You are on page 1of 21

AIN SHAMS UNIVERSITY

FACULTY OF ENGINEERING
Vol. 38, No. 3, September 30, 2003
SCIENTIFIC BULLETIN
Received on : 8/4/2003
Accepted on : 15/9/2003
PP. : 75-94

PREDICTION OF GROUND SUBSIDENCE
ASSOCIATED WITH CONSTRUCTION OF
GREATER CAIRO METRO LINE 3

A.A. Ahmed
1
H.E. Ali
2
S.M. El-Sayed
2
S.M. Nour El-Din
3


ABSTRACT

Substantial length of Greater Cairo Metro Line 3 will be excavated through the Nile
alluviums under a high groundwater head. The planned route passes in close
proximity to many structurally sensitive buildings especially in the downtown area.
Controlling ground subsidence associated with tunneling in these problematic
conditions, is especially consequential when inquiring the success of the new line.
This paper predicts the settlement trough associated with tunneling between Attaba
Station and Bab El-Sharia Station. The analysis is based on two different analytical
approaches. The first approach is a nonlinear three-dimensional finite element
analysis based on the Gap Parameter method. The second approach is a novel method
based on the Artificial Neural Networks (ANN) technology. The ANN was trained
and verified using the observed settlements for many tunnels in Egypt and abroad.
The results show that these approaches compatibly predict the settlement associated
with tunneling although they are based on different modeling techniques.

KEYWORDS: Soft Ground Tunneling, Hydroshield, Three-Dimensional Finite
Elements Analysis, Gap Parameter, Artificial Neural Network (ANN), Genetic
Algorithms (GAs), Monitoring Programs, Settlement Trough, Trough Width
Parameter, Standard Penetration Test (SPT).
"$&
!#% &' )+, -/ 0)2467 8)9:;67 <:+=> ()#?6 A6:C67 EG67 I@ )K2L MNP Q%)RS67 T:2K)#67
UV$ WKS67 0Y7'6 )KDZ$ QK[*\7 ]:K?67 I@ )K2L E_:[ . ( )?% ?67 *:`?67 a@N bR6 7 EG67 d);6:3
QKe:f=\7 QhK2i67 T7c -=:2?67 I@ j%jh67 I@ Q:`V67 Q;iS@ -/ :k'lH( E( QS%j?67 .
( -6:#6:3 / m J bK+S#6 n,:l?67 F'2R67 -/ !4V#67 o+S67 UV$ &()p67 ]b9 QP)V67 >)2#h% !9
-,7'= M:f=> EG67 7b9 . h$ !$ AV267 7b9 -/ K IK 67 EG67 bK+S#6 n,:l?67 F'2R67 -#iV@ IK3 A6:C
\(7 IK#+r#G@ IK#;%)s "7jG#:3 Q%)hf67 d:3( Q2#h67 S2@ :?9 QK 8YjV?67)k:Sh67 Q;%)s -rt
QKiHu67 QK_7)+67 :@> 0)H\7 Q;%)i67 F'2R67 d:`V6 / QKt:Sik\7 QK2lh67 :%uG67 -rt QKS2?
:R2%*j$ !$ -#67 :Rve:#= I@ o;V#67( "7jG#:3 T::Kw r6 n,:l?67 F'2R67 -/ <:+=\7 I@ j%jh
x*:G67( )l@ . IK3 jw( d*:;$ AV267 ye:#= I@ !_)67 -rt F'2R67 z -/ WKrV#67 Q;%)s &u#H
IK#;%)i67 .

1
Professor of Geotechnical Engineering, Ain Shams University, Cairo Egypt
2
Assistant Professor of Geotechnical Engineering, Ain Shams University, Cairo Egypt
3
Graduate Student, Ain Shams University, Cairo Egypt


76
1. INTRODUCTION

As Cairo population has soared in the recent decades from 3.5 millions in 1960 to
about 20 millions today, the need for a new mass transportation system was inevitable.
Studies carried out between 1970 and 1974 suggested to construct the Greater Cairo
Metro to be the spinal mass transportation system (Richards et al., 1997). Madkour et
al. (1999) estimated that about 530 million passengers used the Greater Cairo Metro
in the year 2000.

Metro network, Fig (1), comprises a regional line and two urban lines. The first line of
the Greater Cairo Metro or the regional line was completed in 1989 and was the first
subway metro line in Africa and the Middle East. It is 42.5 km long from El-Marg at
the North of Cairo to Helwan at the South with about 4.5 km underground part
through downtown area using cut-and-cover tunnels (Madkour et al., 1999).

