PHD Unimi R11266

UNIVERSITÀ DEGLI STUDI DI MILANO
Dottorato di Ricerca in Scienze della Terra

Ciclo XXXI
Geomechanical characterization of rock masses

for tunnel excavation with TBM
Ph.D. Thesis
Giacomo Armetti
Matricola R11266
Tutor Academic Year Coordinator

Prof.ssa Tiziana Apuani 2017 - 2018 Prof. ssa Elisabetta Erba
Co- Tutor
Prof. Maria Migliazza
ABSTRACT
The prediction of Tunnel Boring Machine (TBM) performances and the comprehension of the
relationship between the geological, mechanical and structural features of a rock mass and the
machine behaviour are fundamental in order to select the most effective tunnel construction
methods and to estimate the condition of excavation in terms of time and economic costs of the
infrastructure.
The general purpose of this work is to carry out predictive models utilisable during the early stage
of tunnel planning and realisation; for this purpose, many approaches have been considered.
Simple and multiple, both linear and nonlinear, regression analyses have been performed in order to
achieve an empirical equation able to estimate the principal TBM performance indices (ROP and
FPI). In addition, Multiple Correspondence Analysis and clustering have been developed. A
modified version of the Rock Engineering System has been performed and used to obtain a
predictive model.
This study focuses on the exploratory tunnel named “La Maddalena”, located in Northern Italy,
Western Alps, where geological surveys data have been continuously performed during tunnel
construction; this allowed the collection of a large amount of geological, geotechnical and TBM
performance data.
INDEX
1 INTRODUCTION 1
1.1 General purpose 1
1.2 Work organisation and thesis structure 2
2 MECHANISED TUNNELLING 5
2.1 Introduction 5
2.2 Types of Tunnel Boring Machines 6
2.2.1 Open type TBM 7
2.2.2 Shield type TBM 9
2.2.3 Closed mode TBM 11
2.3 Rock breakage mechanism 12
2.3.1 Disc cutters 12
2.3.2 Rock failure 13
2.4 TBM performance indices 16
2.4.1 Rate of penetration 16
2.4.2 Advance rate 16
2.4.3 Coefficient of utilisation 17
2.4.4 Penetration rate per revolution 17
2.4.5 Field penetration index 17
2.4.6 Specific penetration 18
2.4.7 Specific excavation rate 18
2.5 Factors influencing the TBM performances 19

2.5.1 Geological and geotechnical factors 19
2.5.2 Mechanical factors 20
2.5.3 Operational factors 21
3 LA MADDALENA EXPLORATORY TUNNEL PROJECT 23
3.1 Introduction 23
3.1.1 Mont Cenis Base Tunnel 25
3.2 La Maddalena exploratory tunnel 28
3.2.1 General features and purpose 28
3.2.2 Geographical location 28
3.2.3 Tunnel route 29
3.2.4 Encountered rock mass lithologies 30
3.2.5 Rock supports and reinforcements 31
3.2.6 The Tunnel Boring Machine “Gea” 34
4 GEOLOGICAL SETTING 39
4.1 Introduction 39
4.2 Penninic domain 40
4.2.1 Grand Saint Bernard nappe 40
4.3 Tectono-stratigraphic unit of Ambin 42
4.3.1 Clarea complex 44
4.3.2 Ambin complex 46
4.4 Structural evolution of Ambin 48
4.4.1 Pre-Alpine evolution 48
4.4.2 Ductile Alpine evolution 48

4.4.3 Brittle Alpine evolution 49
4.4.4 Neotectonics 50
5 ROCK MASS CHARACTERIZATION 55
5.1 Introduction 55
5.2 Intact rock property determination 57
5.2.1 Lithological description 57
5.2.2 Micro-petrographic description 58
5.2.3 Uniaxial compressive strength 61
5.2.4 Determination of rock abrasivity 69
5.2.5 Determination of rock hardness 73
5.3 Quantitative description of discontinuities 76
5.3.1 Types of discontinuity 80
5.3.2 Orientation 81
5.3.3 Spacing between discontinuities 89
5.3.4 Volumetric joint count 94
5.3.5 Rock Quality Designation index, RQD 96
5.3.6 Rock block volume 98
5.3.7 Groundwater conditions 100
5.3.8 Joint conditions 101
5.4 Rock mass classifications 110
5.4.1 Rock Mass Rating (RMR) system 110
5.4.2 Geological Strength Index (GSI) 117
5.5 Seismic wave propagation 121

5.5.1 Tunnel Seismic Prediction System 121
5.5.2 Tunnel Seismic Prediction results 123
5.6 Tunnel depth 126
5.7 Conclusive remarks 128
5.7.1 Domain I 129
5.7.2 Domain II 130
5.7.3 Domain III 131
6 TBM PERFORMANCE ANALYSIS 133
6.1 Field TBM performance database 134
6.1.1 The first part of the database (DB1) 134
6.1.2 The second part of the database (DB2) 136
6.2 Measured TBM performances 137
6.2.1 Rate of Penetration distribution 137
6.2.2 Field Penetration Index 139
6.3 Conclusion 142
7 COMPARISON BETWEEN DIRECT GEOSTRUCTURAL SURVEY AND SEISMIC

METHODS FOR PREDICTING GEOLOGICAL FRAMEWORK AHEAD TUNNEL FACE
143
7.1 Introduction 143
7.2 Comparison of the results 144
7.2.1 Prediction of significant lithological changes 144
7.2.2 Estimation of fracturing degree of rock masses and presence of shear or weakness
zones prediction 145
7.2.3 Estimation of orientation of the principal elements of discontinuities 147

7.3 Conclusion 152
8 LITERATURE MODELS FOR PREDICTING TBM PERFORMANCES AND

THEIR APPLICATION TO “LA MADDALENA” EXPLORATORY TUNNEL 153
8.1 Approaches for predicting TBM performance 154
8.1.1 Introduction 154
8.1.2 Theoretical models for predicting TBM performance 155
8.1.3 Empirical models for predicting TBM performance 157
8.1.4 Final observation about theoretical and empirical predictive models 166
8.1.5 Artificial intelligence techniques for predicting TBM performance 168
8.1.6 Final observation about artificial intelligence technique 172
8.2 Evaluation of various TBM performance prediction models based on field data from “La
Maddalena” tunnel 174
8.2.1 Cassinelli et al. (1982) model 175
8.2.2 Innaurato et al. (1991) model 177
8.2.3 Khademi Hamidi et al. (2010) models 179
8.2.4 Hassanpour models 182
8.2.5 Delisio et al., models 189
8.3 Conclusion 193
9 DEVELOPMENT OF TBM PERFORMANCE PREDICTIVE MODELS 199
9.2 Theoretical concepts 200
9.2.1 Correlation between variables 200
9.2.2 Simple linear regression analysis 202
9.2.3 Multiple linear regression analysis 206

9.2.4 Nonlinear regression analysis 209
9.3 Effects of single rock/rock mass properties on TBM performances 210
9.3.1 Uniaxial compressive strength 210
9.3.2 Quartz content 212
9.3.3 Alpha angle 213
9.3.4 Spacing between weakness planes 214
9.3.5 Joint volumetric count 216
9.3.6 Rock quality designation index 217
9.3.7 Joint conditions 218
9.3.8 Tunnel Depth 220
9.3.9 Rock Mass Rating 221
9.3.10 Geological Strength Index 223
9.4 Combined effect of rock/rock mass properties on TBM performances 224
9.4.1 Empirical models to predict the rate of penetration 225
9.4.2 Empirical models to predict the field penetration index 230
9.5 Conclusion 236
10 CLUSTER ANALYSIS OF ROCK MASS PARAMETERS FOR TBM

PERFORMANCE ESTIMATION 237
10.1 Multidimensional data analysis 238
10.1.2 Euclidean space 239
10.1.3 Principal component methods 240
10.1.4 Cluster analysis 248
10.1.5 Observations 254

10.2 Application to “La Maddalena” case study 255
10.2.2 Relationship among geological and geomechanical variables 256
10.2.3 Relationship among geological – geomechanical variables and TBM performance

265
10.2.4 Regression analyses on clusters 270
10.3 Conclusion 280
11 ROCK ENGINEERING SYSTEM: FROM THE INTERACTION MATRIX TO TBM

PERFORMANCE ESTIMATION 281
11.1 Theoretical aspects 281
11.1.1 General concepts 281
11.1.2 Matrix resolution 282
11.1.3 Atlas of rock engineering system 284
11.1.4 Matrix coding 285
11.1.5 Cause-effect plot 286
11.1.6 Theorems 1 and 2 289
11.2 RES application to TBM performance estimation 290
11.2.1 Rock mass parameters involved in RES application 290
11.2.2 Interaction matrix 290
11.2.3 Predictive model development 294
11.3 Conclusion 301
12 PREDICTIVE MODEL VALIDATION 303
12.2 Validation of ROP models 306

12.2.1 Validation of RL2 model 306
12.2.2 Validation of RNL model 309
12.2.3 Validation of ROP models carried out for each cluster 311
12.3 Validation of FPI models 315
12.3.1 Validation of FL4 model 315
12.3.2 Validation of FNL model 317
12.3.3 Validation of FPI models carried out for each cluster 319
12.3.4 Validation of FPI model obtained by means of RES approach 323
12.4 Conclusion 324
13 CONCLUSION 327
14 REFERENCES 331
APPENDIX I: Schemes used during geological surveys
APPENDIX II: Histograms of frequency of the principal geological and geotechnical variables
APPENDIX III: Longitudinal geological section of “La Maddalena” tunnel
APPENDIX IV: The generic underground excavation matrix; Hudson, 1992

1 INTRODUCTION
1.1 General purpose
The general purpose of this thesis is to investigate the relationship between the geological,
mechanical and structural features of rock masses and the performances of Tunnel Boring Machine
used for the excavation in order to carry out empirical models able to predict the behaviour of a
machine operating in a specific geomechanical context.
Understanding the influence of geological ground conditions on the TBM performances is

fundamental both to select the most appropriate machine and to estimate the costs and the times of
tunnel realisation to the purpose of making mechanised tunnelling cheaper, safer and more effective
than the traditional method.
This research work focuses on “La Maddalena” exploratory tunnel, which is one of the
infrastructures related to the planned Lion-Turin high speed railway. La Maddalena is located in
Susa Valley, Western Alps, North Italy.
In order to achieve this objective an extensive field work has been performed. The entire tunnel
length (about 7000 meters) has been continuously surveyed and a large amount of information
about discontinuity properties has been collected; moreover, several rock matrix indices have been
acquired. The principal operating parameters have been acquired during tunnelling and analysed to
the purpose of carrying out the performance indices to be relate to rock mass features. A large
database containing the rock mass and intact rock features and the TBM indices has been built and
implemented. Data elaboration allowed for a detailed ground characterisation and TBM
performance description.
The relationship between the geological parameters and the machines ones has been investigated by
means of different approaches with the aim of developing TBM performance predictive models. In
particular, regression analyses, multiple correspondence analyses and clustering, and the Rock
Engineering System (Hudson, 1992) have been performed.
In a first stage, a detailed literature review allowed to select fourteen existing predictive models in
order to evaluate their applicability to La Maddalena tunnel, by comparing the measured
performance and the calculated ones.
1
1.2 Work organisation and thesis structure
Simple and multiple regression analyses, both linear and nonlinear, have been performed to
quantify the correlation between the independent variables (geological and geotechnical properties)
and the response ones (TBM performances).
Multiple correspondence analysis and cluster analysis have been performed in order to objectively
divide the rock mass into homogeneous portions, considered as different classes of excavatability.
The Rock Engineering System (Hudson, 1992) has been applied by providing some modifications
to the original version, in order to define a situ-specific cause-effect matrix able to define the
geomechanical factors predisposing a certain machine response. The final purpose is to define an
empirical TBM performance predictive equation on the basis of the cause-effect matrix.
The predictive capability of the regression models carried out during this work has been tested by
comparing the performances measured during tunnelling and the ones estimated by means of the
predictive model.
This thesis is divided into three parts:
The first part consisted of an extensive field work, with the purpose of collecting a large amount of
geological, structural and mechanical information as well as TMB data collection. In the first part of
this research work, rock mass and machine performance have been described in detail; this part took
a long time.
The second phase was devoted to an accurate statistical analysis of the relationship between rock
mass parameters and machine performances. As a result, empirical predictive modes have been
obtained.
The third phase was dedicated to the validation of the obtained models. In this part, both their
capability to explain the relationship between geological parameters and TBM performances, and to
predict TBM indices have been evaluated.
The working procedure followed during the research is represented in Figure 1.1.
2
DATA COLLECTION
GEOLOGICAL AND TBM

GEOTECHNICAL PARAMETERS
Indirect survey: Direct survey: Personally collected and

seismic method tunnel wall mapping processed Acquired
Analysis Analysis
ROCK MASS TBM

Validation PERFORMANCES
CHARACTERISATION
ANALYSIS OF THEIR RELATIONSHIP
REGRESSION MULTIPLE ROCK ENGINEERING

ANALYSES CORRESPONDENCE ANALYSIS SYSTEM
Simple Multiple Cluster analysis Rock

interaction
matrix
LINEAR AND CLASSES
NONLINEAR OF
EMPIRICAL MODELS EXCAVATABILITY
PREDICTIVE
MODEL
VALIDATION
Figure 1.1 Working flow chart.
A brief summary of each chapter is provided:
Chapter 2: in this chapter a brief overview about mechanised tunnelling is provided. The main
working principles of TBM, the different types of machines, the geological contexts in which they
are involved and the most common TBM performance indices are discussed.
Chapter 3: this chapter describes the general aspects of “La Maddalena”, as the TBM used for the
excavation, the types of supports required, the encountered geological formations, etcetera. The
main features of Lyon-Turin rail and the Mont Cenis Base Tunnel are also described.
Chapter 4: the regional and local geological settings of the area hosting “La Maddalena” tunnel are
described in this chapter. Particular attention was given to the lithological description of the
formations encountered during tunnelling and to neotectonic framework.
3
Chapter 5: the description of geological and geomechanical features of the excavated rock mass has
been provided in this chapter. Here, both the results of tunnel wall mapping, personally carried out,
and the acquired results of laboratory tests on rock samples are presented. Global description of
rock mass is provided.
Chapter 6: this chapter is about the TBM performance indices acquisition, elaboration and analysis.
The statistical description of registered performances has been provided.
Chapter 7: it is devoted to describe the comparison between the information collected during direct
geological surveys, described in Chapter 5, and the one obtained by means of an indirect survey
method: the so called Tunnel Seismic Prediction System.
Chapter 8: this chapter provides an overview about the TBM performance predictive methods and
the description of their advantages and disadvantages. Some of these models have been applied to
this case study and the results have been commented here.
Chapter 9: an overview about the statistical regression methods has been supplied. The results of
simple (single rock parameter correlated with TBM index) and multiple (more than one geological
parameter vs. machine one), linear and nonlinear regression analyses have been described.
Chapter 10: a general description of the multiple correspondence analysis (MCA) and cluster
analysis (CLA) is supplied. In this chapter the MCA and CLA performed on La Maddalena tunnel
data has been set up, moreover, a detailed geological description of each obtained cluster has been
provided.
Chapter 11: in this chapter the Rock Engineering System is introduced. Some modifications with
respect to the classical approach have been proposed and its application to La Maddalena case study
and the obtained results have been discussed.
Chapter 12: the validations of the obtained predictive models are described in this chapter and a
critical discussion about their reliability and their findings is also provided.
Chapter 13: this chapter summarises the general conclusions of the thesis.
4
2 MECHANISED TUNNELLING
2.1 Introduction
Tunnel Boring Machines, TBMs, were introduced in 1950s in order to mechanise tunnelling
operations, improve worker safety conditions and make mechanised tunnelling more convenient and
quicker than the traditional “drill and blast”.
There are different types of TBMs (see Paragraph 2.2) but all of them are composed by a main body
and by a back-up system and consist of four system groups (Maidl et al., 2008):
- Boring system: composed by the cutterhead, the cutters and the cutterhead drive;
- Thrust system: where the main tools are the thrust cylinders; it includes also the clamping
structure and the rear and front support;
- Rock cutting removal system: it has the aim of removing the muck from the tunnel by means
of cutterhead buckets and conveyor belts;
- Lining and support system: this system depends on the type of machine.
The basic operational principle of TBMs is that rotation and thrust are simultaneously applied to the
cutterhead: the first one is guaranteed by the torque generated by TBM motors and the second one
by means of thrusts cylinders. The cutters mounted on the cutterhead are responsible to the rock
mass breakage; the resulting muck is collected by the specific buckets situated on the cutterhead
and moved outside the tunnel by means of the conveyor system.
On the basis of the geological and geomechanical context in which the tunnel has to be realised
different types of rock reinforcements and rock supports can be installed, such as bolts, steel wires,
shotcrete, etc..
5
2.2 Types of Tunnel Boring Machines
Construction companies produce various types of TBMs in order to face the different geological
and geomechanical contexts and different systems are used to classify them. Firstly, it is possible to
distinguish soil/soft ground TBMs and rocky ones.
Generally, the geological and structural setting in which TBM has to work determines its selection;
Figure 2.1 provides a clear example of the suggested machine for each of the most common
geological settings.
Figure 2.1 Geological contexts and suggested TBM (Robbins Company).
One of the most important distinctions in rock TBM classification is between the open (or gripper)
and the shield TBM. In the diagram reported in Figure 2.2, a list of gripper and shield TBM
subcategories has been provided.
Figure 2.2 Types of TBMs (Zou, 2017).
6
2.2.1 Open type TBM
The open type TBM, also called gripper TBM, is widely used for tunnel excavation where rock
masses are characterised by global good quality and stand-up time long enough to put in work the
required supports in safety conditions, given that usually there are not any protection unit for the
worker.
In some cases a partial shield behind the cutterhead can be installed, in particular, it is possible to
install a roof shield (on the crown) or a side steering shoes (on the crown and the walls); moreover,
some machines have a shield protecting the cutterhead.
In Figure 2.3 open TBM without any shields and the three partial shield application solution are
shown.
Figure 2.3 Schemes of open type TBM; a) without any shield, b) with roof shield, c) with roof shield
and side steering shoe, c) with cutterhead shield (Maidl et al., 2008).
The clamping system is composed by the gripper group which includes the gripper plates.
Hydraulic jacks move the grippers.
If a single gripper system is present, we have a main beam TBM (Robbins type); on the contrary if a
double gripper system (named Kelly) is presents, we have a Kelly beam TBM (Jarva type).
Boring cycle of an open TBM is composed by four principal steps.
Firstly, the gripper shoes are pushed against tunnel walls previously excavated (by means of
laterally extendible hydraulic cylinders) and rear legs are lifted, then thrust cylinders push the
cutterhead against the tunnel face and, at the same time, the cutterhead is rotating. In this way
cutters start to cut the rock, and single pieces called chips are broken out of the rock mass. The
excavated material is taken up by buckets situated at the cutterhead and after some passages rock
7
chips are removed from the tunnel. Cutterhead rotation is stopped when the bore stroke is
completed; at this point rear legs are extended. The last stage is called regripping, in this phase
gripper plates are retracted and the retraction of thrust cylinders makes the gripper unit slide
forward.
In Figure 2.4 the most important elements characterising a gripper TBM are reported.
Figure 2.4 Elements characterising a gripper TBM (Maidl et al., 2008).
Open TBM grants free access to the tunnel walls, in this way, when the excavation is realised by
means of one of the TBMs belonging to this group, geologists and engineers have the possibility to
directly collect geological, structural and mechanical data about the encountered rock masses.
For this reason the open TBM is the most adopted in the construction of exploratory tunnels.
8
2.2.2 Shield type TBM
The shield TBM differs from the open ones for the presence of one or two shields. These cover the
main body of the TBM, and not only a limited part of it behind the cutterhead. This group of
machines is utilised in low quality rock masses, in jointed ones, in which stand-up time is limited or
very limited and it is not long enough to allow the miners to put in operation supports and linings.
The shield has the function of avoid tunnel collapse and of making operators work in safe
conditions.
The group of shield TBMs is divided in single shield TBM and double shield TBM.
2.2.2.1 Single shield TBM
This group of TBMs is usually used to excavate jointed rock masses with low stand-up time; in
these contexts single shield TBMs can guarantee high tunnelling performances thanks to the pre-
cast segmental lining or pipe jacking tunnelling methods.
The substantial difference between open and shield TBM is that in the first group the contrast
required by thrust cylinders to transfer the thrust to the cutterhead is provided by rock tunnel walls,
whereas in the second group the contrast is provided by lining, given that rock mass strength in
which single shield TBMs are used generally lower that the one characterising the rock mass in
which open TBM are adopted.
The boring process is cyclic because when bore stroke is completed thrust cylinders have to be
retracted and the ring erector has to install the pre-cast segmental lining.
In Figure 2.5 a typical diagram for a single shield TBM is provided.
Figure 2.5 Diagram for a single shield TBM (Robbins Company).
9
2.2.2.2 Double shield TBM
Double shield TBMs, also called telescopic shields, are frequently used in long tunnel excavation in
hard rocks when high fractured or shear zones and faults may be present. They are suggested for
stable and unstable geological conditions.
The double shield TBMs are considered the most sophisticated ones because they are a combination
of gripper and single shield machines. They guarantee very high tunnelling performances by
working in continuous tunnelling mode.
They consist of two principal elements: a front shield with cutterhead, drive and main bearing; and
a gripper shield with gripping units, auxiliary thrust cylinders and tailskin (Herrenknecht AG). The
two parts of the shield are connected by the main thrust cylinders and a telescopic shield protects
them. When stable rock mass conditions occur the machine is braced against the tunnel walls by
means of the gripper shoes, in this way the front shield can be advanced independently of the
gripper shield thanks to the main thrust cylinders. The segments are installed in the tailskin section
while the machine is tunnelling.
When a bore stroke is completed the gripper shoes are loosened and the auxiliary thrust cylinders
make the gripper shield push behind the front shield.
In Figure 2.6 the principal components of a double shield TBM are shown.
Figure 2.6 Principal components of a double shield TBM (Herrenknecht AG).
10
2.2.3 Closed mode TBM
This type of TBMs is generally used for the excavation of soils, very soft rocks or grounds under
the water table and in general where tunnel face needs to be supported. On the basis of tunnel face
balancing system it is possible to distinguish three main groups of close mode TBMs: the
compressed air machines, the slurry shield machines and the earth pressure balance (EPB) ones:
- In the compressed air machines the element that balances the tunnel face pressure is the
compressed air;
- Slurry shield machines balance the horizontal ground and water pressure by means of a
mixture of bentonite and water; they are used for the excavation of sand and fine gravel
(Figure 2.7);
- In the EPB the balancing material is the one excavated with the addition of liquid or foam
conditioning agents. EPB machines are used to excavate clay, silt and fine sand (Figure 2.8).
There is another type of machines belonging to the closed mode TBM; it is represented by the
crossover TBMs (also called multi-mode TBMs or hybrid machine). This group of machines is able
to work both in open-mode and in closed-mode. In the first case they excavate as a single shield
TBM, in the second one they work as slurry shield or EPB machine. Some cross over TBM are
designed with the purpose of passing from EPB to slurry and vice versa when they are working in
closed-mode. For these reasons the hybrid machine are particularly suggested for the excavation in
highly variable geological contexts.
Figure 2.7 Slurry shield TBM. 1: cutterhead, Figure 2.8 EPB TBM. 1: cutting wheel, 2:
2: excavation chamber, 3: air bubble, 4: excavation chamber, 3: pressure bulkhead, 4:
working chamber, 5: slurry feed line, 6: slurry main drive, 5: thrust cylinders, 6: screw
return line, E: slurry radial shield pressure, F: conveyor, 7: erector, 8: double chamber
grout pressure (Mooney et al., 2016). airlock, 9: tailskin with integrated grout lines
(Institute of Geology of Malaysia).
11
2.3 Rock breakage mechanism
TBMs excavate rock masses through crushing and rotating by applying high pressure on the rock
face with high thrust forces while chipping and rotating by means of a number of tools (disc cutters)
installed on the machine face, the cutterhead. Disc cutters cause the rock failure and the formation
of slabs of rock, also called chips, guaranteeing tunnel advance.
2.3.1 Disc cutters
The cutters (Figure 2.9) installed on the cuttehead are responsible for the rock cutting. There is a
large variety of cutters designed for different ground conditions; in hard rock contests disc tools are
typically used: single discs (the most employed), twin discs (generally mounted in the center of
cutterhead) and double discs. The cutters are mounted on roller bearings and can be replaced in
front of or behind the machine face.
In the Figure 2.10 a schematic representation of a disc cutter with its principal component is
reported.
Figure 2.9 Single disc cutter. Figure 2.10 Disc cutter components (Xia et al
2015).
On the market there is a wide range of cutter diameters up to 6” (150 mm) for microtunnels
realisation, 12” (305 mm), 20” (508 mm), in hard rock tunnelling the most employed diameter are
17” (432 mm) and 19” (483 mm). Cutter geometry seriously affects tunnelling, for example the
increase in the diameter implies an increase in both load limit and linear velocity.
The cutter steel alloy has to be selected on the basis of the rock properties; when soft ground
conditions occur, standard steel should be chosen, in presence of hard and abrasive rocks the best
12
solution could be the heavy duty steel and in case of abrasive soft ground hard-faced version
(Palmieri Company) is frequently adopted.
2.3.2 Rock failure
Before explaining the rock breakage process under TBM rolling cutters it is necessary to introduce
the cutter movements performed during their action and the forces acting on them.
Each cutter performs two rotational movements during its work:
- It rotates around disc axis: this movement is due to the contact with the rock face;
- It rotates around the tunnel axis: this movement is due to the cutterhead rotation.
As for the forces acting on the cutters during their rotation, it is possible to distinguish three forces,
as shown in Figure 2.11.
Figure 2.11 The forces acting on the cutter; R: cutter radius, p: penetration depth, ϕ angle between
cutter and rock (on the left Balci et al., 2009; on the right Wang et al., 2015).
- The normal force (thrust froce), Fn, is the force that the cutterhead applies to the cutter in the
vertical direction. It is the force necessary to maintain the tool at the desired level of
penetration (Balci et al 2009);
- The rolling force, Fr, is the force applied to the disc in the rolling direction by means of
cutterhead torque (Wang et al., 2009). Fr is used to estimate the specific energy requirement,
that is the amount of energy needed to excavate a unit volume of rock (Balci et al., 2009);
- The side force, F’, is represented by the perpendicular action force of the rock and acts on
the contact surface between the cutter and the rock (Wang et al., 2015).
13
The rock breakage mechanism by means of TBM rolling cutters can be divided into two continuous
stages (Gong and Zhao, 2009). In the first one, named indentation, the disc intrudes into the rock
mass, generating internal cracks, large and small fragments.
The second stage is characterized by the propagation of cracks between two adjacent cutters. They
converge towards each other and coalesce, in this way rock chips are generated between two
cutters.
More precisely Cook at al. (1984) described the following phases: the building up of a stress field
followed by the formation of a damaged zone, the generation of surface chipping and finally a
crater formation. As a consequence of the indentation process, three different sectors can be
observed: the crashed zone, the crack and the elastic one (Cook at al.,1984; Chiaia, 2001; Gong et
al., 2006). The penetration process composed of multiple indentation cycles, and each of them
expands the crushed zone and crack propagation allowing the meeting of fractures generated by
adjacent cutters.
Cutters have to be mounted on the cutterhead in order to allow the crack intersection and the
minimum consumption of cutting energy. In order to guarantee an optimum cutting geometry a
parameter given by the ratio between cutter spacing and penetration rate has been introduced and it
should range between 10, for soft rocks and 20 for hard ones (Ozdemir, 1992; Rostami, 2008;
Rostami et al, 2017).
This discussion is graphically represented by Figure 2.12.
Figure 2.12 Rock chip formation under cutter action (Herrenknecht, 2003).
When the spacing between cutters is too small overcrushing occurs, this causes high consumption
of specific energy; whereas when spacing is too large chips cannot be generated (Figure 2.13).
14
Figure 2.13 Cutter spacing and specific energy consumption (Tuncdemir et al., 2008).
The spacing between joints occurring in a rock mass has a great influence on the chip forming
process. The role of joint spacing on rock fragmentation under cutter action has been studied by
many researchers (e.g. Gong et al., 2006; Sun et al., 2011; Liao et al., 2013) by means of field tests,
laboratory tests (e.g. indentation test) and numerical simulation methods.
In particular, Gong et al. (2006) demonstrated that when joints are low spaced (spacing range from
50 mm to 80 mm) the joints were favourable to rock fragmentation; on the contrary, joints had a
little influence on rock breakage mechanism when spacing is higher than 500 mm. In this case,
crack process is similar to the one observed for the single cutter action on the rock without fractures
(Gong et al., 2006).
15
2.4 TBM performance indices
Over the years, many researchers have introduced various indices in order to access the boreability
of a rock mass. These indices are commonly known as TBM performance indices.
Here they have been listed and a brief description for each of them is provided.
2.4.1 Rate of penetration
The rate of penetration (ROP) or penetration rate (PR) is defined as the excavated length in a
continuous phase during the effective boring time; it represents net penetration. This time does not
consider machine breakdown, downtimes for TBM maintenance, excavation stop for reinforcement
or support element or lining installation.
ROP can be calculated as:
𝐿𝑒
𝑅𝑂𝑃 = (2.1)
𝑇𝑒
where Le is the excavated length and Te is the effective boring time.
ROP is generally expressed in [m/h] or in [mm/min].
ROP does not consider any machine specification or operational parameter.
2.4.2 Advance rate
The advance rate, AR, contrariwise to ROP, is the ratio between excavated length and total time,
which includes the time necessary to bore, to install supports and lining, the time needed for the
TBM ordinary and extraordinary maintenance, etc..
AR is calculated in accordance to:
𝐿𝑒
𝐴𝑅 = (2.2)
𝑇𝑡
where Le is the excavated length and Tt is the total time necessary to do all the required tunnelling
operations.
AR is generally expressed in [m/h] or in [mm/min].
AR, like ROP, does not consider any machine specification or operational parameter.
16
2.4.3 Coefficient of utilisation
The coefficient of utilisation is the ratio between AR and ROP:
𝐴𝑅
𝑈= (2.3)
𝑅𝑂𝑃
U is expressed in percent [%].
In stable rock masses the AR is considerably lower than ROP and U ranges between 30 – 50 % due
to machine daily maintenance (Nielsen and Ozdemir, 1993; Boniface, 2000; Sapigni et al., 2002). In
rock masses characterised by low quality ROP could be potentially high, but AR is generally low
due to the reinforcement and support installation necessity, the gripper bearing failure, etcetera. For
this reason, U results as low as 5 – 10 % or less (Barla and Pelizza, 2000; Sapigni et al., 2002).
2.4.4 Penetration rate per revolution
The penetration rate per revolution is the effective bored length for each complete revolutions of the
cutterhead. It can be calculated in accordance to the equation 2.4:
𝑅𝑂𝑃 ∗ 1000
𝑃𝑅𝑒𝑣 = (2.4)
𝑅𝑃𝑀 ∗ 60
where ROP is the rate of penetration and RPM is the number of complete cutterhead revolution in
one minute.
PRev is expressed in [mm/rev].
2.4.5 Field penetration index
The field penetration index, FPI, (Hamilton and Dollinger, 1979; Nelson et al., 1983; Tarkoy and
Marconi, 1991; Klein et al., 1995) is the ratio between the total force acting on the cutterhead or
acting on each cutter and the penetration rate per revolution. It represents the force necessary to
obtain a determinate penetration during a single revolution.
FPI can be calculated according with:
𝐹𝑛
𝐹𝑃𝐼 = (2.5)
𝑃𝑅𝑒𝑣
where Fn is the force acting on each cutter or on the cutterhead.
17
FPI is measured in [kN/cutter/mm/rev] or in [kN/mm/rev] if the total force acting on the cutterhead
is considered and not the force acting on each cutter.
FPI has been introduced in order to represent a normalised measure of ROP, specified in terms of
applied thrust on the cutters for a specific penetration rate per revolution in various geological and
geomechanical situations.
The use of FPI implies a linear relationship between penetration and thrust, however a nonlinear
relationship occurs (Rostami, 1997; Laughton, 1998; Gong and Zhao, 2007). This limitation is
overcome if the machine penetration is higher than the threshold penetration, suggested equal to 1
mm/rev by Gong and Zhao (2007), where ROP and thrust relationship is linear.
2.4.6 Specific penetration
The specific penetration, SP, (Alber, 2000) is the inverse of FPI. SP can be calculated according to
the following formula:
𝑃𝑅𝑒𝑣
𝑆𝑃 = (2.4)
𝐹𝑛
SP is measured in [mm/rev/kN/cutter] or in [mm/rev/kN] if the total force acting on the cutterhead

is considered and not the force acting on each cutter.
SP has been introduced by Alber (2000) with the aim of combining the thrust and the cutterhead
revolution rate with the penetration rate, this allow penetration rate to be normalised against rock
mass strength variations (Farrokh et al., 2012).
Being SP the inverse of FPI, the discussion carried out for FPI limitation can be considered valid
for SP.
2.4.7 Specific excavation rate
The specific excavation rate, SER, was developed by Stevenson (1999) in order to combine the
specific penetration and the tunnel cross area, in fact it can be calculated as the ratio between the
excavated volume per each cutterhead revolution and the thrust per cutter. Final formulation is:
𝑆𝐸𝑅 = 𝐴 ∗ 𝑆𝑃 (2.4)
where A is the tunnel cross section area. SER is measured in [(m3/rev)/(kN/cutter)].
18
2.5 Factors influencing the TBM performances
There are several factors able to influence mechanised tunnelling; they can be divided into three
groups:
- Geological and geotechnical factors: related to the ground that has to be excavated, they can
change considerably during tunnel length. They include both properties related to the intact
rocks (as strength, texture, mineralogy, abrasivity) and to the rock masses (as fracturing
degree, joint orientation, groundwater flow conditions;
- Mechanical factors: these are related to the machine and for this reason they are constant
during the excavation. This group includes the TBM characteristics, like type of TBM,
diameter, cutter specifications;
- Operational factors: as support installation or experience of workers. As the geological

factors, they may be subject to significant changes during the excavation.
2.5.1 Geological and geotechnical factors
The geological, structural, geomechanical and hydrogeological context in which a tunnel

construction takes place is the factor that most significantly influences the TBM performances; for
this reason, the selection of a TBM to be used in a tunnel project should be made on the basis of the
geological condition of the site.
It is necessary to consider that a TBM selection is irreversible, so, in order to guarantee an

appreciable construction level and avoid machine stops, a detailed rock masses characterisation is
fundamental since the project stages.
One of the most influencing parameter is the rock strength; generally when it increases the
penetration decreases and the thrust required to guarantee a penetration is high (Yagiz, 2008; Gong
and Zhao, 2009; Khademi Hamidi et al., 2010; Armetti et al., 2017).
Fracturing degree also affects TBM performance. Penetration is facilitated when the rock mass is
jointed, so this situation could increase ROP, however the occurrence of high fractured ground
requires the installation of supports and reinforcements determining low values of AR. Moreover, if
the rock mass is strongly fractured and the spacing between discontinuities is smaller than disk
spacing, there is no (or little) chip formation, but regrinding of rock blocks. This significantly
reduces the penetration. Also the orientation of discontinuities with respect to the tunnel axis could
influences the penetration, the most favourable angle is of 60° (Bruland et al., 1998; see Chapter 5).
19
Generally it is possible to assert that high quality rock masses are more difficult to be penetrated by
a TBM and as a result low values of ROP and high ones of FPI are measured. Supports are rarely
requested so AR is generally high.
On the contrary, when tunnelling is performed in low quality rock masses net penetration is
generally high and the associated requested thrust is low, however AR is low due to the long time
needed to install supports.
2.5.2 Mechanical factors
The features of machines, cutterheads and cutters are the mechanical factors that influence the TBM
performances. These characteristics are the result of design stages and they cannot be modified
during excavation phases; sometimes modification can be operated on cutter alloy or profile.
The machine type influences tunnelling, for example double shield TBMs allow for continuous
boring cycle (accelerating tunnel construction) in contrast to gripper type (see section 2.2).
Generally machine size affects its performances. The highest performances can be registered with
small TBMs: this is due to the cutterhead rotation speed; which is higher for small diameters than
for large ones. Moreover large size machines require more maintenance time than the small ones.
Some measures as flat cutterhead profile and cutter protection allow the reduction of cutter and
cutterhead wear and breakage, especially when jointed conditions occur (Delisio and Zhao, 2014).
Cutter features (type, profile, material and diameter) seriously affect the performances of a TBM. In
particular, Roxborough and Phillips (1975) and Phillips and Bilgin (1977) assert that larger cutters
require higher thrust in order to obtain the same penetration for a rock mass, being disc -rock
contact area larger. The thrust capacity increases when cutter diameter increases and, as a
consequence, the penetration. Cutters characterised by larger diameters have a longer life because
they rotate slower for a given head rotation speed and they are able to collect more material to wear
before to be replaced (Roby et al., 2008).
20
2.5.3 Operational factors
The operational factors that influence the TBM performance can be divided into two groups: the
constant factors and the variable ones.
The constant ones cannot be modified during the excavation, for example the shape and the
diameter of the tunnel. On the contrary the variable can be modified on the basis of the geotechnical
conditions encountered. This group is represented by machine thrust, torque, power and all TBM
driving choices that can adjusted on the basis of rock mass strength and fracturing degree. For
example in very hard rock thrust is maintained low in order to avoid excessive cutter consumption
and machine damages.
21
3 LA MADDALENA EXPLORATORY TUNNEL
PROJECT
“La Maddalena” exploratory tunnel is one of the infrastructures related to the realization of the high
speed/high capacity rail that aims to connect Lyon (France) to Turin (Italy). For this reason, a
synthetic description Lyon-Turin section and of Mont Cenis Base Tunnel is provided prior to the
description of La Maddalena project.
3.1 Introduction
The “Core Network Europe” was presented by the European Commission in October 2011; it will
connect the regions, the capital, the hubs and the ports over the Alps and Pyrenees, allowing the
modal shift in order to reduce the CO2 emissions.
The objective is to achieve a 17,500 km network of high capacity/high speed rails and Lyon-Turin
Lyne is the fundamental joint of the railway line designed with the purpose of linking Algeciras
(Spain) to Budapest (Hungary); it is located in the central sector of the so called “Mediterranean
Corridor number 3”, of the TEN-T Trans-European conventional rail network. It will connect the
regions, the capital, the hubs and the ports over the Alps and Pyrenees.
The areas along the Mediterranean Corridor (in particular the Lombardy, the Rhone-Alpes region,
Catalonia and Madrid) are considered one of the most important socio-economics regions within the
European Union. For these reasons it is currently one of the most important European projects.
In particular Corridor number 3 will connect Algeciras, Madrid, Tarragona, Seville, Valencia,
Tarragona, Barcelona, Perpignan, Lyon, Turin, Milan, Venice, Ljubljana and Budapest.
In Figure 3.1 the nine network corridors are presented with different colours, the green one in the
Mediterranean Corridor, number 3.
23
3.1 Introduction
Figure 3.1 The Core Network Corridors. Green line: Mediterranean corridor, yellow: Atlantic,
violet: North Sea-Mediterranean, orange: Rhine-Alpine, red: North Sea-Baltic, pink: Scandinavian-
Mediterranean, dark blue: Baltic-Adriatic, light blue: Rhine-Danube, brown: Orient/East-
Mediterranean. (European Commission).
The missing section is the rail link between Italy (Turin) and France (Lyon); nevertheless, it is
highly strategic because it will connect South-West Countries of Europe with Central and Eastern
European regions. Failing Lyon-Turin link means hampering transport relations and so trade
between Spain, France and Italy and between these Countries and Central and Eastern Europe.
As it is possible to observe from Figure 3.1 Lyon-Turin railway line is located in the middle of two
long communication axes that run North-South (North Sea-Mediterranean Corridor) and East-West
(Mediterranean Corridor).
24
3.1 Introduction
3.1.1 Mont Cenis Base Tunnel
The new Lyon-Turin line has a length of 235 km and its core element is represented by a 57.5 km
length base tunnel named Mont Cenis Base Tunnel (MCBT); the promoter and owner company is
TELT, Tunnel Euroalpin Lyon-Turin, previously (LTF, Lyon Turin Ferroviaire).
Today the freight traffic between Italy and France is relevant and the rail mode is supported by the
old Fréjus railway, which is not adequate to answer the need presented. As for Italy-France axis, it
has been estimated that only a little part of goods traffic is guaranteed by rail mode, because of
there are relevant constraints in terms of time and costs associated with the old Fréjus railway
(compared with German axis 29% and Swiss axis 64%). For these reason, Mont Cenis Base Tunnel
is the main project of the whole Corridor 3.
The tunnel has been designed with the purpose of connecting Susa (Turin Area, North Italy) to
Saint Jean de Maurienne (France); it crosses two states, Italy and France, and it will have a total
length of 57.5 km: 12.5 km of tunnel will be realised in Italy and the others 45 km in France. Given
that it is designed as a twin tube tunnel its effective length will be equal to 115 km. Considering that
170 communication bypasses (located each 333 meters), 4 intermediates accesses, 5 ventilation
systems and 3 underground security areas have been planned, the total tunnel length is nearly of 160
km, 80 km of them by means of TBM and 80 with traditional method. The designed tube diameter
is approximately of 10 meters.
A schematic representation of tunnel structure is provided in Figure 3.2.
Figure 3.2 Stucture of Mont Cenis Base Tunnel (TELT).
25
3.1 Introduction
3.1.1.1 Geological prevision
The new Mont Cenis tunnel will cross seven geological zones (Figure 3.3):
- Ultradauphinoise Zone: the principal rocks encountered by the tunnel are sandstones and
schists;
- Sub-Briançonnaise Zone: in this zones tunnel crosses the Croix des Tête Massif,
characterised by charbonatic rocks (limestones, marls and dolostones);
- Houillère Briançonnaise Zone: as for Houillère Briançonnaise Zone MCBT encounters the
Bréquin-Orelle, La Praz and Fourneaux Units; it consists of schist, sandstones and
charconatic levels;
- Inner Briançonnaise Zone (Vanoise): the prevalent lithologies are gneiss and micaschist;
- Gypsum Nappe Zone: the former are evaporitic rocks (like gypsum and anhydrite),
charbonatic rocks, dolostones and calc-schist;
- Briançonnaise Zone of Ambin Massif: characterised by gneiss and micaschist;
- Piemontese Zone: the former is prevalently calc-schist.
Exploratory tunnel
Tunnel access
Tunnel access La Maddalena
Modane
Tunnel access La Praz Length: 7020 m
Saint Martin La Porte Length: 4000 m
Length: 2480 m
Length: 2400 m
1 2 3 4 5 6 7
Figure 3.3 Geological section of Mont Cenis tunnel. 1: Ultradauphinoise Zone, 2: Sub-
Briançonnaise Zone, 3: Houillère Briançonnaise Zone, 4: Inner Briançonnaise Zone, 5: Gypsum
Nappe Zone, 6: Briançonnaise Zone of Ambin Massif, 7: Piemontese Zone (LTF, 2012).
26
3.1 Introduction
3.1.1.2 Access and exploratory tunnel

Prior to the realisation of Mont Cenis, four access/exploratory tunnels had been constructed. Only
one of them is located on the Italian side: “La Maddalena” exploratory tunnel. The other three are
sited on the French side, they are: Saint Martin La Porte, La Praz and Modane tunnels (Figure 3.4).
Figure 3.4 Accesses and exploratory tunnels of MCBT (Tunneltalk, 2013).

