
parameters so the 16-layer models are considered here as more representative. However, the
representation of soil variability could be enhanced extending this to the other dimensions and
considering a spatial variability characterization. As well, 16-layer models RSM results differ from
those obtained with PEM and MCS-LHS methods in 3-layer models because of the way in which the
variability of soil parameters were represented as explained previously.
(6) Some limitations of the results presented in this paper are related with the details of the
numerical models as not including coupled analyses or not modelling the foundations of adjacent
buildings. From the point of view of the probabilistic techniques, the methods of second moments as
the ones used have limitations in the determination of low failure probabilities; also, the RSM where
performed applying Monte Carlo Simulation using the polynomial equations obtained and not
generating new sampling points to adjust the response surface and to define the design point using
FORM as extensively seen in literature. Another limitation appears from the point of view of the
representation of the variability of soil constitutive parameters, because it was only considered the
variation with depth and the models were not based on the realizations of random fields of them.
Those aspects remain to be treated in posterior works.

ACKNOWLEDGEMENTS
This research is supported by Universidad Nacional de Colombia and Colciencias.
Vol. 22 [2017], Bund. 10 3922

REFERENCES
[1] Kempfert, H. G. & Gebreselassie, B. (2006). Excavations and foundations in soft soil.
Berlin Heidelberg: Springer-Verlag.
[2] Ou, C.Y (2006). Deep excavation. Theory and practice. London: Taylor & Francis
Group, 2006.
[3] Ergun, U. (2008). Deep excavations. Electronic Journal of Geotechnical Engineering,
Special Volume Bouquet 08, pp 1-34. Available at ejge.com
[4] Dong, Y. (2014). Advanced finite element analysis of deep excavation case histories.
Doctoral Dissertation, University of Oxford.
[5] Schanz, T., Vermeer, P. A., & Bonnier, P. G. (1999). The hardening soil model:
formulation and verification. Beyond 2000 in computational geotechnics, 281-296.
[6] Hsiung, B-C.B. & Dao, S-D. (2014). Evaluation of constitutive soil models for
predicting movements caused by a deep excavation in sands. Electronic Journal of
Geotechnical Engineering, 2014, 19.Z5, pp. 17325-17344. Available at ejge.com.
[7] Benz, T. (2007). Small-strain stiffness of soils and its numerical consequences. Doctoral
Dissertation, Universität Stuttgart.
[8] Uzielli, M., Lacasse, S., Nadim, F., & Phoon, K. K. (2006). Soil variability analysis for
geotechnical practice. Characterization and engineering properties of natural soils, 3,
1653-1752.
[9] Baecher, G. B., & Christian, J. T. (2003). Reliability and statistics in geotechnical
engineering. John Wiley & Sons, 618pp.
[10] Sánchez, M. (2010). Introducción a la confiabilidad y evaluación de riesgos: Teoría y
aplicaciones en Ingeniería, 2a edición. Universidad de los Andes, 457pp [in Spanish].
[11] Rosenblueth, E. (1975). Point estimates for probability moments. Proceedings, National
Academy of Science 72(10): 3812-3814.
[12] Idris, M. A., Nordlund, E. & Saiang, D. (2016). Comparison of different probabilistic
methods for analyzing stability of underground rock excavations. Electronic Journal of
Geotechnical Engineering, 2016, 21.21, pp. 6555-6585. Available at ejge.com.
[13] McKay, M. D. (1992). Latin hypercube sampling as a tool in uncertainty analysis of
computer models. In Proceedings of the 24th conference on Winter simulation (pp. 557-
564).
[14] Box, G. E. P. & Wilson, K. B. (1951). On the experimental attainment of optimum
conditions. Journal of the Royal Statistical Society, Series B, 13, 1-45.
[15] Myers, M. H., Montgomery, D. C. & Anderson-Cook, C. M. (2016). Response surface
methodology: Process and product optimization using designed experiments. Fourth
edition. John Wiley & Sons Inc.
[16] Catena, A., Ramos, M. & Trujillo, H. (2003). Análisis multivariado - Un manual para
investigadores. Editorial Biblioteca Nueva, Madrid, 414pp [in Spanish].
[17] Montgomery, D. (2005). Diseño y análisis de experimentos. Segunda edición. Limusa
Wiley, Mexico D.F., 700pp [in Spanish].
Vol. 22 [2017], Bund. 10 3923