Line 2 also lies along the citys north/south corridor extending from Shubra El-
Kheima to Giza suburban areas. The line was constructed and put into operation in
several phases. Phase 1A, from Shubra El-Kheima to its intersection with Line 1 at
Mubarak Station. Phase 1B starts from Mubarak Station and terminates at Sadat
Station which forms another interchange with the older first line. Line 2 included the
first-ever tunnel crossing under the Nile River (Ahmed, 2001). Two identical
Herrenknecht Bentonite Slurry TBMs (Hydroshields) of 9.45m diameter were
selected to drive the tunnel in the second line. The details of the employed TBM are
shown in Fig (2). Line 3 is situated in the urban densely-populated areas connecting
Imbaba to the downtown of Cairo at Attaba and Heliopolis district. The first phase of
this line is expected to be from Attaba Station to Abbasia Station with a 4.292 km
double deck circular bored tunnel. This phase includes four stations namely, Attaba,
Bab El-Shariaa, El Guesh, Abdo Basha and Abbassia Stations. It is anticipated that
same tunneling technique of Line 2 will be used in driving the tunnel during the
construction of this phase.

The geological conditions prevailing in both Line 2 and the first phase of Line 3 lie
totally within a Pleistocene deposit known as the young alluvial plain that represents
the majority of the lowland portion of the Nile Valley in the Cairo area. The Nile
River deposits governed the subsurface and groundwater conditions in considerable
area of the project. The sediments in the alluvial Nile plain are generally fairly
consistent with depth, but vary somewhat laterally as a result of the long history of
river meanders, and alternate cycles of sedimentation and erosion before the
construction of Aswan High Dam in Upper Egypt in the 1960's. These sediments are
over 60-90 meters thick in the Cairo area underlain by tertiary sedimentary rocks and
older basement rocks (Shata, 1988). Such geological formations limit the tunnel
construction methods to pressurized full face tunneling machines.

Tunneling through Nile alluviums is commonly associated with unfavorable surface
and subsurface ground subsidence that could affect adversely the feeble buildings in
Cairo. The difficulties of tunneling through these deposits are due to the relatively low
strength, high deformability of soils and the shallow groundwater table. The control of
ground subsidence is considered as one of the main tasks of the tunnel design that
dictate the selection of an appropriate tunneling technology.



77


Fig(1): Greater Cairo Metro Network (after Madkour et al., 1999)


Fig (2): Hydroshield used in Line 2 (after El-Nahhas, 1999)


78
2. SUBSIDENCES DUE TO SOFT GROUND TUNNELING

Ground subsidence associated with tunneling activities in soft grounds results from
the axial ground loss in front of the tunnel, the radial ground loss at the peripherals of
the tunnel and changes of the radial and tangential stresses around the tunnel. Ground
losses describe the difference between the actual and theoretical volumes of the tunnel.
Settlements induced by tunneling are influenced by the affinity of the soil to dilate or
densify when sheared. Dilation results in a small influenced zone localized in the area
above the tunnel, whereas densification is usually coupled with movements expanding
towards the surface of the soil. In undrained cohesive soils, constant volume shearing
is anticipated.

Soil displacements associated with tunneling in most tunnel projects are collected,
documented, interpreted and utilized as an input data in the prediction of future
projects under similar conditions. Gained experiences are documented and statistically
studied; consequently, many empirical relationships and recommendations have been
developed. For estimation of settlement distribution above a single shielded-driven
tunnel, the shape of the settlement trough at the ground surface resembles the shape of
error or normal probability curve. Peck (1969) used the properties of an error function
or the normal probability curve to represent the distribution of the surface ground
settlement. Fig (3) illustrates the observed distribution of the ground settlement curve.
According to Peck (1969), the Settlement value (S) at any point is given by:

-
=
2
2
max
i 2
x
exp S ) x ( S (1)

where (S
max
) is the maximum settlement at the centre of the trough and (i) is the
trough width parameter which represents the offset of the point of inflection of the
normal probability curve.

Many researchers suggested relations between the width parameter, the maximum
settlement and the soil/tunnel characteristics (Peck, 1969; Attwell et al., 1986; Fujita,
1989, Mair & Taylor, 1997).

Modeling via finite elements is considered the most powerful means of the analysis of
tunneling-induced settlement as it allows the main parameters involved in the
tunneling process to be accurately accounted for. Finite element can incorporate the
actual geometry of the tunnel, the ground constitutive behavior, seepage towards and
away from the tunnel, and the different construction phases associated with each
excavation technique. Yet, complications in developing finite models covering the
contemporary tunneling operations and intricate model parameters encourage the use
novel models such as artificial neural networks (Nour El-Din, 2003).

The current research makes use of the successful application of both nonlinear finite
elements and neuronet models to predict the surficial settlement trough associated
with the construction of the proposed Line 3 of the Cairo Metro between Attaba
Station and Bab El-Sharia Station.