- Saint Martin La Porte: the tunnel of Saint Martin La Porte has been characterised by two
working phases. During the first one, from 2003 to 2010, an access tunnel was realized.
Since 2016 an exploratory tunnel has been built down Saint Martin La Porte access with the
purpose to investigate the geological conditions of one of the most complex zones of Mont
Cenis. The excavation has been realised by means of an 11.21 m diameter TBM named
“Federica”. The encountered rock masses consist of schists and sandstones belonging to
Houillère Briançonnaise Zone;
- La Praz: the tunnel access named “La Praz” is 2480 meters long and 11 meters in diameter,
it was constructed between 2005 and 2005 adopting traditional drill and blast method. The
access tunnel of La Praz was excavated into schist and sandstone belonging to Houillère
Briançonnaise Zone;
- Modane: it is an access tunnel long 4000 meters and with a diameter of 10.40 m. it was
constructed between 2005 and 2007 with drill and blast method. The encountered rock
masses consist of quartzite, micaschist belonging to Inner Briançonnaise Zone (Vanoise)
and limited levels of anhydrite. Relevant groundwater circulation has been observed at the
contact between the evaporitic rocks and the quartzite;
- La Maddalena: the exploratory tunnel named “La Maddalena” is described in Paragraph 3.2.
27
3.2 La Maddalena exploratory tunnel

3.2.1 General features and purpose
La Maddalena is one of the four preliminary infrastructures related to MCBT excavation. The
excavation works started in 2012 and concluded in 2017, it is long 7020 meters and it has a
diameter equal to 6.30 meters.
La Maddalena is an exploratory tunnel realised with the purpose to collect geological,

geomechanical, structural and hydrogeological data and all that useful information necessary to plan
the excavation of the Mont Cenis Base Tunnel. Before its planning it is fundamental to understand
the rock mass behaviour as consequence of mechanized tunnelling.
In accordance with the global good quality of La Maddalena rock mass and with the necessity to
directly map the tunnel walls and to install monitoring systems into rock mass, the excavation was
carried out by means of open type Tunnel Boring Machine.
The description of the involved TBM is reported in the following paragraph.
La Maddalena has been planned considering both its primary aim (to gether geological information)
and its future utilization; in fact, during MCBT contruction La Maddalena will be used as access
and ventilation tunnel and during working phase of MCBT it will become a securety tunnel.
3.2.2 Geographical location
The exploratory tunnel of La Maddalena is located in North Italy, in Piedmont Region in Turin
province.
The tunnel access and the yard are sited in Chiomonte Municipality, in the locality named “La
Maddalena” that gives the name to the tunnel; it is at an altitude of 615.5 m on sea level and it is
overhung by “Clarea” viaduct of A32 Torino-Bardonecchia motorway (Figure 3.5).
28
Figure 3.5 Geographical position of the access to La Maddalena tunnel in Chiomonte, Italy
(Google Maps)
3.2.3 Tunnel route
The length of tunnel is 7020.0 meters, its orientation was decided during the planning phase and it
ranges from 307° (from North) at the beginning of excavation to N333° until Pk. 3 + 090 m, at the
end of excavation tunnel axis orientation is N299°. The tunnel rises with a gradient of 0.34% until
pk. 1 + 500.0 m, then it descends with a gradient of 3.32% for 1600 m at the altitude of 573.40 m on
sea level; it is the deeper point of the whole project (Padovese et al, 2017). From this point the
excavation rises with a maximum gradient of 1.20% in order to reach the level of the base tunnel.
A map with La Maddalena and MCBT routes has been reported in Figure 3.6.
Figure 3.6 La Maddalena route (red line) and MCBT one (double blue line); scale 1:5000.
29
3.2.4 Encountered rock mass lithologies
In this section the lithologies encountered during La Maddalena excavation have been listed. The
detailed description is provided in the following chapters.
From Pk. 0 + 198,0 m to 0 + 967,0 m the lithological formation is gneiss belonging to the Ambin
complex (following identified as Domain I, DI, reported in literature as AMC); from Pk. 1 + 360.0
m rock mass consists of micaschist belonging to Clarea complex (following identified as Domain
III, DIII, reported in literature as CLR), between Pk. 0 + 967.0 m and 1 + 360.0 m a transition zone
characterized by an alternation of gneiss and micaschist was encountered (following identified as
Domain II, DII, reported in literature as AMD).
Table 3.1 summarises the acronyms proposed by literature, the ones adopted in this work and the
formal names of geological complex.
Initial (used in
Lithology Formal Name Main lithology Acronym the present
thesis)
Gneiss Ambin complex aplitic gneiss AMC DI
Gneiss- aplitic gneiss, minuti
Transition zone AMD DII
Micaschist gneiss, micaschist
micaschist and gneiss
Micaschist Clarea complex minuti reequilibrated in CLR DIII
blueschist facies
Table 3.1 Lithologies, formal names, acronyms and initials of rock masses encountered during La
Maddalena excavation.
The preliminary geological section has been reported in Figure 3.7.
Figure 3.7 Preliminary geological section of La Maddalena tunnel (Padovese et al., 2017).
30
3.2.5 Rock supports and reinforcements
In order to guarantee long term stability and safety during construction, rock supports and
reinforcements have been provided for La Maddalena excavation. The rock mass stability
conditions changes from stable to local instability conditions of blocks or wedges and the most
relevant support systems involved are rock bolts and steel ribs.
Nine support section types have been applied on the basis of rock mass conditions (Table 3.2).
Section Type Rock Support
STABLE F1
Occasional Bolting
F2
F3a
F3b Systematic Bolting
F3c
F3c1
Light steel ribs
FMV
F4
Heavy steel ribs
INSTABLE F5
Table 3.2 Section types and rock supports and reinforcements applied during La Maddalena
construction.
Sections F1 and F2 are the more light ones and they have been adopted where rock mass are
characterised by good quality and it is self-sustaining. These conditions occur in the gneiss of DI
and part of transition zone (DII) and the first sector of micaschist of Clarea complex.
Sections F3a and F3b require systematic bolting on the roof, they are occasionally utilised in the
transition zone crossing. Section F3c includes systematic bolting both on the roof and on the walls
and steel wire mesh have been applied in addition to the bolts on the roof (Figure 3.8).
This section was adopted in jointet sections of DI and DIII (Padovese et al., 2017).
31
Figure 3.8 Example of F3c section application. Rock bolts and wire mesh have been installed; on
the right bolts are being applied.
F3c1 was involved when jointed conditions make blocks of rock detach from tunnel roof, it was
installed during the excavation of Clarea micaschist, in particular from Pk. 1399.0 m and pk. 4575.0
m. It includes rock bolts, wire meshes and single steel ribs.
The section FMV has been adopted to micaschist of Clarea complex crossing from Pk. 4895 meters.
This section needs the introduction of a specific support (similar to the one proposed by C&M
McNally Engineering Company, known as McNally Support System). Specifically, it is
characterised by systematic bolting and single steel ribs longitudinally jointed by steel bars, steel
bars are parallel to tunnel axis and they are put in the roof and in the upper parts of walls (Figure
3.9).
Figure 3.9 FMV section. On the left, tract of tunnel where FMV section has been installed; on the
right, detail of this section utilization.
32
The bars between ribs work as a shield and they avoid rock wedge and chip falling into the tunnel
void.
F4 and F5 section types have been applied only in those situations where very jointed masses occur,
for example during fault and cataclastic zones crossing, encountered during DIII excavation
(Padovese et al 2017). They require the installation of coupled steel ribs (0.5 m or 1.4 m spaced)
together with bars (previously described) or wire mesh (Figure 3.10).
Figure 3.10 Coupled steel ribs and bars, F4 and F5section types.
The support and reinforcement elements adopted during La Maddalena tunnel construction have
been synthetically described in this paragraph.
All of them, when put to work, allow the operators to observe the tunnel walls, in order to collect
geological, geomechanical and structural rock mass data by direct surveys, given that rock mass is
accessible in accordance with the open TBM type utilized to La Maddalena realisation.
33
3.2.6 The Tunnel Boring Machine “Gea”
Gea is the name of the Tunnel Boring Machine involved in La Maddalena excavation.
It is an open type main beam gripper TBM produced by Robbin’s Company and it has been
specifically designed to be applied in mixed rock masses and therefore it is equipped with tools
suitable for this utilization. It has a total length equal to 240.0 meters.
The specifications of “Gea” have been listed in the Table 3.3.
Machine Diameter 6.30 m

Cutters Series 17 (431.8 mm)
- Number of disc cutters 43
- Max recommended individual cutter load 311.5 kN
Cutterhead
- Normal operating cutterhead thrust 12,756.5 kN
- Periodic maximum cutterhead thrust 13,667 kN
- Cutterhead drive Electric motors
- Cutterhead power 2205 kW
- Cutterhead speed 0.0 – 9.3 rpm
- Cutterhead torque 2,706 kNm
Main thrust cylinders
- Stroke 1.83 m
- Number of main thrust cylinders 4
Hydraulic System – Power
- Maximum system pressure 365 bar
Electrical system
- Motor circuit 690 VAC
- Primary voltage 15 kV
- Secondary voltage 400 V
Machine conveyor
- Width 914.4 mm
TBM weight 250 Ton (approximately)
Table 3.3 specifications of Gea, after Robbins Company, 2005.
34
The selection of this type of machine is due both to global good quality of rock masses and to the
necessity to have a direct access to the tunnel walls and make direct geological surveys possible.
On a rotary cutterhead 43 disc cutter (17”) are mounted, of which 33 are single cutter, 4 are double
ones and 2 are used for overcutting (Figure 3.11); the cutterhead area is covered by a five meters
long roof shield, behind which support and reinforcement installation is performed (where
necessary). Rotary power is supplied by seven three-phase electric motors, they are cooled by water
system.
Figure 3.11 TBM Chutterhead. On the left cutterhead section (Robbins Company, 2005), on the
right a photo of cutterhead taken during its assembly.
The cutterhead thrust is furnished by four main thrust cylinders, each with a stroke of 1.83 meters.
When competent ground conditions occur (as in the case of La Maddalena) cutterhead thrust is
transferred to the tunnel walls through hydraulically extended gripper shoes (Figure 3.12).
The four main thrust cylinders are linked between the gripper shoes and the main beam.
35
Figure 3.12 Gripper and thrust cylinders. In the top left hand corner, gripper section (Robbins
Company, 2005); in the bottom left corner main thrust cylinders, on the right cylinders and gripper.
The TBM is equipped with front and rear laser targets in order to monitor the machine position.
A conveyor belt is used to remove rock cuttings from cutterhead area to the rear of the TBM then
the excavation materials are moved out of the tunnel by means of other conveyors and stocked in
the proper site (Figure 3.13).
36
Figure 3.13 Rock cutting moving. On the left, conveyor belt; on the right, rock cutting provisory
storage site.
Finally, the machine area that includes bolting rigs, rib erector, the rib that allow pre-drilling ahead
the cutterhead (so ahead the tunnel face) and ground treatments is named L1 zone and it is
immediately behind the cutterhead (Figure 3.14).
Figure 3.14 Scheme of Gea L1 zone (Robbin’s Company, 2005).
In this area a vibrating screen is present (Figure 3.15), with the purpose to select rock cuttings in
order not to damage the conveyor belt; in this way the largest pieces of rock can be removed before
to be charged on the conveyor.
37
Figure 3.15 Vibrating screen, on the letf the original version, on the right the modified one.
The area that begins 40 meters away the cutterhead and includes the ribs for shotcrete application is
named L2 (Figure 3.16).
Figure 3.16 Scheme of Gea L2 zone (Robbin’s Company, 2005).
Back – up includes air tanks, electric components, transformers, ventilation assy, evaporators,
booster fans, compressors, rescue chamber, scrubber and scrubber fans, water reels and water tanks.
38
4 GEOLOGICAL SETTING
4.1 Introduction
In this chapter, a general description of the geological contest in which La Maddalena exploratory
tunnel was excavated has been provided. Furthermore, the geological processes that do not directly
affect the tunnel are briefly cited for the sake of comprehensiveness.
For obvious reasons, the geological units excavated during La Maddalena construction and the
events responsible to their formation have been described with more accuracy.
La Maddalena exploratory tunnel has been excavated in the Ambin Massif, which is part of the
Gran San Bernardo Nappe that structurally pertains to the middle Pennidic domain (Figure 4.1).
Figure 4.1 Stucture of Alpine belt. The red arrow indicates the localization of La Maddalena tunnel
(after Spalla et al. 2014).
La Maddalena is located in the central part of the Western Alps named Cottian Alps.
39
4.2 Penninic domain
4.2 Penninic domain
The Penninic domain is one of the four sectors in which the Alps are divided: the Southalpine,
Austoalpine, Penninic and Helvetic domains. The Penninic domain consists of a multi-nappe
system, in fact it is divided into units that preserve continental crust material (Piedmontese zone and
Calc-schist zone) and into a system of three basement nappes; they are (Polino et al., 2002):
- The Upper Penninic nappes: Monte Rosa, Gran Paradiso and Dora Maira nappes;
- The Middle Penninic (Briançonnais) nappes: it consists of Grand Saint Bernard nappe;
- The Lower Penninic (Valais) nappes: Antigorio, Monte Leone and Lebendum,
4.2.1 Grand Saint Bernard nappe
The Grand Saint Bernard multi-nappe system is located in the Middle Penninic nappe. It extends
along the external sector of Western Alps from Valais to Briançon and outcrops close to Acceglio
and to ligurian costs.
Grand Saint Bernard nappe is made up of a pre-Triassic basement and a Triassic succession of
carbonates. The basement consists of carbonaceous schists associated with volcanic materials. It
outcrops in the Acceglio zone, in the Ambin Massif zone, near to Sapey and in the zones of
Southern Vanoise (Chasseforet Massif) and Northern Vanoise (Mont Pourri-Bellecote) and in Aosta
Valley.
4.2.1.1 Pre-Mesozoic basement
The pre-Mesozoic basement is divided into two principal complexes: one is considered
polymetamorphic and one monometamorphic (Bocquet et al., 1974). The first complex is made up
by paraschists and acid and basic magmatic rocks. The monometamorphic complex consists of
successions based on polymetamorphic one; the contact between the two complexes is considered a
primary type contact (Guillot and Raoult, 1984; Gay, 1970).
4.2.1.2 Meso-Cenozoic covers
The Meso-Cenozoic covers confirm the evolution of a subsiding basin under extensional regime
(Stampfli and Marthaler, 1990). At the base of succession there are quartzites, covered by a
succession of carbonate platform and evaporitic and lagoon deposits.
40
4.2 Penninic domain
During Middle and Upper Jurassic the collapse of the carbonate platform and the subsequent
pelagic deposition characterised by large volumes of breccias as result of syn-sedimentary tectonic
activity began.
4.2.1.3 Ambin Massif
The Ambin Massif emerges at the border between Italy and France, and forms a large antiform in
the Cottian Alps. The Massif outcrops in correspondence of a wide axial culmination under more
tectonics elements belonging to Piedmontese nappe.
Argand et al. (1911) related the Massif to the upper Pennidic nappe of Dora Maira on the basis of
the lithological evidences; afterwards, Staub (1937) and Ellenberger (1958), asserted that the Ambin
Massif belongs to Briançonnais zone, according to the Mesozoic cover characteristic. By taking into
account for the lithostratigraphic affinities, the Ambin Massif is jointed to Pontis nappe (Thelin et
al., 1990) and with Chasseforet Massif (Desmons and Mercier, 1993).
41
4.3 Tectono-stratigraphic unit of Ambin
The tectonostratigraphic unit of Ambin represents the lowest unit of Gran Saint Bernardo nappe; it
is formed by a pre-Triassic crystalline basement on which limited Mesozoic metasediments are
rested. The Ambin unit outcrops between Chiomonte and Oulx municipalities, on the left bank of
the Dora Riparia Valley.
The basement is subdivided into two complexes:
- A base complex known as Clarea complex characterised by polymetamorphic schists with

relicts of pre-Alpine metamorphism;
- An upper complex, named Ambin complex, which is mainly composed of Permian

metavolcanic rocks.
The dividing horizon between the two units is in the order of tens of meters in thickness and it is
made up of metaconglomerates (Carboniferous-Permian) with quartz pebbles principally, passing to
metaconglomeratic quartzites. The origin of this contact can be considered of stratigraphic type, in
this case the metaconglomerate level should be considered as the relict of the Late-Varisic
transgression in discordance on the metamorphic basement (Goguel, 1958; Gay, 1970). The
previous interpretation agrees with the monometamorphic evolution of Ambin Massif (Kretz,
1983).
The Mesozoic cover of the Ambin Massif is poorly conserved on its South side and rests in
discordance with the pre-Triassic basement; it consists of a carbonate sequence 20 m thick (Polino
et al., 2002).
The structural scheme of the Ambin Massif area has been reported in Figure 4.2.
42
STRUCTURAL SCHEME Scale 1:200.000
Slope movements
Stratigraphic contact
Thrust
Tectonic contact
Fault
Figure 4.2 Regional structural scheme. 1: Puys complex, 2: Venaus complex, 3: Aigle unit, 4: Vin
Vert unit, 5: Cerogne-Ciantiplagna unit, 6: Lago Nero unit, 7: Albergian unit, 8: Re Magi unit, 9:
Chaberton unit, 10: Valfredda unit, 11: Gad unit, 12: Vallonetto unit, 13: Mesozoic cover (Ambin
unit), 14: Ambin complex (Ambin unit), 15: Clarea complex (Ambin unit), 16: Gypsum, red line: La
Maddalena tunnel. (Geological Map of Italy, Bardonecchia; Polino et al., 2002).
43
4.3.1 Clarea complex
The Clarea complex represents the lowest element belonging to the metamorphic basement of the
Ambin Massif. It emerges in Clarea Valley, Tiraculo Valley, in Rio Pontè and Rio Geronda areas,
in Susa Valley and in Gran Boursey zone. The minimum thickness of the outcrops is about 700-800
metres (Polino et al., 2002).
The Clarea complex is formed by metasediments represented by michaschists, and gneiss minuti.
Some pre-Alpine relicts are present in the lower part of the complex; in these relicts some evidences
of pre-Alpine foliation have been preserved. The Alpine tectonometamorphic deformation becomes
more significant in the upper levels of the complex and it has deleted every trace of pre-Alpine
deformation.
The pre-Alpine foliation, typical of the lowest levels, is considered Cambrian by Desmons and
Fabre (1998) and Desmons (1992). Moniè (1990) conducted radiometric analysis Ar/Ar which
suggested a Varisian Period (310-350 My) and so it is possible to consider Permo-Carboniferous
the monometamorphic complex of Ambin Massif.
The Clarea complex has been subdivided into the following informal units:
4.3.1.1 Micaschist and gneiss minuti reequilibrated in blueschist facies
They emerge in the upper levels of the Clarea complex, close to the contact with the Ambin
complex. The principal forming minerals are quartz, albite, white mica, chlorite, amphibole,
chloritoid and garnet. The schistosity is developed in HP-LT conditions and it is defined by the
orientation of white mica, chlorite and glaucophane.
Levels of quartz are located parallel to the schistosity. The decompression event at LP-LT causes
the mica substitution with albite. This process determines the micaschist transformation into gneiss
minuti albitizated.
4.3.1.2 Micaschist with pre-Alpine parageneses
The micaschist with pre-Alpine parageneses represents the most common unit in the lowest sectors
of Clarea complex. Where the quartz content is relevant rocks, are quite massive and where white
mica content is high the schistosity is dominant.
The principal forming minerals are quartz, muscovite, plagioclase, biotite, chlorite and garnet. The
foliation is defined by the layering of muscovite and biotite levels and quartz-feldspar ones.
44
4.3.1.3 Metabasites with pre-Alpine relicts
This unit consists of dark green metabasites characterised by metamorphic layering defined by the
alternation of amphibolitic domains (in the order of centimetres in thickness) and albite and epidote
ones (in the order of millimetres in thickness).
The unit is characterised by variable thickness, in the order of metres and tens of metres. In some
cases an Alpine foliation has been recognised, due to the HP event.
4.3.1.4 Polymetamorphic ortogneiss
The polymetamorphic ortogneiss unit emerges in the area of Monache Lakes, Clot delle Selle, Rio
Geronda and Rio Clapier (Polino et al., 2002). If Alpine parageneses is preserved, rocks are whitish,
medium fine grained, rich in quartz, white mica and feldspar. In some cases, the original magmatic
fabric is recognisable.
The ortogneiss is characterised by a foliation defined by muscovite and biotite, subsequently

substituted by Alpine aggregates formed by phengite + glaucophane + chlorite.
45
4.3.2 Ambin complex
The Ambin complex is principally formed by albitic-cloritic gneiss characterised by textural and
compositional homogeneity and leucocratic gneiss with jadeite (Gay, 1970; Callegari et al., 1980),
considered of magmatic, volcanic and volcanoclastic origins and belonging to the same magmatic
complex. They are considered as the metamorphic result of effusive or sub-intrusive magmatic
bodies.
There are also levels of quartz-micaschist with chlorite and quartz-micaschist. The
metaconglomerate and the carbonate levels are rarer.
The thickness of the units composing the Ambin complex is extremely variable.
The Ambin complex has been subdivided into the following informal units:
4.3.2.1 Metaconglomerates and conglomeratic quartzites
These whitish rocks are located at the basis of Ambin Massif and they have a thickness of tens of
metres. Both metaconglomerates and conglomeratic quartzites contain clasts of quartz and lithic
fragments (made up by quartz and gneiss) oriented in parallel with respect to the regional foliation.
In some cases, this unit is altered and the global colour can become brownish.
4.3.2.2 Quartz-micaschists, quartzites and marbles
This rock emerges between Chiomonte and Salbertrand municipalities and in Vaccarone area. The
unit is formed by quartz-micaschist, passing to quartzites, the quartzite levels are less of 10 metres
in thickness. Into these levels there are levels (with the same thickness) of micaschist with
glucophane, mica, chlorite and carbonate (Polino et al., 2002). Sometimes the carbonate level is as
relevant as make the rock considered as a calc-schist.
Some levels of micaceous marble have been observed in Ruinas-Maison area.
4.3.2.3 Leucocratic gneiss with jadeite
This gneiss is very compact, whitish and fine grained; it largely emerges in Clarea Valley. These
rocks are homogenous and consist of quartz, albite, white mica and plagioclase. Jadeite (previously
described by Gay in 1972) is very fine grained and often associated with the quartz. The rock is
poorly foliated, due to the mica orientation.
46
Several interpretations of their origin were proposed (Gay, 1970-1972; Callegari et al., 1980); their
composition is alkaline-granitic (Pognante et al 1984).
4.3.2.4 Augen gneiss with albite and chlorite
This is the prevalent lithotype occurring in the Ambin complex. It consists of light green gneiss
formed by quartz, albite, white mica chlorite and relicts of chloritoid.
These rocks are characterised by well-marked foliation, defined by chlorite and white mica. The
texture is characterised by the occurrence of granoblastic aggregates (measurable in tens of
millimetres) made up by albite, quartz, chlorite (large sized) and magnetite. Desmond and Mercier
(1993) considered them as metamorphic result of clastic sedimentary rocks (greywacke), however
at these days they are interpreted as metamorphosed volcanic/volcanoclastic rocks.
4.3.2.5 Quartz-micaschist with chlorite
These rocks outcrop in Gambara Valley, Punta Sommeiller and Vaccarone-Gros Muttet-Col
Clapier. They have medium grain size, they are whitish and very homogeneous. They are affected
by marched foliation and the principal constitutive minerals are quartz, white mica, chlorite,
chloritoid and carbonates (rarely).
Rare boudins of glaucophane schist are detected surveying this lithotype.
4.3.2.6 Metadiorites with magmatic relicts
Tabular corps of metadiorites and metagabbros have been individuated in the Ambin complex and
in Clarea one (Pognante et al., 1984). They have been observed in Susa Valley between Exilles and
La Ramat. These rocks are dark green, massive and medium-fine grained.
This unit is poorly foliated; the foliation is defined by the iso-orientation of amphibole, chlorite and
titanite.
Polino et al. (2002) interpreted them as dike, on the basis of the relationship between these rocks
and the hosting ones.
47
4.4 Structural evolution of Ambin
The structural evolution of Western Alps is characterised by a complex polyphase deformation

history. Each unit is affected by an initial independent evolution, and then the deformation
processes become common to all the units (Polino et al., 1999).
In particular the polyphase evolution of the Ambin unit can be summarised as follows (Lorenzoni,
1965; Gay, 1970; Callegari at al., 1980; Allenbach and Caron, 1986; Borghi and Gattiglio, 1997):
- Pre-Alpine structural evolution;
- Development of the first two phases of Alpine deformation;
- Third phase of Alpine deformation;
- Late phases of massif uplift.
4.4.1 Pre-Alpine evolution
The first foliation consists of microstructural relicts as microlithons and sigmoidal tectonic surfaces.
The successive deformation phase is defined by amphibolite facies minerals, related to an axial
foliation. This defines the schistosity that characterises the lowest structural levels of Ambin unit
(Polino et al., 2002).
4.4.2 Ductile Alpine evolution
The first Alpine deformation developed in blueschist condition. Its intensity is higher in the upper
levels of Clarea complex and it deforms the pre-Alpine schistosity, generating folds North-South
oriented, in the lowest levels.
The second phase has created isoclinal folds and an extensional lineation oriented East-West. The
axial plane foliation is the prevalent one; it was evolved in high pressure-low temperature
conditions (lawsonite blueschists).
The third Alpine deformation phase gives as result the principal planar elements about the upper
levels of Ambin Massif. The axes of the folds developed during the third deformation phase have
orientation North-South. Some linear structures are a consequence of this phase, for instance the
glaucophane (East-West oriented).
The fourth phase is characterised by low grade metamorphism and it is typical of exhumation
events.
48
4.4.3 Brittle Alpine evolution
The brittle Alpine deformation consists of a series of mechanical discontinuities occurring both at
small and large scale. In particular, three main systems of fractures have been distinguished:
- N60 system: it is characterised by planes immerging NW-SE and evidences of sinistral

strike-slip;
- N100/N140 system: it is affected by distensional movements;
- A system ranging between N160/N10 and affected by distensional movements.
One of the most relevant fault systems is the so called “Modane-Termignon-Ruisseau de la

Chavière”, described by Fudral (1998) which takes place along the Arc Valley at the NW border of
the Ambin Massif.
This is delimited by regional fault systems: the Arc Valley Fault, the system with direction ranging
from N120 to N180 (Susa-Condove area) and the Susa Valley deformation zone (N60), the last one
is very persistent in the Ambin Massif (Fudral, 1998).
49
4.4.4 Neotectonics
In the area in which the Ambin Massif takes place, stress field is considered extensional type and it
is characterised by two extensional processes.
The first phase is described as a lateral extrusion of the inner Western Alpine Arc, the second event
is caused by body forces in the belt root and gravitational re-equilibration (Sue et al., 2007).
The neotectonic map of the Western and Central Alps carried out by Sue et al. (2007) is provided in
Figure 4.3.
Figure 4.3 Neotectonic map of Western and Central Alps. Figure B orange lines: faults observed,
blue: seismic alignments, black: possible and probable ones (Sue et al., 2007).
The neotectonic scheme about the area studied during this research work is provided in Figure 4.4.
50
NEOTECTONIC SCHEME
Figure 4.4 Neotectonic map of Ambin Massif. 1: Ambin Massif and cover, 2: Piedmontese Domain,
3: lower tectonostratigraphic units, 4: upper tectonostratigraphic units, 5: gypsum and anhydrite,
6-7: Quaternary, 8: deep seated gravitational slope deformation, 9: stratigraphic limits, 10: thrust,
11: tectonic contacts, 12: strike-slip faults, 13: Susa-Chisone shear zone, 14: subsidence and uplift,
red line: La Maddalena tunnel (Geological Map of Italy, Bardonecchia area; Polino et al., 2002).
The area in which La Maddalena tunnel has been excavated is affected by three main fault systems:
the “Mompantero-Colle delle Finestre” the “Medium Susa Valley-Upper Chisone Valley” and
“Sangone” shear zones.
4.4.4.1 The “Mompantero-Colle delle Finestre” fault system
The “Mompantero-Colle delle Finestre” fault system is a relevant system of faults with mean
direction NNE-SSW, the structure is probably related to “Cenischia-Nizza line”, as described by
Casati and Giovacchini (1977). Mompantero-Colle delle Finestre system is characterise by a mean
direction of N20 and it is observable in the upper sector of Susa Valley and Mourienne, where there
are N-S faults with reject evaluated in the order of kilometres. These faults were dextral strike-slip
during Oligocene and Miocene and then they have been reactivated as normal faults (Polino et al.,
2002; Malusà 2004). This system corresponds to the normal fault system of Modane-Termignon-
Ruisseau de Chaviere and at Bessans-Pont Andagn Fault (Fudral et al., 1998).
51
4.4.4.2 “Medium Susa Valley – Upper Chisone Valley”
The “Medium Susa Valley-Upper Chisone Valley” fault system develops in Medium Susa Valley
and Upper Chisone Valley and has a mean direction of N60. It is characterized by kilometres length
structures that displace the preexisting structural and stratigraphic disposition (Polino et al., 2002;
Giardino and Polino, 1997). The activity of this system was dextral strike-slip during Oligocene-
Miocene and then evolved in normal activity (Malusà, 2004).
Both Col Clapier and Venaus faults are related to this system.
The stereographic projections of the principal neotectonic deformation elements registered in

Medium Susa Valley (Figure 4.5 A) and in Segurét-Vallonetto sector (Figure 4.5 B) have been
reported below.
Figure 4.5 A) Deformation elements observed in Medium Susa Valley. 1: normal faults, 2: strike-
slip faults, 3: random faults, 4: deep seated gravitational slope deformation. B) Deformation
elements observed in Segurét-Vallonetto sector. 1: normal faults, 2: strike-slip faults, 3: random
faults, 4: axes of collapse structures (Geological Map of Italy, Bardonecchia area; Polino et al.,
2002).
52
4.4.4.3 “Sangone” shear zone
The fault system named “Sangone” shear zone is defined by a mean direction of N110-N120. It has
been observed between Oulx and Bardonecchia and it is characterised by extensional movements
(Malusà, 2004). In Sangone Valley, the kinematic indicators detected show normal and sinistral
strike-slip movements.
The maximum extension of this shear zone is about 4 kilometres in length.
53
5 ROCK MASS CHARACTERIZATION
5.1 Introduction
The main purpose of this research work is to carry out the relationships between rock mass features
and TBM performances: for this reason, a detailed rock mass characterisation plays a fundamental
role.
The rock mass characterisation focuses both on intact rock properties determination and on the
mapping and the description of discontinuities.
The first phase of this work consisted into a detailed tunnel wall mapping; in fact, several
geological surveys have been carried out continuously along the entire tunnel length. During each
survey, the geometry and properties of discontinuities and matrix features have been detected. Their
length is equal to the tunnel length excavated the day before of the geological mapping, in this way,
each geological survey generates a “rock mass section”, their mean length ranges between 10 and
20 meters. The decision to make the rock mass section long as a daily excavated tract is due to the
fact that in this way we are able to evaluate the TBM indices for each of them.
The scheme used to perform the tunnel wall mapping is reported in Appendix I.
At the end of the field work, a total of 502 rock mass sections has been surveyed.
For each rock mass section, a series of geological, structural, mechanical and TBM data is
available. The types of data about intact rock, discontinuities, and machine available for each rock
section have been listed in Table 5.1.
For each rock mass section, the geological survey allows to classify rock mass using Rock Mass
Rating (RMR; Bieniawski, 1989) system and Geological Strength Index (GSI; Hoek and Marinos,
2000).
The results of surveys have been provided referring both to the entire tunnel length and to each
encountered geological domain. The geological domains are:
- Gneiss domain, (DI): ranging between Pk. 0 + 198 m and Pk. 0 + 967 m;
- Gneiss-micaschist alternation domain, (DII): ranging from Pk. 0 + 967 m to Pk. 1 + 360 m;
- Micaschist domain, (DIII): ranging between Pk. 1 + 360 m and Pk. 6 + 897 m.
55
5.1 Introduction
All data collected in the Domain I (DI, gneiss), in Domain II (DII, alternation of gneiss and
micaschist) and part of data, from Pk. 1 + 360 m to Pk. 2 + 003 m, collected in Domain III (DIII,
micaschist) have been used in the subsequently regression analysis. All data surveyed in the
remaining part of DIII (from Pk. 2 + 003 m to 6 + 897 m) have been used to validate the obtained
models. For this reason, the description of each geological and geomechanical parameter about the
first tract of DIII is also provided. This part of DIII is identified as DIIIa.
List of parameters available for each rock section
Intact rock Discontinuity TBM
- Lithological - Type of discontinuity; - ROP;

description; - Orientation;
- FPI;
- UCS; - Spacing;
- Quartz content*; - Number of sets; - Prev;
- Thin sections; - Block size;
- SP;
- CAI*; - Persistence;
- CHI*; - Waviness; - Torque;
- Poisson ratio*; - Angle between
- Cutterhead thrust.
- Young modulus*; discontinuity and tunnel
- Density*; axis;
- Seismic wave velocity. - Degree of weathering;
- Aperture;
- Type of infilling
* data measured each material;
200 m - Groundwater condition.
Tunnel data: Tunnel depth, Tunnel axis orientation, Seismic wave velocity.
Table 5.1 List of parameters available for each rock section.
In the following paragraphs a brief description of the methods followed in order to describe the
parameters listed above is provided.
56
5.2 Intact rock property determination

The purpose of laboratory tests on intact rock or rock sample macroscopic and microscopic
description is to organise, from a geological point of view, the rocks in different classes which
represent the mechanical and hydraulic behaviour, and to assess the deformation and strength
features.
A brief description of the basic observation or laboratory measurements that have been conducted in
order to determine the physical and mechanical properties of the rock matrix excavated during La
Maddalena construction has been provided in this section.
The results of laboratory tests have been provided by construction company.
5.2.1 Lithological description
General lithological description of rock has been carried out for each rock section; the description is
based on tunnel walls and samples macroscopic examination.
Particular attention is given to:
- Colour;
- Petrographic composition;
- Texture;
- Grain size and shape;
- Presence of planes of foliation (schistosity), layering, lining, orientation of minerals;
- Presence of folds;
- Intact rock weathering degree.
The macroscopic observations listed above allow to define the lithological type of rock.
57
5.2.2 Micro-petrographic description
The micro-petrographic description of rocks has the purpose to determine the mineral content, to
determinate the presence of microfractures, the degree of weathering, and the grain size and shape.
The grain contacts, cement or matrix have also to be analysed. Generally micro-petrographic
description has the purpose to focus on all parameters which cannot be carried out from
macroscopic observation of a sample of rock.
In order to determine the volumetric proportion of the minerals that make up the excavated
geological units quantitative modal analysis has been used, determined by areal analysis of thin
section.
Thin section studying allows the individuation of dangerous minerals, as fibrous ones, and to
evaluate the content of abrasive mineral, as quartz.
Figure 5.1 shows an example of thin section observed at parallel and crossed Nicols.
Figure 5.1 Thin section in parallel (left) and crossed (right) Nicols.
Thin section examination for rock engineering purpose should be carried out following the
Suggested Methods proposed by ISRM (1978).
58
Results
Quantitative modal analysis by areal analysis of thin section have been carried out during La
Maddalena tunnel excavation in order to determine the volumetric proportion of minerals and the
occurrence of dangerous one.
Here the results of 36 thin section analyses are presented, with particular attention to the quartz
content, considered largely affecting the TBM performance (Table 5.2).
QUARTZ TUNNEL Domain I Domain II Domain III

CONTENT [%] (n = 36) (n = 7) (n = 4) (n = 25)
Range 15.00 – 80.00 37.00 – 80.00 15.00 – 50.00 20.00 – 35.00
Median 30.00 45.00 30.00 25.00
Mean 32.19 49.86 31.25 27.40
Q1-Q3 25.00 – 35.00 40.00 – 52.00 18.75 – 45.00 25.00 – 30.00
Table 5.2 Quartz content measured by areal analysis of thin sections.
The highest content of quartz has been detected into DI, formed by gneiss. In this domain, a portion
with high occurrence of quartz veins has been mapped and in this area a band characterised by 80%
of quartz. DI quartz content is 45.0-50.0%, the others constituting minerals are alkaline feldspars,
muscovite and plagioclase. Quartz content in DII is lower than in DI (30.0%), however the lowest
values have been noted in the third domain (25%). The other minerals forming the rocks of DIII are
white mica, chlorite, and albite (and other minerals in lower percentage).
About DIII, there are not significantly differences between the data collected in DIIIa.
Figure 5.2 shows the quartz content distribution through the tunnel excavation.
59
QUARTZ DISTRIBUTION
100
DI DII DIIIa DIII
80
60
q[%]
40
20
0
0 1000 2000 3000 4000 5000 6000
Pk [m]
Figure 5.2 Quartz content distribution. Red lines represent the domain changes, green line delimits
DIIIa, black ones represent the mean value for each domain.
The quartz content of excavated rocks has been determined also by means of macroscopical
observation of rock samples, during tunnel excavation. This allowed to assign a quartz content
value to each rock mass section detected.
60
5.2.3 Uniaxial compressive strength
Uniaxial compressive strength is one of the most required rock parameters; it can be measured by
means of uniaxial compression test (ISRM, 1979) which is one of the most common laboratory
tests, or by means of the point load test (ISRM, 1985).
5.2.3.1 Uniaxial compression test
This test has the purpose to measure the uniaxial compressive strength of a rock sample and, if
sample is opportunely equipped, this test is able to determine stress-strain curves and Young’s
modulus and Poisson’s ratio in uniaxial loading conditions. The test is mainly intended for strength
and deformability classification and characterisation of intact rock materials.
The diameter of specimen has to be at least 20 times larger than the maximum grain size. The faces
of sample should be perfectly plane parallel and perpendicular to the specimen axis, in particular the
lateral one must be smoothed and regular. The sample has to be measured and the basis area has to
be computed. The number of specimens to be tested has to be representative to take the variability
of the rock strength into account. It is recommended to test 5 specimens for each rock type.
The sample of rock is put between two platens and the axial load is incremented, two loading
procedures can be followed, the first is constant load increment and the second is constant axial
deformation increment. The radial and axial strains are measured by dial gauge.
The rock specimen can be instrumented with electrical gauges for the deformation measurements
and with a control unit to gather the results and, when the interpretation is not manual, with a
computer to interpret the obtained data.
The measurements of axial and radial strains allow to determine the rock pseudo-elastic constants.
The axial force registered during the test is divided by the initial cross-section of the specimen to
get the average axial stress (σa). This latter is plotted versus the overall axial strain (εa) and versus
radial strain (εr).
The stress strain curve shows four stages:
- Crack closure and initial bedding down;
- Elastic deformation until uniaxial stress σci is reached: stable crack propagation begins;
- Crack propagation continues until the axial stress reached σcd: irrecoverable deformations
begin;
61
- Irrecoverable deformations increase until the peak and UCS, σc, is reached.
By analysing stress-strain curve deformation moduli, Young’s modulus and Poisson’s ratio, can be
obtained (Figure 5.3). Since Young’s modulus varies throughout the loading history, stress – strain
curve is not linear so three Young’s moduli have to be determined:
- Tangent Young’s modulus (Et) is the slope of the axial stress-axial strain curve at 50% of the
peak strength;
- Average Young’s modulus (Eav) is the average slope of the straight line portion of the axial
stress-axial strain curve;
- Secant Young’s modulus (Es) is the slope of a straight line linking the origin of the axial
stress-axial strain curve to the point on the curve at 50% of the peak strength.
Poisson’s ratio is defined as:
(∆𝜀𝑟 )
𝜐=− (5.1)
(∆𝜀𝑎 )
The volume variation can be calculated in accordance with the following equation:
∆𝑉
𝜀𝑣 = (5.2)
𝑉
Figure 5.3 Elaboration of strain – stress curve (Brady and Brown, 2004).
Tests have to be carried out following the ISRM suggested methods (1979).
62
Results
During La Maddalena construction uniaxial compressive strength tests have been carried out by
external testing laboratories.
These tests allow the estimation of the uniaxial compressive strength and deformation moduli,
Young’s modulus and Poisson’s ratio. In this work the results of tests on rocks excavated between
Pk. 0 + 198 m and 5 + 700 m have been analysed.
The UCS results are listed in Table 5.3: they are proposed both about the entire tunnel tract and for
each geological domain (ad exception of DII where no tests have been carried out).
TUNNEL Domain I Domain III

UCS [MPa]
(n = 36) (n = 13) (n = 23)
Range 41.20 – 235.98 140.30 – 235.98 41.20 – 232.15
Median 125.76 198.40 100.26
Mean 135.68 190.96 104.43
Q1-Q3 81.73 – 187.190 166.30 – 210.00 71.08 – 124.75
Table 5.3 UCS test results.
Globally, the rocks excavated during La Maddalena realisation could be classified as “high strength
rocks” (Deere and Mille, 1966, classification).
It possible to note that, a significant reduction of UCS values occurs from DI to DIII. DI gneiss can
be considered “high-very high strength rocks” (Deere and Mille, 1966, classification) and
micaschist (DIII) are characterised by “moderately hard rocks”.
Figure 5.4 clearly shows the reduction of UCS in the direction of tunnel excavation.
63
UCS trend through the tunnel

250
200
UCS[MPa]
150
100
50
DI DII DIIIa DIII
0
0 1000 2000 3000 4000 5000 6000
Pk [m]
Figure 5.4 UCS distribution along the tunnel. Red lines represent the domain changes; green line
delimits DIIIa; black horizontal lines represent the mean values in DI and DIII.
In addition, deformation moduli (Et, Es, υt, υs) have been estimated and their average values have
been listed in Table 5.4.
No values have been reported for DII because of no tests have been carried out in this sector.
TUNNEL Domain I Domain III
(mean) (mean) (mean)
Et [GPa] 53.50 57.85 52.36
Es [GPa] 45.35 38.35 47.18
υt [-] 0.19 0.18 0.19
υs [-] 0.13 0.14 0.13
Table 5.4 Mean values of deformation moduli.
64
5.2.3.2 Point load test
Point load test is carried out by applying two punctual loads (loading is manually applied by two
standardized truncated-conic points) to a rock specimen and breaking it. The test allows the indirect
estimation of uniaxial compressive strength. Point load test implies great advantages since it let us
determine UCS with a portable machine, directly on site and on irregular specimens.
The test allows the determination of a strength index IS, which is related to the uniaxial compression
strength. IS is calculated in accordance with:
𝑃
𝐼𝑆 = (5.3)
𝐷2
Where P is the failure loading [kN], it is read on a pressure gauge incorporated on the machine and
D [mm] is the point distance, it is measured by using a graduated pole before and after the failure
loading application.
IS is expressed in MPa.
The specimen shape can be cylindrical or have other shapes and the direction of loading, for
cylindrical core, can be axial or diametral (Figure 5.5).
Specimen shape requirements (ISRM SM 1985)

a) Diametral test:
L> 0.7 D
b) Axial test:
D/L = 1.1+0.05
c) Irregular sample:
D/L = 1.0 / 1.4
Figure 5.5 Specimen shape requirements for a) diametral; b) axial, c) irregular test.
IS varies as function of D, so that a size correction has to be applied to obtain a unique value of I S
that can be involved for purposes of rock strength calculation; the standard diameter is equal to 50
mm. The index IS(50) can be graphically obtained by using the “size correction chart”.
Point load test has to be carried out on 10 (or more) specimens, then it is necessary to calculate the
mean (or median) value of IS50 to use it as input for UCS estimation. The mean (or median) value of
IS(50) is to be calculate deleting the two lowest and highest values from the 10 (or more) valid tests
and computing the mean or median of the remaining values.
65
When a rock sample is anisotropic it has to be tested parallel and normal to the planes of anisotropy,
these directions give the greatest and least strength values respectively.
IS(50) is empirically related to UCS by the following equation (5.4, Brokch and Franklin, 1972; 5.5,
Bieniawski):
𝜎𝑐 = 23.7 ∗ 𝐼𝑆(50) [𝑀𝑃𝑎] (5.4)
𝜎𝑐 = 25.0 ∗ 𝐼𝑆(50) [𝑀𝑃𝑎] (5.5)
66
Results
For each rock section a datum of UCS is available, it has been obtained by Point load test directly
on site. In order to calculate UCS from IS50 value the empirical correlation proposed by Brokch and
Franklin (Equation 5.4) has been used.
Tests have been carried out following the ISRM suggested methods (1985).
The statistical description of UCS values is provided in Table 5.5.
The description was carried out about the entire excavated length and for each domain.
UCS
TUNNEL Domain I Domain II Domain III
[MPa]
(n = 502) (n = 83) (n = 44) (n = 375); *DIIIa
(plt)
Range 60.00 – 255.00 68.00 –219.00 70.00 – 251.00 74.00 – 224.0* 60.00 – 255.00
Median 150.00 143.00 129.00 105.00* 152.00
Mean 149.10 151.02 129.08 112.70* 151.02
Q1 133.50 121.00 101.25 93.00* 143.00
Q3 168.00 174.00 154.25 127.00* 168.50
Table 5.5 UCS (from point load test) statistical description; * DIIIa data: part of DIII data used in
the following predictive model creation.
The mean value referred to the entire excavated material is about 150.00 MPa, and it allows to
classify rock as “high strength” (Deere and Miller, 1966).
In particular, the intact rock samples of gneiss (DI) and of micaschist (DIII) have the same values of
UCS. The strength decreases in the tract of tunnel characterised by the alternation of gneiss and
micaschist.
By plotting the distribution of UCS values vs. the progressive length (Figure 5.6) it is possible to
assert that there are two sectors in micaschist domain in which UCS has values significantly
different from the average of the entire domain.
Between the Pk. 1 + 360 m and Pk. 2 + 003 m the mean value of UCS is 112.70 MPa, it makes
possible to classify this rock as medium strength.
67
The sector between Pk. 2 + 329 m and Pk. 2 + 756 m is characterised by mean value of UCS equal
to 194.10 MPa, so rock can be classified as high strength.
UCS (point load test)

300
250
200
UCS (pl)
150
100
50
DI DII DIIIa DIII
0
0 1000 2000 3000 4000 5000 6000 7000
Pk. [m]
Figure 5.6 UCS (determined by means of point load test) distribution through the tunnel. Red lines
represent the domain changes, green line delimits DIIIa, black lines the mean value of UCS for
each geological domain, black dotted lines represent two zones in which the average value is
significantly different from the mean value of the domain.
68
5.2.4 Determination of rock abrasivity
Rock abrasivity index are frequently used in TBM tunneling and rock drilling for the estimation of
tool wear; the Cerchar abrasivity test is one of the widely used for this purpose. It was introduced
by Centre d’Etudes et Recherches des Charbonages de France in 1970s (Valantin, 1973).
The test is based on a steel pin with defined hardness and geometry that is scratches the surface of a
rock sample under a constant load equal to 70 N.
A scheme of Cerchar testing device is provided in Figure 5.7. Cerchar Abrasivity Index is given by
the ratio between the diameter of the wear flat, d [mm] and the unit correction factor, c = 1 mm
(Figure 5.8); in accordance with the following formulation:
𝑑
𝐶𝐴𝐼 = [−] (5.6)
𝑐
Figure 5.7 Cerchar testing device. 1: weight, 2: pin chuck, 3: steel pin, 4: rock sample, 5: vice, 6:
hand level (Alber et al., 2013; Käslin and Thuro, 2010).
Only worn surface measurements obtained from standard profiles should be used, for example in
Figure 5.8, a and b profiles represent sandard situation, nonstandard worn profile is shown in c, in
this case no measurements have to be recorded.
69
Figure 5.8 Pin shape before and after usage (Alber et al., 2013).
A minimum of five test replications must be made on the rock surface, and then the mean value is
considered.
Rock can be classified on the basis of its abrasivity: the classification is proposed in Table 5.6.
Classification of CAI
CAI Classification
0.1 – 0.4 Extremely low
0.5 – 0.9 Very low
1.0 – 1.9 Low
2.0 – 2.9 Medium
3.0 – 3.9 High
4.0 – 4.9 Very high
> 5.0 Extremely high
Table 5.6 CAI classification, Alber et al., 2013.
70
Results
In this work the results of Cerchar Abrasivity tests carried out during La Maddalena construction,
between Pk. 0 + 198 m and 5 + 700 m have been collected.
The results are listed in Table 5.7, they are proposed both about the entire tunnel tract and for each
geological domain.

CAI [-]
(n = 36) (n = 11) (n = 3) (n = 22)
Range 2.19 – 6.10 3.38 – 6.10 3.00 – 3.88 2.19 – 5.50
Median 3.94 5.30 * 3.60
Mean 4.04 4.96 * 3.65
Q1-Q3 3.28 – 4.74 4.13 – 5.70 * 3.19 – 4.21
Table 5.7 Results of Cerchar Abrasivity tests. (* in DII only three tests have been performed, for
this reason only value range has been provided).
In general, the abrasivity of rocks encountered during La Maddalena excavation is classified as

“high”-“very high”. The most abrasive unit is the gneiss, considered “very high” and “extremely
high” abrasive. Generally CAI decreases from DI to DIII, where rock abrasivity is “high”.
In the transition zone (DII) only three tests have been carried out, and the abrasivity is “high” also
in this tract.
The lowest class of abrasivity detected during La Maddalena realization has been obtained into
micaschist unit, and it is the “medium” one.
The distribution of rock abrasivity reflects the cutter consumption and TBM performance
distribution. Where highest classes of abrasivity have been encountered the maximum excavation
tool consumption occurs and the lowest machine performance has been registered.
Figure 5.9 shows the rock abrasivity distribution through tunnel excavation.
71
ABRASIVITY DISTRIBUTION
7
6
5
4
CAI[-]
3
2
1
0 DI DII DIIIa DIII
0 1000 2000 3000 4000 5000 6000
Pk [m]
Figure 5.9 Rock abrasivity distribution. Red lines represent the domain changes, green line delimits
DIIIa and black ones represent the mean values for each domain.
72
5.2.5 Determination of rock hardness
The hardness of rock units encountered during La Maddalena exploratory tunnel excavation was
determined by means of Cerchar Hardness Test.
The test consists of measuring the indentation time of a bit into the rock sample for 10 mm. A
vertical weight of 200 N provides the compressive force to the bit, which has a diameter of 8.0 mm;
the test is performed at 200 rev/min of drilling rate.
The scheme of Cechar Hardness Index is provided in Figure 5.10.
200 N
Figure 5.10 Cerchar Hardness Test device (Modified after Plinninger Geotechnik).
The test allows to estimate the Cerchar Hardness Index, CHI, expressed in seconds.
Rock can be classified on the basis of its harndness: the classification is proposed in Table 5.8 CHI
classification.
73
Classification of CHI
CHI Classification
0.0 – 20.0 Soft
20.0 – 40.0 Medium hard
41.0 – 80.0 Hard
81.0 – 120.0 Very hard
> 120.0 Extremely hard
Table 5.8 CHI classification.
Results
In this work, the results of Cerchar Hardness tests carried out during La Maddalena construction,
between Pk. 0 + 198 m and 5 + 700 m have been collected.
The results are listed in Table 5.9: they are proposed both about the entire tunnel tract and for each
geological domain. It is necessary to note that some tests carried out in DI give CHI higher than
1200 s. In these cases, a value of 1200 s has been considered, however in statistical analysis these
value have been excluded being considered as out values.

CHI [-]
(n = 36) (n = 11*) (n = 3) (n = 22)
Range 32.30 – 514.70 132.30 – 514.70 42.00 – 55.40 32.30 – 314.30
Median 95.00 362.20 ** 90.55
Mean 134.59 342.85 ** 108.38
Q1-Q3 52.50 – 158.70 170.83 – 495.53 ** 52.85 – 126.28
Table 5.9 Results of Cerchar Abrasivity tests. (* in 7 cases CHI is higher than 1200 s, these results
haven’t been considered; ** in DII only three tests have been performed, for this reason only value
range has been provided).
In general, the rocks excavated during La Maddalena realisation are “very”-“extremely hard”. The
hardest formations have been encountered in DI, where eleven tests have been carried out and seven
74
on them give results of CHI > 1200 s, the gneiss of DI are classified as “extremely hard”. Hardness
decreases in the micaschist domain (DIII) where rocks are classified as “very hard”. A very
important difference between gneiss and micaschist hardness has been noted.
Figure 5.11 shows the rock hardness distribution through tunnel excavation.
HARDNESS DISTRIBUTION
600
DI DII DIIIa DIII
500
400
CHI[s]
300
200
100
0
0 1000 2000 3000 4000 5000 6000
Pk [m]
Figure 5.11 Rock hardness distribution. Red lines represent the domain changes, green line
delimits DIIIa and black ones represent the mean values for each domain.
75
5.3 Quantitative description of discontinuities

Discontinuities represent any separation in rock continuum having zero tensile strength.
The intact rocks are subjected to millions of years of thermal, chemical and mechanical processes
and during these actions discontinuities have been generated into rocks by geological and tectonics
events as a result of different stress levels.
Generally, the geometrical and mechanical properties of discontinuities are determined by the
processes by which the discontinuities have been generated, for this reason structural geology
principles have to be involved in order to understand the discontinuity formation (Price and
Cosgrove, 1990).
In engineering contexts discontinuities can govern the strength, deformability, permeability and
global stability of the rock mass both in underground and in surface. For these reasons it is
fundamental to study the geometrical and mechanical features of discontinuities and the way in
which these influences rock mechanics and hence rock engineering projects.
During this work, La Maddalena tunnel walls have been continuously surveyed and, in this
paragraph, a short description of the methods and techniques that have been adopted to characterise
rock mass discontinuities and the results obtained after discontinuity feature analysis has been
provided.
The described discontinuity properties are:
- Discontinuity type;
- Orientation: dip and dip direction of foliation planes, joints, shear planes, veins, faults and
the orientation of these planes of weakness respect to the tunnel orientation (α angle);
- Fracturing degree expressed by means of spacing, joint volumetric number and rock quality
designation index;
- Groundwater circulation;
- Joint conditions: persistence, aperture, filling, roughness, weathering.
76
The main features describing discontinuity into rock masses are listed in Figure 5.12.
Figure 5.12 Rock mass discontinuity properties (Wyllie, 2014).
There are different methods that allow to carry out discontinuity survey, which are the following:
- Scanline: this method is used to survey discontinuities at small and medium scale. A tape is
fixed on the rock wall and starting from an extreme of the tape to the other one all the
intercepted discontinuities have to be surveyed (Figure 5.13). The number of scanlines
depends on the requested detail level; they should be fixed along 3 orthogonal or linearly
independent lines;
77
- Survey windows: all the discontinuities that outcrop inside a pre-determined area (window)
have to be described (Figure 5.14). This method is used to survey discontinuities at small
and medium scale;
Figure 5.13 Scanline discontinuity survey method Figure 5.14 Window discontinuity survey
(Latham et al., 2006). method (Zhang and Einstein, 2000).
- Drilling: continuous core drilling allows the recovering of intact rock cores and the
occurring discontinuities can be analysed; moreover, boreholes can be monitored by TV
camera measuring the position and the orientation of the discontinuities (Figure 5.15). This
method is used too to survey discontinuities at small and medium scale;
- Digital surface model: rock mass discontinuities can be surveyed by means of digital surface
model (DSM) like laser scanning and photogrammetry (Figure 5.16). These methods are
used to survey discontinuities at large scale and when rock masses are not easy accessible,
for example large slopes.
Figure 5.15 TV camera borehole inspection. Figure 5.16 DSM discontinuity survey.
78
Discontinuities should be described following the “Suggested Methods for the quantitative
description of discontinuity in rock masses” proposed by ISRM (1978).
In the research work the rock mass discontinuities have been surveyed adopting the “window
method”. The area (window) where surveys have been carried out consists of La Maddalena tunnel
walls, all joints, faults, veins and other planes of weakness encountered during tunnel excavation
have been surveyed, then described and analysed.
An example of discontinuity mapping performed during the survey is reported in Figure 5.17, the
window of observation coincides with the tunnel wall.
Figure 5.17 Example of discontinuity mapping performed during La Maddalena survey. The initial
and final progressive length, the graphical representation of the discontinuities, their orientation
and persistence are always noted.
79
5.3.1 Types of discontinuity
The term “discontinuity”, related to a rock mass, indicates any separation in the rock continuum and
any mechanical discontinuity in a rock mass having zero (or very low) tensile strength.
It is a collective term, used without any genetic connotation; the principal most appropriate terms to
indicate rock mass discontinuity are:
- Joint: a break into rock masses along which there has been no visible displacements;
- Fault: a fracture, or zone of fractures, where there has been evidence of displacement;
- Schistosity, cleavage, foliation: discontinuities generated by metamorphism;
- Bedding planes, stratification: discontinuities typical of sedimentary rocks;
- Unconformity, limit: discontinuity which occur when there is change of lithology or

geological unit.
The largest part of discontinuities mapped and described during the geological survey carried out in
this work consists of joints however some faults have been encountered.
Foliation is clearly recognisable into micaschist domain; the principal set of joints develops along
foliation planes.
80
5.3.2 Orientation
The orientation of rock mass discontinuities largely affects the conditions of stability relative to a
rock structure, such as a tunnel or a rock slope. The importance of discontinuity orientation has a
primary role when more deformation conditions occur, for example when shear strength is low or
when a sufficient number of joints for sliding is present.
The orientation of discontinuity in space is described by two angles:
- Dip (Ψ): is the maximum inclination of a discontinuity to the horizontal plane (field range
0.0° – 90.0°);
- Dip Direction (α): is angle between the Northern direction and the horizontal trace of the dip
line measured clockwise from the Northern direction (field range 0.0° – 360.0°).
Another angle is involved in discontinuity orientation definition; it is the strike (β). Strike is the
angle between the Northern direction and the trace intersection of the discontinuity plane with a
horizontal reference plane (field range 0.0° – 180.0°).
Graphical representation of terminology used to describe discontinuities plane orientations has been
provided in Figure 5.18.
Figure 5.18 Terminology describing discontinuity plane orientation (Willie, 2014).
Dip/dip direction system is largely used in field and tunnel walls mapping, in plotting stereonets
and in the analysis of orientation data.
81
During field survey orientation data are usually collected by means of a compass and a clinometer,
as recommended by ISRM (1978).
A method widely used to analyse orientation data of discontinuities is the stereographic analysis, in
fact stereonet allows to identify the discontinuity sets and their orientation in space (by determining
the central value and orientation distribution by cluster analysis); moreover stereographic projection
permits to represent and analyse the three-dimensional orientation data in two dimensions.
They consist of a reference sphere where its equatorial plane is horizontal and its orientation is fixed
about to the North. There are two types of stereographic projections, the equal area stereographic
projections and the equal angle ones:
- Equal area projection (Lambert or Schmidt): the areas are maintained, but not the shapes
because of the circumference arc becomes a different curve. It is adopted for the statistical
analysis of the spatial distribution of the discontinuity planes.
- Equal angle projection (Wulf): in which both the angles and the shapes are maintained. It is
recommended for the visualization of the rock blocks isolated by discontinuities and
excavation surface, and to carry out rock block stability analysis.
A discontinuity plane can be represented as its intersection with the lower half part of the reference
sphere, defining a circular arc called “great circle”, or as a point called “pole”. The pole is the
intersection of the line perpendicular to the plane with the surface of the sphere (Figure 5.19).
Figure 5.19 Stereographic representation of plane (Wyllie, 2014).
The same group of discontinuity poles has to be analysed in order to determine the central value of
the joint set in term of orientation.
82
For this purpose, a great number of discontinuity planes needs to be analyzed, so it is necessary to
consider poles and not great circles. In a spherical projection, a surveyed discontinuity generally
belongs to a cloud of poles which are more or less near the central value.
Results
More than 2100 orientation data in terms of dip and dip direction have been collected during “La
Maddalena” tunnel wall mapping then they were analysed with the purpose to define the main
orientation of each set of joints.
A stereographic projection of all collected orientation data is reported in Figure 5.20.
Figure 5.20 Stereographic projection of discontinuity orientation data.