[18] Husson, F., Lê, S. & Pagès, J. (2013). Análisis de datos con R. Editorial Escuela
Colombiana de Ingeniería, Bogotá, 314pp [in Spanish].
[19] Díaz, L. G. & Morales, M. A. (2015). Análisis estadístico de datos multivariados,
segunda edición. Universidad Nacional de Colombia, Bogotá, 664pp [in Spanish].
[20] Phoon, K. K., & Kulhawy, F. H. (1999). Characterization of geotechnical variability.
Canadian Geotechnical Journal, 36(4), 612-624.
[21] Phoon, K. K., & Kulhawy, F. H. (1999). Evaluation of geotechnical property variability.
Canadian Geotechnical Journal, 36(4), 625-639.
[22] Sainea, C.J. & Torres, M.C. (2016). Selección de parámetros para análisis numéricos y
probabilísticos de excavaciones. In Guzmán, A. & González, A., eds. XV Congreso
Colombiano de Geotecnia & II Conferencia Internacional Especializada en Rocas
Blandas. Oct 5 a 7 de 2016. [in Spanish].
[23] Schuster, M., Kung, G. T. C., Juang, C. H., & Hashash, Y. M. (2009). Simplified model
for evaluating damage potential of buildings adjacent to a braced excavation. Journal of
Geotechnical and Geoenvironmental Engineering, 135 (12), 1823-1835.
[24] Son, M. & Cording, E. J. (2005). Estimation of building damage due to excavation-
induced ground movements. Journal of Geotechnical and Geoenvironmental
Engineering 131(2), p 162-177.
[25] Ingeniería y Georiesgos IGR SAS (2014). Exploración para el proyecto Parque Colina
Colpatria. [in Spanish].
[26] Robertson, P. K., & Cabal, K. L. (2014). Guide to cone penetration testing for
geotechnical engineering. Gregg Drilling & Testing, Inc. 6th Edition December 2014.
[27] Brinkgreve, R. B. J., Engin, E. & Swolfs, W. M. (eds) (2015). Plaxis Material Models
Manual.
[28] Obrzud, R. F. (2010). On the use of the Hardening Soil Small Strain model in
geotechnical practice. Numerics in Geotechnics and Structures.
[29] Obrzud, R. & Truty, A. (2014). The hardening soil model – a practical guidebook. Zace
Services Ltd, Software engineering, Lausanne.
[30] Schweiger, H. F. (2010). Design of deep excavations with FEM - Influence of
constitutive model and comparison of EC7 design approaches. In Earth Retention
Conference 3, ASCE, p 804-817
[31] Mayne (2007). Cone penetration testing. NCHRP Synthesis 368.
[32] Zace Services Ltd, Software Engineering (2016). ZSoil.PC 2016. Available from
www.zace.com.
[33] Bakker, K. J. (2006). A 3D FE model for excavation analysis. Geotechnical Aspects of
Underground Construction in Soft Ground - Proceedings of the 5th International
Conference of TC28 of the ISSMGE, p473-478.
[34] Finno, R. J., Blackburn, J. T., & Roboski, J. F. (2007). Three-dimensional effects for
supported excavations in clay. Journal of Geotechnical and Geoenvironmental
Engineering, 133(1), 30-36.
Vol. 22 [2017], Bund. 10 3924

[35] Schuster, M., Kung, G. T. C., Juang, C. H., & Hashash, Y. M. (2009). Simplified model
for evaluating damage potential of buildings adjacent to a braced excavation. Journal of
Geotechnical and Geoenvironmental Engineering, 135 (12), 1823-1835.
[36] Son, M. & Cording, E. J. (2005). Estimation of building damage due to excavation-
induced ground movements. Journal of Geotechnical and Geoenvironmental
Engineering 131(2), p 162-177.
[37] Nadim, F. (2007). Tools and strategies for dealing with uncertainty in geotechnics. In
Griffiths, D. V. & Fenton, G.A. (Eds) Probabilistic methods in Geotechnical
Engineering. Springer Wien New York. pp. 71-95.
[38] Puła, W. & Bauer, J. (2007). Application of the response surface method. In Griffiths, D.
V. & Fenton, G.A. (Eds) Probabilistic methods in Geotechnical Engineering. Springer
Wien New York. pp. 147-168.
[39] Zhang, J., Chen, H. Z., Huang, H. W., & Luo, Z. (2015). Efficient response surface
method for practical geotechnical reliability analysis. Computers and Geotechnics, 69,
496-505.
[40] Li, D. Q., Zheng, D., Cao, Z. J., Tang, X. S., & Phoon, K. K. (2016). Response surface
methods for slope reliability analysis: review and comparison. Engineering Geology,
203, 3-14.

© 2017 ejge

Editor’s note.
This paper may be referred to, in other articles, as:
Sainea-Vargas, Carlos J. and Torres-Suárez, Mario C.: “Numerical and
Probabilistic Analyses of Deep Excavations in Soft Soils” Electronic
Journal of Geotechnical Engineering, 2017 (22.10), pp 3899-3924.
Available at ejge.com.

View publication stats

You might also like