79

Fig (3): Surface settlement profiles (after Peck, 1969)

3. THREE-DIMENSIONAL FINITE ELEMENT ANALYSIS OF TUNNEL

Realistic analysis of soil-tunneling interaction requires an understanding of details of
the tunneling technology to simulate their effects in the finite element model. The
comprehensive features in tunneling simulation that should be considered in the FE
modeling are illustrated in Fig (4) and summarized as follows:
The nonlinear soil constitutive behavior, which depends mainly on the stress
path, confining pressure and rate of loading,
The unloading forces developed during ground excavation and the potential
seepage towards or away from the tunnel,
The radial and axial ground loss and the overcutting gap,
The pressurized excavation boundaries and ground support measures,
The effect of the potential marginal yield zones around the tunnel,
The tail-skin grouting and the hardening of grouting material with time,
TBM advancement and lining installation,
The mutual interaction between the excavated tunnel and the existing
underground pipelines and tunnels.

Three dimensional analyses seem to be the most suitable numerical framework to
tackle this kind of problems, since the state of stress and strain near the working face
of the tunnel is fully three-dimensional in nature.

The following sections describe the details of the finite element approach. The
employed analysis is based on a rigorous three-dimensional code that was used in the
evaluation of performance of the German Hydroshield in Greater Cairo geological
formations (El-Sayed, 2001). In this algorithm, the soil, shield and liner are modeled
using three-dimensional hexahedral elements.


80






























F
i
g

(
4
)
:

M
o
d
e
l
i
n
g

f
e
a
t
u
r
e
s

o
f

s
o
f
t

g
r
o
u
n
d

s
h
i
e
l
d
e
d

t
u
n
n
e
l
i
n
g

(
a
f
t
e
r

E
l
-
S
a
y
e
d
,

2
0
0
1
)

O
v
e
r
c
u
t

S
e
g
m
e
n
t
a
l

l
i
n
i
n
g

S
h
i
e
l
d

Pressurized face
I
n
s
t
a
l
l
a
t
i
o
n

S
h
a
f
t

P
r
o
b
a
b
l
e

d
e
f
o
r
m
a
t
i
o
n

f
i
e
l
d

S
o
f
t

p
r
e
s
s
u
r
i
z
e
d

g
r
o
u
t


H
a
r
d

g
r
o
u
t

E
x
c
a
v
a
t
i
o
n

T
a
i
l
s
k
i
n

g
r
o
u
t
i
n
g

a
n
d

l
i
n
i
n
g

i
n
s
t
a
l
l
a
t
i
o
n

&

T
B
M

a
d
v
a
n
c
e
m
e
n
t

C
o
n
v
e
r
g
e
n
t

g
r
o
u
n
d

M
u
c
k
i
n
g


d
i
s
p
o
s
a
l

E
x
i
s
t
i
n
g

U
n
d
e
r
g
r
o
u
n
d

S
t
r
u
c
t
u
r
e
s

O
v
e
r
c
u
t

S
e
g
m
e
n
t
a
l

l
i
n
i
n
g

Pressurized face
I
n
s
t
a
l
l
a
t
i
o
n

S
h
a
f
t

P
r
o
b
a
b
l
e

d
e
f
o
r
m
a
t
i
o
n

f
i
e
l
d

H
a
r
d

g
r
o
u
t

E
x
c
a
v
a
t
i
o
n

T
a
i
l
s
k
i
n

g
r
o
u
t
i
n
g

a
n
d

l
i
n
i
n
g

i
n
s
t
a
l
l
a
t
i
o
n

&

T
B
M

a
d
v
a
n
c
e
m
e
n
t

C
o
n
v
e
r
g
e
n
t

g
r
o
u
n
d

F
a
c
e

l
o
s
s
e
s

a
n
d

p
o
t
e
n
t
i
a
l

s
e
e
p
a
g
e

M
u
c
k
i
n
g


d
i
s
p
o
s
a
l

Y
i
e
l
d
e
d

z
o
n
e



81
3.1. Constitutive Modeling
The soil nonlinear behavior is identified by variable modulii dependent on the
confining pressure and the stress level. The soil constitutive relationship is also
expressed in an incremental form to account for the path-dependency. Using the
incremental form of the constitutive matrix [D
et
], the tangential element stiffness
matrix [K
et
] can be written as:

[ ] [ ] [ ][ ]

=
element
e et
t
e et
) Volume ( d B D B K (2)
where the [B
e
] is the element strain-nodal displacement matrix. The loading modulus
(E
t
) and the unloading-reloading modulus (E
ur
) are presented by the following
exponential forms (Duncan and Chang, 1970):

( )( )
n
a
3
2
3
3 1 f
a t
p sin 2 cos C 2
sin 1 R
1 Kp E

f s + f
f - s - s
- =
(3)
n
a
3
a ur ur
p
p K E

s
= (4)

where R
f
is the ratio between the ultimate and the failure deviator stresses, p
a
is the
atmospheric pressure, n is the stiffness exponent, K is the loading stiffness coefficient,
K
ur
is unloading-reloading stiffness coefficient, s
1
& s
3
are the major and the minor
principal stresses and C&f are the soil shear parameters.