Five sets of fractures have been recognised, their mean orientations do not significantly change
through the portion of rock mass investigated during tunnel excavation.
The principal set, named K1, has a mean orientation of 128/47, in terms of dip direction/dip. K1
develops on foliation planes.
The other four sets are less frequent than K1 and their orientation characteristics have been listed in
the following table (Table 5.10).
83
Orientation of principal joint sets
Set Dip Direction Dip
K1 128 47
K2 281 56
K3 348 61
K4 321 57
K5 220 69
Table 5.10 Orientation of principal joint set.
The orientation of the planes of foliation has been surveyed. The mean orientation of foliation
planes is about 123/50, however this orientation change in occurrence of folds with axes oriented
325/50 – 60, in terms of trend and plunge. However, the folded zones are limited and foliation
generally dips towards SE with a medium dip angle, ranging between 30° and 60°.
K4 develops on the planes of folded foliation.
The analysis of data orientation has been performed for each domain.
The stereographic projections related to each geological domain are reported in Figure 5.21
(Domain I), Figure 5.22 (Domain II) and Figure 5.23 (Domain III).
84
Figure 5.21 Stereographic projection of discontinuity orientation data collected in Domain I.
Figure 5.22 Stereographic projection of discontinuity orientation data collected in Domain II.
85
Figure 5.23 Stereographic projection of discontinuity orientation data collected in Domain III.
The summary of joint sets orientation is provided in Table 5.11.
Orientation of principal joint sets (Dip Direction/Dip)
Set Domain I Domain II Domain III
K1 114/49 132/49 128/47
K1b 142/68
K2 295/70 286/62
K3 332/67 350/40 350/60
K4 321/55
K5 222/66
Table 5.11 Mean orientation of joint sets, detected in each domain.
86
5.3.2.1 α angle
The angle between tunnel axis and planes of weakness can affects the TBM performance. This
angle, α [°], was introduced by NTNU prediction model and it is calculated by using the following
equation (Bruland, 1998; Equation 5.7):
𝛼 = sin−1( sin 𝛼𝑓 ∗ sin(𝛼𝑡 − 𝛼𝑠 )) (5.7)
where:
- αf is the dip and αs is the strike of the planes of weakness;
- αt is the strike of tunnel axis;
The best tunnelling conditions occur when α is approximately 60° (Bruland et al, 1998; Macias et
al., 2014).
For each rock mass section α angle has been calculated considering the mean orientation of the
principal set of joints and the tunnel axis orientation in that tract of tunnel. For more information
about the tunnel orientation see section 3.2.3. The results of α calculation have been reported in
Table 5.12.

α [deg]
(n = 502) (n = 83) (n = 44) (n = 375); *DIIIa
Range 2.8 – 85.0 18.4 – 85.0 22.5 – 72.0 7.4 – 69.4* 2.8 – 72.6
Median 45.2 48.3 41.6 39.8* 45.0
Mean 44.8 49.8 40.3 40.7* 45.3
Q1-Q3 36.8 – 52.3 42.0 – 56.8 33.8 – 46.0 29.8 – 52.3* 39.4 – 54.4
Table 5.12 α angle Spacing statistical description; * DIIIa data: part of DIII data used in the
following predictive model creation.
The mean value of α referred to the whole tunnel is 44.8°. The highest mean value has been
calculated in DI where the orientation of the principal set of joints taking in consideration for the
tunnel axis about 50.0°. In DII and DIIIa it decreases, and it is equal to 40.0° in both domains.
However, considering the entire DIII the angle increases and it is equal to 45.0°.
87
The trend of alpha angle through the tunnel is showed in Figure 5.24.
ALPHA ANGLE
90
80
DI DII DIIIa DIII
70
60
50
α [°]
40
30
20
10
0
0 1000 2000 3000 4000 5000 6000 7000
Pk. [m]
Figure 5.24 Alpha trend through the tunnel. Red lines represent the domain changes, green line
delimits DIIIa and black horizontal lines the median values for each domain.
88
5.3.3 Spacing between discontinuities
The spacing is the minimum orthogonal distance between two adjacent fractures, measured on a
scanline. It is an indicator of rock mass fracturing degree, moreover spacing affects the volume of
rock blocks, the strength and deformability properties of rock masses and its stability conditions.
The orientation of the tape used to measure the distances (so the scanline) and the one of the
discontinuity should be perpendicular, however, when it is not possible the operator has to apply the
“Terzaghi correction” in accordance to:
𝑆 = 𝑑 ∗ sin 𝛼 (5.8)
where:
- S is the spacing between two adjacent discontinuities;
- d is the distance between two adjacent discontinuities;
- α is the angle between discontinuity and scanline.
Figure 5.25 reports discontinuities perpendicular to the scanline for which correction equation is not
necessary and sets of discontinuity which form an angle different from 90° with the scanline, for
which Equation 5.8 application is necessary.
60°
tape 90°
89
Figure 5.25 Measurement of joint spacing. The red line represents the measuring tape, the green
lines represent a set of discontinuities perpendicular to the measuring tape and brown and blue
lines represent sets not perpendicular to the tape (modified after ISRM, 1978).
Spacing can be measured for each set of discontinuities, and in this case the average or modal value
of each set of joints are considered, or spacing can be referred to all the discontinuities which occur
into the examined rock mass.
The mean spacing can be calculated in according to:
𝐿
𝑥̅ = (5.9)
𝑁
where L is the scanline length and N is the number of discontinuities occurring in L.
It is also possible to consider the frequency of fracturing as:
𝑁 1
𝜆= = (5.10)
𝐿 𝑥̅
So spacing can be calculated as:
1
𝑆= (5.11)
𝜆
Terzaghi correction has to be applied also when spacing frequency is considered, by using the
following formulation:
𝜆𝑟 = 𝜆𝑎 ∗ sin 𝛼 (5.12)
Where λr is the real frequency of fracturing, considering the angle between scanline and joints, λa is
the apparent λ.
Equation 5.12 has to be applied to each set of joints and the frequency of fracturing related to the
entire rock mass has to be calculated as proposed by:
𝜆𝑟 = ∑ 𝜆𝑖 ∗ sin 𝛼 (5.13)
𝑖=1
were:
- λr is the real frequency of fracturing referred to the entire rock mass;
- n is the number of joint sets;
90
- λi is the frequency of fracturing of each discontinuity set.
Finally, the global spacing between discontinuities can be calculated as the inverse of the frequency
of fracturing referred to the entire rock mass:
1
𝑆= (5.14)
𝜆𝑟
Spacing is measured in metres or in centimetres, the frequency of fracturing in m-1.
On the basis of spacing values, ISRM (1978) suggested to adopt the terminology listed in Table
5.13 in order to describe the classes of spacing of each set.
Spacing [cm] Description
< 2.0 Extremely close spacing
2.0 – 6.0 Very close spacing
6.0 – 20.0 Close spacing
20.0 – 60.0 Mocerate spacing
60.0 – 200.0 Wide spacing
200.0 – 600.0 Very wide spacing
> 600.0 Extremely wide spacing
Table 5.13 Spacing description (ISRM, 1978).
91
Results
During La Maddalena discontinuity survey, spacing has been measured by using scanline method.
In this case spacing is referred to all the discontinuities occurring in the rock mass: firstly we
considered the frequency of fracturing, considering as L the daily excavation advance, an apparent λ
was calculated, so it was necessary to correct it by using the Equation 5.12.
The Terzaghi correction has been applied to each discontinuity set and the real frequency of
fracturing was calculated by means of Equation 5.13.
The total spacing referred to all the occurring discontinuities has been calculated by using Equation
5.14.
The results of the evaluation of the spacing between discontinuity planes have been reported in
Table 5.14.
SPACING
TUNNEL Domain I Domain II Domain III; * DIIIa
[m]
Range 0.20 – 11.80 0.68 – 11.80 0.28 – 10.00 30.0 – 433.30* 0.20 – 4.33
Median 0.90 2.50 1.29 1.00* 0.83
Mean 1.42 3.56 2.00 1.17* 0.88
Q1 0.75 1.50 1.00 0.83* 0.71
Q3 1.29 5.00 2.12 1.40* 1.00
Table 5.14 Spacing statistical description; * DIIIa data: part of DIII data used in the following
predictive model creation.
Given that data distribution not normal it is necessary to consider median values instead of mean
ones.
The distance between discontinuity about the entire investigate tunnel is equal to 0.90 metres and
ranges between 0.75 m and 1.29 m (Q1-Q3).
In particular it has been observed that the spacing decreases significantly through the tunnel
advance. The discontinuities mapped into the gneiss are the highest spaced, measured equal to 2.50
m, it represents a continuous portion of rock mass. The distance between discontinuity planes
decreases into DII, in which a median value of 1.29 m has been calculated. The lowest value of
92
spacing has been computed into the micaschist belonging to the Clarea complex, characterised by
spacing of 0.83 metres.
The results exposed above mean that fracturing degree increase in the same verse of excavation
advance, so that micaschist are considerably more fractured than the gneiss.
The spacing calculated through La Maddalena rock mass has been plotted in Figure 5.26.
SPACING BETWEEN DISCONTINUITIES

5
DI DII DIIIa DIII
4
3
S [m]
0
0 1000 2000 3000 4000 5000 6000 7000
Pk. [m]
Figure 5.26 Spacing trend. It is necessary to consider that the values higher than 5 metres have not
been plotted being them considered as outliers. Red lines represent the domain changes, green line
delimits DIIIa and black lines the mean values for each domain.
It can be observed that the most fractured zones are located at Pk. 3 + 780 m, 3 +860 m, 4 + 440 m
and 5 + 500 m; as deduct by tunnel wall mapping and confirmed by Padovese et al. (2017).
Other fractured zones (less visible by the Figure 5.26 observation) has been encountered at about
Pk. 1 + 150 m and 1 + 320 m (into the DII), and at Pk. 3.450 m (into the micaschist domain).
DIIIa spacing data are not significantly different from the entire DIII spacing features.
93
5.3.4 Volumetric joint count
Palmström (1982, 1985, 1986) introduced the volumetric joint count, Jv, in order to provide a
measure of rock mass jointing.
Jv is a measure of the number of fractures within a unit volume of a rock mass, it is calculated
according to the following equation:
1 1 1 𝑁𝑟
𝐽𝑣 = + + + ⋯+ (5.15)
𝑆1 𝑆2 𝑆3 𝑆𝑟
where:
- S1, S2, S3 are joint spacing related to each joint set;
- Nr is the number of random joints;
- Sr is the random joint spacing, Palmström suggests to set Sr = 5.
Results
Jv related to La Maddalena rock mass have been calculated on the basis of Equation 5.15.
The results have been listed in Table 5.15.

Jv [J*m-3]
(n = 500) (n = 83) (n = 44) (n = 373); *DIIIa
Range 0.00 – 18.20 0.00 – 14.20 0.00 – 14.50 1.50 – 16.00* 1.50 – 18.20
Median 5.40 0.67 2.50 5.20* 6.40
Mean 5.88 1.59 4.01 5.80* 7.05
Q1-Q3 3.43 – 7.98 0.00 – 2.20 1.13 – 6.33 2.70 – 8.40* 4.60 – 9.00
Table 5.15 Jv results. statistical description; * DIIIa data: part of DIII data used in the following
94
As a consequence of spacing distribution, also the frequency distribution of Jv is not normal: for
this reason it is necessary to consider the median value instead of the mean one.
The mean value referred to the entire tunnel is equal to 5.40 m-3, it is possible to observe that the
number of joints per unit of volume increases through the tunnel in the excavation verse, as a result
of the reduction of the distance between discontinuity planes.
In the gneiss, where spacing is high and the number of occurring fractures is low, Jv has a median
value of 0.67 J* m-3, in the gneiss-micaschist alternation is of 2.50 J* m-3, finally it increases to 6.40
J* m-3 into micaschist domain.
By plotting the Jv values vs. Pk. is possible to individuate the most fractured zones, characterised
by high values of Jv (Figure 5.27). These zones correspond to those individuated by tunnel wall
mapping and and by means of spacing analysis (see section 5.3.3).
JOINT VOLUMETRIC COUNT

20
DI DII DIIIa DIII
15
Jv [m^-3]
10
0
0 1000 2000 3000 4000 5000 6000 7000
Pk. [m]
Figure 5.27 Jv distribution through the tunnel. Red lines represent the domain changes and the
black ones represent the mean value for each domain.
95
5.3.5 Rock Quality Designation index, RQD
The Rock Quality Designation index, RQD, was introduced by Deere et al. (1967) with the purpose
of obtaining a quantitative estimate of rock mass quality by analysing drill core logs, especially in
situ. One of the most important RQD use is as parameter of RMR system classification.
RQD is defined as the percentage of intact pieces of core longer than 10.0 cm in the total core
length, in accordance with the following formulation:
∑ 𝐿𝑒𝑛𝑔𝑡ℎ 𝑜𝑓 𝑐𝑜𝑟𝑒 𝑝𝑖𝑒𝑐𝑒𝑠 > 10.0 𝑐𝑚 𝑙𝑒𝑛𝑔𝑡ℎ

𝑅𝑄𝐷 = ∗ 100 (5.16)
𝑇𝑜𝑡𝑎𝑙 𝑙𝑒𝑛𝑔𝑡ℎ 𝑜𝑓 𝑐𝑜𝑟𝑒 𝑟𝑢𝑛
Where no core is available and traces of discontinuities are visible in surface exposure RQD, is
calculable by means of Palmström relationship (Palmström, 1982):
𝑅𝑄𝐷 = 115 − 3.3 ∗ 𝐽𝑣 (5.17)
RQD value can change significantly depending upon the borehole orientation, being it a
directionally dependent parameter. For this reason, the use of Palmström relationship may be useful
in reducing the directional dependence.
96
Results
Any drill core logs were not available about La Maddalena tunnel, for this reason Equation 5.16
cannot be applied in order to calculate the Rock Quality Designation Index.
We decided to evaluate it by means of Palmström relationship (Equation 5.17).
RQD is always very high in the entire investigated tract; its mean value is equal to 93.32%.
However a little difference between DI and DII-DIII has been observed. In particular the highest
values characterise the gneiss (mean value of 98.7%) in accordance with the massive structure of
this rocks. RQD values reduce through the other two domains; there is not difference between DII
and DIII being their mean values 93.36% and 92.13% respectively.
The minimum values of RQD (50%) have been calculated in DIII, it has not a particular
significance being an isolated case.
The areas where RDQ is lower respect to the principal trend represent fractured, faulted or
cataclastic zones.
The trend of Rock Quality Index through La Maddalena tunnel has been reported in Figure 5.28.
RQD trend through the tunnel

100
80
60
RQD
40
20
DI DII DIIIa DIII
0
0 1000 2000 3000 4000 5000 6000 7000
Pk. [m]
Figure 5.28 RQD trend through the tunnel. Red lines represent the domain changes, green line
delimits DIIIa sector.
97
5.3.6 Rock block volume
Block size is considered as one of the most relevant rock mass feature in rock engineering projects,
both underground and surface ones. There are several methods that allow to determine rock block
volume (Vb); in particular, block volume can be measured in situ where the individual blocks are
observable in a surface, moreover, it can be computed from joint spacing or from the volumetric
joint count.
The method that allows to estimate Vb from the volumetric joint count has been synthetically
described.
Vb can be calculated in accordance to the following equation:
𝑉𝑏 = 𝛽 ∗ 𝐽𝑣 −3 (5.18)
where β is the block shape factor, it is usually calculated by means of a method developed by
Palmström (1995), as:
𝑆𝑚𝑎𝑥
𝛽 = 20 + 7 ∗ (5.19)
𝑆𝑚𝑖𝑛
where: Smax and Smin are the maximum and minimum spacing respectively.
98
Results
The rock block volume about La Maddalena rock mass has been calculated by means of Equation
5.18. However, it was not possible to calculated Vb for each rock mass section, in particular for
those ones characterised by high spacing between fractures. This situation frequently occurs in DI
and DII; instead for each rock mass section belonging to DIII Vb was calculated.
For this reason, the statistical description of Vb has not been provided.
It was noted that rock block volume is higher into the gneiss (median value equal to 3.62 m 3) than
in DII and DIII (median value about 0.17 m3).
The median Vb referred to the entire excavate length is of 0.18 m3.
The trend of Rock Block Volume through La Maddalena tunnel has been reported in Figure 5.29.
ROCK BLOCK VOLUME

10
8 DI DII DIIIa DIII
Vb [m^3]
6
4
2
0
0 1000 2000 3000 4000 5000 6000 7000
Pk. [m]
Figure 5.29 Vb trend through the tunnel. Red lines represent the domain changes, green line
delimits DIIIa sector.
99
5.3.7 Groundwater conditions
Generally, the ground water circulation through rock masses is possible thanks to the secondary
permeability, via the discontinuity surfaces.
ISRM (1978) recommend describing seepage from individual or from specific set of discontinuities
taking into account for their filling conditions. It is necessary to consider that a tunnel acts as a
drain for the rock mass; in this case the overall flow into an individual section has to be described.
The description of groundwater condition about La Maddalena tunnel has been carried out from a
qualitative point of view, no inflow or water pressure measurements have been done. Since the
ground water circulation is generally poor and evidences of water flow have been observed in
correspondence of faults or cataclastic zones.
In particular groundwater conditions have been described using the classes proposed by Rock Mass
Rating system (Bieniawski, 1989), described in section 5.4.1.
These classes are:
- Completely dry,
- Damp;
- Wet;
- Dripping;
- Flowing.
As for La Maddalena tunnel the largest part of the surveyed discontinuities belongs to the class
damp.
As already mentioned, water flow has been observed in correspondence of faults and cataclastic
zones.
100
5.3.8 Joint conditions
The parameter named “joint conditions”, Jc, takes into account for five parameters which describe
some features of discontinuity planes, which are: roughness, persistence (or continuity), aperture,
filling and weathering degree.
In this work the parameters listed above have been rated in according to the RMR classification in
order to obtain the parameter “Jc”, described in RMR related paragraph (5.4.1); in this section a
brief description of each of them will be provided.
5.3.8.1 Persistence
As it is well known, discontinuity surfaces are not indefinitely continuous through rock masses and
the persistence represents their area extension or size inside the rock mass. This parameter can be
quantified by observing the trace lengths of discontinuities on the exposure surface; however it is
one of the most difficult rock mass parameter to quantify, for example because in some cases
exposures are smaller than the length (or area) of discontinuity.
Modal trace length for each set of discontinuity should be measured with a tape. Measurements
allow the description of each set according to the classification proposed by ISRM, 1978 (Table
5.16).
Discontinuity persistence
Length [m] Classification
< 1.00 Very low persistence
1.00 – 3.00 Low persistence
3.00 – 10.00 Medium persistence
10.00 – 20.00 High persistence
> 20.00 Very high persistence
Table 5.16 Classes of discontinuity persistence proposed by ISRM, 1978.
101
Results
The structural wall mapping of La Maddalena rock mass allows to determine the persistence of
detected discontinuities. The measurements could be affected by a systematic error due to the fact
that a discontinuity is more extended that the extension of the exposure.
The persistence analysed data regard only the principal set of discontinuity. Persistence has been
expressed in terms of classes and not of modal length.
Generally, the largest part of the detected discontinuities has persistence classifiable as medium,
given that the length of discontinuity traces is between 3.0 and 10.0 metres. In few cases the
persistence is low and these occur in the gneiss (DI) and in the gneiss-micaschist alternation (DII).
There are rare cases of very low persistence, mapped in DI.
No one discontinuity is characterised by high or very high persistence. It is probably due to the fact
that the tunnel wall exposure is limited and the discontinuity trace length might appear shorter than
its real extension (Figure 5.30).
Figure 5.30 Discontinuity persistence distribution (referred to the principal joint set).
102
5.3.8.2 Aperture
The term “aperture” indicates the perpendicular distance separating the adjacent walls of an open
discontinuity, the space between walls is filled by water or air (ISRM, 1978).
The largest aperture can be measured by means of a tape calibrated in mm; however in order to
measure the width of finest one a feeler gauge is necessary.
Apertures can be divided into three classes (and nine subclasses), in according to ISRM SM (1978;
Table 5.17):
Discontinuity aperture
Aperture [mm] Classification
< 0.5 Closed
0.5 – 10.0 Gapped
> 10.00 Open
Table 5.17 Classes of discontinuity aperture proposed by ISRM (1978).
Results
During La Maddalena rock wall mapping the aperture of discontinuities belonging to the principal
set of fractures has been measured by means of tape and feeler gauge. In this work the aperture
values have been divided into three classes in accordance to RMR classification system
(Bieniawski, 1989).
Five aperture classes have been considered; their ranges are: 0 mm, < 0.1 mm, 0.1 – 1.0 mm, 1.0 –
5.0 mm and > 5.0 mm.
Discontinuities have prevalently an aperture ranges between 0.1 and 1.0 millimetres and between
1.0 and 5.0 mm. DII is characterised by the highest occurrence of values > 5.0 mm of discontinuity
surfaces separation.
103
5.3.8.3 Filling
ISRM (1978) defines filling as the material that separates the adjacent rock wall of a discontinuity
and filling width is considered the perpendicular distance between the adjacent rock walls. Filling
materials are usually weaker than the parent rock; however, some exceptions occur when
discontinuities are filled with minerals characterised by high strength and deformation feature such
as calcite, quartz and pyrite.
The most frequent filling materials are sand, silt, clay and cataclastic rocks. The shear strength of
discontinuities is strongly affected by the mechanical features of filling materials.
During a discontinuity mapping the following features about filling material should be recorded:
mineralogy, particle size, over-consolidation ratio, water content, previous shear displacement and
width (ISRM, 1978).
Results
During La Maddalena rock mass surveys, filling materials were analysed on the basis of the
classification proposed by RMR system.
In particular it identifies five classes, these are:
- Absent filling material;
- Hard filling material, width < 5.0 mm;
- Hard filling material, width > 5.0 mm;
- Soft filling material, width < 5.0 mm;
- Soft filling material, width > 5.0 mm.
In the gneiss discontinuities the filling is absent in the largest part of the cases; whereas in the
micaschist (DIII) there are some situations in which filling material is soft and its thickness is
higher than 5.0 mm, generally this happens in correspondence of faults or cataclastic zones.
Filling material is the result of fault activity.
104
5.3.8.4 Roughness
The roughness of discontinuity walls can be observed at large scale and at small one. The term
“waviness” refers to large scale undulations while the term “unevenness” refers small scale
roughness. The roughness of wall discontinuity is one of the most important factors of its shear
strength.
ISRM (1978) proposed suggested methods for sampling roughness (the linear profiling method, the
compass and disc-clinometer one, the photogrammetric method, etc.), however when logistic and
time limitations prevent the use of the raccommended roughness measuring techniques the
roughness description will be limited to the assignment of two descriptive terms. The first one refers
to large scale of observation and the second one to the small observation scale.
In particular, at large scale it can be distinguished three classes or waviness: stepped, undulating
and planar; while at small scale it can be recognised three classes of unevenness: rough, smooth and
slickenside (Figure 5.31).
Figure 5.31 Suggested nomenclature for typical roughness profiles (ISRM, 1978).
105
Results
During La Maddalena tunnel wall mapping the description of discontinuity roughness has been
carried out considering the purpose of survey: the geomechanical classification of rock mass. For
this reason, it was considered both the nomenclature recommended by ISRM SM (1978) and the
classes proposed by RMR system (Bieniawski, 1989). The last ones are:
- Very rough;
- Rough;
- Slightly rough;
- Smooth;
- Slickenside.
In particular, for waviness description the nomenclature suggested by ISRM (1978) has been
adopted while for unevenness description the classes proposed by Bieniawski have been considered.
Generally, the surfaces of the discontinuities belonging to the principal set are slightly rough
(Figure 5.32). In particular, roughness is higher about the gneiss joint surfaces (in which the class
“rough” prevails and “very rough” one occurs) and some cases of smooth or slickenside
discontinuity surfaces have been recognised mapping the micaschist domain (DIII).
Figure 5.32 Discontinuity roughness distribution.
106
5.3.8.5 Weathering degree
Weathering reduces the shear strength of discontinuities and of the rock mass, given that it reduces
also the intact rock strength.
The system used to classify the discontinuity weathering of La Maddalena rock mass is the RMR
one. Bieniawski proposed five classes of alteration that range between unweathered and
decomposed. The description of these classes coincide with the one of the ISRM classes,
Bieniawski includes the ISRM classes “completely weathered” and “residual soil” into a unique
group named “decomposed” (Table 5.18).
RMR class ISRM class ISRM description
Unweathered Fresh No signs of alteration
Slightly weathered Slightly weathred Discoloration
Moderately weathered Moderately weathred Decomposition < 50%
Highly weathered Highly weathred Decomposition > 50%
Completely weathred Decomposition > 100%

Decomposed
Residual soil Rock is converted to soil
Table 5.18 Discontinuity weathering description.
Results
The tunnel surveys allow the description of the weathering condition of the discontinuity surfaces
characterizing La Maddalena rock mass.
Generally, the discontinuities result unweathered or slightly weathered; however, in isolated cases,
especially where faults occur and where dripping is present, they appear moderately weathered.
No cases of highly weathered or decomposed materials have been recognised.
107
5.3.8.6 Joint condition parameter
The description of joint surfaces in terms of persistence, aperture, filling, roughness and weathering
and the rating of this parameter by means of the indication proposed by RMR classification system
allow to carry out the fourth parameter of Bieniawski classification, Jc (see section 5.4.1).
In fact, Jc, joint conditions, is the sum of the rate assigned to each of the parameters listed above.
The results of Jc calculation have been reported in Table 5.19.
Jc TUNNEL Domain I Domain II Domain III; *DIIIa
Range 7.00 – 29.00 11.00 – 29.00 7.00 – 18.00 10.00 – 21.0* 7.00 – 21.00
Median 18.00 20.00 15.00 19.00* 18.00
Mean 17.04 19.42 15.00 18.00* 16.75
Q1 15.00 17.00 15.00 18.00* 14.00
Q3 19.00 22.00 17.00 19.00* 19.00
Table 5.19 Jc statistical description; * DIIIa data: part of DIII data used in the following predictive
model creation.
The median Jc value referred to the entire tunnel is of 18.00, the highest value has been calculated
in DI, where filling is often absent and discontinuity surfaces are rough of very rough. Joint surfaces
show lower quality in DII, where the median value is equal to 15.00.
There is not a significant difference between the features of joint surfaces in the section DIIIa and
the entire DIII. In fact, the median values are 19.00 and 18.00 respectively.
Joint conditions tend to be constant for more or less long tract of tunnel; it means that adjacent rock
mass sections have the same joint conditions. It was particularly observed in DII and DIII, whereas
in DI, Jc changes more rapidly, despite the fact that adjacent tracts characterised by equal values of
Jc can be observed (Figure 5.33).
108
JOINT CONDITIONS
35
DI DII DIIIa DIII
30
25
20
Jc
15
10
5
0
0 1000 2000 3000 4000 5000 6000 7000
Pk. [m]
Figure 5.33 Joint conditions distribution, the red lines represent the domain changes.
The lowest Jc values have been calculated in correspondence of fault zones, where:
- Joint surface asperities have been smoothed by fault movements;
- Discontinuity surfaces can be weathered as result of water circulation;
- Filling materials, generated by fault movements, are present;
- Fault dislocation can make the discontinuities open.
The factors listed above make Jc decrease.
109
5.4 Rock mass classifications

The classification systems are widely applied to the rock masses for the design of civil and mining
underground works, rock foundations and dams as a useful engineering tool. In fact, the rock mass
classification systems are largely used to assess deformation and strength characteristics of a rock
mass considered as a continuous equivalent medium.
The classification methods allow to subdivide the rock masses in different classes of quality: from
extremely poor to excellent or exceptionally good.
They are rarely used for rock slope stability analysis since the rock discontinuities play a
fundamental role and, for this reason, the rock mass could not be not considered as a continuous
equivalent.
Here a brief review of the two rock mass classifications used in this work has been provided. Rock
Mass Rating (RMR) system and Geological Strength Index (GSI) will be described.
5.4.1 Rock Mass Rating (RMR) system
Rock Mass Rating (RMR) system was proposed by Bieniawski in 1976; over the years RMR has
been refined and significant changes in rating assignment have been made. The version most widely
used is the 1989 one: its advantage is that only basic parameter are used in its calculation, they are
related to the mechanical conditions and the geometry of rock mass, they are:
- Uniaxial compressive strength of the rock matrix, σc: it can be evaluated by uniaxial
compression tests (UCS) or point load test;
- Rock Quality Designation, RQD: it is the percentage ratio obtained by adding all the core
rock pieces longer than 10.0 cm and the total length of drill rode;
- Spacing of Discontinuities: the whole rock mass discontinuity spacing has to be considered,
it can be assessed from scanline discontinuity survey;
- Condition of Discontinuities: this parameter considers the overall conditions of

discontinuities in terms of roughness, persistence (or continuity), aperture, filling and
weathered degree;
- Groundwater conditions: this parameter can be considered from a qualitative or from a

quantitative point of view. In the first case a qualitative judgment based on the experience of
the operator in charge of the rock mass characterization is required. In the second case
110
inflow measurement has to be carried out or the ratio between joint water pressure and
major principal stress has to be computed.
A score, called rating, has to be given to each value of the five parameters mentioned above and by
adding the assigned ratings RMRbasic is obtained (section A, Table 5.20). This index is used for the
rock mass mechanical feature assessment.
It is fundamental to note that discontinuity orientations strongly affect the stability conditions of a
“geomechanical work”, for this reason in order to use the RMR system in civil, mining, tunneling or
consolidation works, the orientation of discontinuities with respect to the direction of the excavation
or the existing rock walls have to be taken into consideration. In section B the rating adjustment for
discontinuity orientations have been listed, in particular, in tunneling operations great attention is
given to the effect of discontinuity strike and dip orientation (section F, Table 5.20).
The two extreme situations for the orientation of the principal discontinuity system are:
- Dip direction parallel to the tunnel excavation direction: in this case two situations can
occur. In particular, it is possible to “drive with dip” or “drive against dip”. Drive with dip it
is considered favorable because of rock blocks have not the kinematic possibility to slide
being directed against the rock mass. Drive against dip it is considered unfavorable because
rock blocks can slide from the tunnel crown before to have the time to reinforce the rock
mass (Figure 5.34).
Discontinuity strike is perpendicular to tunnel direction
Excavation Excavation
against dip with dip
1
1
2 3
3
111
Figure 5.34 Excavation against and with dip when discontinuity strike is perpendicular to tunnel
direction. In the stereonet: 1 = tunnel axis, 2 = discontinuity dip direction, 3 = discontinuity strike.
- Dip direction perpendicular to the tunnel excavation direction: in this case the unfavorable
condition occurs when the discontinuities are sub-vertical (Figure 5.35).
Discontinuity strike is parallel to tunnel direction
2
1
Figure 5.35 Excavation in rock mass where discontinuity strike is parallel to tunnel direction. In the
stereonet: 1 = tunnel axis, 2 = discontinuity dip direction.
Considering discontinuity orientation effects RMRcorrected can be obtained.
The qualitative judgment of a given rock mass is subdivided into five classes (I – V), where values
of RMR from 81 to 100 correspond to very good rock, values between 61 and 80 correspond to
good rock, values from 41 to 60 characterise fair rock, values between 21 and 40 correspond to poor
rock and very poor rock are characterised by values ranging between 0 and 20 (section C, Table
5.20).
112
Table 5.20 RMR system (Bieniawski, 1989).

113
In tunnel design, the RMRcorrected can be used to assess the self-support time, the roof span and the
support installation time.
An example is provided by Table 5.21.
Table 5.21 Guidelines for excavation and support of tunnels in accordance with the RMR system
(Bieniawski, 1989).
114
Results
The field surveys carried out during La Maddalena tunnel construction allow the collection of a
series of geological, structural and mechanical data that have been used to classify the excavated
rock mass. For each of the 502 rock sections in which La Maddalena rock mass has been divided,
the collected data and the subsequent analysis allow the rock mass to be classified by means of
RMR system.
The quality of the excavated rock mass results to be principally fair and good. The most frequent
RMR classes are the II and the III; none of the sections belongs to the V class and very few to the
IV. Furthermore, there very few instances of rock mass considered of very good quality (Figure
5.36).
Figure 5.36 Histogram of frequency of RMR distribution.
In particular, the gneiss belonging to Domain I is characterized by higher quality than the quality of
the two other domains. In DI there are few rock sections classified as fairy rocks and the largest part
of them are considered very good and good rocks.
The quality decreases into the gneiss and micaschist alternation (DII), where the very good quality
rock sections are limited; moreover a small decrease can be observed considering RMR values
about the Domain III, in which poor quality rock sections occur (Figure 5.37).
115
Figure 5.37 Histogram of frequency of RMR distribution about each domain.
The mean RMR value calculated for the entire excavated tunnel is equal to 60, the mean RMR
gneiss value is 75 and in the gneiss-micaschist alternation and in the micaschist the values are equal
to 62 and 59 respectively.
The RMR trend through the tunnel is showed in Figure 5.38.
RMR trend through the tunnel

100
80
60
RMR
40
20
DI DII DIIIa DIII
0
0 1000 2000 3000 4000 5000 6000 7000
Pk. [m]
Figure 5.38 RMR trend through the tunnel. Red lines represent the domain changes, green line
116
5.4.2 Geological Strength Index (GSI)
The Geological Strength Index, GSI, is a rock mass classification scheme introduced by Hoek in
1994 and developed over the years (Hoek et al. 1998; Marinos & Hoek, 2000, 2001; Hoek et al.,
2013). It was developed in order to overcome the limitations in rock mass strength criterion that
occur by using RMR system. It has a fundamental role in the Hoek and Brown failure criterion and
it is an input data for numerical analyses of tunnel, slope or foundation problems. The original GSI
chart is based on the field observations. Hoek et al. (2013) proposed a quantification of GSI chart
on the basis of two parameters: joint conditions and RQD.
The GSI method takes into account for those properties that determine the strength and the
deformability of the rock masses: it gives great emphasis on basic geological observation of rock
mass features reflecting the material, its structure and its geological history.
In particular GSI focuses on rock mass structure and on the conditions of the discontinuity surfaces.
As for the structure great attention has to be given to the block interconnection features. Authors
consider six situations, these are: intact or massive, blocky, very blocky, blocky/disturbed/seamy,
disintegrated and laminated/sheared; the description of each classes is provided in Figure 5.39.
The conditions of the discontinuity surfaces are expressed in terms of roughness and weathering;
there are five classes of surface conditions, which are:
- Very good: very rough, fresh and unweathered surfaces;
- Good: rough and slightly weathered;
- Fair: smooth and moderately weathered;
- Poor: slickenside and strongly weathered, covered by patina or filled with angular fragment;
- Very poor: slickenside, strongly weathered, covered by or filled with clay material.
Using the chart proposed in Figure 5.39 and with some experience, GSI can be obtained by
observing rock walls. GSI value is expressed by means of a numerical index associated to every
combination of these two parameters. GSI varies in a 0-100 range.
117
Figure 5.39 GSI chart (Marinos and Hoek, 2000).
GSI does not consider UCS of intact rocks and groundwater conditions.
Marinos and Hoek (2000) published a collection of indicative charts which show the most recurrent
ranges of GSI for rock masses of a large quantity of rock types (Figure 5.40).
118
Figure 5.40 GSI chart for flysch (Marinos and Hoek 2000).
Results
By tunnel walls observation and the utilization of the chart proposed by Marinos and Hoek, 2000,
(Figure 5.39) it was possible to assign a GSI value at each rock sections surveyed. The latest GSI
version (2013) has not been used because it considers Jc and RQD, but in this study, RQD has been
empirically obtained by means of Palmström relationship (Palmström, 1982); so GSI-2000 version
has been considered.
The results of this classification have been reported in Table 5.22.

GSI
(n = 502) (n = 83) (n = 44) (n = 375)
Range 37.00 – 98.00 54.00 – 98.00 43.00 – 84.00 37.00 – 85.00
Median 65.00 80.00 63.50 65.00
Mean 66.11 77.40 63.93 63.86
Q1-Q3 60.00 – 70.00 70.00 – 85.00 52.00 – 74.25 60.00 – 67.00
Table 5.22 GSI statistical analysis.
119
The mean GSI value is about 66; it denotes a good quality rock mass, in particular the highest
values have been assigned to the gneiss of DI (mean value equal to 77.40). The GSI decreases
through the gneiss-micaschist alternation and micaschist domain. In DII and DIII the mean value is
the same (63.93 and 63.86 respectively).
It is possible to assert that on the basis of GSI La Maddalena rock mass can be divided into two
principal groups: DI and DII-DIII respectively.
The trend of GSI through La Maddalena tunnel has been reported in the figure below (Figure 5.41).
GSI trend thorugh the tunnel

100
80
60
GSI
40
20
DI DII DIIIa DIII
0
0 1000 2000 3000 4000 5000 6000 7000
Pk. [m]
Figure 5.41 GSI trend through the tunnel. Red lines represent the domain changes, green line
120
5.5 Seismic wave propagation

The velocity of propagation of seismic waves through a rock mass provides fundamental
information about the geological characteristics and the mechanical features of the ground,
especially when direct survey cannot be done.
Geophysical investigation methods are frequently adopted during excavation works in order to
predict the geological framework ahead the tunnel face, with the purpose of avoiding to
encountered rock mass conditions different from the expected ones.
During La Maddalena construction seismic investigation methods have been involved for this
purpose.
In this work one of them has been taken into account, here a brief description of the methods and
the obtained results is provided; its reliability in terms of geological prediction accuracy will be
discussed in Chapter 7.
The indirect investigation method considered in this work is the so called “Tunnel Seismic
Prediction System”, TSP.
5.5.1 Tunnel Seismic Prediction System
The Tunnel Seismic Prediction System, TSP, is a reflection seismic technique developed by
Amberg Technologies AG (Switzerland) and widely and successfully used in underground
construction (Dickmann and Tang, 2006; Dickmann, 2005; Dickmann and Krueger, 2013).
24 boreholes need to be drilled aligned in the tunnel wall, each of them has a depth between 1.00
and 1.50 metres and they are 1.50 m spaced. Small explosive charges are fired inside these
boreholes, allowing the seismic wave energy to propagate through the rock mass.
Acoustic impedance phenomena, such as significant planes of weakness, faults and shear zones or
changes in mechanical and physical properties of rocks, cause the reflection of a portion of signal.
The reflected part of signal will arrive at the receivers sited in the tunnel walls.
A schematic representation of the position of source points and receivers is provided Figure 5.42.
121
Figure 5.42 Positions of sources (red dots) and receivers (blue dots), Dickmann and Krueger, 2013.
The deceiver units are formed by an ultrasensitive triaxial seismic accelerometer, able to convert the
seismic signal into electrical voltage ones (Dickmann, 2005). The receiver unit provides the
coverage of all three space directions, which allows the recording of the full wave field. The
distinction of compressional (P) and shear (S) waves and then their velocity analysis are performed.
In this way the reflected signal times are converted into distances allowing us to localise the
weakness planes (working as planes of reflection) both in terms of orientation and of distance from
the tunnel front (Figure 5.43).
Figure 5.43 TSP survey result (Dickmann and Krueger, 2014).

TSP allows the estimation of some mechanical and physical properties of rocks (for example
dynamic Young’s modulus, dynamic Poisson’s ratio, density), by means of empirical formulation
based on seismic wave velocity.
122
5.5.2 Tunnel Seismic Prediction results
During La Maddalena construction 22 TSP measurements were carried out with the purpose of
investigating the geological feature of the rock mass ahead of the tunnel face. The seismic
investigated tunnel tract ranges from Pk. 0 + 500 m to Pk. 4 + 980 m.
The results of these measurements have been acquired and a part of them has been used to evaluate
their reliability in geological feature prediction: in particular the analysis takes into account the data
measured between Pk. 0 + 500 m and Pk. 2 + 000 m (see Chapter 7).
The distribution of seismic data obtained from the 22 TBM measurements has been described in
this section.
The following image (Figure 5.44) shows the variation in seismic wave velocities through the
tunnel, both in terms of P waves and of S waves.
SEISMIC WAVE VELOCITY

7000
SW1 SW2 SW3a SW3b
6000
Seismic wave velocity [m/s]
5000
4000
3000 P-Waves
S-Waves
2000
1000
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Pk. [m]
Figure 5.44 Seismic wave velocity measured by TSP technique. Black lines delimit seismic sections.
Rock mass can be divided into three seismic sections, on the basis of the propagation velocity; these
sections are:
- SW1: is the section between Pk. 0 + 500 m and Pk. 1 + 360 m. The measured seismic
velocity is high and constant into the entire tract. The values about P waves range between
5014 m/s and 5826 m/s (mean value is 5597 m/s) and about S waves they range between
2994 m/s and 3565 m/s (the mean value is equal to 3254 m/s). SW1 corresponds to the
geological domain DI and DIII;
123
- SW2: is the rock mass portion from Pk. 1 + 360 m to Pk. 2 + 014 m and it is characterised
by P wave velocity ranging between 3028 m/s and 3677 m/s (mean equal to 3370 m/s) and S
wave velocity from 1947 m/s to 2202 m/s (mean value of 2054 m/s). SW2 coincides with
the first part of DIII;
- SW3: this section covers a total length of 2967 m, from Pk. 2 + 014 m to Pk. 4 + 981 m. It is
possible to observe that SW3 velocity distribution is less homogenous than the SW1 and
SW2 ones. It is possible to recognise a limited portion of rock mass (between Pk. 2 + 166 m
and Pk. 2 + 407 m) in which P and S waves are characterised by lower velocity than the
remaining part of SW3. This subsection has been named SW3a, and the principal sector of
SW3 is called SW3b.
The SW3a mean values for P and S waves are 4232 m/s and 2413 m/s respectively; whereas
referred SW3b the mean values are 5044 m/s, for P waves and 2941 m/s for S waves.
Finally, as far as P waves measured in SW3b are concerned, it is possible to note that a
portion is characterised by mean velocity equal to 5326 m/s (from Pk. 2 + 420 m to Pk. 3 +
266 m) and a sector by mean velocity of 4900 m/s (from Pk. 3 + 266 m). The velocities
referred to S waves are more constant than the P ones.
Seismic wave velocities obtained by the 22 TSP measurements have been compared with the global
quality of the rock mass (expressed by RMR) in which measurements were performed, in order to
investigate the relationship between seismic response and rock mass quality (Figure 5.45).
P WAVES vs. RMR

7000 100
90
6000
P wave velocity [m/s]
80
5000
70
RMR
4000 60
50
3000
40
2000
30
1000 20
450 950 1450 1950 2450 2950 3450 3950 4450 4950
Pk. [m] P Wave
RMR
Figure 5.45 Relationship between P wave velocities and RMR.
124
In the case of the investigated rock mass, the global quality of the ground moderately affects the
seismic wave propagation through it.
Generally, it is possible to observe that when RMR decreases also wave velocity becomes lower, on
the contrary when the quality increases higher velocities can be measured.
The considerations exposed about P waves and rock mass quality relationship are also valid for S
waves.
125
5.6 Tunnel depth
5.6 Tunnel depth

The depth at which a tunnel is excavated is generally used as a preliminary indicator of in situ
stresses; in particular, overbunden is adopted to estimate the lithostatic load σ1.
In situ stresses impact the cracks initiation and failure propagation and the formation of chips of
rock under TBM disc cutter action. However, there are contrasting opinions on the overall effects of
in situ stress and machine performance.
Bordet and Comes (1975), Tarkoy and Marconi (1991), Liu et al. (2002) sustain that if stresses
increase the general TBM performance decreases, on the contrary Klein et al. (1995) asserts that
high confinement stresses make machine performance increase.
Innaurato et al. (2007) demonstrated that lateral stress has limited effects on the thrust necessary to
fragment the rock, therefore on the TBM performances.
The maximum depth reached during La Maddalena construction is over 1990 metres.
The data about tunnel depth have been listed in Table 5.23.
DEPTH [m] TUNNEL Domain I Domain II Domain III; * DIIIa
Range 79.4 –1992.6 79.4 – 569.1 273.8 –588.4 256.8 –616.7* 256.8 –1992.6
Median 951.5 360.9 442.2 448.1* 1083.9
Mean 907.6 360.1 435.6 435.5* 1056.4
Q1-Q3 489.9 –1128.8 211.4 – 520.0 346.1 – 516.9 329.4 –539.0* 1056.4-1156.0
Table 5.23 La Maddalena tunnel depth.
A longitudinal profile is provided by Figure 5.46 in order to clearly visualise the overbunden
achieved by tunnel excavation.
126
5.6 Tunnel depth
TUNNEL DEPTH
2500
2000
depth [m]
1500
1000
500
0
0 1000 2000 3000 4000 5000 6000 7000
Pk. [m]
Figure 5.46 La Maddalena depth. Red line represents the tunnel, the arrow represents the
excavation verse.
127
5.7 Conclusive remarks

The field survey work allows the collection of a large amount of information on the lithology, the
structure, the conditions and the orientations of discontinuity surfaces and groundwater flow of the
Ambin Massif portion hosting La Maddalena tunnel.
These field data, integrated with the results obtained by means of laboratory tests (tests not carried
out in person), allow for the qualitative and quantitative description and the classification, in terms
of quality classes, of the whole investigated rock mass.
In particular, as already mentioned before, three geological domains have been distinguished:
- From Pk. 0 + 198 m to Pk. 0 + 967 m the main lithology is gneiss and it has been classified
as Domain I (DI);
- From Pk. 0 + 967 m to Pk. 1 + 360 m an alternation of gneiss and micaschist has been
surveyed and this portion of rock mass has been described as Domain II (DII);
- From Pk. 1 + 360 m to Pk. 6 + 897 m the main lithology consists of micaschist. This tract of
tunnel has been named Domain III (DIII).
In the following paragraphs a general overview of the recognized geological domain has been
provided.
The histograms for frequency of each parameter detected in each domain are reported in Appendix
II and the geological longitudinal section of the portion of Ambin Massif hosting La Maddalena is
reported in Appendix III.
128
5.7.1 Domain I
The portion of rock mass named (in this work) Domain I ranges between Pk. 0 + 198 m and Pk. 0 +
967 m. It consists of a metamorphic rock, mainly characterised by grey colour, but in some sectors
rocks present greenish coloration as result of the chlorite occurrence.
Rocks are fine-medium grained and present heteroblastic structure, texture is isotropic and weakly
foliated.
The principal forming minerals are quartz, alkaline feldspar, plagioclase, chlorite and muscovite.
Calcite has been recognised in limited portion of rock mass, and quartz veins frequently occur.
The foliation is given by the alternation of sialic levels consisting of quartz and feldspars and
lepidoblastic ones formed by white mica principally. The mean orientation of foliation planes is
about 120/50. On the foliation planes develops the principal set of joints. In some situation metric
folds occur, their axes are characterised by orientation of 325/60, in terms of trend and plunge.
The described rocks are aplitic gneiss belonging to the Ambin complex (Figure 5.47).
The fracturing degree is the lower than the other two domains, the measured total spacing between
weakness planes is 2.50 m (median value), the global joint conditions are good and groundwater
circulation is very limited. The levels of rock hardness and abrasivity hare very high, as also UCS.
The velocity of propagation of seismic waves is high both for P (5500 m/s – 6000 m/s) and S waves
(3000 m/s – 4000 m/s)
Domain I is characterised by high rock mass quality, in fact the largest part of rock mass sections
belong to I and II class of RMR.
Figure 5.47 Aplitic gneiss of Ambin complex.
129
5.7.2 Domain II
The Domain II has been recognised between Pk. 0 + 967 m and Pk. 1 + 360 m.
The former consists of a metamorphic rock, dark grey coloured, and medium-fine grained. The
foliation is clearly evident due to the alternation between layers in which the principal forming
minerals are quartz, albite and white mica (phengite), these rocks can be considered albitic gneiss
minuti; and layers formed prevalently by white mica and then by quartz, classified as quartz-
micaschist (Figure 5.48).
Domain II belongs to the Ambin complex, and represents the transition zone between DI and DIII,
and more precisely the passage from the Ambin complex to the Clarea one.
The foliation planes dips towards SE with and a dip angle ranging between 30° and 60° on which
the most occurring set of fractures develops.
The general quality of rock mass is lower than the previous domain. Spacing between fractures is
shorter than the one of DI (1.30 m), as also the quality of joint surfaces and the mechanical
properties of intact rock. RMR values make the rock mass classify into II and III classes.
The velocity of propagation of compressional and shear seismic waves significantly decreases, the
mean value for P waves is around 3000 m/s – 3500 m/s and for S ones is roughly equal to 2000 m/s.
Figure 5.48 Rock belonging to DII.
130
5.7.3 Domain III
The third geological domain recognised during La Maddalena rock mass survey ranges from Pk. 1
+ 360 m to the end of the excavated tract, it means that DIII is the geological sector more largely
excavated.
The colour of this rocks can variate from light to dark grey and they are generally fine, medium-fine
grained. Foliated texture is clearly recognisable and it is principally due to the alternation of sialic
and micaceous layers. In this domain micaschist and gneiss minuti have been recognised.
Micaschist are formed principally by quartz, albite, white mica (phengite), chlorite, chloritoid,
biotite glaucophane and garnet. The percentage of these minerals is variable, so it was possible to
classify rocks as albitic quartz micaschist, glaucophane micaschist, quartz micaschist and
micaschist. In some local cases rock has been considered as albitic gneiss minuti (Figure 5.49).
These rocks belong to the micaschist and gneiss minuti reequilibrated in blueschist facies of Clarea
complex (see section 4.2.1.1).
Foliation and discontinuity orientation do not differ significantly from the previous two domains
and metric folds have been observed in limited tracts only.
Spacing between planes of weakness is the lowest of the whole tunnel tract (median value equal to
0.83 m), joint conditions are quite good (mean value 18) and the strength properties of intact rocks
are similar to the ones described in DI (UCS mean value 152.00 MPa). The largest parts of rock
mass sections belong to class III and II of RMR classification. The seismic wave velocity is similar
to the one described in DI, nevertheless more than one seismic response has been detected.
Figure 5.49 Micaschist belonging to the Clarea complex.
131
6 TBM PERFORMANCE ANALYSIS
In order to investigate the relationship between the rock mass feature and the TBM performance
obtained in its excavation it has been necessary to create both a rock mass database (see Chapter 5)
and a TBM parameter one.
The machine database is described in this chapter.
It is necessary to anticipate that La Maddalena field TBM database consists of two parts:
- The first part (DB1) concerns TBM data, covering a length of about 2.000 metres, from Pk.
0 + 198 m to Pk. 2 + 010 m.
These data have been used to create the predictive models described in the following
chapters (see Chapter 9, 10, 11) and they have been personally collected;
- The second part (DB2) of La Maddalena field TBM database includes the data referring to a
length of about 5.000 metres from Pk. 2 + 010 m to the end of the excavation.
This part of performance data has been used to validate the predictive models created on the
basis of the information contained in DB1 (see Chapter 12). These TBM data were derived
from the Doctoral Thesis of PhD Andrea Rispoli (Rispoli, 2018); on the contrary the rock
mass data have been personally collected.
133
6.1 Field TBM performance database

During tunnelling, several operational parameters are registered by the automatic acquisition
system, which is able to read, visualise and register them. The sampling rate may vary on the basis
of the needed analyses and the required representativeness: it can range from one datum every five
seconds to one datum every 600 seconds. In this research work, a sampling rate of one datum every
five seconds has been selected.
The outliers in TBM data, with respect to DB1, have been manually removed. The biggest errors
relative to DB2 have been removed by means of a series of filtering codes.
6.1.1 The first part of the database (DB1)
The most important machine parameters (in relation to this study) are listed below:
1) The length of each stroke: it is the length of a continuous advance [mm], measured by the
system detecting the start position and the end one of each continuous boring cycle. In
particular the system monitors the extension of the thrust cylinders, by means of a wire
displacement sensor;
2) The actual time of excavation [s]: the system detects the starting time and the ending time of
each continuous boring cycle;
3) The thrust force [kN];
4) The rotation speed of the cutterhead [rpm].
A number of other TBM parameters has been collected (the average motor power [kW], the average
motor amperage [A], the gripper pressure [bar], the torque [kN*m], etc…). They have not been
involved in the analysis carried out in this research work; some of them are:
The parameters ROP and FPI (see Chapter 2) are calculated taking into account the information
obtained by the acquisition system of TBM operating data. In particular, from parameters 1 and 2 it
is possible to obtain the net rate of penetration (ROP or PR) through the ratio between the effective
bored length and the actual time needed to do it; whereas all the parameters listed above are
involved in order to calculate FPI.
Table 6.1 reports an example of the way in which the data were collected and analysed in order to
carry out the desired TBM performance indices.
134
∆t Stroke Stroke ∆Stroke CHD Force on CHD ROP Prev FPI

In. [mm] Fi. [mm] [mm] Force cutter Speed [mm/min] [mm/rev] [kN/cutter/
[kN] [kN] [rpm] mm/rev]
A B C D = C–B E F = E/43 G H = D/A I = H/G J = F/I

00:05:20 78 282 204 5500 127.91 8.50 38.25 4.50 28.42
00:17:45 1035 1931 896 11000 255.81 9.00 50.48 5.61 45.60
00:26:35 71 1025 954 11800 274.42 8.80 35.89 4.08 67.26
00:07:20 1607 1916 309 11600 269.77 9.20 42.14 4.58 58.90
00:34:40 100 1569 1469 11800 274.42 9.10 42.38 4.66 58.89
00:44:05 73 1922 1849 12000 279.07 8.90 41.94 4.71 59.25
00:39:45 73 1924 1851 12000 279.07 9.10 46.57 5.12 54.51
00:24:35 1101 1925 824 11800 279.42 9.10 33.52 3.68 75.93
00:44:15 75 1092 1017 12000 279.07 8.90 22.98 2.58 108.17
01:03:15 320 1860 1540 10500 244.19 9.10 24.35 2.68 91.12
00:07:25 116 310 194 11400 265.12 9.00 26.16 2.91 91.11
Table 6.1 Example of registration and elaboration of TBM operating data.
The indices calculated as explained above are related to a stroke which is long at maximum 1.80
meters that corresponds to the maximum length of the excavation of the thrust cylinders. It was
considered useful obtaining unique value both of ROP and a unique one of FPI for each rock mass
section, in such a way comparing TBM performance with the rock mass feature was more
immediate. For this reason, a weighted average of all ROP and FPI measurements computed in a
length covering a rock mass section has been calculated.
It was considered a length covered by a rock mass section (LRMS) as the sum of n strokes (St), as
proposed in Equation 6.1:
𝐿𝑅𝑀𝑆 = 𝑆𝑡1 + 𝑆𝑡2 + … + 𝑆𝑡𝑛 (6.1)
At each stroke corresponds a value of ROP, in others words at S t1 corresponds ROP1, …, at Stn
corresponds ROPn, etc..
So the weighted average ROP registered into a rock mass section (ROPRMS) is given in according to
Equation 6.2:
𝑆𝑡1 ∗ 𝑅𝑂𝑃1 + 𝑆𝑡2 ∗ 𝑅𝑂𝑃2 + … + 𝑆𝑡𝑛 ∗ 𝑅𝑂𝑃𝑛

𝑅𝑂𝑃𝑅𝑀𝑆 = (6.2)
𝐿𝑅𝑀𝑆
135
where LRMS is calculated by means of Equation 6.1.
The same procedure has been adopted to calculate FPI for each rock mass section.
The procedure described allowed to assign to each rock mass section a value of ROP and a FPI one.
A complete database including both geological/geomechanical information and machine data has
been created.
6.1.2 The second part of the database (DB2)
The second part of the database was completed with field data obtained from the Ph.D. thesis of
Andrea Rispoli (Rispoli, 2018) and was used to validate the predictive models carried out during
this work.
Rispoli evaluated eight TBM operating parameters; however, ROP and FPI only have been
considered for the purposes of this work.
The Author developed specific calculation codes in order to automatically perform the data filtering
with the purpose of removing the anomalous ones. In particular, these codes filter TBM data in two
stages. A first filtering allows to remove the most important errors, e.g. the values not compatible
with the operating range of the TBM.
During the second filtering phase, a specific code is selected on the basis of the required analysis.
Prior to the utilisation of the acquired data, the Author performed a reliability test that concerns the
excavated length measured and the effective time required to do this.
136
6.2 Measured TBM performances

The ROP and FPI distribution through the tunnel have been analysed, both for DB1 and DB2. Both
the indices have been evaluated for the whole excavated length and for each encountered geological
domain.
As for the portion of DIII ranging between Pk. 1 + 360 m and Pk. 2 + 003 m (named DIIIa), the
machine performances have been calculated on the basis of the modalities discussed for DB1 (see
section 6.1.1).
6.2.1 Rate of Penetration distribution
As already described above, the rate of penetration (ROP) or penetration rate (PR) is the ratio
between a continuously excavated length and the actual boring time (see section 2.4.1).
Figure 6.1 provides the trend of ROP through the whole tunnel.
RATE OF PENETRATION
4
DI DII DIIIa DIII
3
ROP [m/h]
0
0 1000 2000 3000 4000 5000 6000 7000
Pk. [m]
Figure 6.1 ROP distribution through the tunnel. Red lines represent the domain changes, black
horizontal lines represent the median values measured in each domain, and dotted black line
indicates the median value registered in DIIIa. Green line delimitates DIIIa.
The median value of ROP referred to the entire tunnel length is equal to 1.33 m/h. The lowest ROP
have been registered in the gneiss (DI) excavation in which the median values is of 0.94 m/h. This
low value is compatible with the high quality of encountered rock mass, which made tunnelling
difficult. It is in this domain that the minimum ROP has been registered (0.33 m/h).
The rate of penetration increases boring the gneiss-micaschist alternation characterising DII, a
median value of 1.55 m/h has been achieved; that is because in this tract the general quality of rock
137
mass and its strength are lower than those observed in the previous domain (as already discussed in
Chapter 5).
The highest ROP have been calculated in the first portion of the third domain (DIIIa), where a
median of 2.22 m/h has been registered, it is probably due to the decreasing of rock mass quality
and strength, with respect to the DII. In the remaining part of micaschist domain, ROP is similar to
the one measured in DII, in this tract the median value is 1.47 m/h.
The statistical description of ROP distribution through the tunnel has been reported in Table 6.2.
ROP [m/h] TUNNEL Domain I Domain II Domain III; * DIIIa
Range 0.33 – 3.74 0.33 – 2.43 0.85 – 2.64 1.13 – 3.74* 0.62 – 3.74
Median 1.33 0.94 1.55 2.22* 1.47
Mean 1.45 0.96 1.53 2.18* 1.58
Q1-Q3 1.08 – 1.81 0.60 – 1.23 1.19 – 1.78 1.92 – 2.40* 1.18 – 1.97
Table 6.2 ROP statistical description; * DIIIa data: part of DIII data used in the following
138
6.2.2 Field penetration index
The field penetration index (FPI), is calculated as the ratio between the force acting on each cutter
and the penetration rate per revolution (see section 2.4.5).
FPI is strongly affected both by rock mass conditions and by operational ones.
There is in fact a threshold which cannot be exceeded, and it is specific for each machine and that is
independent from geological feature of excavated ground. Moreover, in some situations rock mass
strength could require high thrusts in order to obtain an acceptable advance, however, FPI is
maintained low to avoid machine damages and breakages (also when these thrusts are below the
machine threshold).
During faults or cataclastic zones crossing FPI is intentionally maintained low.
For these reasons, the TBM driving decisions and the operator experience may strongly affect the
TBM performances in terms of FPI and, in minor part, of ROP.
The FPI distribution for the entire analysed tunnel length is provided by Figure 6.2.
FIELD PENETRATION INDEX

300
FPI [(kN/cutter)/(mm/rev)]
DI DII DIIIa DIII

250
200
150
100
50
0
0 1000 2000 3000 4000 5000 6000 7000
Pk. [m]
Figure 6.2 Distribution of FPI through the tunnel, red lines represent the geological domain
changes. Green line delimitates DIIIa.
The values of FPI calculated during La Maddalena tunnel construction range between 40.90 and
101.15 kN/cutter/mm/rev (in terms of Q1-Q3) and FPI has a median value equal to 61.40
kN/cutter/mm/rev.
139
The highest median of FPI has been detected during the excavation of the gneiss domain, where
median FPI is of 136.32 kN/cutter/mm/rev. The index significantly decreases through the gneiss-
micaschist alternation (DII) which are characterised by median value of 78.52 kN/cutter/mm/rev.
About DIII it was noted that the FPI calculated in the first tract of the domain (DIIIa) does not differ
from the index relative to the whole domain, being median values equal to 45.50 and 49.92
kN/cutter/mm/rev respectively.
Table 6.3 provides the statistical description of FPI through the entire excavated length and referred
to each geological domain.
FPI
TUNNEL Domain I Domain II Domain III; * DIIIa
[kN/cutter/mm/rev]
Min. 14.20 26.86 33.52 15.16 14.20
Max. 298.61 298.61 222.98 120.42 165.10
Median 61.40 136.32 78.52 45.50 49.92
Mean 78.29 142.73 94.12 47.30 56.07
Q1 40.90 102.68 55.76 35.66 35.40
Q3 101.15 173.25 138.91 55.12 70.95
Table 6.3 FPI statistical description; * DIIIa data: part of DIII data used in the following
Generally, it was noted that the FPI trend is related to the one concerning the quality, the strength
and the fracturing degree of the rock mass. It is particularly evident by observing the passage from
the massive gneiss, characterised by high quality and strength, to the gneiss-micaschist alternation
domain, in which the quality decreased and fracturing degree increased. The FPI is high in DI
where a high cutterhead force is required to obtain a low penetration and FPI is lower in DII where
a reduced thrust is necessary to register a good penetration rate.
The histograms related to the ROP and FPI distribution through the tunnel have been reported in the
following page (Figure 6.3).
Figure 6.3 ROP and FPI histograms.
140
141
6.3 Conclusion
6.3 Conclusion
During this work, La Maddalena TBM performances have been collected and analysed, in terms of
ROP and FPI.
It is possible to note that performances indices improve in the verse of the excavation, in particular,
the worst boring conditions have been registered during gneiss excavation, where rock mass high
quality make tunnelling difficult.
Generally, in those situations characterised by massive and hard features of ground, TBM
performances are low: the thrust required to bore the rock face is high whereas the penetration is
low. It determines high levels of FPI and low values of ROP.
Working decisions not related to the geological and geomechanical framework may affect the TBM
advance.
142
7 COMPARISON BETWEEN DIRECT
GEOSTRUCTURAL SURVEY AND SEISMIC
METHODS FOR PREDICTING GEOLOGICAL
FRAMEWORK AHEAD TUNNEL FACE
7.1 Introduction
It is well known that during design phase and tunnel excavation the knowledge of the geological,
structural and mechanical properties of rock masses is one of the most important requirements in
order to select the best construction method and guarantee the required safety levels.
This is usually accomplished by means of geomechanical and structural observations with local
sampling (direct surveys) on site, particularly on the tunnel walls when they are accessible, for
example when the excavations is carried out by means of an open type TBM. However, if the
access to the tunnel walls is not allowed and the direct surveys not possible (when the excavation is
realised by means of a closed type TBM or when the ground conditions are so poor that wall
mapping is not safe), indirect methods, like geophysical ones, can be applied. Indirect methods
suffer however of uncertainties and resolution limitations that may affect the obtainable results.
This part of the study aims to compare the geological, structural and geotechnical features of rock
masses gathered by direct geostructural surveys and the results obtained by an indirect seismic
technique; the adopted one is the Tunnel Seismic Prediction (TSP).
In particular, the reliability of TSP in terms of predicting lithological variations, physical and
mechanical properties of intact rock, rock mass fracturing degree and orientation of major
discontinuity elements have been evaluated and discussed.
With this aim, 8 TSP measurements, covering a total length of 1500 meters, between pk. 500 m and
pk. 2000 m, have been analysed. These measurements have been carried out in order to predict the
rock mass features ahead of the excavation front.
143
7.2 Comparison of the results

The purpose of this study is to compare the geological, structural and mechanical data collected
during traditional survey and laboratory analyses with the information obtained by the TSP
investigation. In particular the purpose is to test the reliability of this method in terms of:
- Prediction of significant lithological changes;
- Estimation of mechanical and physical parameters describing intact rock;
- Estimation of fracturing degree of rock masses and presence of shear or weakness zones
prediction;
- Estimation of orientation of the principal elements of discontinuities.
7.2.1 Prediction of significant lithological changes
During “La Maddalena” design phases, many geological field surveys have been performed. These
surveys allowed to depict the geological limit between the gneiss of the Ambin complex and the
micaschists of Clarea complex. However, given the significant depth of the planned tunnel, it was
not possible to establish the exact interaction of this limit with the tunnel axis before tunnel
excavation. The mechanical properties of the two rock types are indeed very different so that their
boreability is not the same; for this reason, the knowledge of limit location was very important.
TSP measurements showed a significant reduction of P waves (from an average wave velocity of
about 5500 m/s to 3500 m/s) and S waves (from about 3250 m/s to 2000 m/s) as shown in Figure
7.1.
P wave velocity S wave velocity

7000 4000
6000
5000 3000
vP [m/s]
vS [m/s]
4000
2000
3000
2000 1000
1000
0 0
400 900 1400 1900 400 900 1400 1900
Pk. [m] Pk. [m]
Figure 7.1 P wave velocity (on the left) and S wave velocity (on the right) along the tunnel (Armetti
et al., 2017).
144
This sudden reduction of P and S wave velocities occurs at pk. 1360 m, where also direct geological
surveys after the tunnel excavation have detected the passage from gneiss to micaschists. P waves
show a little reduction in velocity before the most important one in the transition zone from Ambin
to Clarea complex.
7.2.2 Estimation of fracturing degree of rock masses and presence of shear or weakness
zones prediction
The occurrence of high fractured zones in rock masses generally reduces the global quality of them
and the safety of excavation processes, so that support and reinforcement elements become
fundamental in order to guarantee the advance of works and the safety of workers.
The fracturing degree of a given rock mass cannot be evaluated only by observation of tunnel walls
and it has to be predicted ahead the excavation front.
Both the type of rock and the characteristics of rock mass (e.g. global quality) affect the velocity of
seismic waves; the wave propagation is higher into good quality (low fracturing) grounds and is
lower into poor quality (high fracturing) materials. Generally, a low seismic signal velocity
propagation denoted a fracturing rock mass.
For each section of homogeneous TSP velocity, the rock mass fracturing degree has been evaluated
on the basis of traditional geological survey. The fracturing index has been computed in terms of
distance between joints (spacing), by the following formula:
𝐿
𝑆= [𝑐𝑚] (7.1)
𝑛
where L is the length where the velocity of wave is homogeneous (Figure 7.2) and n is the number
of joints detected in that distance.
145
Figure 7.2 The length L considered in Spacing calculation is the one where homogeneous wave
velocity has been measured.
The velocity of seismic wave propagation and the spacing values have been compared. From these
results, a clear correlation between P and S wave velocity and degree of fracture is not observable in
the two investigated rock types (Figure 7.3), probably because the seismic velocity is most affected
by the intact rock properties than by its fragmentation degree.
vP-S vS-S
600 600
500 500
400 400
S [cm]
S [cm]
300 300
200 200
100 100
0 0
0 2000 4000 6000 8000 0 1000 2000 3000 4000
Vp [m/s] Vs [m/s]
Figure 7.3 P and S wave velocity related to fracture spacing (Armetti et al., 2017).
There was an attempt to evaluate spacing considering a constant L and the calculation has been
carried out adopting different L values (L = 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 m) but did not allow for more
reliable correlations.
146
The degree of fracturing of rock mass has been also evaluated by means of the volumetric joint
count, Jv (Palmström, 1982). It has been related to the wave velocity as reported in Figure 7.4.
vP-Jv
10
9
8
7
Jv [n/m3]
6
5 R² = 0.4324
4
3
2
1
0
4800 5000 5200 5400 5600 5800 6000
vP [m/s]
Figure 7.4 P wave velocity related to Jv.
A more reliable correlation between seismic wave velocity and the fracturing degree of rock mass
has been obtained by using Jv as indicator of rock mass fragmentation. It is due to the fact that
seismic waves propagate into volumes of rock masses and Jv considers the number of
discontinuities per unit volume; in other words, waves propagate into three dimensions and Jv
represents the fracturing degree into three dimension portion of rock. The extremely poor
correlation with spacing is probably because it considers only the linear distance between joints, but
the problem of wave propagation has to be considered as three dimensional one.
7.2.3 Estimation of orientation of the principal elements of discontinuities
The intersection angles of principal discontinuities with tunnel axis are one of the most relevant
aspects involved in tunnel stability analysis. The orientation of the principal elements of
discontinuity (e.g. faults, fractures, cataclastic bans, etc.) is estimated after geostructural surface
survey and tunnel walls mapping; however the TSP method is able to give information about the
weakness plane orientations, since they are planes of acoustic impedance.
Along the considered tunnel tract, TSP survey detected 103 planes of discontinuities; the method
gives the spatial position and the orientation in terms of dip and dip direction for each of them.
The distance between the predicted plane and the mapped one has been evaluated as follows
(Equation 7.2) in order to compare these data with data achieved during tunnel walls mapping:
147
∆𝐿 = |𝐿𝑇𝑆𝑃 − 𝐿𝑀𝐴𝑃 | (7.2)
where LTSP is the plane detected by TSP measurements and LMAP is the plane observed during the
geological survey.
A graphical representation of ΔL is reported in Figure 7.5.
∆𝐿 = |𝐿𝑇𝑆𝑃 − 𝐿𝑀𝐴𝑃 |
ΔL
Tunnel axis
TSP detected Directly detected

fracture fracture
Not in scale
Figure 7.5 Distance between mapped and TSP detected discontinuity.
The fractures have been subdivided into three categories on the basis of ΔL:
- If ΔL < 0.30 m a perfect correspondence among the two methods has considered;
- If 0.30 m < ΔL < 1.00 m the correspondence is considered acceptable;
- If ΔL >1.00 m there are no correspondence.
The concept of fracture division into three classes has been graphically explained by means of the
schemes in Figure 7.6.
148
Perfect correspondence between TSP and directly detected fractures

∆𝐿 < 30 𝑐𝑚
0 cm 30 cm
Tunnel axis
TSP detected Directly detected Not to scale

fracture fracture
Acceptable correspondence between TSP and directly detected

fractures 30 𝑐𝑚 < ∆𝐿 < 100 𝑐𝑚
0 cm 30 cm 100 cm
Tunnel axis
TSP detected Directly detected Not to scale

fracture fracture
No correspondence between TSP and directly detected fractures

∆𝐿 > 100 𝑐𝑚
0 cm 30 cm 100 cm
Tunnel axis
TSP detected Directly detected

Not to scale
fracture fracture
Figure 7.6 Three categories of fractures on the basis of ΔL.
149
The number of occurrences for each category is summarised in Table 7.1.
∆𝑳 Number of fractures
ΔL < 0.30 56
0.30 < ΔL < 1.00 35
ΔL > 1.00 21
Table 7.1 Number of discontinuity elements for each ΔL category (Armetti et al., 2017).
A comparison between directly mapped and TSP detected orientations has been attempted only for
discontinuities with a ΔL < 1.00 m (“perfect” and “acceptable” correspondence). The results carried
out from this comparison show that the dip direction is predicted with good accuracy while the dip
is over estimated, as it can be observed by the two stereographic projections reported in Figure 7.7.
Directly detected fractures TSP detected fractures
Figure 7.7 Stereographic projection of poles of directly detected discontinuities (left) and poles of
discontinuities detected by means of TSP survey (right).
A statistical analysis of joint orientations has been carried out, distinguishing the “perfect
correspondence” class (ΔL < 0.30 m) from the “acceptable correspondence” one (0.30 m < ΔL <
1.00 m). The results are reassumed in the frequency diagrams reported in Figure 7.8.
It is clear that the most frequent difference (Δ dip direction) between observed and TSP estimated
dip direction is 10°.
150
Δ DIP DIRECTION Δ DIP DIRECTION

0.35 0.35
0.30 0.30
Relative frequency
Relative frequency
0.25 0.25
0.20 0.20
0.15 0.15
0.10 0.10
0.05 0.05
0.00 0.00
10 20 30 40 50 60 70 80 90 10 20 30 40 50 60 70 80 90
Class [°] Class [°]
Figure 7.8 Histogram of frequency about the dip direction differences between directly observed
and TSP detected fractures ΔL < 0.30 m on the left and 0.30 m < ΔL < 1.00 m on the right. (Armetti
et al., 2017).
The most common differences (Δ dip) between observed and TSP dips are 30° for ΔL < 0.30 m
class and 50° for 0.30 m < ΔL <1.00 m (as shown in the diagrams of frequency in Figure 7.9).
Δ DIP Δ DIP
0.35 0.35
0.30 0.30
Relative frequency
Relative frequency
0.25 0.25
0.20 0.20
0.15 0.15
0.10 0.10
0.05 0.05
0.00 0.00
10 20 30 40 50 60 70 80 90 10 20 30 40 50 60 70 80 90
Class [°] Class [°]
Figure 7.9 Histogram of frequency about the dip differences between directly observed and TSP
detected fractures ΔL < 0.30 m on the left and 0.30 m < ΔL < 1.00 m on the right. (Armetti et al.,
2017).
The overestimation fracture dips is probably due to the working principles of TSP system that is
most influenced by those planes perpendicular to the tunnel axis than by the parallel and sub-
parallel ones.
151
7.3 Conclusion
7.3 Conclusion
The study summarised in this chapter aims to check the reliability of TSP investigation system,
adopted during “La Maddalena” tunnel construction to predict the geological and mechanical
properties of rock mass ahead the tunnel face, in comparison with direct geostructural surveys.
Three aspects have been investigated during this analysis:
- TSP method was completely reliable as for the individuation of main lithological changes;
in fact, the passage from Ambin complex to Clarea one predicted with accuracy;
- A quite relevant relationship between the joint volumetric count (Jv) and P wave velocity
has been observed. It means that the seismic velocity obtained with TSP method is also
influenced by fracturing degree of rock mass: in other words, the wave velocity can be
involved to predict the degree of fragmentation of a given rock mass.
- As for the orientation and the position of the principal planes of weakness, the method has
been considered satisfactory; however the angle with tunnel axis is overestimated. This
overestimation is because the TSP, due to its principles of working, is mainly affected by the
planes perpendicular to tunnel axis.
In conclusion Tunnel Seismic Prediction method can be considered reliable in the scopes for which
it has been adopted.
152
8 LITERATURE MODELS FOR PREDICTING TBM
PERFORMANCES AND THEIR APPLICATION TO “LA
MADDALENA” EXPLORATORY TUNNEL
This chapter is divided into two parts:
- In the first one, a brief description of the principal TBM performance predictive methods
has been provided. The three most relevant groups of predictive models (based on
theoretical, empirical and artificial intelligence approaches) have been synthetically
illustrated and for each group the models which are mainly involved in tunnel projects have
been described. Two tables show what geological, mechanical or technological parameters
are considered in the proposed model.
- The second section reports the results of the application of the principal models proposed in
literature, to “La Maddalena” case study. In particular the empirical predictive equations
proposed by different authors have been considered to predict TBM performance at “La
Maddalena” tunnel, then the predicted data have been compared with the measured ones.
153
8.1 Approaches for predicting TBM performance

8.1.1 Introduction
In mechanized tunnel excavation, Tunnel Boring Machine (TBM) performance has a very dominant
impact on the times and costs of tunnel realisation: for this reason they must be accurately
predicted. TBM performance prediction is essential to select the most effective method of
tunnelling and to generally estimate time and the economic costs of infrastructure realisation.
In other words, the design of tunnel projects and the selection of construction methods require the
effective TBM performance prediction to make mechanized tunnelling cheaper, safer and more
effective than the “drill and blast” method.
It is well known that the TBM performance is affected by several factors such as geological site
conditions (lithological features, fracture frequency, joint orientation, faults and bedding related to
tunnel axis, rock bursting, swelling, underground water condition, etc.), geotechnical properties
(strength, hardness, brittleness, abrasiveness, etc.) and machine related parameters (machine
diameter, cutter disposition and geometry, disc diameter, thrust, torque, etc.).
TBM were introduced in 1950s and since then many researchers (Tarkoy, 1975; Ozdemir, 1977;
Blindheim, 1979; Nelson and O’Rourke, 1983; Rostami and Ozdemir, 1993; Alber, 1996, 2000,
2008; Barton, 2000; Sapigni et al., 2002; Yagiz, 2002, 2006, 2008; Gong and Zhao, 2009; Hamidi
et al., 2010; Hassanpour et al., 2011; Macias, 2014, Rostami, 2016; Salimi et al., 2016; Adoko et
al., 2017; Armetti et al., 2018, and many others) have developed several methods that allow to
estimate the TBM performance, by analysing cutting forces (theoretical models) or the interaction
between the intact rock and rock mass features and the operating characteristics of the TBM
(empirical models).
The TBM performance predictive models are based on different approaches; however the
theoretical, based on laboratory tests and cutting forces, and the empirical ones, based on field TBM
performance and geotechnical properties of rocks, are the most common, but also artificial
intelligence techniques, including artificial neural network, support vector machine, particle swarm
optimisation and fuzzy inference systems have been developed for this purpose (Alvarez Grima et
al., 2000; Benardos and Kaliampakos, 2004; Yagiz et al., 2009; Yagiz and Karahan, 2011;
Mahdevari et al., 2014; Ghasemi et al., 2014).
154
8.1.2 Theoretical models for predicting TBM performance
The theoretical models are based on the analysis of breakage mechanisms and on the study of the
cutting forces acting on cutters to define a force equilibrium relation (Crow, 1975; Ozdemir et al.,
1978; Roxborough and Phillips, 1975; Sanio, 1985; Sato et al., 1991; Snowdon et al., 1982). The
first developed theoretical models focused on full-scale laboratory cutting tests or indentation tests
in various rock types and gave an estimation of cutting forces based on cutter and cutting geometry,
spacing and penetration of the cut (Khademi Hamidi et al., 2010). The estimated cutting forces are
considered to evaluate cutterhead thrust and torque and to estimate the performance of a TBM with
certain specifications (Rostami, 2016).
Nevertheless, these models are difficult to apply, both because they require several laboratory tests,
which sometimes do not represent the real rock masses behaviour during tunnels excavation and
because the test equipment required is not often available in the research centres. A positive aspect
of these methods is that they provide information about rock fragmentation and into applied force-
TBM penetration behaviour of rock masses (Khademi Hamidi et al., 2010).
The main advantages and disadvantages of theoretical modes are reported in Table 8.1.
Advantages Disadvantages
- Flexible with machine - Unable to easily consider rock

characteristic and cutter geometry; mass and intact rock parameters;
- Can be adopted for thrust and - Unable to take into account the
torque optimization; presence of joints;
- Can be used for cutterhead design - Advised against in jointed rock

and its optimization; masses;
- Can be involved to understand the - Unable to account for required

effective working condition of the field adjustment.
cutters and related forces.
Table 8.1 Principals advantages and disadvantages of theoretical modes (modified after Rostami,
2016).
The model developed by the Colorado School of Mines, CSM (Rostami, 1997; Rostami and
Ozdemir, 1993) is the most frequently used one of the group. CSM can be considered as a semi-
theoretical model. It begins from the individual cutter force calculation in to define the overall
thrust, torque and power requirement of the entire cutterhead (Delisio and Zhao, 2014).
155
The original model is based only on intact rock properties and later Yagiz (2002) and
Ramezanzadeh et al. (2008) modified it taking into account the influence of rock mass parameters
as the angle between tunnel axis and weakness planes (α).
The CSM output is the penetration of TBM into rock mass (m/h or mm/min). The list of intact rock
and rock mass characteristics and machine related parameters involved in CSM model are
summarised in Table 8.2.
Rock and rock mass

Machine parameters Output
parameters
- Uniaxial compressive - Cutter load - Penetration

strength (UCS); capacity(FN); [m/h; mm/min]
- Brazilian tensile - Cutter spacing

strength (BTS); (Scutter);
- Cerchar Abrasivity - Cutter diameter
Index (CAI); (Øcutter);
- Rock quartz content - Cutter number (ncutter);
(q); - Cutter tip width (Wtip)
- Joint spacing (Js); - TBM thrust (FTBM);
- Angle between tunnel - TBM torque (MTBM).
axis and the planes of
weakness (α).
Table 8.2 List of intact rock and rock mass characteristics and machine related parameters
involved in CSM model.
156
8.1.3 Empirical models for predicting TBM performance
Empirical models are based on the back-analysis of geological, geotechnical and TBM performance
collected during a tunnel realisation and the predictive equations proposed are carried out from the
statistical analysis of the dataset. In other words, they consist of empirical correlations between
geological-geomechanical parameters and the actual TBM performance. This group of models
allows to estimate the TBM performance in a geological site with given features and are generally
site-specific.
The main advantages and disadvantages affecting empirical models have been listed in Table 8.3.
Advantages Disadvantages
- Based on field performance of - Very sensitive at joint properties;

TBM and of rock features; - Unable to consider machine
- Consider the TBM as the whole specifications, cutter geometry
system; and cutter spacing;
- Particular importance is given at - Applicable only to similar tunnel
geological site condition; project.
- Account of many of field
adjustment.
Table 8.3 Principals advantages and disadvantages of empirical modes (modified after Rostami,
2016).
The geological, geotechnical and mechanical properties of intact rocks and rock masses are
considered as independent variables (input) and the TBM performance as dependent ones (output).
The influences of the first ones on the machine operational parameters are evaluated using statistical
regression analysis.
Empirical predictive models can be divided into two large groups: the single parameter models and
the multiple parameter ones.
The first type of models is easy to apply because it includes a single intact rock parameter such as
the uniaxial compressive strength, UCS (Graham, 1976; Farmer and Glossop, 1980; Hughes, 1986;
Armetti et al., 2018), or rock matrix index, such as the drilling rate index (DRI), punch penetration,
Taber abrasion, brazilian tensile strength, shore harness and point load index (Tarkoy, 1973;
Blindheim, 1979; Nelson et al., 1983, 1985; Bamford, 1984; Wijk, 1992; Dollinger et al., 1998) and
one machine index. They can be utilised in a limited range of cases and generally tend to
157
underestimate the actual TBM performance due to the lack of some relevant geological parameters
(i.e. because they are hard to collected or not available).
Since the TBM performance does not depend on a single rock feature but by the combination of
more of them several multiple parameter models have been proposed later (Sundin and Wanstedt,
1994; Bruland, 1998; Nelson et al., 1999; Alvarez Grima et al. 2000; Zhao et al., 2007; Yagiz, 2008
and many others). This type of approach is able to account for both more than one rock parameters
and TBM ones, however the typical disadvantage is to understand the relation type between the
rock and machine parameters. Researchers commonly carry out linear or non-linear multiple
regression analyses and the best combination of rock mass parameters to predict the TBM
behaviour in tunnelling is selected (Alber, 1996; Yagiz, 2008; Gong and Zaho, 2009; Hassanpour et
al., 2011; Farrock et al., 2012). The coefficient of determination (R2) is adopted as indicator of
correlation strength.
Some predictive models aim to correlate rock mass geomechanical quality estimating by rock mass
classification systems with TBM performances:
- Rock Structure Rating (RSR, Wickham et al., 1972) was considered by Cassinelli et al.
(1982) and Innaurato et al. (1991);
- Rock Mass Rating (RMR, Bieniawski, 1989) was used by Sapigni et al. (2002), Ribacchi
and Lembo Fazio, (2005), Hassanpour et al. (2009), Armetti et al. (2018);
- Rock Mass Quality Index (Q, Barton et al., 1974) by Burton (2000);
- Geological Strength Index (GSI, Hoek, 1994; Marinos and Hoek, 2000) by Hassanpour et al.
(2009) and Armetti et al. (2018).
Bieniawski et al. (2006) developed the Rock Mass Excavability index, RME, based on RMR
classification system. RME was carried out from over 400 sections of tunnels and has been updated
many times with more case histories up until now (Bieniawski et al., 2007 a,b; Bieniawski and
Grandori, 2007).
RME index is based on five input parameters: UCS of intact rock, abrasivity, rock mass fracturing
degree, stand-up time of the excavation and groundwater conditions.
The ratings for each of the five parameters are summarised in Table 8.4.
158
Table 8.4 RME index input ratings (Bieniawski and Celada).
The assigned ratings are summed in order to give the RME value. Several statistical correlations
have been carried out between RME and TBM performance index, as coefficient of utilization,
average rate, penetration rate and specific energy.
One of the empirical models most widely used nowadays around the world is the Norwegian
University of Science and Technology method (NTNU) developed by Bruland (1998); it is based on
empirical correlation between geological and geotechnical properties of rock masses and the actual
TBM performance. NTNU method takes into account some machine parameters and provides the
penetration rate, the advance rate and the utilization factor as the output indices.
The list of intact rock and rock mass characteristics and machine related parameters involved in
NTNU method is summarised in Table 8.5.
159
Rock and rock mass

parameters
- Uniaxial compressive - Cutter thrust (FN); - Penetration Rate

strength (UCS); - Cutter spacing (PR);
- Cerchar Abrasivity (Scutter); - Advance Rate (AR);
- Utilization Factor (U).
Index (CAI); - Cutter diameter
- Cutter Life Index (Øcutter);
(CLI); - Cutterhead revolution
- Drilling Rate Index per minute (RPM);
(DRI); - Installed cutterhead
- Porosity; power.
- Rock quartz content
(q);
- Joint spacing (Js);
- Angle between tunnel
axis and the planes of
weakness (α).
involved in NTNU model.
QTBM method proposed by Barton (2000), is based on Q-system with the addition of some
parameters as the quartz content, cutter life index and the stress on tunnel face.
QTBM method is able to predict TBM performance; however it is difficult to use it in practical
applications because it includes several parameters, in addition some of them are overlapped (Gong
and Zhao, 2009).
The list of intact rock and rock mass characteristics and machine related parameters involved in
QTBM method are summarised in Table 8.6.
160
Rock and rock mass

parameters
- Rock Quality - Average cutter thrust - Penetration Rate (PR);

Designation (RQD); (F); - Advance Rate (AR).
- Number of joints (Jn);
- Joint roughness (Jr);
- Joint alteration (Ja);
- Joint water condition
(Jw);
- Stress Reduction
Factor (SRF);
- Quartz content (q);
- Rock Mass Strength
(SIGMA).
- Cutter Life Index
(CLI);
- Average biaxial
strength on tunnel face
(σθ)
involved in QTBM method.
Delisio et al. (2014) developed a predictive model suitable when “blocky rock condition” occurs.
“Blocky rock condition” means that there are face instabilities in jointed rock masses, where the
combined effect of the structure of rock mass and the in-situ stress conditions causes a
fragmentation process of the tunnel face, which can became “blocky” (Delisio et al., 2013).
Blocky rock conditions may entail a significant impact on the TBM performance such as the
increase of cutter/cutterhead wear caused by the dynamic impact of the rock blocks against cutters
or the damages to the muck conveyor system generated by blocks in the muck.
These conditions may generally result in negative effects that increase the TBM maintenance
requirements and lead to higher construction times and costs. (Delisio et al., 2014).
Some of the most common empirical formulations are reported in Table 8.7.
161
−0.2
𝑃𝑅 = 5 ∗ 𝑄𝑇𝐵𝑀
Barton, 2000
𝑅𝑄𝐷 𝐽𝑟 𝐽𝑤 209 𝑆𝐼𝐺𝑀𝐴 20 𝑞 𝜎𝜃
𝑄𝑇𝐵𝑀 = ∗ ∗ ∗ ∗ ∗ ∗
𝐽𝑛 𝐽𝑎 𝑆𝑅𝐹 𝐹10 𝐶𝐿𝐼 20 5
Bieniawski, 2007 𝐴𝑅𝐴 = 0.422𝑅𝑀𝐸07 − 11.61
Cassinelli et al., 1982 𝑃𝑅 = −0.0059𝑅𝑆𝑅 + 1.59
Gong and Zhao, 2009 𝐵𝐼 = 37.06 ∗ 𝑈𝐶𝑆 0.26 ∗ 𝐵𝑖 −0.10 ∗ (0.84𝑒 −0.05𝐽𝑉 + 𝑒 −0.09∗sin(𝛼+30) )
𝐹𝑃𝐼 = 9.401 + 0.397 ∗ 𝐿𝑜𝑔𝛼 + 0.011 ∗ 𝐽𝑐2 + 1.14𝐸 − 5 ∗ 𝑅𝐷𝑄3 + 1.32𝐸
Hamidi et al., 2010
− 8 ∗ 𝑈𝐶𝑆 4
𝐹𝑃𝐼 = 0.222𝐵𝑅𝑀𝑅 + 2.755
Hassanpour et al. 2009 𝐹𝑃𝐼 = 9.273𝑒 0.008𝐺𝑆𝐼
𝐹𝑃𝐼 = 11.718𝑄0.098
𝐹𝑃𝐼 = 𝑒𝑥𝑝(0.008𝑈𝐶𝑆 + 0.015𝑅𝑄𝐷 + 1.384)
Hassanpour et al., 2011 0.06 ∗ 200 ∗ 7
𝑅𝑂𝑃 =
𝐹𝑃𝐼
Innaurato et al., 1991 𝑃𝑅 = 𝜎𝑐−0.437 − 0.0437𝑅𝑆𝑅 + 3.15
−0.66
𝑆𝑃 = 250 ∗ 𝜎𝑐𝑚
Ribacchi and Fazio, 2005

𝑅𝑀𝑅 − 100
𝜎𝑐𝑚 = 𝜎𝑐 ∗ 𝑒𝑥𝑝 ( )
18
𝑅𝑂𝑃 = 1.093 + 0.029 ∗ 𝑃𝑆𝐼 − 0.003𝑈𝐶𝑆 + 0.437 ∗ 𝐿𝑜𝑔(𝛼) − 0.219
Yagiz, 2008
∗ 𝐷𝑃𝑊
PR = Penetration Rate [m/h]; ROP = Rate of Penetration [m/h]; Bi = Boreability Index
[(kN/cutter)/(mm/rev)]; CLI = Cutter Life Index; SP = Specific Penetration [(mm/rev)/(kN/cutter)]; FPI =
Field Penetration Index [(kN/cutter)/(mm/rev)]; ARA = Average Rate of Advance; Q = Rock Tunnelling
Quality Index; QTBM = Q-system classification applied to TBM; RMR =Rock Mass Rating; RME = Rock
Mass Excavability; BRMR = Basic Rock Mass Rating; GSI = Geological Strength Index; RSR = Rock
Structure Rating; UCS = Uniaxial Compressive Strength [MPa]; RQD = Rock Quality Designation [%]
(Deere et al., 1967); Jn = Joint Set number; Jr = Joint Roughness number; Jw = Joint Water Reduction
factor; Ja = Joint Alteration number; SRF = Stress Reduction Factor; SIGMA = Rock Mass Strength
[MPa]; F10 = average cutter force [tnf]; q = quartz content [%]; σθ = average biaxial stress on tunnel face
[MPa]; Jv = total number of joints per m3 [m-3]; Jc = Rating for Joint conditions; PSI = Peak Slope Index
[kN/mm]; DPW = Distance between Planes of Weakness (spacing) [m]; σc = rock material uniax. compr.
strength [MPa]; σcm = r. m. uniax. compressive strength [MPa]; α = angle fractures/tunnel axes [deg].
Table 8.7 Principal empirical predictive models.
162
8.1.3.1 Selection of the most appropriate model
An empirical predictive model is carried out by analysing the relationship between rock and rock
masses features (independent variables) and TBM specification measured on site (dependent
variables). The literature proposes a large variety of relationships but it is impossible to know
beforehand what kind of correlation occurs in each specific case. For example, in a certain case the
relation between UCS and ROP is of a linear type while in other cases it may be nonlinear;
furthermore, the parameters involved in a model may be different from those of another one. For
these reasons several models are usually developed for the same site, trying different combination
of input and different relation type: generally the influence of each independent variable on the
dependent ones is evaluated using forward stepwise regression analyses; at this point it is necessary
to select the most representative model.
A predictive model is reliable if the measured and the calculated data are comparable. There are
different ways to choose a formulation:
- Comparison of different coefficient of determination (R2). Generally, the model with the
higher R2 is considered the most representative; however, this coefficient increases with the
number of inputs. This is why the adjusted coefficient determination, R2adj, that takes into
account the number of independents variables, has to be considered;
- T-test is adopted to evaluate the significance of R2 for each variable comparing the
computed and the tabulated t-value, using the null hypothesis. When the computed value is
higher than the tabulated one the null hypothesis (that is a testable statement opposed to the
hypothesis considered) is rejected and the R2 is significant. If the greater value is the
tabulated one, null hypothesis cannot be refused, so R2 is not significant and the model
should be modified;
- The significance of the regression is tested by means of the analysis of variance, F-test. If
the computed F-value is higher than the tabulated one, the null hypothesis will be refused
and that signifies a real relationship between independent and dependent variables and the
generated model can be used. Otherwise it is not reliable;
- The statistical significance of a model or of an independent variable involved into a model is

also evaluated considering the significance value, p-value. If the p-value is less than the
significance level the resulted formulation is considered statistically significant, otherwise a
new selection of input and relation type must be carried out. Generally, the statistical level is
163
fixed at 0.05; in other words, it means that a statistically significant results has less than 5%
of occurring from chance;
- Exclusion of a formulation as a consequence of multicollinearity that occurs when two or

more independent variables are related. In these cases, erroneous results can be obtained
because of the redundancy of input. Multicollinearity is evaluated by means of the variance
inflation factor, VIF that is the coefficient of determination of each variable to the others.
VIF ranges from 1 to infinity; VIF equal to 1 means no linear relation between independent
variables and VIF higher than 10 indicates multicollinearity problems (Montgomery and
Peck, 1992);
- Some authors (Johnson and Omland, 2004; Feng and Jimenez, 2015) recommend using a
selection criterion able to consider the fitness and the complexity of the model, that is a
selection criterion based on information criterion (Akaike information criterion, AIC), the
deviance information criterion (DIC, Spiegelhalter et al., 2002; Lunn et al., 2012) and the
Bayesian information criterion (BIC);
- Graphical method: finally, as additional test the predicted and the measured parameters (as
ROP or AR) are plotted on the same Cartesian plane. The empirical model is more able to
predict the TBM performances more the calculated and the measured data are closely
aligned on the bisector. The Figure 8.1 shows a reliable equation (point closely aligned on
the bisector) and an equation that must be refused (calculated and observed data are not
comparable).
3.00
ROP predicted [m/h]
2.00
1.00
0.00
0.00 1.00 2.00 3.00
ROP measured [m/h]
Figure 8.1 Graphical results of reliable equation, on the left (Yagiz, 2007) and not reliable one on
the right.
164
Table 8.8 shows the principal theoretical and empirical predictive model proposed in literature. For
each of them the intact rock, rock mass and machine considered parameters are specified.
It is observable that there is a high utilisation of mechanical rock properties and rock masses site
characteristics and a low frequency of machine indices is registered.
Uniaxial compressive strength is the most involved parameter, both for its easy determination and
representativeness. Great importance is given to the joint sets both in terms of frequency (RQD and
spacing) and of their orientation in comparison the tunnel’s one.
The largest part of models, in particular the empirical group, does not take into account TBM or
cutter specifications to carry out predictive formulation. However, given that also machine features
influence the excavation processes, some models consider them and provide a most complete study
of all parameters implicated in the boring (CSM, 1993; Gehring, 1995; NTNU, 1998).
The most relevant models and the parameters that they involve have been listed in Table 8.8.
165
8.1.4 Final observation about theoretical and empirical predictive models
In the table reported in the following page (Table 8.8), the most adopted theoretical and empirical
TBM performance predictive models have been listed with the purpose of showing what parameters
are more frequently considered into TBM performance prediction equations.
13 studies have been considered and the most of them is based on the empirical approach.
As for intact rock parameters the most utilized is the uniaxial compressive strength followed by the
abrasivity index and the quartz content. Other parameters are sometimes considered, such as the
Brazilian tensile strength and the peak slope index.
The most frequently considered rock mass features are fragmentation, expressed both by means of
spacing and RQD, and the angle between the most relevant set of discontinuities and the orientation
of tunnel.
The machine specifications which are more frequently taken in consideration by theoretical and
empirical approaches are the diameter of machine, the force acting on cutterhead and the number of
cutter.
The most carried out indices are the net penetration rate, the advance rate, FPI and coefficient of
utilization are less evaluate.
Barton, CSM and NTNU are the models that involve the largest quantity of rock, rock mass and
machine related parameters.
166
REFERENCES METHOD ROCK INPUT TBM INPUT OUTPUT
INTACT ROCK PARAMETERS ROCK MASS PARAMETERS
TE EM UCS BTS PSI CAI q RQD DPW α Jc JW GSI other Øm Øc N TF RS Scutt PR AR Prev FPI SP
[mm/rev]
kN/mm
[MPa]
[MPa]
[mm]
[m/h]
[m/h]
[deg]
[m/s]
[kN]
[cm]
[%]
[%]
[m]
[m]
[m]
[-]
[-]
[-]
*
Alber (1998)
Barton (2000)
Cassinelli et al. (1982)
CSM (1993,2006,2010)
Farrokh et al. (2012)
Ghering (1995)
Gong and Zhao (2009) BI