3.2. Effect of Stress Path
During the finite element analysis, it is not possible to pre-determine the regions
subjected to loading or unloading in order to conclude whether the loading modulus or
unloading modulus should be used. Since the loading modulus is lower than the
unloading modulus as shown in Fig. (5), the use of the loading modulus can lead to
numerical divergence when unloading occurs. Using the unloading-reloading modulus
during the first iteration of every loading step will underestimate the displacement in
the first iteration if loading occurs but the correct modulus will be used in subsequent
iterations according to a parameter called the stress level (Duncan et al., 1984)
depending on deviator stress, the shear parameters and the confining pressure as
follows:

4
a
3
3
3 1
p sin cos C
) (
SL
s

f s + f
s - s
= (5)

The stress level (SL) is calculated for each gauss point and compared to the maximum
value reached during the loading history at the same gauss point (SL
max
). The
modulus (E) depends on the parameter SL as following:
If (
max
SL SL ), loading is taking place and the used modulus E=E
t

If (
max
SL 75 . 0 SL ), unloading is taking place and used modulus E=E
ur

If (
max max
SL 75 . 0 SL SL > > ), neutral loading is taking place and the modulus
E is calculated by interpolation between E
t
and E
ur
as shown in Fig (5).



82







Fig (5): Effect of stress path on soil stiffness (after Duncan et al., 1984)

3.3. Interface Modeling
The shield-soil interface is modeled using a hyperbolic gap element. The liner-grout-
soil interface is modeled by introducing grout elements with incremental hardening
strength parameters and initial hydrostatic pressure equal to the grouting pressure.
Details of the tail-grout characteristics and modeling are elaborated by El-Sayed
(2001) as shown in Figs (6) and (7); the normal stiffness is considered a constant of
very high value for the contact and of trivial value for the open gap. The tangential
shear stiffness (k
t
) of the gap is defined by:

ns
a
n
n
sf
i w ti t
p
R
A k k

- =
s
d s
t
g
2
tan
1 (6)
where k
ti
is the stiffness coefficient, g
w
is the water unit weight, s
n
is the stress normal
to the interface surface, A
i
is the contact area and R
sf
& n
s
are defined similarly to the
soil hyperbolic model parameters.



Fig (6): Interface element
SL
max
0.75 SL
max

E
t

E
ur

SL
E
e
1

s
1
-

s
3

E
t

E
ur

Local axes for
convergent gap
Normal
stiffness
Tangential
stiffness


83

























Fig (7): Interface modeling for shielded tunneling (after El-Sayed, 2001)





Overcut
Shield elements
Gap elements
Soil elements
(a) Shield/soil interface modeling
Liner elements
Soil elements
(b) Liner/grouting/soil interface modeling
Tail gap
Shield/Liner
Enfolding ground
Initially prestressed grout elements


84
3.4. Modeling of Staged Construction
Tunnel excavation is modeled by removing a cluster of ground elements from the
finite element meshing; conversely, lining elements are new elements that are added
to the mesh; this algorithm is equivalent of the Stress Reversal Approach (Ahmed,
1991). The required changes in the mesh are applied to reconstruct the residual vector
{R} resulting from the difference between the applied force and the straining forces
and the tangential stiffness matrix [K
t
]. The residual vector and the stiffness matrix
are calculated at the beginning of each iteration (Newton-Raphson), i.e. the (i+1)
th

iteration is described by the following equation:

[ ] { } { } R U K
t t
1 i
t t
1 i t
t t
i
D +
+
D +
+
D +
=
&
{ } [ ] { }


=
D + D +
s - =
elements of No.
1 e
element
e
t t
i
T
e
t t
) Volume ( d B F (7)
where {F} is the nodal force vector. The left subscript denotes the iteration process
and the left superscript donates a sequential time index. If the iteration superscript is
zero, the matrix or vector is calculated at the end of the previous time step. The stress
increment can be calculated from the strain {e} using the following integration:

{ } { } { } { } [ ] { }
{ }
{ }

e
e
D + D +
+
D + D +
+
D +
+
D +
e + s = s + s = s
t t
1 i
t t
i
d D
et
t t
i
t t
1 i
t t
i
t t
1 i
& (8)
Numerical integration is used to evaluate the integral in the stress calculations.
Employing a predictor-corrector method (Modified Euler scheme) as follows:

{ } { } [ ] { } { } [ ] [ ] { } U B D
2
1
D
2
1 t t
1 i e et
t t
i
t t
i
t t
1 i et
t t
i
t t
i
t t
2 / 1 i
&
&
D +
+
D + D + D +
+
D + D + D +
+
+ s = e + s @ s (9)

then is calculated using the predicated stress . The stress is updated as follows:

{ } { } [ ] [ ] { } U B D
t t
1 i e et
t t
2 / 1 i
t t
i
t t
1 i
&
D +
+
D +
+
D + D +
+
+ s @ s (10)

3.5. Geotechnical Data
Tunnel depth at the considered section is 20.5m as shown in Fig(8). According to
Hamza Associates (2002), the soil in the region of the bored tunnel between Attaba
Station and Bab El Shariaa has the following distinctive layers:
1. Man-made ground: This layer is encountered at the ground surface and
extends down to a depth from the ground surface of 5.50 m. This layer
constitutes mainly of sand, stone pieces, red bricks, asphalt pieces, silt and
clay.
2. Clay: Firm to stiff brown micaceous calcareous CLAY with little calcareous
pebbles. This layer is encountered below the first layer and extends down to a
depth from the ground surface of 10.45 m. This layer is interbedded at a depth
of 9.0 by a 0.45 m thick layer of firm Silt.
3. Sand: Dense to very dense yellowish brown micaceous calcareous slightly
silty to silty poorly graded SAND with different percentage of gravel at some
depths varying from slightly gravelly to gravelly. This layer is encountered
below the second layer and extends down to the end of the boring.

The groundwater was encountered at a depth of 1.1m. The geotechnical properties of
the encountered soil strata are listed in Table (1).


85


Fig (8): Tunnel configuration and soil stratification

Table (1): The geotechnical parameters
Stratum
SPT
(blows)
C
(kPa)
f
o
K K
u
R
f
n R
sf
n
s

g
(kN/m
3
)
FILL 7 0 25 50 150 0.8 0.6 0.8 0.6 17
CLAY 15 50 0 100 300 0.8 0.5 0.8 0.5 18.5
SAND 50 0 37 600 1800 0.8 0.5 0.8 0.5 19

3.5. Ground Loss
The face loss is estimated to be 0.09%, while the loss along perimeter of the TBM
(which is mainly due to the overcutting) is estimated to be 0.14%. These data are
based on the results of the monitoring programs performed on two test sections that
were heavily instrumented during the construction Line 2 - Phase 1A (Hamza
Associates, 1995). Campenon Bernard-SGE (1999) reported comparable values of
ground losses during the driving of Al-Azhar Twin Roadway Tunnels.

3.6. Results of Finite Element Simulation
The data input required several hours to be completed and verified during all stages of
tunnel construction. The mesh of the finite element comprises 3015 element and 3528
joint. Twenty two steps were employed in the analysis. Each step is divided into 10
substeps. Newton-Raphson iterative procedure was used in the analysis. About 4
hours were required to complete the analysis using Pentium IV-1.7 GHz computer.
The predicted settlement from the finite element analysis is shown in Fig (9). The
maximum settlement is 13.8 mm and the trough extends to about 35m.


86


Fig (9): The spatial distribution of settlement using finite element method


4. THE NEURONET MODEL

Artificial Neural Network (ANN) is a simulation of human-like response within the
computer hardware and specialized software using multiple layers of interconnected
processing elements called neurons. ANN systems are not programmed, they just
learn by examples. The learning process of the neural networks is accomplished by
adjusting the strengths of the binds between the neurons to cause the overall network
to output proper results. The learning algorithm is based upon the backpropagation
algorithm (Werbos, 1974; Rumelhart et al., 1986) by which the weight of the
connections between the neurons are adjusted to reduce the error between the desired
output and the actual output.

Artificial neural networks are formed by clustering of the primitive artificial neurons
into layers, which are then connected to one another. Some of the neurons interface
with the external environment to receive the inputs and other neurons provide the
networks outputs through the input and the output layer. All the rest of the neurons
are included in a number of hidden layers between the input and the output layers
(Tsoukalas and Uhrig, 1997).

4.1. Network Topology
The model comprises two-staged ANN model similar to procedures of hand
calculations of tunnel-induced settlements. The configuration of the neural network is
shown in Fig (10). The first stage (A), indicated by light solid arrows, is used to
estimate the maximum settlement using the following input parameters: the diameter
of tunnel, the depth of the tunnel, depth of the groundwater table, thickness of
different soil layers, and the average values of the standard penetration test (SPT)
counted in each layers. The second stage (B), indicated by bold dashed arrows, is used
to estimate the trough width from the following input variables: the tunnel depth, the
excavated diameter of tunnel and, the maximum surface settlement obtained from the
first stage.


87




Fig (10): The ANN topology
(* indicates multiple nodes)


*
*


88
A key motive to divide the ANN model into two stages is the difference in size of data
of the maximum settlements and trough widths. Most instrumentation programs
exercise frequenter measurements of the longitudinal maximum settlement than
acquiring the transverse troughs (Murray, 1990).