Hamidi et al. (2010)
Hassanpour et al. (2011)
Innaurato et al. (1990)
NTNU (1998)
Ribacchi and Lembo Fazio (2005)
Yagiz (2008)
FREQUENCY 2 11 12 3 1 4 3 5 10 9 2 3 2 1 4 2 3 5 1 2 7 1 1 2 1
TE = theoretical models; EM = empirical models; UCS = Uniaxial Compressive Strength [MPa]; BTS = Brazilian Tensile Strength [MPa]; PSI = Peak Slope
Index [kN/mm]; CAI = Cerchar Abrasivity Index; q = quartz content [%]; RQD = Rock Quality Designation [%] (Deere et al., 1967); DPW = Distance between
Planes of Weakness (spacing) [m]; α = angle between fractures and tunnel axes [deg]; Jc = Rating for Joint conditions; Jw = Joint Water Reduction factor; GSI =
Geological Strength Index; Øm = TBM diameter [m]; Øc = cutter diameter [mm]; N = number of cutter [-]; TF = Total Force to the cutterhead [kN]; RS =
Rotational Speed [m/s]; Scutt = Cutter Spacing [cm]; PR = Penetration Rate [m/h]; AR = Average Rate [m/h]; Bi = Boreability Index [(kN/cutter)/(mm/rev)];
FPI = Field Penetration Index [(kN/cutter)/(mm/rev)]; Prev = Penetration per Revolution [mm/rev]; SP = Specific Penetration [(mm/rev)/(kN/cutter)]; (other:
CLI = Cutter Life Index; Jn = Joint Set number; Jr = Joint Roughness number; Ja = Joint Alteration number; SRF = Stress Reduction Factor; SIGMA = Rock
Mass Strength [MPa]; F10 = average cutter force [tnf]; σθ = average biaxial stress on tunnel face [MPa]; Jv = total number of joints per m3 [m-3]; σcm = rock mass
uniaxial compressive strength [MPa]).
Table 8.8 List of intact rock, rock mass and machine parameters involved in the most relevant theoretical and empirical models.
167
8.1.5 Artificial intelligence techniques for predicting TBM performance
Apart from the theoretical and empirical models, artificial intelligence technique (AI) has been
introduced to predict the TBM performance by several researches (Alvarez Grima et al., 2000;
Benardos and Kaliampakos, 2004; Yagiz et al., 2009; Yagiz and Kaharan, 2011; Mahdevari et al.,
2014; Ghasemi et al., 2014).
Artificial intelligence techniques include:
- Artificial neural network, ANN;
- Particle swarm optimization, PSO;
- Support vector machine;
- Fuzzy logic.
Many authors developed hybrid algorithms such as neuro-fuzzy or PSO-ANN: the aim is to
improve the performance of the artificial neural network through the use of algorithms of
optimization such as PSO, generic algorithm, GA, or the imperialism competitive algorithm, ICA
(Alvarez Grima et al., 2000; Bashir and El-Hawary, 2009; Lin and Hsieh, 2009; Yagiz and Karahan,
2011; Marto et al., 2014; Gordan et al., 2016).
The AI techniques are very flexible and this aspect makes them powerful tools in solving
geotechnical problems, in particular when the problem is complex and nonlinear (Armaghani et al.,
2017).
168
8.1.5.1 Artificial Neural Network ANN
Artificial neural network methods are simplified mathematical models that have the purpose of
simulating some organisational principles of the nervous system. In tunnelling ANN methods are
adopted to predict TBM performances. ANN is able to learn automatically, from provided training
patterns, to find the relationship between input and output data (Zurada, 1992).
The constitutive elementary units of an ANN system are the artificial neurons (or nodes): they
develop a process of information very similar to the biological brain ones. The elementary units are
interconnected in a predefined topology known as layer and neurons work in parallel layers
(Alvarez Grima et al.,2000).
Usually, neural network topology is composed by input layer, inner layers and output layer. It is
possible to train the network and during the training process the architecture of network and
connection weights are modified in order to reduce the output error respecting every output layer
node (Armaghani et al., 2017).
On the basis of the network architecture ANN methods are divided into two groups: the feed-
forward and the feed-back respectively. Generally speaking, hidden and output neurons are able to
accomplish some specific activation functions of network input with the purpose to produce neuron
outputs. Each neuron output is considered as input of the next neuron layer. Mathematical equations
are able to give the total net input and output to every hidden or output neuron.
Among all artificial intelligence technique, ANN (in combination with other adjusting algorithm,
have received great attention because of they are able to solve several geotechnical problems
(Momeni et al., 2014; Gordan et al., 2015).
169
8.1.5.2 Particle swarm optimization
Kennedy and Eberhart in 1995 first developed a sub-field of the swarm and computational
intelligence called particle swarm optimization, PSO. It is an optimization algorithm and it has been
inspired by the behaviour of some animals (like fish and birth) with regards to food acquisition in
nature (Yang, 2010; Brownlee, 2011; Armaghani et al., 2014, 2015). PSO has the advantage of
being easier than the other swarm intelligence algorithm. That is because PSO does not consider
mutation/crossover and pheromone, as other algorithms do, and it utilizes the real-number and the
global communication between swarm particles (Kennedy and Eberhart, 1995). In particle swarm
optimization algorithm, the particles (that are the entries) are scattered in the space of objective
function (Armaghani et al., 2017).
The algorithm aims to put the particles in optimal spaces; the particles move inside the global
current towards their own best position. Figure 8.2 schematically represents the movement of
particles in PSO, xi is the current best for the specific element i and p’ is the global best.
Xi
Possible
directions
Particle i
p’
Figure 8.2 Representation of the movement of particles in PSO.
After moving, an element finds a better location than the previous one and the algorithm stops after
a certain number of interactions (Armaghani et al., 2017). PSO algorithm is commonly used to
predict TBM performance: in this way one of the most important studies was conducted by Yagiz
and Karahan in 2011. PSO is usually utilized to optimize the artificial neural network, so ANN and
PSO are frequently adopted in combination.
Atashpaz-Gargari and Lucas in 2007 developed one of the best optimizing algorithms available in
science and engineering field: the imperialism competitive algorithm, ICA.
For the sake of brevity, in this text the explanation of ICA is not provided. ICA procedure can be
seen in other works such as Atashpaz-Gargari et al., 2008; Marto et al., 2014; Armaghani et al,
2015, 2017.
170
8.1.5.3 Hybrid techniques
Many researchers (Alvarez Grima et al., 2000; Bashir and El-Hawary, 2009; Lin and Hsieh, 2009;
Gordan, 2015; Armaghani et al., 2017) have developed the so called “Hybrid technique” with the
purpose of overcoming the limitations that affect the singular techniques. For example, many
algorithms (genetic algorithm and optimization one) are often combined with ANN in order to
predict penetration rates with more accuracy.
Alvarez Grima et al. (2000) combined logic and numeric method, the logical component consists of
fuzzy set and the numerical one consists of neural network and data analysis (Figure 8.3).
Figure 8.3 Neuro-fuzzy modelling technique (Alvarez Grima et. al, 2000).
Logical component takes into account fuzzy logic, fuzzy sets and approximate reasoning, whereas
artificial neural network considers neural nets, data clustering and data analysis, as it is schematised
in Figure 8.3.
Fuzzy logic component allows to incorporate expert knowledge, dealing with uncertainty and
imprecision, making decision and performing inference. On the other hand, ANN permits to
recognise patterns and to adapt methods in case of environment changes (Jang et al. 1997).
171
8.1.6 Final observation about artificial intelligence technique
Table 8.9 reports the most utilised artificial intelligence predictive methods, summarised with the
purpose of showing what parameters are more frequently considered into TBM performance
prediction calculation.
Seventeen studies have been considered and the majority of them is based on the artificial neural
network technique.
As for intact rock parameters the most utilized is the uniaxial compressive strength followed by the
Brazilian tensile strength and by the rock brittleness index. Other parameters are sometimes
involved such as the quartz content and the peak slope index.
The most frequently considered rock mass features are fragmentation (expressed by means of
spacing and RQD) and the angle between the most relevant set of discontinuities and the orientation
of tunnel.
The machine specifications are less considered than rock-rock mass one and some models (Torabi
et al., 2013; Yagiz and Karahan, 2011, 2015; Yagiz, 2009; Yavari Mahdavi 2005) do not consider
any TBM characteristic. The other ones involve cutter geometry features and operational
parameters.
The most carried out index is the net penetration rate, whilst the advance rate and coefficient of
utilization are less frequently evaluated.
172
REFERENCES TECHNIQUE ROCK INPUT TBM INPUT OUTPUT
INTACT ROCK PARAMETERS ROCK MASS PARAMETERS
ANN FIS other UCS BTS PSI BI q p RQD DPW α Jc RMW W RMR deepth other Øm Øc N TF CT RPM RS SE PR AR UI
rev/min
kN/mm
[KNm]
[MPa]
[MPa]
[MPa]
[m^2]
[J/kg]
[mm]
[m/h]
[m/h]
[deg]
[m/s]
[kN]
[%]
[%]
[%]
[m]
[m]
[m]
[m]
[-]
[-]
[-]
[-]
Alvarez Grima et al. (2000) ANFIS
Benardos and Kaliampakos (2004)
Eftekhari et al. (2010)
Ghasemi et al. (2014)
Gholami et al. (2012)
Gholamnejad and Tayarani (2010)
Mahdevari et al. (2014) SVR
Mikaeil et al. (2009)
Oraee et al. (2012) ANFIS
Salimi and Esmaeili (2013)
Shao et al. (2013) ELM
Simoes and Kim (2006)
Torabi et al. (2013)
Yagiz and Karahan (2011) PSO
Yagiz and Karahan (2015)
Yagiz et al. (2009)
Yavari and Mahdavi (2005)
FREQUENCY 10 3 6 16 6 3 6 1 1 7 12 9 1 1 2 3 1 2 1 1 1 2 2 1 1 1 15 2 2
ANN = artificial neural network; FIS = fuzzy inference system; ANFIS = neuro – fuzzy inference system; SVR = support vector regression; ELM extreme
learning machine; PSO = particle swarm optimization; UCS = Uniaxial Compressive Strength [MPa]; BTS = Brazilian Tensile Strength [MPa]; PSI = Peak
Slope Index [kN/mm]; BI = Brittelness Index [MPa]; q = quartz content [%]; p = porosity [m2]; RQD = Rock Quality Designation [%]; DPW = Distance
between Planes of Weakness [m]; α = angle between fractures and tunnel axes [deg]; Jc = Rating for Joint conditions; RMW = Rock Mass Weathering [-]; W =
position water table; RMR = Rock Mass Rating (Bieniawsi, 1989); depth = tunnel depth [m]; Øm = TBM diameter [m]; Øc = cutter diameter [mm]; N =
number of cutter [-]; TF = Total Force to the cutterhead [kN]; CT = Cutterhead Torque [kNm]; RS = Rotational Speed [m/s]; SE = Specific Energy [J/kg]; PR =
Penetration Rate [m/h]; AR = Average Rate; UI = Utilization Index [%]; (other: υ = Poisson ratio [-], ϕ = friction angle [deg]).
Table 8.9 List of intact rock, rock mass and machine parameters involved in the most relevant artificial intelligence predictive techniques.
173
Maddalena” tunnel
8.2 Evaluation of various TBM performance prediction models based on field
data from “La Maddalena” tunnel
Various approaches for predicting field penetration index, FPI, penetration rate, PR or ROP,
specific penetration, SP, and other machine performance parameters have been developed by
several researchers as mentioned in the previous paragraphs.
In this section, an evaluation of existing performance prediction model will be presented; this
evaluation is based on field data from “La Maddalena” tunnel with the purpose of testing the
prediction capability of some of the more frequently involved formulations.
Fourteen predictive equations have been selected, based on the availability of required parameters.
The models have been tested on the tunnel sector ranging between Pk. 198 m and 2010 m. This
length, as previously described, is subdivided into 186 sections.
The prediction capability of existing models is graphically evaluated; the actual and the calculated
machine parameter are plotted into a graph where the bisector line represents the line where the
predicted and measured indices are the same. If the plotted points are close to the dashed line, the
model is reliable and it is good at predicting TBM performance. If plotted points are under or over
the line, the formulation tends to underestimate or overestimate the machine performance,
respectively.
The percentage of prediction error has been calculated for each rock section by means of the
following equation:
𝑀𝑒𝑎𝑠𝑢𝑟𝑒𝑑 𝑃𝑒𝑟𝑓𝑜𝑟𝑚𝑎𝑛𝑐𝑒 𝐼𝑛𝑑𝑒𝑥 − 𝑃𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 𝑃𝑒𝑟𝑓𝑜𝑟𝑚𝑎𝑛𝑐𝑒 𝐼𝑛𝑑𝑒𝑥

𝐸(%) = | | (8.1)
𝑀𝑒𝑎𝑠𝑢𝑟𝑒𝑑 𝑃𝑒𝑟𝑓𝑜𝑟𝑚𝑎𝑛𝑐𝑒 𝐼𝑛𝑑𝑒𝑥
The error distribution has been evaluated along the investigated tract.
174
Maddalena” tunnel
8.2.1 Cassinelli et al. (1982) model
Cassinelli et al. (1982) used Rock Structure Rating (Wickham et al. 1972) and the observed TBM
performances to develop an empirical predictive method. They carried out the following equation:
𝑃𝑅 = − 0.0059 ∗ 𝑅𝑆𝑅 + 1.59 (8.2)
where RSR is empirically derived from RMR by means of 8.3, Bieniawski (1989):
𝑅𝑆𝑅 = 0.77 ∗ 𝑅𝑀𝑅 + 12.40 (8.3)
This formulation has been involved to compute the theoretical TBM performance for “La
Maddalena” tunnel; it tends to underestimate the actual penetration per revolution rates, as shown in
Figure 8.4.
CASSINELLI et al., 1982

5
𝑃𝑅=−0.0059∗𝑅𝑆𝑅+1.59
4
PR predicted [m/h]
0
0 1 2 3 4 5
PR measured [m/h]
Figure 8.4 Comparison between the measured and predicted PR from the Cassinelli et al., (1982)
model.
The results obtained in this study have been compared with the results carried out by the research
conducted by Farrokh (Figure 8.5).
175
Maddalena” tunnel
Figure 8.5 Result of comparison for Cassinelli et al., 1982 model. Farrokh et al., 2012.
It is observable that the Cassinelli model produced the same type of prediction both for “La
Maddalena” and Farrokh database tunnels.
The predicted value of PR is always the same; it could be due to the fact that RSR is derived by
means of an empirical formulation, and not obtained by direct site investigation. Equation 8.3
probably is not appropriate to evaluate RSR about La Maddalena rock mass.
Moreover, the value of RSR, empirically obtained, is more or less constant through the investigated
tunnel length, for this reason, the predicted PR results constant, given that RSR is the unique
independent variable of the model.
176
Maddalena” tunnel
8.2.2 Innaurato et al. (1991) model
Innaurato et al. (1991) modified the model proposed by Cassinelli et al. (1982) by including UCS.
The utilized database consists of 19 km bored in five tunnels excavated in igneous, sedimentary and
metamorphic rocks with UCS values in the range 50-150 MPa. They found a strong relation
between PR, uniaxial compressive strength and rock structure rating, RSR, as reported in Equation
8.4:
𝑃𝑅 = 𝑈𝐶𝑆 −0.437 − 0.047 ∗ 𝑅𝑆𝑅 + 3.15 (8.4)
where RSR has been calculated according to Equation 8.3
The machine performance parameter carried out is PR, measured in m/h.
The prediction results carried out by means of Equation 8.4 are reported in Figure 8.6.
INNAURATO et al., 1991

14
12
10
PR predicted [m/h]
𝑃𝑅=(𝑈𝐶𝑆)^(−0.437)−0.047∗𝑅𝑆𝑅+3.15
0
0 2 4 6 8 10 12 14
PR measured [m/h]
Figure 8.6 Comparison between the measured and predicted PR from the for Innaurato et al.
(1991) model.
The formulation proposed by Innaurato tends to overestimate the actual TBM performances of “La
Maddalena” tunnel; all points are clearly plotted above the bisector line. The use of RSR as an
independent variable implies the same limitations discussed in section 8.2.1.
177
Maddalena” tunnel
The median value of the percentage of error (evaluate according to Equation 8.4) is equal to
278.16%.
Sapigni et al. (2002) use the Innaurato model to compute the theoretical PR for Mean, Pieve and
Varzo tunnels in Italy and then compared them with the recorded ones. They observed that the
predicted parameters are much higher than the registered ones (Figure 8.7 left). Farrokh (2012) used
the Innaurato model to predict the TBM performance for more than 300 tunnels and observed an
underestimation of penetration rates (Figure 8.7 right).
Figure 8.7 Difference between actual and predicted penetration rates for the Innaurato model. Left
Sapigni et al. (2002); right Farrokh et al. (2012).
Farrokh asserts that one of the potential problems that affects the Cassinelli and the Innaurato
models may be the absence of any parameters able to take into account the TBM diameter (Farrokh
et al., 2012). Generally, if the diameter of TBM is large, the penetration rates will be lower than in
those of small machines.
The empirical formulation proposed by Innaurato cannot be used to predict the TBM performance
in “La Maddalena” tunnel and it could not be adopted to estimate the penetration rates both in the
tunnels studied by Sapigni and by Farrokh.
178
Maddalena” tunnel
8.2.3 Khademi Hamidi et al. (2010) models
Khademi Hamidi et al. (2010) collected rock mass and TBM performance data about 8.5 km of the
Zagros tunnel excavated in sedimentary rocks by means of double shield TBM. Authors carried out
linear and nonlinear multiple regression analysis based on RMR system and the angle between the
principal set of discontinuities and tunnel axes, α.
8.2.3.1 Linear model
“La Maddalena” parameters have been entered in the linear type equation (Equation 8.6) proposed
by Khademi Hamidi et al. (2010).
𝐹𝑃𝐼 = 4.161 + 0.091 ∗ 𝑈𝐶𝑆 + 0.077 ∗ 𝑅𝑄𝐷 + 0.117 ∗ 𝐽𝐶 + 1.077 ∗ 𝐿𝑜𝑔𝛼 (8.6)
RQD related to La Maddalena rock mass has been empirically calculated by means of the equation
proposed by Palmström, 1982, (Equation 5.19: RQD=115-3.3*Jv).
The comparison between actual and computed FPI has been plotted in Figure 8.8.
KADEMI HAMIDI et al., 2010

350
𝐹𝑃𝐼=4.161+0.091∗𝑈𝐶𝑆+0.077∗𝑅
300 𝑄𝐷+0.117∗𝐽𝐶+1.077∗𝐿𝑜𝑔𝛼
FPI predicted [kN/cutter/mm/rev]
250
200
150
100
50
0
0 50 100 150 200 250 300 350
FPI measured [kN/cutter/mm/rev]
Figure 8.8 Comparison between the measured and predicted FPI from the Khademi Hamidi et al
(2010) linear model.
The formulation (8.6) proposed by Khademi Hamidi et al. (2010) is not able to estimate the actual
performance of “La Maddalena” TBM.
179
Maddalena” tunnel
8.2.3.2 Nonlinear models
Khademi Hamidi et al. (2010) proposed also nonlinear formulation, exponential (Equation 8.7) and
polynomial (Equation 8.8) type respectively.
𝐹𝑃𝐼 = 𝐸𝑋𝑃(1.995 + 0.004 ∗ 𝑈𝐶𝑆 + 0.005 ∗ 𝑅𝑄𝐷 + 0.01 ∗ 𝐽𝑐 + 0.068 ∗ 𝐿𝑜𝑔𝛼) (8.7)
𝐹𝑃𝐼 = 9.401 + 0.397 ∗ 𝐿𝑜𝑔𝛼 + 0.011 ∗ 𝐽𝑐 2 + 1.14𝐸 − 5 ∗ 𝑅𝑄𝐷 3 + 1.32 − 8 ∗ 𝑈𝐶𝑆 4 (8.8)
The prediction results obtained by means of Equation 8.7 and Equation 8.8 are reported in Figure
8.9 and Figure 8.10 respectively.

350
𝐹𝑃𝐼=𝐸𝑋𝑃(1.995+0.004∗𝑈𝐶𝑆+0.005∗𝑅𝑄𝐷
300 +0.01∗𝐽𝑐+0.068∗𝐿𝑜𝑔𝛼)
250
200
150
100
50
0
0 50 100 150 200 250 300 350
Figure 8.9 Comparison between the measured and predicted FPI from the Khademi Hamidi et al.
(2010) exponential model.
180
Maddalena” tunnel

350
𝐹𝑃𝐼=9.401+0.397∗𝐿𝑜𝑔𝛼+0.011∗(𝐽𝑐
300 )^2+1.14𝐸−5∗(𝑅𝑄𝐷)^3+1.32−8∗(𝑈𝐶𝑆)^4

250
200
150
100
50
0
0 50 100 150 200 250 300 350
Figure 8.10 Comparison between the measured and predicted FPI from the Khademi Hamidi et al.
(2010) polynomial model.
The median percentage of error relative to Equation 8.7 and 8.8 prediction is equal to 68.19% and
73.32% respectively. Both the exponential and polynomial formulations proposed by Khademi
Hamidi et al. (2010) are not able to predict the actual TBM performance in the case of “La
Maddalena”. In all three cases a significant inability in prediction can be observed. RQD is
empirically calculated and it could determine the reliability of input data.
Farrokh et al. (2012) evaluated these models based on field data from more than 300 tunnels and an
overestimation of penetration rate can be observed, Figure 8.11.
Figure 8.11 Result of comparison for Khademi

Hamidi et al. (2010) linear model carried out by
Farrokh et al. (2012) based on field data from
more than 300 tunnels. Farrokh et al. (2012).
181
Maddalena” tunnel
8.2.4 Hassanpour models
Hassanpour et al., carried out four empirical predictive equations based on four different tunnel
projects, each of them realized in different geological and mechanical conditions; Authors
developed a “general model” on the basis of these four models.
8.2.4.1 Zagross tunnel
In 2009 Hassanpour et al., the researchers analysed the Zagross water conveyance tunnel
excavation, in Iran. The tunnel passes through the Zagross Simply folded zone and the encountered
lithologies are limestone, shale and limy shales belonging to the Carbonate-Argillaceous rocks of
Padbeh, Gurpi and Ilam Formations (Hassanpour et al., 2009). The UCS values range from 15 to
150 MPa and the maximum overburden is 650 meters. The excavation was realized by means of a
double shield TBM. The empirical equation (Equation 8.9) carried out by Hassanpour et al. (2009)
allows to predict FPI:
𝐹𝑃𝐼 = 𝑒𝑥𝑝(0.004 ∗ 𝑈𝐶𝑆 + 0.008 ∗ 𝑅𝑄𝐷 + 2.077) (8.9)
“La Maddalena” FPI values have been calculated according the previous formula, then the predicted
and measured values have been compared (Figure 8.12).
HASSANPOUR et al., 2009

350
𝐹𝑃𝐼=exp(0.004∗𝑈𝐶𝑆+0.008∗𝑅𝑄𝐷+2.077)
300
250
200
150
100
50
0
0 50 100 150 200 250 300 350
Figure 8.12 Comparison between the measured and predicted FPI from the Hassanpuor et al.
182
Maddalena” tunnel
The median of percentage of error generated by using Equation 8.9 in predicting “La Maddalena”
performance is 64.42%.
The inability of this method to predict “La Maddalena” FPI values is probably due to the different
lithology of excavated rock masses and to the different (lower) UCS range.
8.2.4.2 Karaj tunnel
Karaj water conveyance tunnel, Iran, is located in Central Alborz geological zone and passes
through the pyroclastic rocks of Karaj formation that consist of tuffs, agglomerate, shaly and sandy
tuffs. The UCS values range from 30 to 150 MPa and the maximum overburden is 600 meters.
The excavation is realized by means of a 4.65 m in diameter double shield TBM (Hassanpour et al.,
2010). Hassanpour et al., collected and analysed rock mass and TBM parameters and developed a
new predictive model in order to predict FPI. The following equation was carried out:
𝐹𝑃𝐼 = exp(0.005 ∗ 𝑈𝐶𝑆 + 0.002 ∗ 𝑅𝑄𝐷 + 2.129) (8.10)
The previous formula (Equation 8.10) has been used to calculate “La Maddalena” FPI values, the
error that affects the prediction is defined by a median value equal to 75.98%. The calculated and
measured values have been compared (Figure 8.13).

350
𝐹𝑃𝐼=exp(0.005∗𝑈𝐶𝑆+0.002∗𝑅𝑄𝐷
300
250
200
150
100
50
0
0 50 100 150 200 250 300 350
(2010) exponential model
183
Maddalena” tunnel
In this case too, the formulation proposed by Hassanpour et al. (2010) tends to underestimate the
actual values of FPI about “La Maddalena” tunnel.
This lack is probably due to different lithology of excavated rock masses and to the different
(lower) UCS range.
8.2.4.3 Ghomrood tunnel
Ghomrood water conveyance tunnel (Iran) is 24.5 km length and passes through the Sanandaj-
Sirjan metamorphic belt characterized by Jurassic metamorphic rocks (low to medium grade) and
Cretaceous limestone.
The encountered lithologies are limestone, shale, slate, sandstone, phyllite and schist with quarzitic
veins (Hassanpour, 2010). The tunnel is constructed by means of a double shield TBM (4.52 m in
diameter), it is 700 meters deep (maximum value).
The minimum value of UCS is 25 MPa and the maximum one is 150 MPa (Hassanpour, 2010).
Hassanpour analysed both geological-geomechanical features and TBM parameters in order to

obtain an empirical predictive equation to pre-estimate FPI index, he proposed the following
formula:
𝐹𝑃𝐼 = exp(0.004 ∗ 𝑈𝐶𝑆 + 0.023 ∗ 𝑅𝑄𝐷 + 1.003) (8.11)
The proposed formulation (Equation 8.11) has been utilized to predict FPI of “La Maddalena”
tunnel. The error calculated for this model is 53.84%. The comparison between measured and
predicted FPI values is shown in Figure 8.14.
184
Maddalena” tunnel
HASSANPOUR, 2010
350
𝐹𝑃𝐼=exp(0.004∗𝑈𝐶𝑆+0.023∗𝑅𝑄𝐷+1.003)
300

250
200
150
100
50
0
0 50 100 150 200 250 300 350
Figure 8.14 Comparison between the measured and predicted FPI from the Hassanpuor (2010)
exponential model.
The comparison between actual and computed FPI values gives a result similar to the previous
ones: a significant underestimation is induced.
8.2.4.4 Manapouri tunnel
Manapouri Second Trailrace tunnel in New Zeland is excavated into gneiss, intrusive rocks like
gabbro and diorite, calc-silicate and quartzite belonging to the Paleozoic metamorphic and igneous
rocks of the Fiordland Complex (Hassanpour et al., 2011). The encountered rocks are characterised
by UCS values ranging from 100 to 225 MPa.
Tunnel maximum deep is 1200 meters from surfaces. The excavation is realized by means of a main
beam open TBM, 10 meters in diameters.
In this case study Authors carried out a new equation to predict TBM performance, it is as follows:
𝐹𝑃𝐼 = exp(0.005 ∗ 𝑈𝐶𝑆 + 0.020 ∗ 𝑅𝑄𝐷 + 1.644) (8.12)
Equation 8.12 has been involved to predict FPI of “La Maddalena” tunnel and the comparison
between measured and predicted values is shown in Figure 8.15.
185
Maddalena” tunnel

350
𝐹𝑃𝐼=exp(0.005∗𝑈𝐶𝑆+0.020∗𝑅𝑄𝐷+1.644)
300

250
200
150
100
50
0
0 50 100 150 200 250 300 350
The plot shows that there is a group of points plotted close to the bisector line. This means that in
some cases the proposed formulation has been able to predict FPI, however this method cannot be
used to calculate the TBM performance of “la Maddalena” because in several cases they are
underestimated. The error in prediction (35.79%) is much more reduced in comparison to the other
models.
8.2.4.5 Hassanpour general method
As it is known, predictive models are usually specific-site, they are carried out for specific
condition and they cannot be used to predict TBM performance in conditions different from those in
which they have been produced. For this reason, data from the previous projects with different rock
mass features were collected and compiled in a database in order to develop a more accurate TBM
performance prediction model that can be used in different geological and geotechnical conditions.
Hassanpour et al., in 2011, developed a model based on Zagross, Karaj, Ghomrood and Manapouri
Tunnels database.
A multivariable regression analysis was used to relate the machine parameters and the geological
ones.
186
Maddalena” tunnel
As a result, an empirical equation was introduced; it is as follows:
𝐹𝑃𝐼 = exp(0.008 ∗ 𝑈𝐶𝑆 + 0.015 ∗ 𝑅𝑄𝐷 + 1.384) (8.13)
Equation 8.13 has been used to predict PFI for “La Maddalena” tunnel, and calculated and actual
FPI have been compared, an error in prediction equal to 48.00% has been observed. The result is
shown in Figure 8.16:

350
𝐹𝑃𝐼=exp(0.008∗𝑈𝐶𝑆+0.015∗𝑅𝑄𝐷
300 +1.384)
250
200
150
100
50
0
0 50 100 150 200 250 300 350
Figure 8.16. Comparison between the measured and predicted FPI from the Hassanpuor et al.
In some cases, the FPI values are correctly predicted applying the previous formulation, in other
cases there is an underestimation in prediction. In particular low FPI are best predicted than higher
one.
The error that affects predictions is defined by a median values equal to 44.19%.
Authors developed a new formulation (Equation 8.14) in order to calculate ROP, based on FPI, by
assuming practical values for:
- Average disc cutter load, Fn = 200 kN/cutter;
- Revolution per Minute, RPM = 7 rev/min.
187
Maddalena” tunnel
ROP is calculated as:
0.06 ∗ 200 ∗ 7
𝑅𝑂𝑃 = (8.14)
𝐹𝑃𝐼
The FPI measured at “La Maddalena” tunnel has been involved in Hassanpour predictive
formulation, then computed and registered values have been compared, the error in calculation is
reduced (30.96%).
The result of comparison is shown in Figure 8.17.
HASSANPOUR ET AL., 2011

6
5
ROP predicted [m/h]
𝑅𝑂𝑃=(0.06∗200∗7)/𝐹𝑃𝐼
0
0 1 2 3 4 5 6
ROP measured [m/h]
Figure 8.17 Comparison between the measured and predicted ROP from the Hassanpuor et al.
(2011) linear model.
As it can be seen, the equation proposed by Hassanpour is able to calculate ROP with accuracy.
Most of the points are plotted close to 1:1 line (bisector) and the measured and the predicted ROP
are very similar.
188
Maddalena” tunnel
8.2.5 Delisio et al., models
Delisio et al. (2013, 2014) developed TBM predictive models for jointed rock masses, where
“blocky conditions” occur. They introduced a new TBM performance index, the FPIblocky.
FPIblocky is defined according to:
𝐹𝑁
𝐹𝑃𝐼𝑏𝑙𝑜𝑐𝑘𝑦 = (8.15)
𝑝
where FN is the TBM thrust force [kN] and p is the penetration per revolution [mm/rev].
Authors proposed a deviation from standard definition of the Field Penetration Index, which is as
follows:
𝐹𝑛
𝐹𝑃𝐼 = (8.16)
𝑝
where Fn is the average cutter load [kN/cutter].
Delisio proposed to consider TBM thrust force, FN, instead of average cutter load, Fn, where
blocky conditions are verified. The modification of standard FPI definition is justified by Authors
by the fact that especially in blocky rock masses tunnel face is not regular and plane so not all
cutters are in contact with the face. If some discs are not in contact with the rock, the load will be
distributed on the cutters touching the face and Fn cannot be evaluated to be constant anymore
(Delisio et al., 2013).
8.2.5.1 Lötschberg Base Tunnel model
In 2013, Delisio et al., analysed field geological and machine related data collected during the
excavation of Lötschberg Base Tunnel in Switzerland. The considered section is composed by rock
units belonging to Aar Massif. It is formed by three lithological units (Delisio et al., 2013):
- Gneiss and granidiorite, Baltschieder granodiorite;
- Grained granite, Central Aar granite;
- Crystalline sector composed of granite, gneiss and amphibolite.
In the Aar Massif blocky conditions occurred, because of the volumetric joint count, Jv, ranges
between 4 to 25 joints/m3 (Delisio et al., 2013). The mean UCS ranges between 108 and 220 MPa,
189
Maddalena” tunnel
BTS between 11 and 19 MPa, CAI between 3.3 and 5.2; these characteristics define very strong and
highly rock masses.
Maximum depth is 1950 meters and the excavation was realized by means of two gripper TBMs.
Authors performed a multivariate regression analysis to determine an empirical relationship to

predict the FPIblocky, taking into account the joint counts (n/m3) and UCS of intact material. The
following equation has been carried out.
𝐹𝑃𝐼𝑏𝑙𝑜𝑐𝑘𝑦 = 5952 − 1794 ∗ 𝑙𝑛𝐽𝑣 + 𝑈𝑆𝐶 (8.17)
The gneiss domain of “La Maddalena” tunnel consist of massive gneiss where blocky conditions do
not occur; for this reason, the previous formulation has been applied to predict FPI in gneiss-
micaschist and micaschist domains. The comparison between calculated and predicted values is
shown in Figure 8.18.
DELISIO ET AL., 2013

7000
6000
FPI predicted [kN/mm/giro]
5000
4000
3000
2000
1000
0
0 2000 4000 6000
FPI measured [kN/mm/giro]
Figure 8.18 Comparison between the measured and predicted FPIblocky from the Delisio et al.
(2013) model.
As shown in Figure 8.18, the predicted FPIblocky ranges between the same ranges of measured
FPIblocky furthermore the median value of calculated error is 34.71%; however Delisio formulation
could not be utilised for “La Maddalena” case study because of significant data dispersion and only
few cases able to predict FPI with high accuracy.
190
Maddalena” tunnel
Generally, Delisio model tends to overestimate FPI, this is probably due to the fact that “La
Maddalena” rock mass Jv is lower than the Lötschberg one.
8.2.5.2 Lötschberg and Second Manapouri Trailrace Tunnels model
In 2014 Delisio and Zhao developed a new model based on field data collected at Lötschberg Base
Tunnel (see 8.2.5.1 paragraph) and at Second Manapouri Trailrace Tunnel (see 8.2.4.4 paragraph).
Authors have carried out two predictive empirical equations, the first (Equation 8.18) suitable for
rock masses where blocky conditions are verified and the second (Equation 8.19) that allow us to
evaluate FPI starting from FPIblocky.
𝐹𝑃𝐼𝑏𝑙𝑜𝑐𝑘𝑦 = 𝑒 6 ∗ 𝐽𝑣−0.82 ∗ 𝑈𝐶𝑆 0.17 (8.18)
𝐷
𝐹𝑃𝐼 = ∗ 𝐹𝑃𝐼𝑏𝑙𝑜𝑐𝑘𝑦 (8.19)
𝑐
where D is the diameter of TBM [m] and c is the number of cutters put on the cutterhead.
Equation 8.18 has been used to predict FPI at “La Maddalena” tunnel, then the comparison between
predicted and measured FPI have been carried out (Figure 8.19).
DELISIO and ZHAO, 2014

10000
𝐹𝑃𝐼𝑏𝑙𝑜𝑐𝑘𝑦=𝑒^6∗𝐽𝑣^(−0.82)∗𝑈𝐶𝑆
8000
FPIblocky predicted [kN/mm/rev]
6000
4000
2000
0
0 2000 4000 6000 8000 10000
FPIblocy measured [kN/mm/rev]
Figure 8.19 Comparison between the measured and predicted FPIblocky from the Delisio and Zhao
(2014) model.
191
Maddalena” tunnel
The equation proposed by Delisio and Zhao (2014) produces a very relevant underestimation of FPI
characterised by median percentage of error equal to 86.34%: for this reason, it is not utilisable to
predict FPI at “La Maddalena” tunnel.
The formulation 8.19 carried out to estimate the FPI starting from FPIblocky was also used to predict
FPI at “La Maddalena”.
This formulation considers both the TBM diameter (6.30 meters) and the number of cutters installed
on the machine cutterhead (43 cutters). A comparison between predicted and registered FPI has
been carried out, the result is shown in Figure 8.20.
DELISIO and ZHAO, 2014

300
𝐹𝑃𝐼=𝐷/𝑐∗𝐹𝑃𝐼𝑏𝑙𝑜𝑐𝑘
250
200
150
100
50
0
0 50 100 150 200 250 300
Figure 8.20 Comparison between the measured and predicted FPI from the Delisio and Zhao
(2014) model.
As it is possible to note in Figure 8.20, the predicted FPI ranges between the same ranges of
measured FPI, the median error in prediction is evaluated equal to 34.86%, however the plot shows
a great data dispersion that made it not utilizable to predict FPI for “La Maddalena” exploratory
tunnel. In some cases, points are plotted close to the 1:1 line and it means that in these situations the
model has been able to correctly calculate FPI.
192
8.3 Conclusion
8.3 Conclusion
During this study, fourteen empirical TBM predictive models have been selected and their capacity
to predict the TBM performances at “La Maddalena” exploratory tunnel has been tested.
Three models aim to predict the net penetration rate (ROP or PR) of a TBM and eleven are intended
to calculate FPI. The number of FPI models is higher than the ROP one because of FPI is a
composite parameter, for this reason, it is considered more representative. It is based both on rock
masses parameters and machine ones (diameter and number of cutters).
One model gives significantly overestimation of the selected indices as a result, nine generate a
constant result and in three cases great data dispersion is observable; in these situations, the ranges
of predicted and observed values are comparable, however these models are not applicable for “La
Maddalena” case study because the differences between actual and computed values can be very
high. Only one model, proposed by Hassanpour in 2011, was able to predict with accuracy “La
Maddalena” TBM performances.
The errors generated by using literature proposed models to predict “La Maddalena” TBM
performances are generally high, as expressed by median values in the previous paragraphs.
As already explained, the error in prediction of TBM performance, computed by using the
mentioned models, has been evaluated for each rock section. Then the spatial distribution of the
errors has been evaluated.
Generally, error in prediction is higher in gneiss domain (it occurs eight in cases); however, some
models show constant distribution of percentage of error (two cases) and in some cases the higher
percentage of error has been measured into micaschist domain (four cases).
An example for each of the three types of distribution has been reported in Figure 8.21.
1500
INNAURATO et al., 1991
1000
Error [%]
500
0
100 300 500 700 900 1100 1300 1500 1700 1900
Pk. [m]
193
8.3 Conclusion
120
100
DELISIO et al., 2013
80
Error [%]
60
40
20
0
100 300 500 700 900 1100 1300 1500 1700 1900
Pk. [m]
300
250
200
Error [%]
150
100
50
0
100 300 500 700 900 1100 1300 1500 1700 1900
Pk. [m]
Figure 8.21 Spatial distribution of error in prediction. Red points represent the passage from DI to
DII and from DII to DIII.
The use of empirically calculated parameters as independent variables is not suitable, because the
formulation selected to calculate the parameter could be not appropriated.
It is confirmable that model limitations are caused by limited database used in the development of
the models; in fact, the acceptable formulation was carried out focusing on a large database (four
tunnels and 55.4 km of available data). Farrokh asserts that utilizing a large database of machine
field performance records that considers an ample range of ground conditions and tunnel diameters
can generate models able to offer better accuracy than the existing one.
Some common causes of the inability of a model to predict TBM performance in a tunnel different
from that where the model has been carried out are:
- High site-specificity: they are valid only if the geological and mechanical contests similar to
the ones in which they have been developed;
- Exclusion of relevant parameters or introduction of inappropriate parameters: related to the

decisions of the operators and to the availability of the data;
- Introduction of too many inputs in the model;
194
8.3 Conclusion
- Lack of distinction between the rock masses conditions and the work constraints that affect
TBM performance.
Nevertheless, the site specificity of prediction models do not have to be considered as a negative
aspect only because they are the best method to estimate the TBM performance in a given tunnel
project (Hassanpour et al., 2016).
The following graphs show:
- The ranges of ROP values obtained from Cassinelli, Innaurato and Hassanpour models
(Figure 8.22);
- The ranges of predicted FPI obtained from Kademi Hamidi and Hassanpour models (Figure
8.23);
- The ranges of predicted FPIblocky from Delisio et al. models (Figure 8.24).
These ranges have been compared with measured one; the actual range of TBM performance has
been reported.
Measured ROP 4
Cassinelli et al.
(Eq. 8.2) 3
Innaurato et al.
(Eq. 8.5) 2
Hassanpour et al.1
(Eq. 8.14)
0
0 2 4 6 8 10 12
ROP [m/h]
Figure 8.22 Ranges of values of ROP predicted from the models listed on the left. Asterisks
represent the 25th and 75th percentile of ROP, diamond boxes represent the median. Measured
values of ROP have been represented in red. Red dashed line demarcates the ranges of observed
values.
195
8.3 Conclusion
Measured FPI 10
Kademi Hamidi et al.

9
(Eq. 8.6)

8
(Eq. 8.7)

(Eq. 8.8) 7
Hassanpour et al.
(Eq. 8.9) 6
Hassanpour et al.
(Eq. 8.10) 5
Hassanpour
(Eq. 8.11) 4
Hassanpour et al.
3
(Eq. 8.12)
Hassanpour et al.
(Eq. 8.13) 2
Delisio and Zhao

1
(Eq. 19)
0
0 20 40 60 80 100 120 140
FPI [kN/cutter/mm/rev]
Figure 8.23 Ranges of values of FPI predicted from the models listed on the left. Asterisks represent
the 25th and 75th percentile of FPI, diamond boxes represent the median. Measured values of FPI
have been represented in red. Red dashed line demarcates the ranges of observed values.
196
8.3 Conclusion
Measured FPIblocky3
Delisio et al.
2
(Eq. 8.17)
Delisio and Zhao.

1
(Eq. 8.18)
0
0 1000 2000 3000 4000 5000 6000 7000
FPIblocky [kN/mm/rev]
Figure 8.24 Ranges of values of FPIblocky predicted from the models listed on the left. Asterisks
represent the 25th and 75th percentile of FPIblocky, diamond boxes represent the median. Measured
values of FPIblocky have been represented in red. Red dashed line demarcates the ranges of observed
values.
197
9 DEVELOPMENT OF TBM PERFORMANCE
PREDICTIVE MODELS
9.1 Introduction
The relationship between the geological, mechanical and structural frameworks concerning a given
rock mass and the performance of the TBM used to construct a tunnel have been investigated with
the purpose of carrying out empirical models able both to explain the relationship and to predict the
behavior of the TMB working into a specific ground.
For this purpose, simple and multivariate regression analyses (both linear and nonlinear) have been
carried out and discussed in this chapter.
The models obtained have been validated and the results of validation have been reported in
Chapter 12.
The models have been developed using geological and machine data collected from Pk. 0 + 197 m
to Pk. 2 + 010 m, as described in Chapter 5. The geological information contained in each of the
186 rock mass section surveyed has been input into the model as independent variables and the
machine parameters, attributed to each section, as independent ones.
The reliability of the obtained models in TBM performance prediction has been tested on the data
collected from Pk. 2 + 010 m to Pk. 6 + 776 m, this length is divided into 212 rock mass sections
A brief theoretical introduction about statistical methods and tests involved in the analyses has been
provided, prior to the presentation and discussion of the obtained models.
199
9.2 Theoretical concepts
When two or more variables related to the same problem are detected, the relationship between
them can be generally investigated in two different ways.
If it is necessary to study the level of association between two variables, the statistic technique
adopted is the correlation; instead, if it is necessary to carry out an equation able to link two or more
variables, the statistical procedure adopted is the regression analysis. The best way has to be
selected on the basis of the purpose of the study.
9.2.1 Correlation between variables
The correlation aims to investigate if a relationship between two variables occurs, and to
numerically quantify the association level. Generally, the first step is to create a scatterplot and to
observe the disposition of the points.
If the points are principally plotted into the quadrant I and III, the association between the two
variables is positive; on the contrary, if the points are distribute between II and IV the correlation is
negative (Figure 9.1).
Figure 9.1 Positive (blue dots) and negative (red dots) correlation.
If the points are equally distributed among the quadrants, there are not any associations.
By observing the point distribution, it is possible to understand whether the correlation is linear (the
points are aligned on a line) or not.
If the correlation is linear and the distribution is Gaussian type, it can be quantified by means of
Pearson coefficient, calculated as:
200
𝐶𝑂𝑉𝑋𝑌
𝑟= (9.1)
𝑆𝑋 𝑆𝑌
where: COVXY is the covariance between X and Y and SX and Sy are the standard deviations of X
and Y respectively.
The Pearson coefficient ranges between – 1 and + 1. If r is equal to + 1 or – 1, there is a perfect

correlation and points are perfectly aligned on a line; the more distant the points from the line are,
the weaker the correlation is, values near to 0 indicate no correlation between the studied variables
(Figure 9.2).
Figure 9.2 Example of data scatterplots with a variety of correlation values. In a,g,h,i points are
perfectly aligned on a line and in these case r is +1 or -1. In d, l, m, p there is no correlation ad
r=0. In n and o there is correlation but r=0 because of the correlation is not linear (Boracchi et al.,
2015).
When the relation between variables is not linear, or at least one variable is not normally
distributed, or at least one of them is an ordinal variable, the Pearson coefficient cannot be
considered. In these cases, the Spearman rank correlation coefficient ρs should be used. ρs is
calculated as:
6 ∗ ∑𝑖 𝐷𝑖2
𝜌𝑠 = 1 − (9.2)
𝑁(𝑁 2 − 1)
where Di is the difference between the ranks and N is the number of the observations. ρs ranges
from – 1 (negative correlation) to + 1 (positive correlation); ρs equal to zero means absence of
correlation.
201
Correlation analysis represents a preliminary stage in the investigation of the relationship between
variables: in fact at the end of the analysis it is possible to assert if relationship occurs, and to
understand the type and the level of correlation.
If a relationship between variables is recognised by plotting data, the study can be continued. It has
a dual purpose:
- Explanation: the aim is to comprehend and evaluate the effects of the independent variable
on the dependent one, on the basis of a mathematical function;
- Predictive: the aim is to individuate a combination of dependent variables in order to predict

the value of the dependent one.
For these purposes, regression analyses are largely adopted.
9.2.2 Simple linear regression analysis
Regression analysis is a group of statistical techniques that allows to explain the relationship
between one or more independent variables and a dependent one.
The simple regression analysis considers one independent variable only and its effects on the
dependent one, on the contrary, if the purpose is to investigate the relations between more than one
independent variables and a dependent parameter a multiple regression analysis has to be
performed.
The objective of linear regression analysis is to find a line that minimises the distance between the
observed data and the line; which is calculated by means of the least square method, as follows:
𝑦𝑖 = 𝑎 + 𝑏𝑥𝑖 + 𝑒𝑖 (9.3)
where:
b is calculated as:
2
𝑏 = ∑(𝑥𝑖 − 𝑥̅ )(𝑦𝑖 − 𝑦̅)⁄∑(𝑥𝑖 − 𝑥̅ ) (9.4)
a is calculated as:
𝑎 = 𝑦̅ − 𝑏𝑥̅ (9.5)
and ei is the distance between the point and the line and it should be as small as possible:
202
𝑚𝑖𝑛(∑ 𝑒𝑖2 ) (9.6)
A graphical explanation of least square method has been reported in Figure 9.3.
Figure 9.3 Least square method.
9.2.2.1 Coefficient of determination
The coefficient of determination, R2, is one of the measures used to assess the ability of a regression
in explaining the relationship between a dependent and an independent variable. R2 expresses the
proportion of variance (the variance is the square of the standard deviation of a set of data) in the
dependent variable that is explained by the regression and the predictor variable.
The coefficient of determination is calculated as follows:
𝑆𝑆𝑅
𝑅2 = (9.7)
𝑆𝑆𝑇
where:
- SSR is the sum of squares of the model or the explained sum of squares, and it is obtained by
the total sum of squares minus the residual sum of squares;
- SST is the total sum of squares.
R2 ranges between 0 and 1; R2 equal to zero means that the regression does not explain any
percentage of variance, on the contrary R2 of 1 means that all the variance is explained by the
regression.
203
9.2.2.2 Model assumptions
In order to carry out a linear regression model it is necessary that three assumptions are verified,
which are:
- Independence: the response variables (yi) are independent;
- Normality: the values of the response variables are normally distributed;
- Homoscedasticity: the response variables have the same variance.
The assumptions are usually tested on the random errors, they have to be independent and normally
distributed, they have to have a constant variance, σ2, for all settings of the independent variables
(homoscedasticity) and their mean has to be equal to zero. The verification of the assumption can be
performed considering the residuals (ei); however considering the standardised residuals is suitable.
Normality can be checked graphically or numerically.
In the first case box-plots (normal distribution occurs when they are symmetric), QQ-norm
diagrams (normal distribution occurs if the points are aligned on the bisector line) and histograms
(observing the probable distribution) are widely used. Figure 9.4 reports an example of QQ plot,
which shows a normal distribution and two histograms: the first one is about a normal distribution,
while normality is not verified in the second one.
Figure 9.4 From left to right: normal QQ plot, histogram showing normal distribution, histogram
showing not normal distribution.
204
Normal distribution can be numerically tested by means of tests that compare the shape of a given
distribution to the one of a normal curve, a significant result means that the distribution is not
shaped as a normal curve.
The normality test which is most frequently applied is the “Shapiro-Wilk normality test”. Its null
hypothesis is that data are normally distributed, so if p-value is higher than 0.05 the null hypothesis
has to be rejected and data are normally distributed, on the contrary the population is not normally
distributed. Shapiro-Wilk, W, is calculated as follow:
2
(∑𝑛𝑖=1 𝑎𝑖 𝑥(𝑖) )
𝑊= (9.8)
∑𝑛𝑖=1(𝑥𝑖 − 𝑥̅ )2
where ai is a constant.
Homoscedasticity can be checked producing a scatterplot of the standardised residuals vs. the fitted
values or standardised values vs. each of the independent variables. When the assumptions are
satisfied, residuals vary randomly around zero and their spread is the same throughout the plot
(Figure 9.5).
Homoscedasticity Heteroscedasticity
Figure 9.5 Homoscedasticity verified (on the left), and not verified (on the right); Boracchi et al.,
2015.
Homoscedasticity is numerically tested by means of the White test. It carries out a regression
analysis on the residual values as follows:
𝑒𝑖2 = 𝑎𝑟 + 𝑏𝑟 𝑥𝑖 + 𝑏𝑟2 𝑥𝑖2 (9.9)
Where a and b have been obtained by means of leas square method.
205
The R2 about Equation 9.9 has to be calculated and if n* R2 is higher than x2 there is evidence of
heteroscedasticity.
9.2.3 Multiple linear regression analysis
A multiple regression analysis is carried out when the combined effect of more independent
variables on the dependent one is investigated; it is considered as an extension on the simple
regression.
There are many situations, in statistics, in which one variable is able to predict very accurately
another single variable. However, the predictions usually improve when additional relevant
information is considered. The most widely used mathematical functions with the purpose of
expressing the relationships among more than two variables are equations as:
𝑦 = 𝑏0 + 𝑏1 𝑥1 + 𝑏2 𝑥2 +. . . +𝑏𝑘 𝑥𝑘 (9.10)
where:
- y is the variable of response (the dependent one);
- x1, x2, … , xk are the explanatory variables (the independent ones);
- b0, b1, … , bk: are the regression coefficients. They are obtained by means of the least-
squares method.
The sum of the squared difference between the measured y1 and the residual (y1 predicted by the
model) has to be as low as possible:
∑[𝑦1 − (𝑏0 + 𝑏1 𝑥1𝑖 + 𝑏2 𝑥2𝑖 +. . . +𝑏𝑘 𝑥𝑘𝑖 )]2 𝑚𝑖𝑛𝑖𝑚𝑢𝑚 (9.11)
where y1 is the ith data value for y, x1i is the ith data value for x1, … , xki is the ith data value for xk.
The value of b0, b1, … , bk which satisfy this property are the least square estimates of the
regression.
9.2.3.1 T-test and F-test
T-test is frequently used to evaluate the influence of each predictor and its significance. The index t
is equal to the ratio between the estimated value and the standard deviation.
T-test is used to assess only one regression coefficient at a time; on the contrary, the F-test can
assess multiple coefficients simultaneously.
206
The F-test is carried out to evaluate the utility of the regression, comparing the fits of different
linear models. F-tests assess the overall significance of the regression model, based on the following
hypothesis:
- Null hypothesis: the fit of the obtained model is equal of the intercept- only model (in which
no coefficients are present);
- Alternative hypothesis: the fit of the regression model is significantly improved compared to
the intercept-only one.
If the p-value related to the F-test is less than the selected significance level the null hypothesis can
be rejected, and it means that at least one of the independent variables significantly affects the
dependent variable.
Generally, if t-test is not significant, the F-test will not be significant either.
R2 provides some information about the strength of the relationship between the regression model
and the dependent variable and it does not provide a hypothesis test about this relation. On the
contrary, the F-test determines if this relationship is statistically significant. When the p-value
related to the F-test is less than the selected significance level the R2 is significantly different from
zero.
9.2.3.2 Coefficient of determination
When a multiple regression analysis is carried out, there is more than one dependent variable, so it
is necessary to adopt a modified version of the coefficient of determination, the so called R 2adjusted,
which considers the number of predictors, because the coefficient of the determination tends to
increase with the number of dependent variables included into the model.
9.2.3.3 Multicollinearity
Multicollinearity occurs when two or more independent variables are associated or highly correlated
to each other, if it is present, the performed regression is not reliable. In geomechanics
multicollinearity occurs if a variable is computed from another one in the same dataset, or when two
variables represent the same aspect. When this problem takes place, the partial regression
coefficient may not be estimated correctly, and it makes difficult to assess the relative influence that
an independent variable has on the dependent one.
207
Multicollinearity can be detected by means of tolerance and the variance inflation factor, VIF,
which is the reciprocal of the tolerance. When tolerance is less than 0.1 – 0.2 VIF is higher than 10
and it indicates that there are problems of multicollinearity.
The researchers have to select with great accuracy the independent variables that will be input into
the regression analysis in order to avoid the problem above explained. When VIF is > 10 it is
necessary to revise all the input parameters and remove the ones considered redundant.
9.2.3.4 Regression model building
There are different methods to select and enter the variables into a regression model. The most
commonly used are: the “Backward elimination”, the “Forward selection” and the “Stepwise
regression”.
Backward elimination: it is the so called “top down approach”. Firstly it is necessary to decide a
significance level, SLS, which makes the variables stay into the model; this level can be 0.05
however; if it causes too many parameters to be removed, it can be elevated at 0.20. All
independent variables are input into the model, then the predictor characterised by the highest p-
value has to be considered. If it is higher than SLS the variable has to be removed and the model
has to be fitted without it. On the contrary if the highest p-value is < SLS the variable has to be
maintained.
The procedure continues until all independent variables have p-values less than SLS.
Forward stepwise: it is also known as “bottom up approach”. In this case too, it is necessary to
choose a significance level, SLE, so that the variables are inserted in the model, it can be 0.05 but
for this method it may be too low and cause too few predictors to be entered; for this reason the
suitable SLE is equal to 0.20. Then the variables are added to the equation one at a time: it is
necessary to consider the parameter with the lowest p-value. If it is less than SLE this predictor has
to be included into the model, whereas if p-value > SLE it is necessary to stop and consider the
previous model.
The procedure has to be continued until none of the new independent variables have p-values less
than SLE.
Stepwise regression: this method combines the backward and forward ones. It is necessary to
consider both SLS and SLE and choose them in such a way that SLE < SLS. The procedure starts as
the bottom up process (forward selection).
208
The new variable must have p-value < SLE to be entered, at this point all the variables that have
already been input have to be re-tested and their p-values have to be less than SLS to allow them to
stay into the model.
The process continues until none of the new variables can be input and none of the old ones have to
be removed.
9.2.3.5 Model assumption
The assumptions for multiple linear regression models are that the response variables have to be
independent, the values of the response variable have to be normally distributed and they have to be
characterized by the same variance.
In order to check these assumptions, the same methods described for simple linear regression
analysis are adopted (see section 9.2.2.2).
9.2.4 Nonlinear regression analysis
Generally, the linear regression analysis is used in order to investigate the relation between
independent variables and a dependent one; however, if the relationship between the dependent
variable and at least some of the independent ones is not linear, or when the assumptions for linear
regressions are not satisfied, a nonlinear regression analysis should be performed.
Nonlinear regression requires estimating the coefficients in a not linear relation between
independent variables and dependent one.
The nonlinear models considered in the commercial statistical software are: logarithmic, inverse,
quadratic, cubic, polynomial, compound, logistic and exponential mathematical forms. The
mathematical function that best represents the relation between dependent and independent variable
can be individuated by analysing a scatterplot of the collected data.
Once the mathematical function has been determined, the coefficients are generally calculated on
the basis of the Levenberg-Marquardt nonlinear least squares algorithm (Nash, 1987).
Obviously, the assumptions of normality are not required and the unique assumption is that the data
are well represented by the selected model.
209
9.3 Effects of single rock/rock mass properties on TBM performances
The effect of single intact rock properties or rock mass parameters on the TBM performances has
been investigated by means of correlations and simple regression analyses.
The geological parameters involved in this study are: UCS, quartz content, α, spacing Jv, RDQ, Jc
RMR, GSI and tunnel depth; these are the independent variables. ROP and FPI are the dependent
ones.
All the analyses have been performed by means of the commercial statistical software “IBM SPSS
Statistics 25”.
9.3.1 Uniaxial compressive strength
The relationship between the uniaxial compressive strength of the intact rock and ROP and FPI has
been analysed and the result is reported in Figure 9.6.
UCS - ROP UCS - FPI

4.0 350.0
FPI [(kN/cutter)/(mm/rev)]]
3.5 300.0
3.0 250.0
ROP [m/h]
2.5
200.0
2.0
150.0
1.5
1.0 100.0
0.5 50.0
0.0 0.0
0.0 100.0 200.0 300.0 0.0 100.0 200.0 300.0
UCS [MPa] UCS [MPa]
Figure 9.6 UCS vs. ROP (on the left); UCS vs. FPI (on the right), modified after Armetti et al. 2018.
As it is possible to expect, when rock strength increases the performances of the TBM decrease. In
particular when high values of UCS characterise the intact rock, the net penetration rate is generally
low because of the chip rock formation is more difficult; on the contrary FPI increases because a
high thrust has to be applied in order to guarantee the penetration. It is necessary to consider that
there is a threshold of thrust which cannot be exceeded in order to avoid machine damages.
Linear and exponential mathematical functions have been selected as the most representative ones
to explain the relation between UCS and ROP-FPI. Previous studies described linear relationship
210
(e.g. Khademi Hamidi et al., 2010) and the exponential one (Graham, 1976; Farmer and Glossop,
1980; Hughes, 1986; Gong and Zhao, 2009; Hamidi et al., 2010).
The assumptions for linear regression analysis have been checked and they result satisfied. The
regression analyses allow to carry out 4 empirical equations (linear and exponential equations both
for ROP and FPI) listed in Table 9.1.
Equation r R2 F Reference
𝑅𝑂𝑃 = −0.010 ∗ 𝑈𝐶𝑆 + 2.829 0.550 0.306 81.305 9.12
𝑅𝑂𝑃 = 3.809 ∗ 𝑒 −0.080∗𝑈𝐶𝑆 - 0.322 87.424 9.13
𝐹𝑃𝐼 = 0.912 ∗ 𝑈𝐶𝑆 − 20.973 0.554 0.307 81.502 9.14
𝐹𝑃𝐼 = 21.553 ∗ 𝑒 0.010∗𝑈𝐶𝑆 - 0.324 88.048 9.15
Table 9.1 Results of regression analyses. For each equation the coefficient of determination and F
values are reported. Coefficient of correlation is listed only for linear models.
The coefficients of determination R2 are medium-low, it means that UCS alone cannot totally
determine the performances of a TBM and then UCS cannot be utilised as unique TBM
performance predictor.
211
9.3.2 Quartz content
The quartz content of the rocks excavated during La Maddalena construction has been related to the
obtained TBM performances. The relationship is shown in Figure 9.7.
QUARTZ - ROP QUARTZ - FPI

4.0 350.0
3.5 300.0
3.0 250.0
ROP [m/h]
2.5
200.0
2.0
150.0
1.5
1.0 100.0
0.5 50.0
0.0 0.0
0.0 20.0 40.0 60.0 0.0 20.0 40.0 60.0
q [%] q [%]
Figure 9.7 q vs. ROP (on the left); q vs. FPI (on the right), modified after Armetti et al. 2018.
It is possible to observe that if quartz content increases, the performances of TBM decrease. This
mineral is seriously responsible for the hardness, the strength and the abrasivity of the rocks, so its
determination is fundamental in order to predict tool consumption and machine penetration.
In this case, selecting a mathematical function is not completely correct in order to describe the
relation between rock quartz content and machine performance, due to the distribution of data.
However, if it is necessary to select a regression model, the linear one can be considered only for
ROP, given that the assumption of linear regression not satisfied for FPI.
The power function could be considered as the regression model able to examine the relation
between quartz content and FPI.
The models obtained by means of the regression analyses are listed in Table 9.2.
𝑅𝑂𝑃 = −0.055 ∗ 𝑞 + 3.408 0.747 0.558 232.330 9.16
𝐹𝑃𝐼 = 0.213 ∗ 𝑞1.689 - 0.468 161.656 9.17
Table 9.2 Results of regression analyses. r, R2 and F value are reported.
212
9.3.3 Alpha angle
The alpha angle represents the orientation of discontinuities with respect to the tunnel axis; its
calculation has been explained in Chapter 5. Here the relation between α and TBM performances
has been proposed (Figure 9.8).
ALPHA ANGLE - ROP ALPHA ANGLE - FPI

4.0 350.0
3.5 300.0
3.0 250.0
ROP [m/h]
2.5
200.0
2.0
150.0
1.5
1.0 100.0
0.5 50.0
0.0 0.0
0.0 20.0 40.0 60.0 80.0 0.0 20.0 40.0 60.0 80.0
α [deg] α [deg]
Figure 9.8 Alpha angle vs. ROP (on the left); α vs. FPI (on the right).
The relation between alpha and ROP-FPI is weak, since the effects of discontinuity orientation has
not to be evaluated alone but combined with the distance between the discontinuity planes. It means
that the role of the discontinuity planes has to be investigated in terms of orientation and frequency
at the same time.
It was noted that when α increases the machine performances decrease. The assumptions for the
linear regression have been verified and two simple linear regression analyses have been carried
out: the results obtained have been reported in Table 9.3.
𝑅𝑂𝑃 = −0.014 ∗ 𝛼 + 2.123 0.296 0.088 17.668 9.18
𝐹𝑃𝐼 = 1.357 ∗ 𝛼 + 40.326 0.310 0.096 19.498 9.19
Table 9.3 Results of regression analyses. For each equation r, R2 and F value are reported.
As expected, the coefficients of regression and F value are very low.
213
9.3.4 Spacing between weakness planes
The rock mass fracturing degree may seriously affect the TBM performances: for this reason, in this
work three parameters representing the ground fracturing conditions have been correlated to the
measured ROP and FPI.
The scatterplots about the spacing between the discontinuity planes and TBM indices are provided
in Figure 9.9.
SPACING - ROP SPACING - FPI

4.0 350.0
3.5 300.0
3.0 250.0
ROP [m/h]
2.5
200.0
2.0
150.0
1.5
1.0 100.0
0.5 50.0
0.0 0.0
0.0 5.0 10.0 15.0 0.0 5.0 10.0 15.0
SPACING [m] SPACING [m]
Figure 9.9 Spacing between discontinuities vs. ROP (on the left); S vs. FPI (on the right).
When the distance between the planes of weakness (spacing) is low, ROP covers a large range of
values, since the machine could potentially advance rapidly and easily given that fractures
frequently occur and chips formation is advantaged by weak planes. However, in case of high
fragmentation, the machine cannot be driven at maximum speed due to the fact that the muck
removal system would not be able to move too much muck in a short time.
FPI generally increases when fractures are highly spaced; however data distribution shows the
presence of a threshold that cannot be exceeded even when rock mass is compact and potentially a
high thrust might be requested. On the contrary, fractured rock mass sections do not require high
levels of thrust on the cutterhead.
The mathematical forms selected to explain the relation between spacing and ROP are the linear
and the polynomial ones, whereas for FPI linear model and logarithmic one (on the basis of the
threshold occurrence) have been considered.
The assumptions for linear regression are satisfied, so the empirical equations related to the
correlation between spacing and TBM performances have been reported in Table 9.4.
214
𝑅𝑂𝑃 = −0.139 ∗ 𝑆 + 1.820 0.477 0.227 54.126 9.20
𝑅𝑂𝑃 = 0.020 ∗ 𝑆 2 − 0.338 ∗ 𝑆 + 2.069 - 0.268 33.569 9.21
𝐹𝑃𝐼 = 15.654 ∗ 𝑆 + 62.824 0.593 0.352 100.034 9.22
𝐹𝑃𝐼 = 50.871 ∗ ln 𝑆 + 72.669 - 0.412 128.778 9.23
Table 9.4 Results of regression analyses. For each equation R2 and F value are reported.
Coefficient of correlation is listed only for linear models.
The relation between FPI and fracture spacing is stronger than the ROP-spacing one.
215
9.3.5 Joint volumetric count
The joint volumetric count, Jv, has been calculated on the basis of spacing data, as reported in
Chapter 5. The influence of Jv on the TBM performances has been investigated.
An appreciable correlation between Jv and TBM indices has been noted (Figure 9.10).
Jv - ROP Jv - FPI
4.0 350.0
3.5 300.0
3.0 250.0
ROP [m/h]
2.5
200.0
2.0
150.0
1.5
1.0 100.0
0.5 50.0
0.0 0.0
0.0 5.0 10.0 15.0 20.0 0.0 5.0 10.0 15.0 20.0
Jv [n*m^-3] Jv [n*m^-3]
Figure 9.10 Jv vs. ROP (on the left); Jv vs. FPI (on the right).
As it is possible to observe, the highest machine performances occur when the number of joints per
unit of volume ranges between 5 and 10 joints * m-3. In this situation the net rate of penetration is
high and the thrust which has to be applied to guarantee a specific rate per revolution is low. When
Jv is lower than 5.0 – 10.0 joints * m-3, rock mass is compact and difficult to penetrate, on the
contrary, when Jv is higher than 5.0 – 10.0 joints * m-3, the degree of fracturing is elevated, so the
same observation exposed for low levels of spacing can be considered. The best mathematical
function able to explain the relationship between Jv and ROP-FPI is the polynomial one. In this
case, Jv represents the influence of the rock mass fracturing degree on the performances of the
TBM more clearly than spacing. A polynomial regression analysis has been carried out both with
ROP and FPI as dependent variables. The results have been listed in Table 9.5.
Equation R2 F Reference
𝑅𝑂𝑃 = −0.018 ∗ 𝐽𝑣 2 + 0.279 ∗ 𝐽𝑣 + 0.949 0.312 41.579 9.24
𝐹𝑃𝐼 = 1.701 ∗ 𝐽𝑣 2 − 28.014 ∗ 𝐽𝑣 + 156.665 0.428 68.420 9.25
Table 9.5 Results of polynomial regression analysis. R2 and F values are reported.
216
9.3.6 Rock quality designation index
The Rock Quality Designation index, RQD, (Deere et al., 1967) has been also considered in order to
evaluate the influence of rock mass fracturing degree on the TBM performances.
No one correlation between RQD and ROP or FPI has been observed, it is clearly represented by
the scatterplots in Figure 9.11.
RQD - ROP RQD - FPI

4.0 350.0
3.5 300.0
3.0 250.0
ROP [m/h]
2.5
200.0
2.0
150.0
1.5
1.0 100.0
0.5 50.0
0.0 0.0
0.0 25.0 50.0 75.0 100.0 0.0 25.0 50.0 75.0 100.0
RQD [%] RQD [%]
Figure 9.11 RQD vs. ROP (on the left); RQD vs. FPI (on the right).
The RQD index about La Maddalena rock mass has been calculated by means of the empirical
equation proposed by Palmström, 1982. The results obtained are almost always very high, even in
those situations in which an intense fracturing degree has been observed, as explained in Chapter 5.
By relating RQD with TBM performance measured during the excavation, it is not possible to
observe the machine behavior where jointing level is high, so where RQD should be low. For this
reason, RQD cannot be used as prediction in a regression analysis.
Moreover, when RQD is empirically calculated (for example by means of the Palmström equation),
its use as predictor in an empirical model is not suitable because using an empirically derived
variable into a regression analysis might not useful.
217
9.3.7 Joint conditions
The joint conditions parameter, Jc, takes into account for five discontinuity surface features
(persistence, aperture, roughness, filling and weathering); for this reason, relating it with the TMB
performance indices, it may be useful to investigate the effect of the general conditions of weakness
planes on the behavior of the machine involved in tunnel excavation.
The relationship between joint conditions and ROP-FPI has been represented by the scatterplots
reported in Figure 9.12.
JOINT CONDITIONS - ROP JOINT CONDITIONS - FPI

4.0 350.0
3.5 300.0
3.0 250.0
ROP [m/h]
2.5
200.0
2.0
150.0
1.5
1.0 100.0
0.5 50.0
0.0 0.0
0.0 10.0 20.0 30.0 0.0 10.0 20.0 30.0
Jc [-] Jc [-]
Figure 9.12 Jc vs. ROP (on the left); Jc vs. FPI (on the right).
It is observable that the Jc parameter and the machine ones are barely related; however, it is
possible to recognise that where the global quality of joint surfaces increases the TBM
performances decrease.
The effect of Jc on the machine performances should not be evaluated alone, but together with the
other parameters describing the geometry (in terms of orientation respect to the tunnel axis) and the
frequency of joint occurrence. The quality of fracture surface gives limited information if
considered alone.
Regression analysis has been performed even through the low correlation between the independent
variables and dependent ones. Only the results obtained for FPI have been reported, given that the
coefficient of regression related to Jc-ROP relationship is very low (Table 9.6).
218
𝐹𝑃𝐼 = 5.600 ∗ Jc + 0.514 0.335 0.107 23.277 9.26
𝐹𝑃𝐼 = 0.455 ∗ 𝐽𝑐 2 − 11.060 ∗ 𝐽𝑐 + 146.646 - 0.141 15.054 9.27
Table 9.6 Result of linear and polynomial regression analysis. For each equation R2 and F value
are reported. Coefficient of correlation is listed only for linear model.
219
9.3.8 Tunnel Depth
The depth of excavation is a first indicator of in situ stress condition: it is clear that this parameter is
not enough alone to predict the performance of a TBM boring at a certain depth.
The correlation between the depth achieved by La Maddalena tunnel and the performances
measured at the relative depths has been studied (Figure 9.13).
TUNNEL DEPTH - ROP TUNNEL DEPTH - FPI

4.0 350.0
3.5 300.0
3.0 250.0
ROP [m/h]
2.5
200.0
2.0
150.0
1.5
1.0 100.0
0.5 50.0
0.0 0.0
0.0 200.0 400.0 600.0 0.0 200.0 400.0 600.0
Depth [m] Depth [m]
Figure 9.13 Tunnel depth vs. ROP (on the left); Tunnel depth vs. FPI (on the right).
By observing the depth-ROP scatterplot, the rate of penetration appears to increase with the
overbunden. The correlation is poor and it means that the vertical stress affects the performance of
the TBM but its effect has to be evaluated in addition to other geological and mechanical
parameters.
Tunnel depth appears not to influence FPI; however this absence of evident correlation does not
mean that FPI is not influenced by vertical stress, but it means that the vertical stress alone does not
significantly affect boring machine. Linear and nonlinear regression analyses have been carried out
in order to explain the relation between tunnel depth and the rate of penetration. The results are
summarised in Table 9.7.
𝑅𝑂𝑃 = 0.002 ∗ 𝑑𝑒𝑝𝑡ℎ + 0.858 0.314 0.098 20.063 9.28
𝑅𝑂𝑃 = 0.053 ∗ 𝑑𝑒𝑝𝑡ℎ0.541 - 0.207 48.038 9.29
Table 9.7 Results of linear and not linear regression analysis. R2 and F value are reported.
220
9.3.9 Rock Mass Rating
The effect of the global quality of rock mass on the TBM performances has been evaluated by
means of correlate the RMR index with the machine parameters. The scatterplots obtained have
been reported in Figure 9.14.
ROCK MASS RATING - ROP ROCK MASS RATING - FPI

4.0 350.0
3.5 300.0
3.0 250.0
ROP [m/h]
2.5
200.0
2.0
150.0
1.5
1.0 100.0
0.5 50.0
0.0 0.0
0 20 40 60 80 100 0 20 40 60 80 100
RMR [-] RMR [-]
Figure 9.14 RMR vs. ROP (on the left); RMR vs. FPI (on the right), modified after Armetti et al.,
2018.
By increasing the rock mass quality, here expressed by means of RMR system, the performances of
the TBM generally decreases, in particular ROP reduces and FPI increases. However; the machine
performances become lower not only when rock mass sections characterised by high quality occur,
but also in low quality grounds because highly jointed or weak rock walls cannot contrast the
gripper thrust without make tunnel walls instable. In these situations, FPI is reduced so also ROP is
low.
The top TBM performances have been measured in class II (good) as also observed by Sapigni et al.
(2002).
The TBM behaviour in poor quality classes is not describable due to the general good or very good
quality of La Maddalena rock mass. The performances in lower classes are predictable by observing
the polynomial regression curve related to the RMR-ROP analysis.
In order to quantify the correlation between RMR and the performance indices two linear regression
analysis and two polynomial ones have been carried out. The results have been proposed in Table
9.8.
221
𝑅𝑂𝑃 = −0.023 ∗ 𝑅𝑀𝑅 + 3.038 0.425 0.180 40.516 9.30
𝑅𝑂𝑃 = −0.001 ∗ 𝑅𝑀𝑅 2 + 0.089 ∗ 𝑅𝑀𝑅 − 0.665 - 0.228 27.000 9.31
𝐹𝑃𝐼 = 2.900 ∗ 𝑅𝑀𝑅 − 97.590 0.599 0.359 103.093 9.32
𝐹𝑃𝐼 = 0.052 ∗ 𝑅𝑀𝑅 2 − 4.222 ∗ 𝑅𝑀𝑅 + 137.889 0.383 56.696 9.33
Table 9.8 Result of linear and polynomial regression analysis. For each equation R2 and F values
are reported. Coefficient of correlation is listed only for linear models.
222
9.3.10 Geological Strength Index
The correlation between the Geological Strength Index and ROP and FPI has been evaluated, and
the results obtained are very similar to the ones observed for RMR-ROP and RMR-FPI correlations.
For this reason, in this section only the scatterplots (Figure 9.15) and the results of linear and
nonlinear regression analyses have been reported (Table 9.9); as for the discussion related to the
global rock mass quality effect on the TBM performance, see section 9.3.9.
GSI- ROP GSI - FPI

4.0 350.0
3.5 300.0
3.0 250.0
ROP [m/h]
2.5
200.0
2.0
150.0
1.5
1.0 100.0
0.5 50.0
0.0 0.0
0 20 40 60 80 100 0 20 40 60 80 100
GSI [-] GSI [-]
Figure 9.15 GSI vs. ROP (on the left); GSI vs. FPI (on the right), modified after Armetti et al.,
2018.
𝑅𝑂𝑃 = −0.022 ∗ 𝐺𝑆𝐼 + 3.010 0.374 0.140 29.936 9.34
𝑅𝑂𝑃 = −0.001 ∗ 𝐺𝑆𝐼 2 + 0.141 ∗ 𝐺𝑆𝐼 − 2.495 - 0.218 25.450 9.35
𝐹𝑃𝐼 = 2.920 ∗ 𝐺𝑆𝐼 − 105.901 0.560 0.313 84.011 9.36
𝐹𝑃𝐼 = 0.079 ∗ 𝐺𝑆𝐼 2 − 8.074 ∗ 𝐺𝑆𝐼 + 265.835 - 0.357 50.769 9.37
Table 9.9 Result of linear and polynomial regression analysis. For each equation R2 and F values
are reported. Coefficient of correlation is listed only for linear models.
223
9.4 Combined effect of rock/rock mass properties on TBM performances
The combined effect of the geological and geotechnical features of a rock mass on the TBM
performances registered during its excavation has been investigated by means of multiple regression
analyses, which allow to carry out empirical equations.
These equations are the most practical way to explain the relationship between more independent
variables and a response one, and are used to predict a parameter from collected data: for this reason
they have a great importance during early phases of tunnel design.
During this work, several regression analyses have been performed in order to predict both ROP
and FPI. Initially a ROP model and an FPI one has been carried out both for each recognised
geological domain and for the entire tunnel length considered to this purpose (models have been
obtained on the tunnel tract ranging between Pk. 0 + 197 m and Pk. 2 + 010 m).
It was noted that the most performing models were the ones carried out by considering the
geological domains all together and not separately. This is because considering the whole domains
more data are involved into the computation and data range is larger, so different situation can be
considered at the same time, on the contrary the models related to singular domain take into account
for prevalently homogeneous situations.
The results of the regression analyses reported in the following paragraphs have been developed
considering the domains all together.
224
9.4.1 Empirical models to predict the rate of penetration
As mentioned above, in order to obtain an empirical correlation able to represent the relation
between rock mass features and ROP, both linear and not linear regression analyses have been
performed.
Linear regression models should be preferred in order to investigate the relationship between more
than one independent variable and a dependent one, because linear models are the simplest to be
carried out and used. For this reason, nonlinear regression analyses have to be developed when
linearity assumptions are not satisfied.
9.4.1.1 Multiple linear regression
The geological parameters, considered as predictor, to enter as input in multiple regressions

analysis, have been selected on the basis of the single regression results. For this purpose the
selected parameters are:
- UCS: it gives fundamental information about intact rock strength and affects TBM
performances;
- Quartz content: it influences the abrasivity and the hardness of rock, so it seriously affects
tunnelling;
- Spacing or Jv: they are strongly related to each other because they represent the same
phenomenon; for this reason they cannot be both inserted in the same model. In this study it
was decided to input the spacing between discontinuity as fracturing degree indicator, being
Jv a derived parameter;
- Alpha angle, Jc, Tunnel depth: these parameters alone are not strongly related to ROP; in
this analysis we test their influence when combined with other factors.
RMR and GSI are not included because they consider some of the parameter which have been
entered separately.
The model has been generated adopting a stepwise backward selection; at the beginning of the
analysis all predictors are included, then the variables that have not significant influence on the
response are removed one at a time. The variable is excluded when the null hypothesis (the
dependent variable does not influence the independent one) cannot be rejected (p-value >0.05).
225
In this case the parameter that does not affect the rate of penetration is α (p-value >0.05): for this
reason a second linear model, named RL2, has been performed.
The results of the analysis have been reported in Table 9.10.
Coefficients
Unst.ed Std.ed Collinearity
coefficients coeff. Sign. 95,0% C.I. for B statistics
Model
Std. (p- Bottom Top
B error Beta t value) Limit Limit Toler. VIF
RL (Constant) 2.944 0.248 11.864 0.000 2.455 3.434
1
UCS -0.004 0.001 -0.236 -4.578 0.000 -0.006 -0.002 0.725 1.380
q -0.040 0.004 -0.555 -10.043 0.000 -0.048 -0.033 0.633 1.580
α -0.002 0.002 -0.042 -0.896 0.371 -0.006 0.002 0.889 1.124
S -0.056 0.016 -0.192 -3.572 0.000 -0.087 -0.025 0.668 1.496

Jc 0.030 0.009 0.164 3.250 0.001 0.012 0.049 0.761 1.314
depth 0.001 0.000 0.114 2.445 0.015 0.000 0.001 0.884 1.310
RL (Constant) 2.889 0.240 12.027 0.000 2.415 3.363

2
UCS -0.004 0.001 -0.239 -4.632 0.000 -0.006 -0.002 0.727 1.376
q -0.041 0.004 -0.565 -10.447 0.000 -0.049 -0.033 0.660 1.516
S -0.057 0.016 -0.195 -3.636 0.000 -0.088 -0.026 0.671 1.491
Jc 0.030 0.009 0.163 3.230 0.001 0.012 0.048 0.762 1.313
depth 0.001 0.000 0.116 2.475 0.014 0.000 0.001 0.885 1.130
Table 9.10 Results of multiple linear regression analysis, red box underlines the no significant
parameters.
The second regression gives the coefficient of those variables that significantly influence the rate of
penetration (p-value < 0), the null hypothesis is rejected. They are: UCS, quartz content, spacing,
joint conditions and tunnel depth.
Problems of multicollinearity do not occur (VIF < 10) and the linear regression assumptions are
verified, by testing the random errors; they are independent and normally distributed, they have a
mean equal to zero and homoscedasticity has been observed.
226
R2 (the correlation strength indicator), R2adj. (that takes into account for the number of predictors)
and F-value (related to the utility of the overall regression model) are listed in Table 9.11.
r R2 R2adj F Sign.
0.809 0.653 0.643 67.683 0.000
Table 9.11 RL2 Model summary.
R2adj is 0.643, it means that the regression model is able to explain the 64.30% of the total variance
of the dataset. F-value is equal to 67.683 and Sign. is 0.000, so the null hypothesis can be rejected.
The equation able to explain the relationship between the five geological parameters and ROP is as
follows:
𝑅𝑂𝑃 = −0.004 ∗ 𝑈𝐶𝑆 − 0.041 ∗ 𝑞 − 0.057 ∗ 𝑆 + 0.030 ∗ 𝐽𝑐 + 0.001 ∗ 𝑑 + 2.889 (9.38)
By observing the regression coefficients, it is possible to assert that the equation can be adopted to
explain the simultaneous influence of five geological properties on the TBM performance. In
Chapter 12, the capability of the model to predict the rate of penetration will be discussed.
9.4.1.2 Multiple nonlinear regression analysis
It has been observed that the relationship between some geological parameters and ROP (or FPI),
seems to be nonlinear, on the basis of the analyses of the scatterplots reported in section 9.3.
Carrying out multiple nonlinear regression analyses may be useful in order to evaluate the nonlinear
relationship between the geological and geotechnical features of the rock mass and the rate of
penetration.
Nonlinear regressions can be performed in two steps: in the first one the mathematical function that
best represent the relation between a given independent variable and a dependent one has to be
selected. This selection has to be performed for each variable that is necessary to input into the
analysis. Then the mathematical functions have to be entered into the model in order to evaluate
their combined effects.
In this study the more representative mathematical functions have been recognised by means of the
scatterplots (section 9.3). Moreover, to this purpose simple nonlinear regression analyses have been
performed, the results have been listed in Table 9.12.
227
Independent
Variable
Model R2 F Sign. Constant b1 b2
UCS Linear 0.306 81.305 0.000 2.829 -0.010
Logarithmic 0.290 75.243 0.000 7.823 -1.306
Quadratic 0.306 40.435 0.000 2.864 -0.011 1.811E-6
Power 0.303 79.802 0.000 196.760 -1.033
Exponential 0.322 87.424 0.000 3.809 -0.008
q Linear 0.558 232.330 0.000 3.408 -0.055
Logarithmic 0.572 246.176 0.000 8.293 -1.930
Quadratic 0.583 127.693 0.000 6.688 -0.250 0.003
Power 0.565 238.526 0.000 2.465 -1.485
Exponential 0.560 234.192 0.000 5.824 -0.042
α Linear 0.088 17.688 0.000 2.123 -0.014
Logarithmic 0.086 17.356 0.000 3.472 -0.533
Quadratic 0.089 8.952 0.000 2.287 9.226E-5 -0.023
Power 0.100 20.530 0.000 6.897 -0.445
Exponential 0.107 22.127 0.000 2.263 -0.012
S Linear 0.227 54.126 0.000 1.820 -0.139
Logarithmic 0.244 59.433 0.000 1.722 -0.434
Quadratic 0.268 33.569 0.000 2.069 -0.338 0.020
Power 0.262 65.460 0.000 1.584 -0.348
Exponential 0.256 63.185 0.000 1.725 -0.114
Jc Linear 0.035 6.660 0.011 2.101 -0.035
Logarithmic 0.024 4.578 0.034 2.882 -0.489
Quadratic 0.073 7.234 0.001 0.240 0.178 -0.006
Power 0.050 9.701 0.002 6.205 -0.544
Exponential 0.066 13.095 0.000 2.532 -0.037
Depth Linear 0.098 20.063 0.000 0.858 0.002
Logarithmic 0.138 29.573 0.000 -1.901 0.572
Quadratic 0.195 22.095 0.000 -0.430 0.010 -1.122E-5
Power 0.207 48.038 0.000 0.053 0.514
Exponential 0.158 34.410 0.000 0.709 0.002
Table 9.12 Linear and nonlinear models between geological parameters and ROP. In bold the
selected models.
228
The equations selected to be entered into the nonlinear model are:
𝑅𝑂𝑃 = 3.809 ∗ 𝑒 −0.008𝑈𝐶𝑆 (9.39)
𝑅𝑂𝑃 = −1.930 ∗ 𝑙𝑛(𝑞) + 8.293 (9.40)
𝑅𝑂𝑃 = 2.263 ∗ 𝑒 −0.012𝛼 (9.41)
𝑅𝑂𝑃 = 1.584 ∗ 𝑆 −0.348 (9.42)
𝑅𝑂𝑃 = 2.532 ∗ 𝑒 −0.037𝐽𝑐 (9.43)
𝑅𝑂𝑃 = 0.053 ∗ 𝑑0.514 (9.44)
The global multiple nonlinear equation is as follows (model RNL):
𝑅𝑂𝑃
= 22.316𝑒 −2.120𝐸−4𝑈𝐶𝑆 − 1.426𝑙𝑛(𝑞) + 7.806𝑒 −0.0003𝛼 + 31.171𝑆 −0.003 − 1.412𝑒 −0.094𝐽𝑐
+ 0.001𝑑1.024 − 53.952 (9.45)
The coefficient of determination, R2, related to RNL is equal to 0.655.
229
9.4.2 Empirical models to predict the field penetration index
9.4.2.1 Multiple linear regression
Multiple linear regression analyses have been developed in order to investigate the influence of the
principal rock mass parameters on the field penetration index.
The procedures adopted are the same discussed in the paragraphs about ROP models (see section
9.4.1.1), for this reason here only the analyses results and a brief comment on them are reported.
The input parameters are: UCS, quartz content of the rock, spacing between fractures, alpha angle,
joint condition parameters, and the depth of the tunnel; these are the independent variables.
The response variable is FPI.
The model has been generated adopting a stepwise backward selection; for this reason, at the
beginning all predictors are included, then the not influencing variables (p-value >0.05) are
removed one at a time.
In this case the parameter that does not affect the field penetration index are: Jc (Sign. 0.926),
tunnel depth (Sign. 0.726) and alpha angle (Sign. 0.143).
The results of the regression analyses have been reported in Table 9.13.
230
Coefficients
coefficients coeff. Sign. 95,0% C.I. for B statistics
Model
Std. (p- Bottom Top
B error Beta t value) Limit Limit Toler. VIF
FL1 (Constant) -78.297 24.605 -3.182 0.002 -126.850 -29.744
UCS 0.386 0.093 0.234 4.134 0.000 0.202 0.570 0.725 1.380
q 2.472 0.400 0.375 6.185 0.000 1.683 3.261 0.633 1.580
α 0.330 0.224 0.075 1.473 0.142 -0.112 0.773 0.889 1.124
S 8.498 1.556 0.322 5.461 0.000 5.427 11.569 0.668 1.496
Jc 0.086 0.924 0.005 0.094 0.926 -1.737 1.910 0.761 1.314
depth 0.008 0.023 0.018 0.360 0.719 -0.037 0.053 0.884 1.131
FL2 (Constant) -76.917 19.647 -3.915 0.000 -115.686 -38.148
UCS 0.386 0.093 0.235 4.165 0.000 0.203 0.569 0.729 1.372
q 2.474 0.398 0.375 6.215 0.000 1.688 3.259 0.635 1.576
α 0.331 0.223 0.075 1.480 0.141 -0.110 0.772 0.890 1.124
S 8.549 1.456 0.324 5.872 0.000 5.676 11.421 0.759 1.317
depth 0.008 0.023 0.018 .351 0.726 -0.037 0.052 0.905 1.105
FL3 (Constant) -72.394 14.803 -4.891 0.000 -101.602 -43.186
UCS 0.386 0.093 0.235 4.175 0.000 0.204 0.569 0.729 1.372
q 2.439 0.385 0.370 6.339 0.000 1.680 3.199 0.676 1.480
α 0.328 0.223 0.075 1.473 0.143 -0.111 0.768 0.891 1.122
S 8.547 1.452 0.324 5.885 0.000 5.681 11.413 0.759 1.317

FL4 (Constant) -63.491 13.555 -4.684 0.000 -90.236 -36.745
UCS 0.394 0.093 0.239 4.247 0.000 0.211 0.576 0.731 1.368
q 2.564 0.377 0.389 6.807 0.000 1.821 3.307 0.710 1.408
S 8.707 1.453 0.330 5.993 0.000 5.840 11.573 0.764 1.310
Table 9.13 Results of multiple linear regression analysis, red boxes underline the no significant
parameters.
231
Four models have been created; FL1, FL2 and FL3 involve parameters (Jc, tunnel depth and α) that
do not significantly affect the response variable. The model FL4 considers only the influencing
independent variables, they are: UCS, quartz content and the spacing between fractures. FL4 is not
affected by multicollinearity problems and the linearity assumptions are satisfied.
The carried out empirical equation is as follows:
𝐹𝑃𝐼 = 0.394 ∗ 𝑈𝐶𝑆 + 2.564 ∗ 𝑞 + 8.707 ∗ 𝑆 − 63.491 (9.46)
The regression indices are summarised in Table 9.14.
r R2 R2adj F Sign.
0.761 0.568 0.571 83.222 0.000
Table 9.14 FL4 model summary.
The regression model named FL4 is able to explain the 57.10% of the total variance, it modestly
represents the relationship between UCS, q, S and FPI.
232
9.4.2.2 Multiple nonlinear regression
Multiple nonlinear regression analyses have been carried in order to investigate the effects of the
geological and mechanical parameters on the FPI.
The procedure is the same of the one discussed about ROP model (see section 9.4.1.2). In this case
too, the most representative nonlinear functions have been individuated on the basis of the
scatterplots and by means of simple nonlinear regressions.
The results are listed in Table 9.15.
233
Independent
Variable
UCS Linear 0.307 81.502 0.000 -20.973 0.912
Logarithmic 0.310 82.756 0.000 121.997 -491.460
Quadratic 0.317 42.554 0.000 -91.105 1.967 -0.004
Power 0.324 88.374 0.000 0.125 1.336
Exponential 0.324 88.048 0.000 21.553 0.010
q Linear 0.414 129.729 0.000 -48.921 4.240
Logarithmic 0.426 136.557 0.000 -430.079 150.462
Quadratic 0.444 72.990 0.000 -377.938 23.836 -0.272
Power 0.468 161.656 0.000 0.213 1.689
Exponential 0.450 150.420 0.000 15.444 0.047
α Linear 0.096 19.498 0.000 40.326 1.357
Logarithmic 0.089 17.887 0.000 -81.448 48.792
Quadratic 0.096 9.708 0.000 36.297 1.558 -0.002
Power 0.073 14.416 0.000 14.045 0.473
Exponential 0.082 16.399 0.000 45.215 0.013
S Linear 0.352 100.034 0.000 62.824 15.654
Logarithmic 0.412 128.778 0.000 72.669 50.871
Quadratic 0.423 67.109 0.000 33.244 39.216 -2.431
Power 0.408 126.951 0.000 60.941 0.543
Exponential 0.316 85.002 0.000 55.965 0.159
Jc Linear 0.112 23.277 0.000 0.514 5.600
Logarithmic 0.092 18.645 0.000 -145.474 86.042
Quadratic 0.141 15.054 0.000 146.646 -11.060 0.455
Power 0.070 13.841 0.000 8.247 0.804
Exponential 0.086 17.235 0.000 32.202 0.052
Depth Linear 0.034 6.529 0.011 134.180 -0.083
Logarithmic 0.045 8.621 0.004 274.663 -29.366
Quadratic 0.061 5.969 0.003 195.753 -0.477 0.001
Power 0.057 11.200 0.001 677.540 -0.356
Exponential 0.046 8.771 0.003 124.179 -0.001
Table 9.15 Linear and nonlinear models between geological parameters and FPI. In bold the
selected models.
234
On the basis of the scatterplots reported in section 9.3 and considering the results listed in Table
9.15 alpha and depth have been excluded from the multiple analyses. The equations entered into the
model are:
𝐹𝑃𝐼 = 0.125 ∗ 𝑈𝐶𝑆 1.336 (9.47)
𝐹𝑃𝐼 = 0.213 ∗ 𝑞1.689 (9.48)
𝐹𝑃𝐼 = 50.871 ∗ 𝑙𝑛(𝑆) + 72.669 (9.49)
𝐹𝑃𝐼 = 0.514 ∗ 𝐽𝑐 + 5.600 (9.50)
The equations 9.45, 9.46, 9.47 and 9.48 have been considered in order to evaluate the nonlinear
effect of multiple parameters on the FPI. The empirical model, named FNL, which has been
obtained, is as follows:
𝐹𝑃𝐼 = 8.408𝑈𝐶𝑆 0.509 + 7513.902𝑞 0.011 + 27.552𝑙𝑛(𝑆) + 0.080𝐽𝑐 − 7818.256 (9.51)
The coefficient of determination related to FNL is 0.590.
235
9.5 Conclusion
9.5 Conclusion
In order to evaluate the relationships between the geological and mechanical parameters that
characterise the rock mass excavated during La Maddalena construction and the TBM performances
registered during tunnel boring regression analyses have been carried out.
The bivariate regression analyses, linear and not linear type, between single rock/rock mass
parameters and TBM performance indices (ROP and FPI) have allowed to investigate the effects of
the geological conditions on the machine involved to excavate into a specific ground. During this
study twenty correlations have been tested: in three cases no correlation has been observed.
When correlation (more or less strong) occurs, simple regression analyses have been developed.
These permit both to evaluate the correlation between a single rock parameter and a single TBM
one, and the mathematical form that best describes this relation.
Thirteen regression equations have been obtained about the relationship between the geological
features of La Maddalena rock mass and the measured ROP; seven of them are linear and six not
linear. About FPI, six linear regression equations and seven not linear ones have been carried out.
The geological parameters more related to the TBM indices are UCS, quartz content, spacing and Jv
and the parameters which represent the global quality of rock mass (RMR and GSI).
The simple regression analyses do not allow for evaluating the combined effects of all the
parameters characterising a rock mass; for this purpose it is necessary to carry out multiple
regression analyses. Linear regression analyses have been developed both for ROP and FPI. The
R2adj. of ROP model is slightly higher than the FPI one. Linearity assumptions are satisfied in both
cases, for this reason the linear models can be considered able to explain the influence of geology
on tunnel excavation.
Furthermore nonlinear multiple regression analyses have been considered and a model has been
created for both TBM indices. The type of nonlinear relationship has been imposed: in other words,
it has been assumed that the geological parameters and the TBM ones are linked by a specific type
of mathematical function. It might be considered as an error, and for this reason it is recommended
to avoid the use of nonlinear multiple models. Linear models are simpler to create and use,
moreover, when the assumptions that they imply are verified, they are statistically more reliable
than the nonlinear ones.
236
10 CLUSTER ANALYSIS OF ROCK MASS PARAMETERS
FOR TBM PERFORMANCE ESTIMATION
In this chapter a method able to estimate the rock mass excavatability is presented.
As it is well known, the performances of a TBM employed to bore a given rock mass depend on a
large quantity of geological, structural and mechanical rock mass parameters and on machine
specification. The most common way to analyse the combined effect of all these parameters is the
multiple regression analysis (as described in Chapter 9); however during this research a new
multidimensional geological-TBM data analysis tools has been involved: they are represented by
multiple correspondence analysis (MCA) and by clustering.
These statistical tools allowed to divide the rock masses encountered during the construction of “La
Maddalena” exploratory tunnel in three homogeneous portions characterised by specific
geomechanical features and TBM performances. In this way three classes of excavatability have
been defined.
Multiple regression analyses for each cluster have been carried out in order to provide a
mathematical function able to relate the geological characteristics of clusters and the performances
measured during its boring.
This chapter is divided in two parts: in the first one some theoretical notes on the involved statistical
methods have been provided and in the second one their application at “La Maddalena” case study
and the obtained results are explained.
237
10.1 Multidimensional data analysis
10.1.1 Introduction
Survey results are usually summarized in tables by counting the response to a series of questions
and the observations (the individuals) are described by several variables; in other words, a group of
individuals is described by a set of variables. Very often the specific goal is to describe the data
structure under two aspects. The first one is to determine which observations are similar to each
other and the second one is to understand the relationship between the variables.
In order to reach this objective, a set of statistical methods is available; they are listed in Figure
10.1.
Figure 10.1 Multivariate data analysis methods (Husson et al., 2017).
Exploratory Data Analysis (EDA) and in particular Multivariate Data Analysis (MDA) refer to all
descriptive methods for multivariate data set which allow to visualise and describe the data set. In
particular these methods permit to study the similarities and the dissimilarities between individuals
from a multidimensional point of view (Husson et al., 2010). EDA is fundamental in all statistical
analysis both as main aim and as preliminary study.
Two kinds of methods are able to study the similarities between individuals:
238
- The first one is the Principal Component Methods such as Principal Component Analysis
(PCA) adopted for continuous variables or Multiple Correspondence Analysis (MCA) for
the categorical ones (Greenacre, 2006);
- The second one is the clustering analysis. In particular hierarchical and partitional clustering
methods can be distinguished. The first one requires to define a distance (Euclidean,
Manhattan, etc.) and an agglomeration criterion (Ward, centroid, etc.) (Husson et al., 2010).
A tree, named dendogram, represents the indexed hierarchy. The partitional clustering
method is another way to study the similarities between subjects. The most utilised
algorithm of this group is based on the Euclidean distance and it is the K-mean algorithm.
The clusters of individuals are then described by the variables (Husson et al., 2010).
Principal component methods are frequently used as pre-processing step of clustering with the aim,
for example, of denoising data or transform categorical data in continuous ones.
Principal component and clustering methods ones are often combined in order to better understand
both the characteristics of collected data and the similarities/dissimilarities between individuals.
10.1.2 Euclidean space
A dataset is a more or less a large collection of information, that are represented as points belonging
to some spaces. The most common one is the Euclidean space which has a set of considerable
properties useful for clustering, however also the non-Euclidean space has to be considered (see
section 10.1.4.4). The points of the Euclidean space are vectors of real numbers; theirs length is the
number of the space dimensions. The coordinates of the represented point are the components of the
vector. All spaces have a distance measure giving a distance between any two points in the space. In
the Euclidean space the most frequently calculated distance is the Euclidean one, computed as the
square root of the sum of the squares of the differences between point coordinates in each
dimension.
Others distance measures are the Manhattan one (calculated as the magnitude sum of the differences
in each dimension, Birant and Kut, 2007) and the L∞ - distance (calculated as the maximum
difference magnitude in any dimension).
A relationship between two points, in order to be the measuring distances function, has to satisfy the
following requirements:
- The distance has to be positive;
239
- Only the distance between a point and itself can be null;
- Distance is symmetric (it is independent from the order that the points are considered during
the calculation of distance);
- Distance measures respond on the triangle inequality (it means that the distance from x to y
to z is always more than the distance going from x to z directly).
10.1.3 Principal component methods
When one of the principal component methods is chosen in order to perform a statistical analysis,
individuals are considered in an Euclidean space of high dimensions and the study of the shape of
the point cloud permits to study the similarities between individuals (Husson et al., 2010). Principal
component methods approximate the point cloud into a low dimensional Euclidean subspace while
maintaining the same distance between individuals.
The main purpose common to all principal component methods is to describe a data set (with
individuals and variables) utilizing a small number of variables uncorrelated each other with the aim
of obtaining as much information as possible. A brief description of principal component methods
has been reported.
- Correspondence analysis, CA: it is a multivariate graphical technique that allows to study

the relationships between two variables (Abdi and Valentin, 2007). They are compared in
pairs and they are not considered all together. As outcome, CA shows a graphic display of
the rows and columns of a contingency table that allows to visualize the major relationships
among the variable responses in a low-dimensional space (Abdi and Valentin, 2007);
- Principal component analysis, PCA: it can be considered as a denoising method which

separates noise and signal. In fact, the first dimension extracts the essential information and
the second one is related to noise. The following clustering is more stable without the noise
than the one carried out from the original distance. The relationships between variables are
studied all together, but this approach presents some limitations indeed it considers the
linear relationship between variables as the only possible one (actually more complex links
can also exist) and it focuses on a Gaussian distribution of variables;
- Multiple correspondence analysis, MCA: In this case too, the relationships between
variables are studied all together, but the limitations that affect the PCA do not occur. It
240
focuses both on linear and nonlinear relationship and without distributional assumptions,
unlike traditional multivariate technique.
In this study only MCA has been applied, for this reason in the following paragraphs only MCA
will be described.
10.1.3.1 Multiple Correspondence Analysis
Multiple Correspondence Analysis is one of the most widespread statistical methods involved to
inspect and describe multivariate association among variables and /or patterns of similarity among
statistical units or “individuals”. MCA is commonly used to analyse data collected through surveys
and it is applied to tables with individuals (in the rows) and categorical variables (in the columns).
The target is to identify:
- Groups of individuals with similar profile in their answer to the questions;
- The associations between categorical variables.
MCA is an extension of CA for summarizing a data table containing more than two variables (Abdi
and Williams, 2010); in other words MCA is the specific application of correspondence analysis to
tables with individuals and their answers to a number of categorical variables (Husson et al. 2017).
MCA can be considered as a way to code categorical variable onto a set of continuous ones, the
principal components. As for principal component analysis, only the first dimension is usually kept
in consideration to stabilize the eventual following clustering by deleting the noise from the data
(Husson et al., 2010).
MCA is designed for the analysis of categorical variables; however, numerical variables can be also
analysed by subdividing their values in a finite set of classes. In MCA, both the modalities of the
variables and the individuals are represented as clouds of points, vectors, in Euclidean spaces,
typically of high dimensionality. Then the clouds are projected in low-dimensional spaces (usually
characterized by two or three dimensions); however, preserving the highest possible amount of data
variability is requested.
As a result, MCA allows to represent a multitude of variables and/or statistical units in few
graphical displays (usually one or two), thus achieving the goal of “data reduction”. The displays
allow to investigate the association/similarity patterns by the interpretation of the configurations of
the clouds of points represented within.
Two essential rules of interpretation are:
241
- Concerning the association among variables: when two or more points are close to one
another, this suggests the possible presence of at least a subgroup of statistical units
characterised by these modalities;
- Concerning the similarity among individuals: the lower the distance among two or more
points, the higher the similarity among the corresponding units.
As a consequence of the second rule, the coordinates of the points can be involved to identify
subgroups of individuals with similar characteristics, or, in other words, to detect the presence of
homogeneous clusters.
In MCA, the overall quality of representation of the displays is measured by the index of explained
inertia, which quantifies the relative amount (expressed as percentage) of data variability preserved
by the low-dimensional spaces. As a rule, a relatively high proportion of explained inertia (e.g. >
50%) usually indicates good quality of representation.
This method does not provide a systematic procedure able to assign the statistical units to the
clusters; that is to establish, for each individual, to which cluster it belongs. For this purpose, a
frequently adopted procedure consists in performing cluster analysis on the coordinates of the
points in the respective displays.
Finally, it must be pointed out that, since MCA is based on a merely geometrical approach, the
groups of individuals with similar profiles and the associations between categorical variables can be
deducted by observing the graphical displays carried out by MCA method; however, it is known
that, sometimes, the displays could show spurious or unrealistic relationships.
Therefore, to rule out the possibility of faulty conclusions, it is of fundamental importance that the
results should be evaluated by field experts and in order to make these considerations more
objectives clustering procedure can be applied.
Study of individuals
An individual is a row of the data table, composed by a set of K numerical values that evolves
within the so called individual’s space, a space RK (Husson et al., 2017). The scatterplot of the
individuals in RK is represented in Figure 10.2.
242
Figure 10.2 Scatterplot of the individuals in RK; (Husson et al., 2017).
The study of individuals aims to understand the similarities between the individuals in terms of all
variables (which are the most similar/dissimilar individuals) and point out if there are consistent in
terms of their similarities.
The Euclidean distance between individuals depends on their characteristics and it gives
information about their similarity: for example, if two individuals have similar values within the
table of all K variables they are also “close”. The distance is expressed by using an indicator matrix1
as:
𝐾
′)
(𝑥𝑖𝑘 − 𝑥𝑙𝑘 )2
𝑑(𝑖, 𝑖 = √𝐶 ∑ (10.1)
𝐼𝑘
𝑘=1
where:
- (xik - xlk) is the difference between categories;
- Ik is the number of individuals carrying the category k;
- C is a constant.
1
From a data table of individuals X categorical variables an indicator matrix with individuals in the rows and
all of the categories for every variable in the columns has been constructed. In the table the element xij has a
value 1 if the individual i carries the category k, vice versa it has value 0. For this reason the table I X K is
composed by 1 and 0 only. (Husson et al., 2017).
243
In order to obtain a graphical representation of the cloud of individuals, NI (Figure 10.3) they are
projected on a plane P of RK. NI is made up of many points and it is built to see if it is identifiable a
specific shape or particular groups of individuals.
During the representation it is necessary to minimise the NI distortion: for this reason the distance
between the projected points has to be as close as possible to the distance between the original ones.
Figure 10.3 Representation of the cloud of individuals, for the two first principal components: Dim
1 and Dim 2. The percentage regards the percentage value of the explained inertia (Husson et al.,
2017).
The cloud is constructed maximizing the inertia of individuals projected onto a series of orthogonal
axes (Dim 1 and Dim 2 in Figure 10.3).
The study of individuals can be difficult, due to their large number, so commonly it is generalized
by studying the categories through the individuals that they represent (Husson, 2017).
Study of categories
The information obtained by a variable can be evaluated in terms of its categories: for this reason, in
MCA it is necessary to study the categories as categories present both variables and a group of
individuals (Husson et al., 2017). In order to understand how close a category is to others it is
fundamental to consider the distance between each other. It has to be taken into account that two
categories k and k’ are each associated with a group of individuals. If few individuals select, both
categories these will be more distant, so more different, contrariwise they will be similar. The
244
distance between k and k’ is based on the count of the number of individual belonging both the two
categories. It can be calculated as:
2 𝐼𝑘≠𝑘′
𝑑𝑘,𝑘 ′ = 𝐶′ (10.2)
𝐼𝑘 𝐼𝑘′
where:
C’ is a constant and Ik≠k’ is the number of individual belonging either k and k’.
Being xik =0 or 1, the number of individuals selecting only one category is:
𝐼𝑘≠𝑘′ = ∑(𝑥𝑖𝑘 − 𝑥𝑖𝑘′ )2 (10.3)