As the complex interaction among the network architectural units makes model
optimization a difficult task, the network configuration has been determined using the
evolution principle of Genetic Algorithms (GAs). Genetic Algorithms combine
selection, crossover, and mutation operators with the goal of finding the best solution
to the problem. The introduction of GAs in ANNs is implemented by introducing both
the structure and parameters of the neural network as a fixed-length string and a
population of such string is evolved to minimize the error between the predicted
outputs and the training set. There are four stages in the genetic search process:
initialization, evaluation, selection, crossover and mutation. In the initialization stage,
a population of genetic structures which are randomly distributed in the solution space
is selected as the starting point of the search. In the second stage, each structure is
evaluated using a fitness function. On the basis of their relative fitness values,
structures in the current population are selected for reproduction.. The selected
structures are recombined using crossover. A mutation operator, which arbitrarily
alters one or more components of a selected structure, provides the means for
introducing new information into the population (Nour El-Din, 2003).

4.2. Generalization Technique
One of the primary goals in training neural networks is to ensure that the network
performs well on data that is has not been trained on; this principal is called
generalization. The standard method of ensuring good generalization is to divide
the training data into multiple data sets. The most common data sets are the training,
testing, and validation data sets. Testing (cross-validation) data set is used by the
network periodically during training as a test for performance. During testing, the
weights are not trained, but the performance of the network on the cross-validation set
is saved and compared to past values. If the network is starting to over-train on the
training data, the cross-validation performance will begin to degrade. Thus, the cross
validation data set is used to determine when the network has been trained as well as
possible without over-training. About 50% of the data in the input and desired files
will be used for training, 25% are used for cross validation (testing) and 25% for
validation of the model. The network is optimal when the error in the cross
validation set is at its minimum position.

4.3. Training and Testing Databases
The training database was collected from field measurements for surface settlement
associated with soft ground tunneling and the corresponding geotechnical information.
The database covers a wide range of variation in geotechnical data and the tunnel
geometrical data (depth and diameter). The database used in developing the stages (A)
and (B) are detailed in Table (2).

The considered cases comprise the following tunneling techniques: Hydroshield
Tunneling (Greater Cairo Metro, Al-Azhar Roadway Tunnels, El-Salam Syphon),
Earth Pressure Balance TBM (Alexandria Wastewater Tunnel), Compressed Air TBM
(Cairo Wastewater Tunnel) and Open Face TBM (Alberta Experimental Tunnels).
Also, the data bases include tunnel diameters ranging between 2.00 and 9.45m with


89
depth ranging between 5 to 25m. The geotechnical conditions range between soft
alluvial deposits with shallow groundwater depths for Cairo projects, marine
geological deposits of El-Salam Syphon, fractured limestone for Alexandria waste
water projects, and glacial clay till for Alberta tunnels.


Table (2): Database of the tunnels used to develop the ANN
Project and Location
Depth of
springline (m)
Tunnel
Diameter
(m)
No. of
Cases
(A*)/(B**)
Greater Cairo Metro Egypt
(after Hamza Associates, 1995 & 1997)
13.4-25 9.45 17/2
North Tunnel Al-Azhar Roadway
Tunnels Egypt
(after Campenon Bernard-SGE, 1999)
19.5-23 9.45 4/1
Alberta Experimental Tunnels Canada
(alter El-Nahhas, 1980)
27& 11.7 2.56 & 6.2 2/9
Alexandria Wastewater Tunnel Egypt
(after Kotait, 2001)
14 2.81 1/1
Spinal Tunnel Cairo Wastewater
Tunnel Egypt
(after El-Nahhas et al., 1990)
15.6 5.15 1/1
First Tunnel Al-Salam Syphon
Egypt
(after Esmail, 1997)
24 6.6 1/1
Soft Soil Tunnels
(after Lee et al., 1999)
5 2.5 0/7
* Used in ANN for stage (A)
** Used in ANN for stage stage (B)

4.4. Validation of the Proposed ANN Model
The performance of the model has to be checked for an independent validation data
set that was previously unseen by the model. The coefficient of correlation (R
2
), as
defined by Chapra and Canale (1990), is the key criteria to evaluate the performance
of analytical models. The value of R
2
generally range between zero and one. The
model behavior can be categorized according to the value of (R
2
) as shown in Table
(3).

Tables (4) and (5) show that the value of the coefficient of correlation between the
validation (production) data set and the ANN results are generally less the correlation
with training or cross-validation sets. However, strong correlation is observed
between all data sets and ANN results.