𝑖=1
So:
2 1
𝑑𝑘,𝑘 ′ = 𝐶′ ∑𝐼𝑖=1(𝑥𝑖𝑘 − 𝑥𝑖𝑘′ )2,
𝐼 𝑘 𝐼𝑘′
∑𝐼𝑖=1 𝑥𝑖𝑘 2 +∑𝐼𝑖=1 𝑥𝑖𝑘′ 2 −2 ∑𝐼𝑖=1 𝑥𝑖𝑘 𝑥𝑖𝑘′

= 𝐶′ (10.4)
2
Being 𝑥𝑖𝑘 =0 or 1; 𝑥𝑖𝑘 2 = 𝑥𝑖𝑘 so ∑𝑖 𝑥𝑖𝑘 = ∑𝑖 𝑥𝑖𝑘 = 𝐼𝑘
2 1 1 ∑𝐼𝑖=1 𝑥𝑖𝑘 𝑥𝑖𝑘′

𝑑𝑘,𝑘 ′ = 𝐶′ ( + −2 ) (10.5)
𝐼𝑘′ 𝐼𝑘 𝐼𝑘 𝐼𝑘′
However:
1 𝐼𝑘 ∑𝐼𝑖=1 𝑥𝑖𝑘
2
= 2= (10.6)
𝐼𝑘 𝐼𝑘 𝐼𝑘2
Developing the equation, the distance between two categories can be expressed as:
𝐼
2 𝑥𝑖𝑘 𝑥𝑖𝑘′ 2
𝑑𝑘,𝑘 ′ = 𝐶′ ∑ ( − ) (10.7)
𝑖=1
245
In MCA the categories can be depicted at the barycentre 2 of the individuals with those categories.
The best use of representation is made when the representation corresponds to the one carried out
by maximizing the inertia of the cloud of categories on a sequence of perpendicular axes, as
represented in Figure 10.4.
Figure 10.4 Representation of the cloud of categories (Husson, et al., 2017).
The inertia of a category is usually evaluated from the distance from k to the centre of gravity of the
categories cloud with coordinates 1/I, as:
𝐼
2
𝑥𝑖𝑘 1 2
𝑑𝑘,𝐺 = 𝐼∑( − )
𝐼
𝐼𝑘 𝐼
𝑖=1
𝐼
= −1 (10.8)
𝐼𝑘
The distance increases when very few individuals have taken the category k. The inertia can be
expressed as:
2
The barycentre of all of categories for a variable is located at the centre of gravity of the cloud of
individuals (Husson et al., 2017).
246
2
𝐼𝑘
𝐼𝑛𝑒𝑟𝑡𝑖𝑎(𝑘) = 𝑑𝑘,𝐺 ∗
𝐼
𝐼𝐽
𝐼𝑘 𝐼
= ( − 1)
𝐼𝐽 𝐼𝑘
𝐼 − 𝐼𝑘
=
𝐼𝐽
1 𝐼𝐾
= (1 − ) (10.9)
𝐽 𝐼
The inertia of variable j, which is the inertia of all of the Kj categories for a variable j is equal to:
𝐾𝑗
1 𝐼𝑘
𝐼𝑛𝑒𝑟𝑡𝑖𝑎(𝑗) = ∑ (1 − )
𝐽 𝐼
𝑘=1
𝑗𝐾
With ∑𝑘=1 𝐼𝑘 = 𝐼
𝐾𝑗 − 1
𝐼𝑛𝑒𝑟𝑡𝑖𝑎(𝑗) = ( ) (10.10)
𝐽
The variable inertia only depends on the category number that makes it up (Husson et al., 2017).
It is recommended to have an equal number of categories for each variable therefore if a variable
has a large number of categories the categories are shared over many different dimensions and this
variable does not influence the principal axes construction.
The inertia associated with all of the categories is the same of the inertia of the category cloud.
247
10.1.4 Cluster analysis
PCA, CA and MCA allow to visualize the association between variables and to identify the features
of studied characteristics on the basis of graphical displays. In order to make objective this
graphical interpretation the cluster analysis method can be adopted.
The objective of cluster analysis is to assign observations to groups (the clusters) so that:
- Observations within each group are similar to one another with respect to variables or
attributes of interest;
- Groups themselves stand apart from one another.
Cluster analysis seeks to discover the number and the composition of the groups. The methods that
can be used to carry out a cluster analysis can be divided into two groups: the hierarchical methods
and the non-hierarchical ones.
10.1.4.1 Hierarchical clustering methods
The hierarchical clustering method goal is to visualize the links between objects and in order to do
that a hierarchical tree (or dendrogram) has to be built (Figure 10.5).
Figure 10.5 Hierarchical tree summarising the associations between 5 objects (Husson et al.,
2017).
The hierarchical methods can be divided into two families, based on different algorithms: the
agglomerative one and the divisive one. In the first one, subject start in their own separate cluster,
then the most similar are combined and this procedure is repeated until all subjects are located in
one cluster. At last, the best cluster number is chosen out of all cluster solutions.
248
There are several algorithms based on the Agglomerative Hierarchical Clustering, AHC; the main
ones are:
- Nearest neighbour method, also called single linkage method: the distance (similarity)
between two clusters is defined with the aim to be the one between the two closest groups,
the neighbour. This approach is simple to use and it is suitable when the cluster shapes are
not spherical or elliptical; as for negative aspects it is criticized because it does not consider
the cluster structure;
- Furthest neighbour or complete linkage method: the distance (similarity) between two
clusters is defined with the aim to be the maximum one between members. It is sensitive to
outliers and it doesn’t consider the cluster structure, it produces compact and similar in size
cluster;
- Average linkage method: it is a quite robust method where the distance between two groups
is evaluated as the average distance between all pairs of members in the two considered
clusters;
- Centroid method: this approach is based on the distance between centroids (that is the main
value for each variable) of each cluster. Groups whose centroids are nearest together are
connected. Also this method is robust;
- Ward’s method: this algorithm is the most largely adopted in AHC. It consists of regrouping
two elements by maximizing the quality of the obtained partition (Husson et al., 2017). All
possible group couples are arranged and the distance (squared) within each cluster is
calculated and summed over all of theme. The combination that offers the lowest sum of
square is considered. This method is sensitive to outliers and it generally returns equal size
clusters.
The divisive methods, sometimes called top-down clustering, are less used and conceptually more
complex than the agglomerative ones. All subjects start into one cluster (at the top) then the cluster
is split through a flat clustering algorithm. The procedure is applied until each subject is in its own
group (Christopher et al., 2008).
249
Number of clusters
When the cluster analysis has been obtained, it is necessary to choose the optimal cluster solution;
the optimal number of clusters can be selected with several methods (Jolliffe, 2002); however in
order to do this operation it is useful to take into account some important concepts:
- The graphical aspect of the dendrogram (Figure 10.6) suggests into how many clusters the
partitioning has to be carried out. In fact the tree shows which clusters have been joined at
each analysis passage and the distance between clusters at the time of connecting. If a large
gap in the distance between clusters occurs from one stage to the following one it means that
at one passage clusters that are close together were linked and in the following step the
joined clusters were far. This observation suggests that the suitable number of clusters is the
one observed just before the significant gap in distance;
- A bar chart can visualize the decrease in the node levels (Figure 10.6), each irregularity in
this decrease educes another division;
- The cluster number should not be too high in order not to prevent the succinct nature of the
approach;
- Cluster interpretability: it is strongly advisable to maintain only the easy interpretable

subdivisions.
Figure 10.6 Dendrogram of hierarchical clustering, in the top right-hand corner the bar chart
representing the level indices (Husson et al., 2017).
250
At the end a visual examination both of hierarchical tree (dendrogram) and of bar char of level
indices proposes a division into a certain number of cluster (Q) when the inertia between cluster at
the passage for Q-1 to a Q cluster partition is much considerable than that from Q to Q + 1 cluster
partition (Husson et al., 2017).
10.1.4.2 Non-hierarchical clustering methods
This group of clustering methods, also known as partitioning clustering, is based on the algorithm
class of point assignment. Each point is assigned to the cluster into which it best fits. This operation
usually comes after a phase in which initial clusters are approximated. Variations allow to split or
combining clusters or permit points not to be assigned if there are outliers (elements too far from
any of the considered clusters). The most important family of clustering algorithms in this category
is known as k-means clustering method. These methods assume a Euclidean space. The desired
number of clusters, k, is specified in advance; it is possible to decide the number of clusters by trial
and error.
The steps of such a method are synthetically described in the following rows:
1) Choosing the initial cluster centres;
2) Assigning each individual to the cluster that is nearest to;
3) Finding the centroids of the formed cluster;
4) Re-calculating the distance between each centroid and each individual;
5) Moving the subjects that are not in the cluster that they are nearest to;
6) Continuing this procedure until the centroids become stable.
The k-means algorithm procedure for a simple case is graphically shows in Figure 10.7.
Figure 10.7 Simple illustration of k-means algorithm (Husson et al., 2017).
251
In the case above, the subjects have to be divided into the circle cluster and into the squared one.
Firstly, the individuals are divided into two random clusters (1); secondly, the gravity centres are
calculated for each of the cluster, the centres of gravity are represented by the white square and
circle (2).
Each subject is assigned to the group it is closest to (3) and, in the end, the gravity centres foe the
new generated clusters are calculated. If step three is reapplied and there are not any changes, it
means that the algorithm has converged.
10.1.4.3 Comparison between Hierarchical and Non-hierarchical methods
The non-hierarchical methods present some advantages when compared to hierarchical ones:
- they are suitable when a large number of data are involved into the analysis;
- they permit individuals to move to a cluster to another one while in hierarchical approach
once an individual is assigned it cannot move to a different cluster.
On the other hand, non-hierarchical methods present two critical aspects:
- it is complicated to know how many clusters are necessary to carried out a significant
analysis, so the procedure may have to be repeated many times;
- the choice of the initial cluster centre can influence too much the analysis. It may be useful
to try to use different centroids to understand their impact.
A suitable solution is represented by mixed algorithms, which combine the two approach to obtain a
methodology that includes the advantages of each.
For example, it is possible to adopt a hierarchical approach in the first stages to quantify how many
clusters the data must be divided and then to use the centres of the obtained clusters as initial cluster
centres in the non-hierarchical method.
Where there is a very large number of individuals, another mixed algorithm can be developed. In a
first phase the individuals are partitioned into several groups. On the one hand many of these groups
are very close together and on the other hand they are homogeneous and contain subjects that
cannot be separated (Husson et al., 2017).
252
10.1.4.4 Clustering in a Non-Euclidean space
When the space is non-Euclidean it is needed to use some measures of distance that are calculated
from points, for example cosine distance, Jaccard or edit distance. In a non-Euclidean space it is not
possible to base distance on the location of points. Some problems can emerge if it is needed to
represent a cluster, because it is not possible to replace a point collection by their centroid.
In a non-Euclidean space is not permitted to combine points in a cluster but it is only allowed the
selection of one of the points of the cluster itself to represent it. The selected point is near to the
other ones of the cluster and occurs in the “centre”. This representative point is named “clustroid”.
The clustroid can be selected in various ways in order to minimize the distance between it and the
other cluster points. The clustroid in the non-Euclidean space is used in place of the centroid
(Euclidean space), and the point selected as clustroid has to satisfy the following requirements:
- minimisation of the sum of the distance to the other cluster points;
- minimisation of the maximum distance between itself and another point in the cluster;
- minimisation of the sum of the distance squares to the other points in the cluster.
There are no considerable changes in the option of stopping criteria (the options that make possible
to determine when the merging process can be stopped when it has k clusters) from Euclidean to
non-Euclidean space.
253
10.1.5 Observations
As already mentioned, combining the principal component methods and the clustering ones allows
to better visualise and to comprehend the data. PCM is not only involved in pre-processing steps (as
denoising data, transformation of categorical variables in continuous ones and balancing of the
influence of groups of variables) but also to carry out a framework to visualize data. The clustering
methods can be represented onto the map carried out by the PCM (Husson et al., 2010).
In other words, the simultaneous use of the principal component methods and the clustering ones
enriches the analysis of dataset. Finally, the two described approaches complete one another
because: on one hand it is possible to represent both the continuous view (provided by PCM) and
the discontinuous one (the clusters) of the same dataset in a unique framework; on the other hand
the dendrogram and the clusters, defined from more than two dimensions, provide some
information “outside of the map” (Husson et al., 2010), whereas the 2D map does not provide
information concerning the position of the subjects in the other dimensions.
254
10.2 Application to “La Maddalena” case study
10.2.1 Introduction
The statistical tools described in the previous paragraphs have been adopted to analyse the
geological, mechanical and TBM related data in order to consider the combined effect of
geotechnical rock masses properties on the machine performance.
The analyses were performed using the software R version 3.4.1 (R Core Team, 2017), with the
packages FactoMineR (Lê et al., 2008) and quantreg (Koenker, 2017), added.
The considered tract of tunnel has a length of 2010 m and the dataset consist of 186 rock mass
section, corresponding to a single site survey (as already described in Chapter 5); each record
includes three kinds of variables.
The first one is represented by the geological domain (the lithology) to which the section belongs.
The domain has been codified as categorical variable with three modalities I, II, III);
a) DI gneiss;
b) DII alternation of gneiss and micaschist;
c) DIII micaschist.
The second one consists of geological and geomechanical characteristics of detected rock mass. In
particular, the geotechnical features selected to carry out the following presented analysis are:
d) uniaxial compressive strength, UCS [MPa];
e) quartz content, q [%];
f) spacing of fractures, S [m];
g) alpha angle, α [°];
h) Joint conditions, Jc [-];
i) tunnel depth [m];
The third type of variables consists of TBM performance indices (considered dependent variables):
j) rate of penetration (ROP);
255
k) field penetration index (FPI).
A detailed description of above mentioned parameters have been provided in Chapter 5.
The subsequent analyses have been divided in two steps: the first one aims to assess the association
among geological variables (letters from d to i); in this step MCA and cluster analysis have been
carried out. The second one aims to evaluate the relationship between geological and mechanical
features of excavated rock masses (from d to i) and the TBM performance registered during tunnel
construction (from j to k).
10.2.2 Relationship among geological and geomechanical variables
The main goal was to assess the association among geological variables, detect the presence of
subgroups of rock sections, and describe the distinctive features of the subgroups possibly found.
To such aim a MCA of the geological variables was performed, followed by cluster analysis. The
geological variables were converted from numeric variables to categorical ones. The modalities
obtained after the conversion are indicated in Table 10.1.
The uniaxial compressive strength values have been divided in four classes; each class has a range
of 50 MPa, considering that minimum UCS value is equal to 68.0 MPa.
UNIAXIAL COMPRESSIVE STRENGTH,

UCS [MPa]
Class Ranges
Low 0.0 – 100.0
Medium 101.0 – 150.0
High 151.0 – 200.0
Very high 201.0 – 250.0
Table 10.1 Classes of UCS utilised in MCA.
Quartz content percentage has been divided in four classes with different amplitudes, as reported in
Table 10.2. The class amplitudes have been decided on the basis of the quartz percentage
distribution.
256
QUARTZ CONTENT, q [%]
Class Ranges
Low 0.0 – 25.0
Medium 26.0 – 30.0
High 31.0 – 45.0
Very high 46.0 – 100.0
Table 10.2 Classes of quartz content utilised in MCA.
As for discontinuity spacing, both terminology and class ranges proposed by ISRM (Burton, 1978)
have been adopted. The classes described by the Suggested Methods as “extremely close spacing”
(spacing minor than 0.02 m), the one defined as “very close spacing” (spacing between 0.02 and
0.06 m), the “close” class (0.06 – 0.20 m) and the “moderate spacing” one (0.20 – 0.60 m) have
been as a unique class characterised by spacing minor than 0.60 m and named “close-moderate”
class.
This decision has been taken because of the minimum spacing observed during tunnel wall mapping
was equal to 0.27 m, for this reason extremely close, very close and close spacing have not been
considered (Table 10.3).
SPACING OF FRACTURES, S [m]
Class Ranges
Close – Moderate spacing 0.00 – 0.60
Wide spacing 0.60 – 2.00
Very wide spacing 2.00 – 6.00
Extremely wide spacing > 6.00
Table 10.3 Classes of discontinuity spacing utilised in MCA.
The angle between tunnel axis and the main set of fractures orientation, α, can range between 0.0
and 90.0 degrees, this range has been divided in four classes with the same amplitude (22.5°) as it is
reported in Table 10.4.
257
ALPHA ANGLE, α [deg]
Class Ranges
Low 0.0 – 22.5
Medium 22.6 – 45.0
High 45.1 – 67.5
Very high 67.6 – 90.0
Table 10.4 Classes of α utilised in MCA.
The parameter identified as “joint conditions” is computed in accordance with the RMR
(Bieniawski 1989), in which Jc represents the fourth parameter. Its values have been divided into
four classes of different amplitude, as showed in Table 10.5.
JOINT CONDITIONS, Jc [-]
Class Ranges
Low 0.0 – 10.0
Medium 11.0 – 20.0
High 21.0 – 25.0
Very high 26.0 – 30.0
Table 10.5 Classes of Jc utilized in MCA.
Tunnel depth has been involved in the present analysis as indicator of in situ stress.
The class have been named taking into account that the maximum depth reached in the considered
tunnel tract is equal to 616.7 meters. For this reason the depth values ranging between 501.0 m and
700.0 m have been classified as “very high”.
Actually it is clear that a tunnel 700.0 meters depth cannot be considered as a very high deep tunnel.
“La Maddelena” depth classes are listed in Table 10.6.
258
TUNNEL DEPTH, [m]
Class Ranges
Low 0.0 – 100.0
Medium 101.0 – 250.0
High 251.0 – 500.0
Very high 501.0 – 700.0
Table 10.6 Classes of tunnel depth utilized in MCA.
10.2.2.1 Multiple Correspondence Analysis
As the parameter classes as been defined the MCA can be performed.
The number of relevant dimensions was determined by examining the percentages of explained
inertia, using the re-evaluation formula illustrated in Benzecri (1979).
Only the dimensions provided of a meaningful geological interpretation were retained in the
analysis.
In the MCA the first two dimensions were selected, with proportions of explained inertia equal to
51.2% and 15.7%, respectively; therefore, the overall measure of quality of representation is 66.9%
(Table 10.7).
Explained Inertia
Table 10.7 Percentage of explained inertia.
The plot of variable modalities is shown in Figure 10.8.
259
The main feature emerging from the figure is that the higher modalities of the variables (such as:
alpha very high, or Jc high and very high), with the exception of the variable depth, lie on the right
side of the figure, while the lowest modalities lie on the left side. This indicates an overall positive
association among these variables.
In addition, the modalities depth low (i.e. 0 – 100 m) and quartz content very high (46% – 100%)
have high coordinates on the y axis, and are thus separated from the rest. This suggests the possible
presence of a subgroup of rock sections characterised by these modalities.
4
Axis II (in. 15.7%)
-1
-2
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
Axis I (in. 52.2%)
Figure 10.8 Variable map determined in the MCA of geological characteristics. Red symbols: very
high modalities, orange: high modalities, green: medium ones, light blue: low modalities.
Triangles: UCS, circles: quartz content, crosses: alpha angle, x: Jc, diamonds: depth, asterisks:
spacing.
10.2.2.2 Clustering
The considerations presented above are mainly geometric and semi-quantitative, so in order to
classify the sections into homogeneous groups (clusters), a cluster analysis was performed on MCA
displays. The clustering algorithm was agglomerative nesting with Ward linkage (see section
10.1.4.1).
The number of clusters was chosen by visual inspection of MCA displays (Figure 10.8), of cluster
analysis displays, i.e. the dendrogram (Figure 10.9), and more formally by evaluating the
percentage gains of between cluster inertia (Figure 10.9, upper right corner).
260
Figure 10.9 Dendogram of cluster analysis, in the upper right corner the inertia gain.
After these considerations the chosen number of clusters is three. According to the percentage gains
of inertia (first four: 25.9%, 24.4%, 15.9%, 11.9%) rock mass sections were divided in three
clusters, composed by 111, 66 and 9 rock mass sections respectively (Figure 10.10).
Figure 10.10 Factor map. The three colours identify the three clusters, it is possible to note that
cluster I and II are more similar between each other respect to cluster III.
261
The position of clusters is observable from the factor map. Cluster III is composed by few elements
and it is distant, so it is different, from cluster I and II, as it is possible to see by the three-
dimensional view of the factor map (Figure 10.11).
Figure 10.11 Three-dimensional view of the hierarchical clustering on the factor map.
To describe the main characteristics of the clusters, the geological variables were summarized
separately for each cluster, using: mean, median, minimum and maximum value, standard
deviation, 25th and 75th percentiles (Q1 and Q3) and interquartile range (Q3-Q1).
Furthermore, to evaluate to what extent clusters correspond to distinct geological domains, the
frequencies of rock sections belonging to each domain were reported for each cluster (Table 10.8).
The clusters obtained from clustering process and the geological domains defined after traditional
surveys have been compared and a good correspondence has been noted.
In particular, cluster I matches the third domain and include part of micaschist sections belonging to
the alternation gneiss-micaschist relate to second domain. Cluster II coincides with domain I, in fact
60 sections belong to Ambin gneiss and includes 6 gneiss sections belonging to the gneiss-
micaschist alternation.
Cluster III is formed by 9 geological sections belonging to the domain I; this portion of rock mass
was not considered as a separated domain but it was surveyed as a very hard tract of gneiss.
262
MCA showed the significant difference of this portion from the two others both in terms of
geotechnical features and TBM performance. In other words: domain I constitutes cluster II,
domain II is divided in cluster I and cluster II and the hardest portion on domain I was separate as
cluster III. A most detailed description of cluster is reported in Table 10.8.
Cluster 1 (n=111) Cluster 2 (n=66) Cluster 3 (n=9)

UCS [MPa]
range 68.00 – 251.00 103.00 – 219.00 200.00 – 217.00
median 115.00 153.00 200.00
Q1-Q3 93.00 – 135.00 130.00 – 168.25 200.00 – 200.00
Alpha [°]
range 7.40 – 69.40 29.80 – 85.00 42.50 – 56.40
median 39.90 48.45 44.90
Q1-Q3 29.80 – 49.00 44.00 – 59.98 44.75 – 52.30
Spacing [m]
range 0.27 – 5.50 0.75 – 11.80 0.86 – 6.00
median 1.13 3.25 1.88
Q1-Q3 0.84 – 1.56 2.39 – 6.00 1.29 – 2.48
Jc
range 7.00 – 21.00 12.00 – 29.00 18.00 – 29.00
median 17.00 20.00 20.00
Q1-Q3 15.00 – 19.00 17.00 – 23.00 20.00 – 20.00
quartz[%]
range 25.00 – 45.00 30.00 – 45.00 50.00 – 80.00
median 25.00 45.00 50.00
Q1-Q3 25.00 – 30.00 45.00 – 45.00 50.00 – 50.00
depth[m]
range 190.00 – 616.70 133.00 - 569.10 79.40 - 215.30
median 442.00 440.25 103.50
Q1-Q3 329.40 – 531.60 271.03 – 523.20 85.95 – 127.50
geological domain: n (%)
I 14 (13%) 60 (91%) 9 (100%)
II 38 (34%) 6 (9%) 0 (0%)
III 59 (53%) 0 (0%) 0 (0%)
ROP[m/h]
range 0.50 – 3.74 0.38 – 2.30 0.33 – 1.17
median 1.88 0.95 0.63
Q1-Q3 1.42 – 2.28 0.60 – 1.17 0.49 – 0.87
FPI[kN/cutter/mm/rev]
range 15.16 – 190.97 27.15 – 279.10 79.39 – 298.61
median 55.10 146.64 131.03
Q1-Q3 39.19 – 84.70 117.81 – 198.33 96.38 – 141.40
Table 10.8 Statistical description of clusters.

263
Cluster I is principally composed (53%) by geological sections belonging to the micaschist of

Clarea complex (third geological domain), and for the 34% it is composed by rock mass portions
defined as an alternation of gneiss and micaschist lithology. Only the 13% of rock sections is
characterised by gneiss belonging to Ambin complex.
All range parameters are reported as first quartile (Q1) and third quartile (Q3).
All geomechanical parameters relating to the rock portions characterising cluster I are lower than
the same parameters evaluated in the others rock mass sections. Cluster I is characterised by
medium values of uniaxial compressive strength, ranging between 93.0 MPa and 135.0 MPa
(respectively Q1 and Q3). The angle between principal discontinuity set orientation and tunnel axis
ranges between 29.8° and 49.0° and quartz content ranges between 25% and 30%. Fractures are
medium spaced (0.84 m – 1.56 m) and generally joint conditions are rated 15.0 – 19.0. Generally,
rock mass quality of this cluster is lower than the quality of clusters II and III. These geomechanical
and geological features determine TBM performance better than the ones measured in cluster II and
cluster III rock masses, ROP is high (1.42 m/h – 2.28 m/h) and FPI is low (39.19 kN/cutter/mm/rev
– 84.70 kN/cutter/mm/rev); it means that encountered rocks are easy drillable: the net penetration
achieved is good and the thrust required to guarantee a penetration per revolution is low.
Cluster II is mainly characterised by geological section belonging to the Ambin complex, described
like domain I (91%). The remaining 9% is formed by portions of rock masses lithologically
considered as an alternation of gneiss and micaschist (domain II). Cluster II does not include
micaschist sections related to the Clarea complex.
Rock mass portions belonging to cluster II are characterised by general good quality, UCS and
quartz content are higher than the ones in cluster I (respectively 130.0 MPa – 168.3 MPa and 45%)
and discontinuities are more spaced: Q1 and Q3 are 2.39 m and 6.00 m respectively, they form an
angle between 44.0° – 60° with tunnel axis. Jc parameter is slightly higher than the previous one. In
general rock mass is compact and hard and it results in a low boreability: ROP is less than 1 m/h
(0.6 m/h – 1.2 m/h) and FPI is high (from 117.8 kN/cutter/mm/rev to 198.3 kN/cutter/mm/rev).
Cluster III is defined only by nine geological sections of gneiss belonging to the first domain
(Ambin complex); it includes no one portions of micaschist or gneiss-micaschist. It represents the
hardest zone of tunnel with the highest levels of UCS (maximum value 217.0 MPa) and of quartz
content (50.0%). Discontinuities are medium-high spaced (1.29 m – 2.48 m). The TBM
performance in this cluster are the lowest registered, ROP is very low, ranging between 0.5 m/h and
0.9 m/h and FPI is high (96.4 kN/cutter/mm/rev – 141.4 kN/cutter/mm/rev).
264
10.2.3 Relationship between geological-geomechanical variables and TBM performance
In the second step of the analysis, the relationships between TBM performances (ROP and FPI) and
geological characteristics have been assessed through a regression models. In order to account for
the joint effect of geotechnical site characteristics on TBM performances, the clusters identified in
the previous step were used as independent variable.
The distributions of ROP and FPI were evaluated for each cluster through histograms and the
hypothesis of Gaussian distribution was assessed by the Shapiro-Wilks test.
Since the hypothesis of Gaussianity was rejected in some cases, quantile regression models
(Koenker and Portnoy, 1996) were fitted for estimating the median values of ROP and FPI for the
distinct clusters of rock sections and the respective differences.
Histograms
In order to evaluate the statistical distribution of ROP and FPI for each cluster, histograms of
frequency have been created. As we can observe by graphs the normal distribution of data occurs
only in three cases (ROP in cluster I and III; FPI in cluster II).
In the other cases TBM performance indices are not normally distributed (Figure 10.12).
265
Figure 10.12 Histograms of frequency distribution of ROP and FPI for each cluster.
266
Normality test
In addition to the histograms, in each cluster “Shapiro-Wilk normality test” has been carried out in
order to test the normal distribution of ROP and FPI. It is a normality test that tests if a sample
comes from a normally distributed population. The null hypothesis is that the population is
normally distributed.
So if the p-value is less than 0.05 the null hypothesis will be rejected and the population will be
considered not normally distributed; on the contrary when the p-value is higher than 0.05 the
population is considered normally distributed.
The results of “Shapiro-Wilk normality test” are reported in Table 10.9.
TESTED NUMBER OF W p-value NULL

PARAMETER CLUSTER HYPOTHESIS
ROP I 0.98678 0.348200 accepted
ROP II 0.93292 0.001489 rejected
ROP III 0.91223 0.331800 accepted
FPI I 0.87470 3.188E^-08 rejected
FPI II 0.98898 0.827600 accepted
FPI III 0.72358 0.002673 refused
Table 10.9 Results of Shapiro-Wilk normality test.
As reported in Table 10.9, in some cases p-values are higher than 0.05 so in these cases the null-
hypothesis cannot be refused and population is considered normally distributed; however, in three
cases p-values are lower than 0.05, it means that population is not normally distributed.
Since the hypothesis of Gaussianity was rejected in some cases, we used quantile regression models
(Koenker and Portnoy, 1996) for estimating the median values of ROP and FPI for the distinct
clusters of rock sections and the respective differences.
267
10.2.3.1 Quantile regression
A quantile regression analysis of deviance table was carried out in order to compare the medians
both of ROP and FPI of the three clusters and test if they are dissimilar from each other. ANOVA
(analysis of variance) test was performed and the considered null hypothesis is that the medians of
clusters are all the same. The results obtained by quantile regression (Table 10.10) show significant
differences of performance indices between cluster I and II (p < 0.0001, for both indices) and
between cluster I and III (p < 0.0001 for both indices).
Differences between medians Test of hypothesis

Estimate (95% C. I.) Global test: p < 0.0001
ROP[m/h]
Cluster 2 VS Cluster 1 -0.92 (-1.15, -0.69) p < 0.0001
Cluster 3 VS Cluster 1 -1.25 (-1.52, -0.98) p < 0.0001
Cluster 3 VS Cluster 2 -0.33 (-0.60, -0.06) p = 0.0508
Cluster 2 VS Cluster 1 144.3 (124.5, 164.1) p < 0.0001
Cluster 3 VS Cluster 1 131.0 (96.3, 165.7) p < 0.0001
Cluster 3 VS Cluster 2 -13.3 (-58.5, 31.9) p > 0.999
Table 10.10 Comparison of median TBM performances among clusters.
The results obtained by quantile regression show that the median values of ROP between cluster I
and II, cluster I and III, cluster II and III are significantly different, being p-values less than 0.05
and null hypothesis refused. Also the median values of FPI between cluster I and II, and cluster I
and III have been considered dissimilar (p-values < 0.05).
Higher values of ROP belong to rock sections in cluster I (IQR: 1.42 to 2.28 m/h), followed by
those in clusters II and III (with IRQs lower than 1.00 m/h). Concerning FPI, cluster I has lower
values (IQR: 39.2 to 84.7) than cluster II and II, which overlap one another (IQR: 117.80 to 198.30
and 94.40 to 141.40 kN/cutter/mm/rev, respectively).
The regression analysis shows that FPI of cluster II (146.65 kN/cutter/mm/rev) and FPI of cluster
III (131.00 kN/cutter/mm/rev) are not significantly different. Theoretically, keeping in
consideration the hardness of cluster III, FPI would have to be higher than the FPI measured during
268
cluster II excavation because of the excavation of hard rock masses requires high TBM thrust in
order to guarantee a penetration per each cutterhead revolution. The same median value of FPI has
been registered in cluster II and III is because it is not possible to impart a too high force on very
hard rock faces in order to avoid cutter breakage and TBM damages. In other words, there is a force
threshold above which TBM cannot be driven in relation to the rock hardness.
At each cluster correspond different TBM performances, so the clusters can be considered as
classes of excavatability, characterised by specific geological and geomechanical features
associated with different ranges of ROP and FPI, as summarised in Table 10.11.
Geomechanical ROP FPI

ranges [m/h] [(kN/cutter)/(mm/rev)]
UCS = 93.0 – 135.0 MPa 1.42 – 2.28 39.19 – 84.70

CLUSTER
α = 29.8° – 49.0°
I
S = 0.84 – 1.56 m
q = 25 – 30 %
Jc = 15 – 19
UCS = 130 – 168.3 MPa 0.60 – 1.17 117.81 – 198.33

CLUSTER
α = 44° – 60°
II
S = 2.39 – 6.00 m
q = 45%
Jc = 17 – 23
UCS = 200 MPa 0.49 – 0.87 96.38 – 141.40

CLUSTER
α = 44.75° – 52.30°
III
S = 1.29 – 2.48 m
q = 50 %
Jc = 20
Table 10.11 Ranges (Q1;Q3) of geomechanical properties for each cluster and the corresponding
TBM performances.
269
At this point, three classes of excavatability have been delineated; they allow the prediction of the
TBM performances when the geomechanical features describing each class occur.
In order to carry out an empirical formulation able to describe the relationship between geological
properties and TBM performance multiple regression analyses have been developed for each
cluster.
10.2.4 Regression analyses on clusters
Multiple linear regression analysis has been selected as statistical approach to obtain empirical
formulation able to describe the relationship between geological and geotechnical features
describing each cluster and the performances of the TBM measured in the same cluster.
The analysis is divided in two groups: in one the dependent variable is ROP, in the other the
dependent one is FPI.
In both cases the independent variables are: UCS, q, spacing, alpha angle, Jc, tunnel depth.
The models were generated adopting the stepwise backward selection, and following the procedures
described in Chapter 9.
10.2.4.1 Relationship between geological conditions of clusters and ROP
Cluster I
About cluster I the assumptions for linear regression are satisfied. For this reason a multiple linear
regression analysis has been carried out. The excluded variables are the spacing between
discontinuities and α, they have Sign. 0.996 and Sign. 0.887, respectively.
All coefficient statistics, concerning each model, have been listed in the Table 10.12.
270
Coefficients
coefficients coeff. 95,0% C.I. for B statistics
Model Sign.
Std. (p- Bottom Top Toleranc
B error Beta t value) Limit limit e VIF
C1a (Constant) 2.305 0.408 5.644 0.000 1.495 3.115
UCS -0.006 0.001 -0.348 -4.200 0.000 -0.009 -0.003 0.743 1.346
α 0.000 0.003 -0.010 -0.141 0.888 -0.006 0.005 0.963 1.039
S 0.000 0.001 0.000 0.005 0.996 -0.001 0.001 0.922 1.085
Jc 0.056 0.014 0.303 3.882 0.000 0.027 0.085 0.838 1.193
q -0.041 0.007 -0.473 -6.300 0.000 -0.054 -0.028 0.905 1.104
depth 0.001 0.000 0.258 3.083 0.003 0.000 0.002 0.726 1.378
C1b (Constant) 2.305 0.406 5.672 0.000 1.499 3.111
UCS -0.006 0.001 -0.348 -4.305 0.000 -0.009 -0.003 0.773 1.293
alpha 0.000 0.003 -0.010 -0.142 0.887 -0.006 0.005 0.968 1.033
Jc 0.056 0.014 0.303 3.945 0.000 0.028 0.084 0.857 1.167
q -0.041 0.006 -0.473 -6.366 0.000 -0.054 -0.028 0.916 1.092
depth 0.001 0.000 0.258 3.114 0.002 0.000 0.002 0.734 1.363
C1c (Constant) 2.295 0.399 5.754 0.000 1.504 3.086
UCS -0.006 0.001 -0.347 -4.334 0.000 -0.009 -0.003 0.782 1.279
Jc 0.056 0.014 0.301 3.981 0.000 0.028 0.084 0.873 1.146
q -0.041 0.006 -0.473 -6.413 0.000 -0.054 -0.028 0.919 1.089
depth 0.001 0.000 0.257 3.138 0.002 0.000 0.002 0.747 1.338
parameters.
C1c includes the variables that effective have affected ROP, as Sign. parameter is able to
demonstrate. The geotechnical properties involved in this model are: UCS, q, Jc and depth. There
are not any problems of multicollinearity. C1c model summary is reported in Table 10.13.
271
r R2 R2adj F Sign.
0.686 0.470 0.450 23.494 0.000
Table 10.13 C1c model summary.

The regression analysis allows to explain only the 45.00% of the total variance, in other words the
model modestly represent the relationship between the geological characteristics of cluster I and
ROP.
On the basis of the statistical analyses, the proposed empirical equation is as follows:
𝑅𝑂𝑃 = 2.295 − 0.006 ∗ 𝑈𝐶𝑆 − 0.041 ∗ 𝑞 + 0.056 ∗ 𝐽𝑐 + 0.001 ∗ 𝑑 (10.11)
Cluster II
In cluster II, linear regression model about ROP and geological parameters cannot be performed,
being the assumptions not satisfied.
The relation between ROP and the geomechanical parameters have been investigated by means of
linear, logarithmic, quadratic, polynomial and exponential regression models. For each model
statistical tests have been carried out in order to select the model better describing how dependent
and independent variables are related. The results and the parameter coefficients are reported in
Table 10.14.
272
Independent
Variable
UCS Linear 0.186 14.586 0.000 1.909 -0.006
Logarithmic 0.197 15.709 0.000 5.980 -1.006
Quadratic 0.204 8.070 0.001 3.482 -0.027 6.491E-5
Power 0.167 12.866 0.001 116.897 -0.981
Exponential 0.161 12.294 0.001 2.227 -0.006
q Linear 0.011 0.722 0.399 1.383 -0.010
Logarithmic 0.011 0.722 0.399 2.349 -0.373
Quadratic 0.011 0.722 0.399 1.383 -0.010 0.000
Power 0.025 1.657 0.203 8.062 -0.594
Exponential 0.025 1.657 0.203 1.730 -0.016
S Linear 0.084 5.875 0.018 1.125 -0.042
Logarithmic 0.067 4.638 0.035 1.149 -0.163
Quadratic 0.086 2.974 0.058 1.184 -0.072 0.003
Power 0.045 2.993 0.088 1.026 -0.140
Exponential 0.073 5.010 0.029 1.027 0.041
α Linear 0.066 4.513 0.038 1.405 -0.009
Logarithmic 0.067 4.568 0.036 2.782 -0.468
Quadratic 0.069 2.348 0.104 1.842 -0.026 0.000
Power 0.068 4.667 0.035 6.135 -0.501
Exponential 0.065 4.430 0.039 1.395 -0.009
Jc Linear 0.028 1.867 0.177 1.289 -0.018
Logarithmic 0.025 1.617 0.208 1.887 -0.319
Quadratic 0.034 1.100 0.339 0.665 0.046 -0.002
Power 0.015 0.969 0.329 1.869 -0.263
Exponential 0.018 1.194 0.279 1.153 -0.015
Depth Linear 0.206 16.602 0.000 0.412 0.001
Logarithmic 0.223 18.323 0.000 -1.740 0.455
Quadratic 0.238 9.825 0.000 -0.152 0.005 -5.041E-6
Power 0.300 27.396 0.000 0.032 0.559
Exponential 0.286 25.649 0.000 0.443 0.002
Table 10.14 Linear and nonlinear models between geological parameters and ROP observed in
cluster II. Red boxes represent the excluded variables and in bold the selected models.
Quartz content and joint conditions do not affect ROP, being the null hypothesis cannot be rejected
for any regression model.
273
The mathematical functions that best describe the relation between UCS, spacing, α, tunnel depth
and ROP are listed below:
𝑅𝑂𝑃 = −1.006 ∗ ln 𝑈𝐶𝑆 + 5.980 (10.12)
𝑅𝑂𝑃 = −0.042 ∗ 𝑆 + 1.125 (10.13)
𝑅𝑂𝑃 = −0.468 ∗ ln 𝛼 + 2.782 (10.14)
𝑅𝑂𝑃 = 0.032 ∗ 𝑑0.559 (10.15)
There are three models that can be used to describe the relation between alpha angle and ROP, they
are: linear, logarithmic and power (F values are very similar: F = 4.513, 4.568 and 4.667
respectively). In this study, all of them have been considered and it was observed that logarithmic
function works better than the others; in fact the regression model developed entering alpha as
logarithmic related gave a higher coefficient of determination than the ones obtained considering
linear or power functions.
The equations 10.12, 10.13, 10.14, 10.15 have been entered in the multiple nonlinear regression
model in order to investigate the simultaneous nonlinear relationship between the geological
conditions characterising cluster II and ROP.
The carried out equation is as follows (model C2a):
𝑅𝑂𝑃 = −0.650 ∗ ln 𝑈𝐶𝑆 − 0.024 ∗ 𝑆 − 0.249 ∗ ln 𝛼 + 2.317𝐸 −5 ∗ 𝑑1.540 + 9.608 (10.16)
The coefficient of determination for this model is equal to 0.338.
Cluster III
Cluster III is defined by nine elements only. For this reason carry out a regression analysis on its
data might not useful be; however, in this study linear analysis has been developed.
Quartz content, alpha angle and tunnel depth were excluded from inputs due to their p-values higher
than 0.005, C3c takes in consideration only those variables determining ROP. All coefficient
statistics, concerning each model, have been listed in Table 10.15.
274
Coefficients

coefficients coeff. 95,0% C.I. for B statistics
Model Sign.
Std. (p- Bottom Top Tolera
B error Beta t value) Limit limit nce VIF
C3a (Constant) -20.392 26.904 -0.758 0.504 -106.012 65.228
UCS 0.148 0.168 3.112 0.878 0.444 -0.387 0.682 0.002 628.979
alpha 0.012 0.017 0.215 0.737 0.515 -0.040 0.065 0.235 4.247
S 0.503 0.417 -5.218 1.206 0.314 -0.817 1.862 0.003 285.951
Jc -0.453 0.308 -5.318 -1.470 0.238 -1.434 0.528 0.002 655.688
depth -0.009 0.016 -1.341 -0.551 0.620 -0.059 0.041 0.003 296.840
C3b (Constant) -6.251 7.345 -0.851 0.443 -26.645 14.143
UCS 0.059 0.045 1.244 1.326 0.256 -0.065 0.183 0.019 53.444
alpha 0.008 0.013 0.141 0.600 0.581 -0.029 0.045 0.298 3.360
S 0.227 0.069 1.587 4.033 0.016 0.086 0.468 0.106 9.401
Jc -0.288 0.062 -3.374 -4.641 0.010 -0.460 -0.116 0.031 32.059
C3c (Constant) -9.388 4.820 -1.948 0.109 -21.778 3.003
UCS 0.079 0.028 1.657 2.775 0.039 0.006 0.151 0.040 24.787
S 0.258 0.057 1.478 4.538 0.006 0.112 0.404 0.136 7.379
Jc -0.306 0.050 -3.593 -6.124 0.002 -0.435 -0.178 0.042 23.952
parameters.
The geotechnical properties involved in C3c model are: UCS, spacing and Jc. VIF is higher than 10;
this value is not due to the correlation between independent variables but to the limited number of
data. The proposed empirical equation is as follow:
𝑅𝑂𝑃 = 0.079 ∗ 𝑈𝐶𝑆 + 0.258 ∗ 𝑆 − 0.306 ∗ 𝐽𝑐 − 9.388 (10.17)
The model summary is provided in Table 10.16.
r R2 R2adj F Sign.
0.963 0.928 0.885 21.526 0.003
Table 10.16 C3c model summary.
275
10.2.4.2 Relationship between geological conditions of clusters and FPI
The same procedures of analysis described in the previous paragraph have been utilized to
investigate the relationship among the geomechanical conditions of each cluster and the FPI
registered in the considered tunnel portion.
Cluster I
As for cluster I, the assumptions of linearity have been satisfied, so a linear model has been carried
out. At the beginning of analysis all geological variables (UCS, q, S, Jc, α and depth) have been
entered then, joint condition parameter was excluded from inputs due to is p – value.
The statistical results have been provided by Table 10.17.
Coefficients
coefficients coeff. statistics
Model
Std. Bottom Upper Toleran
B error Beta t Sign. Limit Limit -ce VIF
C1 (Constant) -61.268 26.435 -2.318 0.022 -113.684 -8.852
d
UCS 0.544 0.093 0.467 5.846 0.000 0.360 0.729 0.749 1.335
q 2.700 0.426 0.459 6.334 0.000 1.855 3.545 0.909 1.100
S 0.110 0.035 0.227 3.156 0.002 0.041 0.179 0.927 1.079
Jc -0.426 0.940 -0.034 -0.453 0.651 -2.289 1.437 0.856 1.168
depth -0.052 0.026 -0.158 -1.960 0.053 -0.104 0.001 0.737 1.356
C1 (Constant) -69.901 16.812 -4.193 0.000 -103.819 -37.157

e
UCS 0.576 0.093 0.494 6.160 0.000 0.390 0.761 0.753 1.328
q 2.991 0.421 0.469 7.107 0.000 2.156 3.826 0.996 1.004
S -0.065 0.027 -0.195 -2.469 0.015 0.040 0.176 0.945 1.058
depth -0.049 0.026 -0.150 -1.915 0.030 -0.100 0.002 0.773 1.293
Table 10.17 Results of multiple linear regression analysis, red box underlines the no significant
parameters.
276
C1e includes the factors that effective have influenced FPI, they are: UCS, q, S and depth. There are
no problems of multicollinearity.
The regression analysis allows to carry out a new predictive equation:
𝐹𝑃𝐼 = 0.576 ∗ 𝑈𝐶𝑆 + 2.991 ∗ 𝑞 + 7.651 ∗ 𝑆 − 0.065 ∗ 𝑑 − 69.901 (10.18)
C1e summary has been reported in Table 10.18.
r R2 R2adj F Sign.
0.703 0.494 0.475 25.624 0.000
Table 10.18 C1e model summary.
Cluster II
The model assumptions for linear regression for geology-FPI relationship in cluster II have not been
verified. For this reason the same procedure previously adopted to carry out multiple nonlinear
regression analyses (see section 10.2.4.1) has been applied.
Linear, logarithmic, quadratic, power and exponential regression models have been considered.
About quartz content, alpha angle, Jc and tunnel depth no correlations have been recognised, on the
contrary power and exponential relations have been observed for UCS and S respectively, as
reported in Table 10.19.
Independent
Variable
UCS Linear 0.069 4.747 0.033 77.608 0.503
Logarithmic 0.078 5.419 0.023 -257.537 82.189
Quadratic 0.098 3.429 0.039 -179.796 3.865 -0.011
Power 0.095 6.695 0.012 4.347 0.698
Exponential 0.085 5.958 0.017 74.405 0.004
S Linear 0.094 6.608 0.012 129.166 5.703
Logarithmic 0.074 5.149 0.027 125.976 22.239
Quadratic 0.094 3.278 0.044 124.824 7.875 -0.183
Power 0.115 8.337 0.005 41.045 0.213
Exponential 0.118 8.549 0.005 115.431 0.049
Table 10.19 Linear and nonlinear models between geological parameters and FPI observed in
cluster II. In bold the selected models.
277
The mathematical function that better describes the relation between UCS and FPI is the power one:
𝐹𝑃𝐼 = 4.347 ∗ 𝑈𝐶𝑆 0.698 (10.19)
The function that better describes the relation between S and FPI are exponential one:
𝐹𝑃𝐼 = 115.431 ∗ 𝑒𝑥𝑝0.049∗𝑆 (10.20)
The equations 10.19 and 10.20 have been entered in the multiple nonlinear regression model in
order to investigate the simultaneous nonlinear relationship between UCS and spacing
characterising cluster II and FPI. The obtained equation is as follows (model C2b):
𝐹𝑃𝐼 = 5.522 ∗ 𝑈𝐶𝑆 0.568 + 42.618 ∗ 𝑒𝑥𝑝0.069∗𝑆 (10.21)
Cluster III
The results of linear regression analysis are reported in Table 10.20.
Coefficients
Std.ed Collinearity
Unst.ed coefficients coeff. 95,0% C.I. for B statistics
Model Sign.
Std. p- Bottom Top Toler
B error Beta t value Limit limit ance VIF
C3d (Const.) -3864.774 2480.367 -1.558 0.217 -11758.40 4028.86
UCS 22.302 15.493 1.970 1.439 0.246 -27.004 71.607 0.002 628.979
alpha -5.877 1.523 -0.434 -3.859 0.031 -10.724 -1.030 0.235 4.247
S 0.012 0.384 0.028 0.030 0.978 -1.212 1.235 0.003 285.951
Jc -6.204 28.425 -0.305 -0.218 0.841 -96.665 84.257 0.002 655.688
depth -0.797 1.451 -0.517 -0.549 0.621 -5.413 3.819 0.003 296.840
C3e (Const.) -3797.190 901.642 -4.211 0.014 -6300.551 -1293.8
UCS 21.877 5.457 1.933 4.009 0.016 6.726 37.028 0.010 104.010
alpha -5.901 1.123 -0.436 -5.253 0.006 -9.020 -2.782 0.325 3.080
Jc -5.388 7.169 -0.265 -0.752 0.494 -25.292 14.516 0.018 55.593
depth -0.754 0.228 -0.489 -3.311 0.030 -1.387 -.122 0.102 9.759
C3f (Const.) -3144.566 231.987 -13.555 0.000 -3740.909 -2548.2
UCS 17.902 1.283 1.581 13.957 0.000 14.604 21.199 0.159 6.294
alpha -5.374 0.839 -0.397 -6.407 0.001 -7.531 -3.218 0.532 1.881
depth -0.643 0.165 -0.417 -3.894 0.011 -1.067 -.218 0.178 5.607
Table 10.20 Results of multiple linear regression analysis, red boxes: no significant parameters.
278
UCS, alpha angle and tunnel depth have been influenced the FPI during cluster III excavation, the
mathematical form of this relationship is given by:
𝐹𝑃𝐼 = 17.902 ∗ 𝑈𝐶𝑆 − 5.374 ∗ 𝛼 − 0.643 ∗ 𝑑 − 3144.566 (10.22)
The summary about this model is provided in Table 10.21.
r R2 R2adj F Sign.
0.995 0.990 0.984 161.731 0.000
Table 10.21 C3f model summary.
279
10.3 Conclusion
10.3 Conclusion
Multiple correspondence analyses and cluster analysis have been carried out on the geological,
geomechanical and TBM data collected during La Maddalena excavation between the Pk. 0 + 198
m and the Pk. 2 + 010 m.
These statistical methods allow to objectively divide the rock mass into three homogeneous groups,
named cluster I, II, and III respectively. Each of them can be considered as a class of excavatability
at which correspond specific geological-geotechnical properties and ROP and FPI ranges. The
identified clusters are comparable to the geological domains defined after rock mass surveys.
In order to obtain the mathematical representation of the relationship between the rock mass
features and the TBM performances registered during its excavation multiple regression analyses
have been carried out for each clusters.
When the assumptions for linear regression were satisfied, multiple linear regression method has
been adopted; on the contrary, nonlinear regressions have been performed. In these cases, simple
nonlinear regression analyses were considered in order to obtain the best mathematical function
able to describe the nonlinear relationship between a single rock mass property and the selected
TBM index. The nonlinear analyses have been obtained on the basis of this information.
The explanatory capability of the obtained empirical equations has been tested and the results are
presented in Chapter 12.
280
11 ROCK ENGINEERING SYSTEM:
FROM THE INTERACTION MATRIX TO TBM
PERFORMANCE ESTIMATION
11.1 Theoretical aspects

11.1.1 General concepts
The interaction matrix, proposed by John Hudson in 1992, is defined as the tool able to identify and
link all the rock mass parameters that describe rock engineering systems. The Figure 11.1 shows a
simplified scheme of interaction matrix.
Figure 11.1 Interaction Matrix (After Hudson, 1992).