The plots of the measured and predicted settlements for training, cross validation and
production are shown in Figs (11) and (12). The results indicate that the model
performs well in obtaining the Gaussian settlement distribution characteristics





90
Table (2): ANN performance versus (R
2
) values (after Chapra and Canale, 1990)
Category Value of R
2

Strong correlation exists between output of the ANN and actual data More than 0.8
Medium correlation exists between output of the ANN and actual data Between 0.2 and 0.8
Weak correlation exists between output of the ANN and actual set Less than 0.2

Table (3): Correlation Coefficient for stage (A)
Data set R
2

All data 0.92
Training data 0.95
Cross-validation data 0.98
Production data 0.80

Table (4): Correlation Coefficient for stage (B)
Data set R
2

All data 0.97
Training data 0.99
Cross validation data 0.99
Production data 0.84


0
3
6
9
12
15
18
21
24
27
30
0 3 6 9 12 15 18 21 24 27 30
Actual maximum settlement (mm)
P
r
e
d
i
c
t
e
d

m
a
x
i
m
u
m

s
e
t
t
l
e
m
e
n
t

(
m
m
)
Greater Cairo Metro
North Tunnel Al-Azhar
Road Tunnels
Alberta Experimental
Tunnels
Alexandria Waste Water
Project
Spinal Tunnel Cairo
Wastewater
First Tunnel Al-Salam
Syphon
Predicted = Measured

Fig (11): Stage (A) performance
Roadway Tunnels


91
0
2
4
6
8
10
12
14
16
18
20
0 2 4 6 8 10 12 14 16 18 20
Actual trough width parameter (m)
P
r
e
d
i
c
t
e
d

t
r
o
u
g
h

w
i
d
t
h

p
a
r
a
m
e
t
e
r

(
m
)
Greater Cairo Metro
North Tunnel Al-Azhar
Road Tunnels
Alberta Experimental
Tunnels
Alexandria Waste Water
Project
Spinal Tunnel Cairo
Wastewater
First Tunnel Al-Salam
Syphon
Soft Soil Tunnels
Predicted = Measured

Fig (12): Stage (B) performance


4.4. ANN Prediction of Line 3 Settlement
The ANN model was feed the tunnel configuration and SPT data to analyze the
tunneling status between Attaba and Bab El-Sharia Stations. The time consumed in
feeding the input and production of the output was only a few minutes. The model
gives a maximum settlement of 13.8mm and a trough width parameter of about 8.5m.
A comparison of results of the two models is shown in Fig (13) which implies that
almost typical results can be obtained using both approaches. The maximum
difference between the two models is about 1mm near the end of the trough. Variation
in the parameters of the numerical model or geotechnical data may give differences
larger than predicted discrepancies in the analyses.

0
2
4
6
8
10
12
14
16
-50 -40 -30 -20 -10 0 10 20 30 40 50
Distance from the tunnel CL (m)
S
e
t
t
l
e
m
e
n
t

(
m
m
)
ANN model
FE model

Fig (13): The results of the finite element model and the neuronet model

Roadway Tunnels


92
5. SUMMAY AND CONCLUSIONS

In the present study, two dissimilar analytical approaches are utilized to predict the
settlement trough associated with construction of Line 3 bored tunnel between Attaba
Station and Bab El-Sharia Station. The first approach is based on nonlinear three-
dimensional finite element simulation of the Hydroshield tunneling technology. The
model adapts the main factors affecting the pressurized bentonite slurry tunneling
such as unloading forces due to excavation, ground nonlinearly, interface conditions,
engineering properties of shield, rate of advance, machine overcutting, face pressure,
yielding zones and the tail pressurized grouting process. The second approach is based
on the back-propagated supervised ANN enhanced by the evolution capabilities of
GAs. Many projects in Egypt and abroad were utilized to train, test and validate the
proposed model. The database used in model training covers a wide spectrum of
geological conditions, configurations and construction techniques. The two
approaches give compatible results concerning the maximum settlement and the
trough width.

The ANN model proved to be reliable and robust in determining the settlement
associated with tunneling despite the small effort and time to develop the model.
Results of mentoring programs of the future tunneling projects cab be employed in
developing more yielding ANN models to compete with the current practice of
complicated finite element methods.

6. ACKNOWLEDGEMENT

The authors are deeply grateful to Dr. A. Abu-Krisha of the National Authority of
Tunnels (NAT) for providing invaluable monitoring and geotechnical information for
the Greater Cairo Metro Project.

7. REFERENCES

1. Ahmed, A.A., 1991, "Interaction of Tunnel Lining and Ground", Ph. D. Thesis,
Ain Shams University, Cairo, Egypt.
2. Ahmed, A.A., 2001, Three-Dimensional Modeling of the Nile crossing, 9
th

Int. Colloquium on Structural & Geotechnical Eng., Ain Shams Univ., Cairo,
10-12 April.
3. Attewell, P., Yeates, J. and Selby, A., 1986, "Soil Movements Induced by
Tunnelling and Their Effects on Pipielines and Structures", Blackie & Sons
Ltd., Glasgow, UK.
4. Campenon Bernard-SGE, 1999, "Tunneling at the CWO Crossing, Results of
Montoring", El Azhar Road Tunnels Project, Detailed Design, NAT, Egypt.
5. Chapra, S.C. and Canale, R.P., 1990, "Numerical Methods for Engineers",
McGraw-Hill Publishing Co., NY, USA.
6. Duncan, J. M. and Chang, C. Y., 1970, "Nonlinear Analysis of Stresses and
Strains in Soils", Journal of Soil Mech. and Found. Div., ASCE, Vol. 96, No.
SM5, pp. 1629-1653.


93
7. Duncan, J. M., Seed, R. B. Wong, K. S. and Ozawa, Y., 1984, "FEADAM84,
A Computer Program for Finite Element Analysis of Dams", Virginia
Polytechnic Inst. and State University, Dept. of Civil Engineering, USA.
8. El-Nahhas, F., El-Kadi, F. and Shalaby, A., 1990, Field Measurements
During Construction of a Compressed Air Tunnel in Cairo, Proc. of Intl.
Congress on Tunnels and Underground Works Today and Future, Chengdu,
China.
9. El-Nahhas, F.M, 1980, "The Behaviour of Tunnels in Stiff Soils", Ph.D.
Thesis, Alberta University, Edmonton, Canada.
10. El-Nahhas, F.M., 1999, "Soft Ground Tunnelling In Egypt: Geotechnical
Challenges and Expectations", Tunnelling and Underground Space
Technology, Vol. 14, No. 3, pp. 245-256.
11. El-Sayed, S.M., 2001, "Elasto-Plastic Three-Dimensional Analysis of Shielded
Tunneling with Special Application on Greater Cairo Metro", Ph.D. Thesis,
Ain Shams University, Faculty of Engineering, Cairo Egypt.
12. Esmail, K.A., 1997, "Numerical Modeling of Deformation around Closed Face
Tunneling", Ph. D. Thesis, Ain Shams University, Cairo, Egypt.
13. Fujita, K., 1989, "Underground Construction, Tunnel, Underground
Transportation", Proc. of 12th ICSMFE, Rio de Janeiro, pp. 2159-2176.
14. Hamza Associate, 1995, "Greater Cairo Metro: Phase (2) Tunnel Monitoring",
Comprehensive Report, NAT, Egypt.
15. Hamza Associates, 1997, "Greater Cairo Metro Line 2 Phase 2, Phase 3
Tunnel Monitoring", Lot 42 (El-Dokki/ El-Gezira, NAT, Egypt.
16. Hamza Associates, 2002, Geotechnical Investigation Report : Greater Cairo
Metro Line (3), NAT, Egypt.
17. Kotait, H, 2001, "Effect of Tunneling on Adjacent Structures", Ph.D. Thesis,
Alexandria University, Alexandria, Egypt.
18. Lee, Chung-Jung; Wu, Bing-Ru; and Chiou, Shean-Yau, 1999, "Soil
Movement around a Tunnel in Soft Soils", Proc. Natl. Sci. Counc. ROC (A),
Vol. 23, No. 2, pp.235-247.
19. Madkour, A., Hudson, M. A. and Bellarosa, A., Construction of Cairo Metro
Line 2, Proc. Inst. Civ. Engrs, Civ. Eng, 1999, 132, May/August, pp. 103-117
20. Mair, R. and Taylor R., 1997, "Bored Tunneling in the Urban Environment",
Thema Lecturer, Plenary Session 4, 14th Int. Conf. On SMFE, Hamburg, 6-12
Sept.
21. Murray, R.T., 1990, "Rapporteur's Paper", in Geotechnical Instrumentation in
Practice, Proceedings of the Conference of Geotechnical Instrumentation in
Civil Engineering Projects, Thomas Telford, London, UK, pp. 75-85.
22. Nour El-Din, S.M., 2003, "Neuronet Prediction of Settlement Associated with
Soft Ground Tunneling", M.Sc. Thesis, Ain Shams University, Faculty of
Engineering, Cairo, Egypt.


94
23. Peck, R.B., 1969, "Deep Excavation and Tunneling in Soft Ground", State-of-
the-Art, Proceeding of the 7th International Conference on Soil Mechanics and
Foundation Engineering, Mexico City, Mexico, pp. 225-290.
24. Richards, D.P., Ramond, P. and Herrenkenecht, M., 1997, "Slurry Shield
Tunnels on the Cairo Metro", General Report, RETC, Las Vegas, USA.
25. Rumelhart, D.E., Hinton, G.E. and Williams, R.J., 1986, Learning Internal
Representations by Error Propagation, in Parallel Distributed Processing,
Rumelhart and McClelland (Eds.), Vol 1, Cambridge, MA: MIT Press, USA.
26. Shata, A.A., 1988, Geology of Cairo, Egypt, Bulletin of the Association of
Engineering Geologists, Vol. XXV, No. 2, pp. 149-183.
27. Tsoukalas, L.H. and Uhrig, R.E., 1997, "Fuzzy and Neural Approaches in
Engineering", John Wiley & Sons Inc, NY, USA.
28. Werbos, P.J., 1974, "Beyond Regression: New Tools for Prediction and
Analysis in the Behavioral Sciences", Ph.D. Thesis, Dept. of Applied
Mathematics, Harvard University, Cambridge, Mass, USA.

You might also like