Interaction matrix was proposed by the Author with the aim of representing the total system
behaviour, the relevant factors and their interactions. Even if the interaction matrix can be applied at
all systems, Hudson restricted his studies to interactions both between rock/rock masses features
and interactions induced by construction in natural ground and to rock slope stability problems.
281
The subjects, also called parameters, have to be listed along the leading diagonal of the matrix (top
left to bottom right) and we have to consider the interactions between subjects in the off-diagonal
boxes, using the convention of a clockwise rotation; as it is observable in Figure 11.1 where an
example of 2*2 matrix has been reported. In an n*n matrix there are n*(n-1) off-diagonal terms.
If the off-diagonal terms are the same (for example in a 2*2 matrix), the influence of a on b, and the
influence of b on a is the same, vice versa a influences b in a different way as b influences a.
In other words:
- when the complementary off-diagonal boxes are the same the matrix is symmetric; it means
that Boxij = Boxji. Hudson proposed the relation between stress and strain for the elastic
cases as example of symmetric matrix;
- when the complementary off-diagonal boxes are not the same the matrix is asymmetric, the
information contained in Boxij is different to the one listed in Boxji, an example of the
relationship between stress and strain for the inelastic cases.
11.1.2 Matrix resolution
The term “matrix resolution” focuses on how many lading diagonal terms are involved in an
interaction matrix; generally when the number of subjects increases the quality of resolution
improves (Hudson, 1992).
Hudson introduced an algebraic method, called (AB)C concept to indicate matrix resolution when
two or more subjects have been combined. To better understand this passage it might be helpful to
look at Figure 11.2.
A AB AC
BA B BC
CA CB C
Binary Interaction Ternary Interactions
282
(AB)C
AB
C(AB) C
Mixed Resolution Matrix Ternary Interactions reduced to Mixed

Resolution Binary Interactions
Figure 11.2 Matrix resolution methods (After Hudson, 1992).
The AB component, upper left corner, represents binary interaction (AB and BA) and each off-
diagonal term is a device combining two of the subjects with the independent variable determined
by the clockwise rotation (Hudson, 1992). The ternary (or higher level) interactions can be
considered both as a pathway through binary interaction matrix and as an interaction where a
number of parameters have been combined.
The dependent variable is inserted in the bottom right corner box of the matrix.
Figure 11.3 considers the simplest interaction matrix including rock mass properties, site features
and their combined effect on a generic project engineering.
R RS RP
SR S SP
PR PS P
Figure 11.3 Simplest interaction matrix involving rock mass properties (R), site conditions (S) and
project engineering components (P); (Modified after Hudson, 1992).
283
On the basis of R-S-P matrix the resolution could be increased and coarse, medium and finest
resolution, corresponding to top, intermediate and lowest level matrix respectively, are obtainable;
as reported in Figure 11.4.
Figure 11.4 Example of coarse, medium and fine resolution matrices (Hudson, 1992).
Increasing the matrix resolution means that a sub-matrix within each element of the higher level
matrix exists; furthermore a suite of cascading matrices is created and in the lowest level some
measurable quantity is usually achieved.
11.1.3 Atlas of rock engineering system
Hudson proposed an atlas of rock engineering system both for slopes and underground excavation.
Firstly, Author selected the appropriate leading diagonal terms; secondly, he identified the
mechanism in the off-diagonal ones. The matrix has been created with the aim to provide a
checklist of most important rock mechanisms and their possible interaction.
The twelve leading diagonal terms selected by Hudson for a generic underground excavation matrix
have been listed below.
1. Excavation dimension;
2. Rock support;
3. Depth of excavation;
4. Excavation method;
5. Rock mass quality;
284
6. Discontinuity geometry;
7. Rock mass structure;
8. In situ stress;
9. Intact rock quality;
10. Rock behaviour;
11. Discontinuity aperture;
12. Hydraulic condition.
The listed leading diagonal terms generate 132 off-diagonal interactions.
The generic underground excavation matrix has been reported in Appendix IV.
11.1.4 Matrix coding
Hudson proposed five methods in order to code the interactions between matrix subjects:
- Binary approach: if a parameter influences another one it is necessary to assign 1 value to

the corresponding off-diagonal box, vice versa code 0 will be assigned. If there is a higher
number of off-diagonal boxes with a value of 1, the system can be considered interactive,
contrariwise the system will not very interactive. This method is the simplest one, however
it is not able to explain the intensity of interactions, and it can only say if a parameter
influences another one or not.
- The expert semi-quantitative method (ESQ) considers five categories of interaction, ranging
from 0 to 4. The rate equal to 0 means “no interaction”, 1 means “weak interaction”, rate of
2 “medium”, rate 3 means “strong interaction” and rate equal to 4 is synonym of “critical
interaction”. It is an extension of binary system and it is affected by greater sensitivity than
the previous approach. ESQ is the most adopted method to code matrices. The rate of each
interaction parameter is usually evaluated based on the point of view of field-expert, for
example Frough and Toraby, 2013; adopted this approach for estimating TBM downtimes.
- The Pi vs. Pj slope method consists of plotting Pi and Pj values and to evaluate the slope of
the line that gives information on the rapidity by which Pj changes by function of Pi. This
method requires the knowledge of parameter values and to carry out the curve Pi vs. Pj for
all the off-diagonal boxes.
- The fourth method requires to represent by partial differential equations all the mechanisms
in the boxes and to compute the constants associated with the solution. This approach could
be difficult to apply, but there is potential in this method.
285
- The explicit method is based on numerical analysis of mechanisms; it requires large

knowledge of the investigated phenomena. The leading diagonal parameters have to have
the same quality, for example they must be expressed in the same units. The fifth method
can be linked to the ESQ and fourth ones.
11.1.5 Cause-effect plot
Once the matrix has been numerically rated, it is possible to sum the rating values of the rows and
the values of the columns through each parameter to evaluate the cause and effect coordinates. In
particular it is necessary to consider the sum of the row values as the “cause”: it means that it
represents the way in which Pi influences the system; on the other hand, the sum of column values
has to be considered as “effect”: it is the effect that Pi has on the system.
The boxes of leading diagonal parameters have no value because of they do not affect each other.
The cause and effect coordinates, calculated for each parameter, have to be plotted into a cause-
effect space. The Figure 11.5 shows an example of C, E plot creation.
P1 I12 I13 I14 I1i I1N C1

P2
I21 P2 I23 I24 I2i I2N C2 Pi
P3
I31 I32 P3 I34 I3i I3N C3
EFFECT (E)
I41 I42 I43 P4 I4i I4N C4 PN

P1
Ii1 Ii2 Ii3 Ii4 Pi IiN Ci
P4
IN1 IN2 IN3 IN4 INi PN CN
E1 E2 E3 E4 Ei EN CAUSE (C)
Figure 11.5 Matrix interaction for N parameter (on the left) and constellation of N parameter
points (on the right).
The coordinates of causes and effects are an indicator of the interaction intensity and dominance of
a parameter (Hudson, 1992). The distance along the diagonal and the perpendicular one from the
diagonal to the parameter (the point in the figure) represent the parameter interaction intensity and
dominance respectively, as shown in Figure 11.6.
286
Figure 11.6 Parameter interaction intensity and dominance.
The most interactive parameter is the one characterised by the highest value of C + E; on the
contrary, the least interactive one is the parameter affected by the least C + E value.
The most dominant parameter is the one defined by the highest C – E value, it means that this
parameter affects the system more than the system influences itself. The least C – E value is typical
of the most subordinate parameter (in these evaluations it is necessary to consider the sign).
Four extremes values of C + E and C – E might occur, they are located at the four corner point of
the C, E plot, as shown in Figure 11.7.
Figure 11.7 The extreme cases of parameter interaction intensity and dominance.
287
Point 1 represents the extreme case of the C + E and it is equal to zero. It corresponds to the “zero
interaction intensity” condition and it means that the considered parameter does not affect the
system and the system does not affect the parameter. On the contrary, the “maximum interaction
intensity” condition is represented by point 3; it occurs when each box of the associated row and
column has the maximum rate.
Point 2 represents the extreme case of dominance by the system on the parameter, row sum is zero
and column is maximized. Point 4 explains the opposite situation: row is maximized and column is
zero.
Generally, the four extreme cases are named as follow:
- Point 1: “Zero interaction intensity”;
- Point 2: “System dominates the parameter”;
- Point 3: “Maximal interaction intensity”;
- Point 4: “Parameter dominates system”.
288
11.1.6 Theorems 1 and 2
Theorem 1 establishes that the mean of C coordinates is equal to the mean of E coordinates and the
mean of C and E related to all parameters lies on the C = E line (the diagonal). Given that each
cause coordinate is the sum of a row, the result obtained summing all the cause coordinates together
is equal to all the boxes in the matrix added together; it is the same for effect coordinates.
The sum of all the causes is equal to the one of all the effects: for this reason, if it is divided by the
dimension of the matrix, the resultant mean cause and the resultant mean effect are exactly the
same. In conclusion, theorem 1 asserts that the gravity centre of point constellation in a C, E space
lies on the C = E line.
Theorem 2 asserts that parameter dominance or subordinacy is expressed by the distance of the
point from the C = E line and the maximum parameter dominance or subordinacy is verified when
the parameter is 50% interactive.
On the basis of theorem 1 and 2, Author concluded that each constellation of points in a C, E space
has the gravity centre on diagonal line (C = E line); furthermore if the constellation is along this
line, the parameters can be ranked in order of interactive importance; contrariwise, if the parameter
are scattered on the line perpendicular to the C = E line, they can be listed in order of dominance (or
subordinacy) intensity.
289
11.2 RES application to TBM performance estimation

As it is well known, several parameters characterise ground conditions and the performance of
TBM working in a given rock mass depends on them. These parameters are influencing each other,
and they simultaneously affect TBM performance.
The simultaneous influence of geological site features on FPI and ROP has been evaluated by
means of statistical approaches (Chapters 9 and 10) and in this chapter it is investigated using rock
engineering system (RES) approach.
To establish as more as possible machine-rock mass interaction, 186 rock sections belonging to “La
Maddalena” tunnel (for a total length of 2010 m) have been studied. The length investigated in this
study involves the gneiss of Ambin complex (identified as domain I, DI), the micaschist of Clarea
complex (domain III, DIII) and the transition zone formed by an alternation between gneiss and
micaschist (domain II, DII). The detailed description of these units has been provided in Chapter 5.
11.2.1 Rock mass parameters involved in RES application
In order to apply RES at “La Maddalena” TBM performance analyses, five rock mass parameters
have been selected as leading diagonal terms: UCS, quartz content (q), spacing between fractures
(S), the angle between the principal set of discontinuities (α), the parameter describing joint
conditions (Jc) and tunnel depth (d). Some parameters selected by Hudson for a generic excavation
matrix have been excluded from the present analysis because they represent information not
available in this study (for example in situ stress and rock behaviour) or because considered not
relevant for this specific excavation (as excavation method and dimension or support). The quartz
content and joint condition information have been added to the parameters selected by Hudson.
The indexes related to TBM performances is the field penetration index (FPI).
11.2.2 Interaction matrix
The matrix coding method adopted in this study considers both the semi-quantitative coding method
developed by Hudson (Hudson, 1992; Mazzoccola and Hudson, 1996) and the method according to
the slope of Pi vs. Pj relationship (instead of slope the coefficients of correlation between Pi and Pj
have been used).
In this study six class of interaction have been proposed and used; in fact, in order to make the
evaluation of the effect of a geological parameter on another one more objective, bivariate
290
correlations have been carried out and coefficients of correlation, r, have been calculated.
Consequently, the parameter interaction ratings have been assigned on the basis of r value.
If between two parameters there is no relationship we will assign rating equal to 0.00, if R ranges
from 0.01 to 0.20 the assigned rate is 1 (very weak interaction), if r ranges from 0.21 to 0.40 the
assigned rate is 2 (weak interaction), if coefficient of regression is between 0.41 and 0.60 a rate
equal to 3 has been considered (medium interaction), if r ranges from 0.61 to 0.80 the rating is 4
(strong interaction) and r between 0.81 and 1.00 involves a rate equal to 5 (very strong interaction).
Ratings and classes are listed in Table 11.1.
r Description Rating
0.00 No interaction 0
0.01 – 0.20 Very weak 1
0.21 – 0.40 Weak 2
0.41 – 0.60 Medium 3
0.61 – 0.80 Strong 4
0.81 – 1.00 Very strong 5
Table 11.1 Rating and classes of interaction.
Bivariate correlation analyses between parameters have been carried out and r values related to each
correlation between Pi and Pj are resumed in Table 11.2.
FPI UCS q S Alpha Jc Depth

FPI 1.00 0.55 0.66 0.59 0.31 0.34 0.19
UCS 0.55 1.00 0.48 0.40 0.21 0.28 0.15
q 0.66 0.48 1.00 0.43 0.31 0.34 0.31
S 0.59 0.40 0.43 1.00 0.21 0.45 0.14
Alpha 0.31 0.21 0.31 0.21 1.00 0.15 0.13
Jc 0.34 0.28 0.34 0.45 0.15 1.00 0.22
Depth 0.19 0.15 0.31 0.14 0.13 0.22 1.00
Table 11.2 Coefficients of correlation between Pi vs. Pj.
The rates of each effect have been assigned based on r values reported in Table 11.2 and
considering the classification proposed in this study and summarised in Table 11.1.
The interaction matrix has been completed as proposed in Figure 11.8. The seven leading diagonal
terms are the Uniaxial Compressive Strength, the spacing between discontinuities, the quartz
291
content, the joint condition parameter, the tunnel depth, the alpha angle and FPI. Interactions
between parameters have been placed into the 42 off-diagonal boxes and 80 is the sum of ΣCi and
ΣEi.
effects causes
ΣCi
Ei↓ Ci →
UCS 2 0 2 0 1 3 8
0 Spacing 0 3 0 0 3 6
3 3 Quartz 2 0 2 3 13
0 0 0 Jc 0 0 2 2
1 1 2 1 Depth 1 1 7
0 1 0 1 0 alpha 2 4
0 0 0 0 0 0 FPI 0
ΣEi 4 7 2 9 0 4 14 80
Figure 11.8 “La Maddalena” interaction matrix.
It is observable that the off-diagonal terms are not symmetric to the leading diagonal, it means that
the way in which a parameter influences another one is not the same as the second one influences
the first one. In order to identify the “less interactive” and the “more interactive” parameters, the
“most dominant” and the “most subordinate” ones, the coordinates of causes (Ci) and those of
effects (Ei) have been plotted into a cause-effect space. The cause effect diagram carried out for “La
Maddalena” case study is reported in Figure 11.9.
15
FPI
UCS
10
Jc spacing
effect (E)
quartz
spacing
Jc
5
depth
alpha UCS
alpha
quartz FPI
depth
0
0 5 10 15
cause (C)
Figure 11.9 Cause-effect “La Maddalena” diagram.
292
The most interactive parameters fall in the upper right region, contrariwise the less interactive ones
are plotted in the lower left region.
By observing cause-effect diagram, it can be observed that FPI is totally affected by the system. As
we know FPI cannot influence rock mass conditions and it depends on ground features and on
machine specification (not included in this RES analyses).
Tunnel depth has only effect on the system, for this reason it is located on the cause axes, because
there is no one geotechnical parameter able to affect tunnel depth.
As we could expect quartz content has the maximum effect on the system followed by UCS, in
particular the content of quartz results the most interactive parameter.
Spacing between fractures and alpha angle both influence the system and are influenced by it. The
parameter describing joint conditions results largely affected by the system, this is in accordance
with the nature of this index.
The histogram in Figure 11.10 has shown the role of system interactivity, plotting the interactive
intensity obtained from the sum of causes and effects vs. the parameters.
16
14
12
10
8 CAUSES
EFFECTS
6
0
UCS S Q Jc DEPTH ALPHA FPI
Figure 11.10 Histogram of interactive intensity.
In conclusion it is possible to assert that the most interactive parameter is quartz content (C + E =
15) and the least one is tunnel depth (C + E = 7). The most dominant element is quartz content (C –
E=11) and the most subordinate, excluding FPI, is Jc (C – E = – 7).
293
11.2.3 Predictive model development
In order to carry out a TBM performance predictive model based on RES approach, each rock mass
section (186) has been rated to calculate an index called, in this study, FPIR (FPI-RATED). Each
rock mass section is described by five geomechanical parameters and depth tunnel (UCS, q, S, α, Jc
and depth); for each parameter a rating has to be assigned.
In order to assign objectiveness the rates to parameters, in a first moment ISRM recommendations,
RME and Deere and Miller classifications have been involved to divide parameter values into
classes; however this approach does not result effective in the following analyses. For this reason a
different approach has been adopted.
The statistical distribution of collected values has been investigated for USC class determination.
Rating equal to 1 has been assigned to those values ranging between the minimum value and the
first quartile (68 – 105 MPa), rating equal to 2 affects the class ranging from first quartile to median
value (105 – 130 MPa), rating of 3 has been assigned to the UCS values between median value and
third quartile (130 – 160 MPa) and the values higher than 160 MPa have been affected by rating
equal to 4.
Also quartz content classes have been defined considering first, second and third quantile of data
distribution. The rating assigned to the class containing the values under first quartile (Q1 = 25%) is
equal to 1, rating of 2 has been assigned to the q values ranging between first quartile and median
value (25 – 30%), values between second and third quartile (30 – 45%) are affected by rating equal
to 3 and rating of 4 has been assigned to quartz content values major than 45%.
The same considerations adopted for UCS and quartz content have been utilised for spacing
classification. Rating equal to 1 corresponds to spacing values less than 100 cm (first quartile),
rating of 2 corresponds to those values ranging between first and second quartile (100 – 150 cm),
the spacing values between median and third quartile (150 – 287.5 cm) have been rated 3 and the
values higher than 278.5 cm have been characterized by rating equal to 4.
The alpha angle ratings have been assigned considering the influence of discontinuity orientations
on tunnel excavation. In this study the strike of the principal set of fracture is perpendicular to the
tunnel axis and excavation drives against dip. For this reason the discontinuities characterized by
dips less than 20° and major than 45° determine “fair” conditions of excavation and the assigned
rating is equal to 1; the dips ranging between 20° and 45° result in unfavourable excavation
conditions, for this reason the assigned rating is equal to 2.
294
About joint condition parameter (Jc) the classes proposed by RMR classification (Bieniawski,
1989) have been involved. High values of Jc coefficients mean a good global rock masses quality
and it determines high values of FPI. For this reason the Jc less than 10 are characterized by rating
equal to 1, Jc between 10 and 20 are affected by rating equal to 2, the rating of 3 has been assigned
to Jc values ranging between 20 and 25 and values higher than 25 are related to rating equal to 4.
At the entire range of tunnel depth values has been assigned a rating equal to 0.3.
The parameter ratings have been listed in Table 11.3.
RATING
1 2 3 4
UCS [MPa] 68-105 105-130 130-160 >160
q [%] <25 25-30 30-45 >45
S [cm] <100 100-150 150-287.5 >278.5
alpha [°] <20; >45 20-45
Jc 0-10 10-20 20-25 25-30
depth [m] <1000 rating 0.3
Table 11.3 Rating of affecting parameters in FPI.
As a result we obtained a table where in rows there are rock sections and in the columns the rated
parameters describing each rock mass section (Table 11.4).
Rock UCS alpha S Jc q depth

mass
section [MPa] [°] [cm] [%] [m]
1 4 1 2 2 4 0.3
2 4 2 2 2 4 0.3
3 4 2 2 2 4 0.3
4 3 2 2 2 4 0.3
5 3 2 3 2 4 0.3
6 3 2 3 2 4 0.3
... 3 1 2 2 4 0.3
N 3 1 4 3 4 0.3
Table 11.4 Example of rated parameters for each rock section.
295
The interactive intensity (C + E) related to each parameter and carried out by the interaction matrix
has been transformed into a percentage to obtain a weighting coefficient able to express the
proportional contribute of each parameter on FPIR, according to the following equation (Hudson,
1992):
(𝐶 + 𝐸)
𝐼𝐶 = % (11.1)
∑𝑖(𝐶 + 𝐸)
Ic is typical for each parameter and it has been normalised by dividing it with the maximum rating
assigned to that parameter. This operation gives ai as result, in according to the following equation
(Hudson, 1992):
1 (𝐶 + 𝐸)
𝑎𝑖 = ∗ % (11.2)
𝑀𝑃𝑖𝑗 ∑𝑖(𝐶 + 𝐸)
At this point, ai for each parameter has been established and this allows us to compute FPIR
according to the following formulation:
6
𝐹𝑃𝐼𝑅 = ∑ 𝑎𝑖 ∗ 𝑃𝑖𝑗 (11.3)
𝑖=1
where
- i refers to the six geomechanical parameters;
- j refers to the rock mass section (1 to 186);
- ai is the value of C + E computed for each parameter;
- Pij is the rate assigned to different class parameters.
In order to better understand the calculation of FPIR practical explanation has been reported in
Figure 11.11.
296
Rock section UCS alpha S Jc q depth FPIR

RATINGS
161 1 2 1 2 1 0.3 38.13
162 3 2 1 2 1 0.3 45.63
163 2 2 2 2 1 0.3 45.94
164 3 2 1 2 1 0.3 45.63
165 1 1 2 2 1 0.3 37.19
166 3 2 2 2 1 0.3 49.69
167 2 1 1 1 1 0.3 33.44
... 2 1 1 1 1 0.3 33.44
N 3 1 1 2 1 0.3 40.63
Max. value assigned to
each parameter 4 2 4 4 4 0.3
(𝐶 + 𝐸)
𝐼𝐶 = %
∑𝑖(𝐶 + 𝐸) 15 10 16.25 13.75 18.75 8.75
𝐼 (15/4)
𝑎𝑖 = 𝑐⁄𝑚𝑎𝑥 3.75 5.00 4.06 3.44 4.69 29.17
FPIR related to each rock section has been calculated in according to the following formula:
𝐹𝑃𝐼𝑅 = 𝑎𝑖−𝑈𝐶𝑆 ∗ 𝑈𝐶𝑆𝑖𝑗 + 𝑎𝑖−𝛼 ∗ 𝛼𝑖𝑗 + 𝑎𝑖−𝑆 ∗ 𝑆𝑖𝑗 + 𝑎𝑖−𝐽𝑐 ∗ 𝐽𝑐𝑖𝑗 + 𝑎𝑖−𝑞 ∗ 𝑞𝑖𝑗 + 𝑎𝑖−𝑑 ∗ 𝑑𝑖𝑗 (11.4)
An example about FPIR calculation, for rock section number 161, has been provided below:
𝐹𝑃𝐼𝑅 = 3.75 ∗ 1 + 5.00 ∗ 2 + 4.06 ∗ 1 + 3.44 ∗ 2 + 4.69 ∗ 1 + 29.17 ∗ 0.3 = 38.13 (11.5)
Figure 11.11 Practical example of FPIR calculation.
At this point the FPI value measured in each rock mass section has been related to the FPIR
computed as below described. Regression analyses have been carried out where FPI is the
dependent variable and FPIR the independent one.
Regression analyses were carried out on:
- the entire investigated tract (consisting of gneiss, domain I; gneiss and micaschist
alternation, domain II; and micaschist, domain III);
- gneiss section (DI);
- gneiss and micaschist alternation (DII);
- micaschist (DIII).
For a detailed description of geological domains see Chapter 5.
297
In order to select the best mathematical function able to describe the relationship between
dependent and independent variables, linear, logarithmic, power, exponential relationships have
been investigated by means of IBM SPSS Statistic 25.
Graphical representations of the relationship between computed FPIR and observed FPI have been
presented and statistical tests have been carried out with the purpose to individuate the best
mathematical function describing the relationship (Figure 11.12).
DI+DII+DIII
-Linear:
F=204.037
R2=0.526
-Logarithmic:
F=196.436
R2=0.516
FPI
-Power:
F=238.614
R2=0.565
-Exponential:
F=239.059
R2=0.565
FPIR
DI
-Linear:
F=15.039
R2=0.157
-Logarithmic:
F=14.389
R2=0.151
FPI
-Power:
F=20.415
R2=0.201
-Exponential:
F=20.623
R2=0.206
FPIR
298
DII
-Linear:
F=32.156
R2=0.434
-Logarithmic:
F=32.530
R2=0.436
FPI
-Power:
F=40.067
R2=0.488
-Exponential:
F=39.364
R2=0.484
FPIR
DIII
-Linear:
F=8.027
R2=0.123
-Logarithmic:
F=6.587
R2=0.104
FPI
-Power:
F=3.102
R2=0.052
-Exponential:
F=3.911
R2=0.064
FPIR
Figure 11.12 Relationships between observed FPI and FPIR on the left and F and R2 values for
each tested mathematical function, on the right.
It is observable that the relationship between FPI and FPIR are low in domain I and very low in
domain III: the relationship observed in the transition zone is acceptable; however, the higher level
of correlation between FPI and FPIR has been observed involving data related to the entire tunnel
tract (DI+DII+DIII). For these reasons, the subsequent analyses have been developed considering
the domains all together.
299
The correlation index between FPIR and FPI measured into all domains are significantly high than
the ones measured into separately considered domains. The mathematical function that best
represents the relationship between FPIR and FPI is exponential type.
At this point, we carried out an exponential regression model where FPIR has been selected as
independent variable and FPI has been considered as dependent one, in according to the following
formula:
𝐹𝑃𝐼 = 8.422 ∗ 𝑒 0.043∗𝐹𝑃𝐼𝑅 (11.6)
The relationship between FPI values measured in situ and the predicted by means of RES approach
is quite strong (r = 0.706 and R2 = 0.498).
300
11.3 Conclusion
11.3 Conclusion
The purpose of this study was not to propose a new predictive formulation only, but to carry out a
new approach able to reduce the subjectivity that usually affects the assignment of ratings to
parameters considered in RES application.
After this study, it is possible to advise to assign ratings, for matrix interaction compilation, on the
basis of Pearson correlation index (r) between the parameters. By considering the modulus of r
values the operators should divide the r range into four or five classes and assign a rating to each of
them. In this way, the point of views of tunnelling expert could be confirmed or rejected end
parameter interactions can be objectively quantified.
Predictive models can be carried out on the basis of RES approach, as Frough and Torabi (2013)
were able to demonstrate. In their research they rated parameters using classes as proposed by
Bieniawki in Rock Mass Excavatability classification. In this study, it was observed that using
rating proposed by common geomechanical classification (i.e. ISRM, RME, Deere and Miller) the
obtained results were not appreciable. It is possible to conclude that for a given specific site the best
method to identify classes of parameter is to consider their statistical distribution, then to divide
them on the basis of first, second and third quartile; therefore ratings will be assigned to each class.
The proposed method is applicable to different geological and geotechnical situations and it allows
to carry out a predictive model specific for a given tunnel project.
301
12 PREDICTIVE MODEL VALIDATION
12.1 Introduction
The utility of the models carried out in this study and presented in the previous chapters (Chapters
9, 10 and 11) has been tested, focusing on its explanatory and predictive capabilities.
Explanatory capability of models
The aim of this phase is to evaluate whether a regression equation is appropriate to represent the
relationship between the geological and geotechnical features of a given rock mass and the
performances of the TBM involved to bore it. In other words, being that the regression model
construction implies assumptions and choices about the cause-effect relation and input parameters,
it is necessary to test the reliability of the regression equation and so, the reasonableness of the
assumptions and the selected inputs.
The capability of a regression model to represent the relationship between the geological and
geotechnical parameters and the TBM indices is largely expressed by means of the coefficients r, R2
and R2adj. Moreover, this capability has been evaluated comparing the ROP or FPI values got by the
data recorded during the excavation phases and the ones calculated by means of the equation
developed in the present work. Then the measured and the evaluated data have been plotted in a
Cartesian space in order to graphically evaluate the correlation between them.
To this purpose, the dataset built with data collected from Pk. 0 + 198 m to Pk. 2 + 010 m has been
used firstly to obtain the regression equations and then to compare the recorded data and the model
results.
Predictive capability of models
The dataset referring to the rock mass encountered between Pk. 0 + 198 m and Pk. 2 + 010 m has
been used to obtain regression models; another dataset, different from the first one, has been used to
validate them. In other words, the model and the following equation are obtained by the analysis of
data recorded and gathered along a specific tunnel tract (from Pk. 0 + 198 m to Pk. 2 + 010 m), and
then they are validated, to verify the predictive ability of the equations, by using a dataset gathered
in a different tunnel section, ranging between Pk. 2 + 0.10 m and Pk. 6 + 897 m.
303
12.1 Introduction
The capability of the obtained predictive models has been evaluated by comparing the TBM
performances index ROP and FPI measured (by the TBM acquisition system recording data) and
computed (with the regression models) gathered and obtained, respectively, along the tunnel section
between Pk. 2 + 010 m to Pk. 6 + 897 m (for a total length of about 4800 m).
In this case too, the coefficient of determination and the graphical representation are adopted to
evaluate the relation between measured and calculated data.
Table 12.1 reports the comparison between the main characteristics of the rock mass portion used to
obtain the regression models (from Pk. 0 + 198 m to Pk. 2 + 010 m) and the one used to validate
them (from Pk. 2 + 010 m to Pk. 6 + 897 m).
Model Creation Model Validation

Pk. 0 + 198 m – Pk. 2 + 010 m Pk. 2 + 010 m – Pk. 6 + 897 m
UCS [MPa]
range 68.00 – 251.00 60.00 – 255.00
median 129.00 143.00
Q1-Q3 104.70 – 158.00 152.00 – 168.50
Alpha [°]
range 7.40 – 85.00 2.80 – 72.60
median 45.20 45.00
Q1-Q3 36.80 – 52.30 39.40 – 54.14
Spacing [m]
range 0.27 – 11.80 0.20 – 4.33
median 1.50 0.83
Q1-Q3 1.00 – 2.76 0.71 – 1.00
Jc
range 7.00 – 29.00 7.00 – 21.00
median 18.00 18.00
Q1-Q3 15.00 – 20.00 18.00 – 19.00
quartz[%]
range 25.00 – 50.00 20.00 – 35.00
median 30.00 25.00
Q1-Q3 25.00 – 45.00 25.00 – 30.00
depth[m]
range 79.40 – 616.70 256.80 – 1992.60
median 439.30 1083.9
Q1-Q3 285.20 – 522.10 1056.4 – 1156.00
Table 12.1 Comparison between the rock mass portion used to obtain the regression models (Pk. 0
+ 198 m – Pk. 2 + 010 m) and the one use to validate them (Pk. 2 + 010 m – Pk. 6 + 897 m).
304
12.1 Introduction
The Table 12.2 summarises the statistical description of ROP and FPI measured both in the tract of
tunnel used to create models and in the one considered for their validation.
Model Creation Model Validation

Pk. 0 + 198 m – Pk. 2 + 010 m Pk. 2 + 010 m – Pk. 6 + 897 m
ROP[m/h]
range 0.30 – 3.74 0.62 – 3.74
median 1.40 1.47
Q1-Q3 1.00 – 2.00 1.18 – 1.97
range 15.16 – 298.6 14.20 – 165.10
median 84.80 49.92
Q1-Q3 50.70 – 140.90 35.40 – 70.95
Table 12.2 Comparison between ROP and FPI measured in the rock mass portion used to obtain
the regression models (Pk. 0 + 198 m – Pk. 2 + 010 m) and the one use to validate them (Pk. 2 +
010 m – Pk. 6 + 897 m).
The most reliable regression models descripted in Chapters 9 and 10, and validated in this Chapter,
are reported in Table 12.3.
Equation ID
𝑅𝑂𝑃 = −0.004 ∗ 𝑈𝐶𝑆 − 0.041 ∗ 𝑞 − 0.057 ∗ 𝑆 + 0.030 ∗ 𝐽𝑐 + 0.001 ∗ 𝑑 + 2.889 9.38
𝑅𝑂𝑃 = 22.316𝑒 −2.120𝐸−4𝑈𝐶𝑆 − 1.426𝑙𝑛(𝑞) + 7.806𝑒 −0.0003𝛼 + 31.171𝑆 −0.003
9.45
− 1.412𝑒 −0.094𝐽𝑐 + 0.001𝑑1.024 − 53.952
𝑅𝑂𝑃 = 2.295 − 0.006 ∗ 𝑈𝐶𝑆 − 0.041 ∗ 𝑞 + 0.056 ∗ 𝐽𝑐 + 0.001 ∗ 𝑑 10.11
𝑅𝑂𝑃 = −0.650 ∗ ln 𝑈𝐶𝑆 − 0.024 ∗ 𝑆 − 0.249 ∗ ln 𝛼 + 2.317𝐸 −5 ∗ 𝑑1.540 + 9.608 10.16

𝑅𝑂𝑃 = 0.079 ∗ 𝑈𝐶𝑆 + 0.258 ∗ 𝑆 − 0.306 ∗ 𝐽𝑐 − 9.388 10.17
𝐹𝑃𝐼 = 0.394 ∗ 𝑈𝐶𝑆 + 2.564 ∗ 𝑞 + 8.707 ∗ 𝑆 − 63.491 9.46
𝐹𝑃𝐼 = 8.408𝑈𝐶𝑆 0.509 + 7513.902𝑞 0.011 + 27.552𝑙𝑛(𝑆) + 0.080𝐽𝑐 − 7818.256 9.51
𝐹𝑃𝐼 = 0.576 ∗ 𝑈𝐶𝑆 + 2.991 ∗ 𝑞 + 7.651 ∗ 𝑆 − 0.065 ∗ 𝑑 − 69.901 10.18
𝐹𝑃𝐼 = 5.522 ∗ 𝑈𝐶𝑆 0.568 + 42.618 ∗ 𝑒𝑥𝑝0.069∗𝑆 10.21

𝐹𝑃𝐼 = 17.902 ∗ 𝑈𝐶𝑆 − 5.374 ∗ 𝛼 − 0.643 ∗ 𝑑 − 3144.566 10.22
𝐹𝑃𝐼 = 8.422 ∗ 𝑒 0.043∗𝐹𝑃𝐼𝑅 11.6
Table 12.3 List of validated predictive models.
305
12.2 Validation of ROP models
12.2.1 Validation of RL2 model
This model is described in section 9.4.1.1; here the obtained empirical equation has been reported:
𝑅𝑂𝑃 = −0.004 ∗ 𝑈𝐶𝑆 − 0.041 ∗ 𝑞 − 0.057 ∗ 𝑆 + 0.030 ∗ 𝐽𝑐 + 0.001 ∗ 𝑑 + 2.889 (9.38)
The formula is able to explain the 64.30% of the total variance of the dataset (R2adj = 0.643).
Firstly its explanatory capability has been verified comparing the ROP values measured between
Pk. 0 + 198 m and Pk. 2 + 010 m, and the ones calculated by means of RL2 model. Then its
predictive reliability has been verified comparing data registered from Pk. 2 + 0.10 m to Pk. 6 + 897
m, and the ones calculated in this tract by means of the equation.
RL2 explanatory capability
The comparison between the measured ROP values and the ones calculated by means of the
equation 9.38, is reported in Figure 12.1.
RL2 Model
4.0
3.5
3.0
ROP calculated [m/h]
2.5
2.0
1.5
1.0
0.5
0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
ROP measured [m/h]
Figure 12.1 Comparison between measured and calculated ROP (RL2 model).
306
The correlation between the measured and calculated ROP is quite good, as it is possible to note the
points are roughly aligned along the bisector. It means that the model is capable to explain the
relationship between the geological parameters and the TBM one.
Use of RL2 model to predict ROP
In order to evaluate its capability to predict ROP, RL2 has been used to calculate the rate of
penetration in the tract of tunnel that has not been considered during model construction. By
comparing predicted and measured performance, it was noted that the model is not able to predict
the rate of penetration for the whole tunnel. It can be due to the fact that the depth of tunnel sector
in which model was developed is significantly lower than the one of the tract in which model has
been tested. Moreover, it is necessary to consider that the depth parameter has been considered as
an indicator of in situ stress conditions; however it gives a first indication of vertical stress
conditions and does not provide any information about horizontal stress. In situ stress levels could
be increased with depth increasing and this situation might affect the machine response.
The RL2 model could be used to predict ROP in those portions of tunnel characterised by
overbunden lower than 1000 meters. The comparison between measured and predicted values is
shown in Figure 12.2.
RL2 Model - prediction

4.0
3.5
3.0
ROP predicted [m/h]
2.5
2.0
1.5
1.0
0.5
0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
ROP measured [m/h]
Figure 12.2 Measured vs. predicted ROP.
307
The correlation coefficient, r, is equal to 0.441 ad R2 is of 0.195. It denotes a moderate correlation

between the observed and evaluated data.
The model is not applicable to the entire length because of the significant increasing of depth; this
remarks the fact that the empirical predictive models can be used only in site conditions similar to
the ones in which the models has been built.
Moreover, the rock mass portion considered to obtain the model is characterised by spacing
between fractures higher than the sector used to validate it.
308
12.2.2 Validation of RNL model
The nonlinear regression analyses carried out to describe the nonlinear relation between
independent variables and response parameter allow the selection of the equation presented in
section 9.4.1.2 and here reported:
𝑅𝑂𝑃 = 22.316𝑒 −2.120𝐸−4𝑈𝐶𝑆 − 1.426𝑙𝑛(𝑞) + 7.806𝑒 −0.0003𝛼 + 31.171𝑆 −0.003 − 1.412𝑒 −0.094𝐽𝑐
+ 0.001𝑑1.024 − 53.952 (9.45)
The coefficient of determination related to Equation 9.45 is equal to 0.655 so the correlation is quite
good.
Also in this case the explanatory and predictive capability of the model has been tested, following
the same approach described in the previous paragraphs.
RNL explanatory capability
The comparison between measured and calculated data (Figure 12.3) shows that the regression
model generates an overestimation of ROP.
RNL Model
4.0
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
ROP measured [m/h]
Figure 12.3 Comparison between measured and calculated ROP (nonlinear model).
309
The linear model best represents the influence of geomechanical conditions on ROP. It is
reasonable to consider that the input parameters have been entered with the correct mathematical
nonlinear function and an overestimation of the constant value has been made.
Use of RNL model to predict ROP
The equation 9.45 has been used to predict ROP in the tract of tunnel not considered during model
construction. In this case too, the formulation has been applied only in those rock mass portions
where the maximum depth is of 1000 meters (for the reasons discussed about the linear model); the
correlation is moderate (R2 = 0.114) and the points distribution is similar to the one observed for
RL2.
The comparison is graphically shown in Figure 12.4.
RNL Model - prediction

4.0
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
ROP measured [m/h]
Figure 12.4 Measured vs. Predicted ROP (nonlinear model).
The overestimation of the constant of the nonlinear equation makes the predicted values
overestimated with respect to the real ones.
310
12.2.3 Validation of ROP models carried out for each cluster
For each cluster described in Chapter 10, the relationship between the geological parameter and
ROP has been investigated; this allows to carry out three empirical equations. Firstly their
explanatory ability has been tested, then the equation obtained for cluster I has been considered to
predict ROP. Equation 10.11 has been selected to forecast the TBM index because the cluster that
best represents the Domain III (micaschist) is the cluster I.
Explanatory capability
Cluster I
Cluster I is defined by the following ranges of values: UCS = 93.0 – 135.0 MPa; α = 29.8° – 49.0°;
S = 0.84 – 1.56 m; q = 25 – 30 %; Jc = 15 – 19 (the detailed description is provided in Chapter 10).
The linear multiple regression analysis allows to define the geological parameters-ROP
relationship; it is as follows (Model identified as C1c):
𝑅𝑂𝑃 = 2.295 − 0.006 ∗ 𝑈𝐶𝑆 − 0.041 ∗ 𝑞 + 0.056 ∗ 𝐽𝑐 + 0.001 ∗ 𝑑 (10.11)
The comparison between the observed and the evaluated data is proposed in Figure 12.5.
C1c Model
4.0
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
ROP measured [m/h]
Figure 12.5 Measured vs. calculated ROP, cluster I.
311
In the case of cluster I, the relationship between calculated and measured ROP is appreciable, points
are plotted close to the 1:1 line. The coefficient of correlation is 0.686 and the determination one is
of 0.470.
Cluster II
Cluster II, as described in Chapter 10, is defined by: UCS = 130 – 168.3 MPa; α= 44° – 60°; S =
2.39 – 6.00 m; q = 45%; Jc = 17 – 23. When this situation occurs, the equation describing the
relation between geomechanical properties and ROP is of nonlinear type and it is as follows:
𝑅𝑂𝑃 = −0.650 ∗ ln 𝑈𝐶𝑆 − 0.024 ∗ 𝑆 − 0.249 ∗ ln 𝛼 + 2.317𝐸 −5 ∗ 𝑑1.540 + 9.608 (10.16)
ROP observed and the same index calculated by means on 10.16 has been compared (Figure 12.6).
C2a Model
6.0
5.0
4.0
3.0
2.0
1.0
0.0
0.0 1.0 2.0 3.0 4.0 5.0 6.0
ROP measured [m/h]
Figure 12.6 Measured vs. calculated ROP, cluster II.
As it is possible to observe, a significant overestimation of ROP has been generated. Two possible
causes have been individuated. The first is related to the constant value proposed by the regression
analysis; reducing it the predicted values are included in the range of the observed ones. It is
possible that, by a geological point of view, the rate of penetrations might be high but technical
factors, operator decisions or in general not geological conditions make ROP maintained low;
although ROP values equal to 5.00 – 6.00 m/h is quite impossible to be achieved (when tunneling is
performed in rock masses similar to La Maddalena one).
312
Cluster III
Cluster III is characterised by the following properties: UCS = 200 MPa; α= 44.75° – 52.30°; S =
1.29 – 2.48 m; q = 50 %; Jc = 20. Also in this case a linear multiple regression analysis has been
carried out, as result, the following empirical formulation has been obtained:
𝑅𝑂𝑃 = 0.079 ∗ 𝑈𝐶𝑆 + 0.258 ∗ 𝑆 − 0.306 ∗ 𝐽𝑐 − 9.388 (10.17)
This equation was used to calculate the rates of penetration in cluster III then they have been
compared with the measured data, a very high correlation has been noted, as it is possible to see in
Figure 12.7.
C3c Model
3.0
2.5
2.0
1.5
1.0
0.5
0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0
ROP measured [m/h]
Figure 12.7 Measured vs. calculated ROP, cluster II.

In this case it is possible to conclude that the TBM performances have been determined by
geological conditions only, being the R2 equal to 0.929. It is necessary to consider that the
regression analyses, in this case, were performed on nine observations only.
313
Use of C1c model to predict ROP
The tract of rock mass ranging between Pk. 2 + 0.10 m and Pk. 6 + 897 is not very similar to any
cluster recognised in the first tunnel sector. However the most similar one is cluster I: for this
reason the model obtained for cluster I (C1c) has been used to predict ROP in the tract of tunnel
from Pk. 2 + 0.10 m to Pk. 6 + 897. The depth of tunnel excavation is considered by the model: this
makes it not applicable to the entire tunnel tract but only to those portions affected by maximum
overbunden equal to 1000 meters.
The scatterplot of the data registered during the excavation and the ones calculated by means of the
C1c model (Equation 10.11) is reported in Figure 12.8.
C1c Model - prediction

4.0
3.5
3.0
ROP predicted [m/h]
2.5
2.0
1.5
1.0
0.5
0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
ROP measured [m/h]
Figure 12.8 Measured vs. predicted ROP (C1c model).

The equation obtained from the analysis developed on data belonging to the cluster I is quite able to
predict ROP. Points are aligned near to the bisector line; r and R2 are equal to 0.319 and 0.102
respectively. The rock mass sections considered to develop the model are characterised by UCS
ranges between 93.0 MPa and 135.0 MPa (in terms of Q1-Q3) which is very different from the UCS
values measured in the validation rock mass portion (Q1 and Q2 equal to 152.00 MPa and 168.50
MPa, respectively). If the UCS values had been similar, the prediction would have been more
accurate.
314
12.3 Validation of FPI models
The procedure adopted to validate the FPI models is the same described for ROP model validations;
so the capability of the models in geology-machine relationship has been verified (comparing the
FPI values measured between Pk. 0 + 198 m and Pk. 2 + 010 m, and the ones calculated by means
of obtained model). Then their predictive reliability has been verified comparing data registered
from Pk. 2 + 0.10 m to Pk. 6 + 897 m, and the ones calculated in this tract by means of the
generated equations.
12.3.1 Validation of FL4 model
This model has been discussed in section 9.4.2.1; the obtained empirical equation is as follows:
𝐹𝑃𝐼 = 0.394 ∗ 𝑈𝐶𝑆 + 2.564 ∗ 𝑞 + 8.707 ∗ 𝑆 − 63.491 (9.46)
FL4 explanatory capability
The model explains the 57.10% of the total variance of the dataset (R2adj = 0.571). The comparison
between the FPI measured during tunnel construction and the indices calculated by means of the
equation 9.46, is represented in Figure 12.9.
FL4 Model
300
FPI calculated [(kN/cutter)/(mm/rev)]
250
200
150
100
50
0
0 50 100 150 200 250 300
FPI measured [(kN/cutter)/(mm/rev)]
Figure 12.9 Measured vs. calculated FPI, FL4 model.
315
The correlation between the registered and calculated FPI is appreciable: in fact, the points are
approximately disposed along the 1:1 line and the coefficient of correlation, r, is equal to 0.761 and
R2adj. is 0.571. It means that the model is capable to explain the relationship between the geological
parameters and the FPI.
Use of FL4 model to predict FPI
In order to evaluate the model capability to predict FPI, it has been used to calculate the field
penetration index on the tract of tunnel that has not been considered during model building. By
comparing predicted and observed performances it was noted that the model is not suitable to
predict FPI (Figure 12.10).
FL4 Model
200
150
100
50
0
0 50 100 150 200
Figure 12.10 Measured vs. predicted FPI, FL4 model.
This lack may be due to the fact that only three geological factors have been entered into the
regression analysis. Probably in the tract where the model has been tested, the most influencing
variables have changed and one or more than one of those had not been considered by the equation.
Also this evidence underlines the situ-specificity of the empirical predictive models.
316
12.3.2 Validation of FNL model
The nonlinear effects of the geological parameters on FPI are represented by the model described in
9.4.2.2. The empirical equation describing this relationship is as follows:
𝐹𝑃𝐼 = 8.408𝑈𝐶𝑆 0.509 + 7513.902𝑞 0.011 + 27.552𝑙𝑛(𝑆) + 0.080𝐽𝑐 − 7818.256 (9.51)
FNL explanatory capability
The measured data and the FPI obtained by means of equation 9.51 have been compared; the result
is shown in Figure 12.11.
FNL Model
300
250
200
150
100
50
0
0 50 100 150 200 250 300
Figure 12.11 Measured vs. calculated FPI, nonlinear model (FNL).
The ability of the nonlinear model to explain the geology-FPI relation is very similar to the one
observed for the linear regression model. It allows to assert that the nonlinear functions considered
during nonlinear multiple regressions were correctly selected.
317
Use of FNL model to predict FPI
Linear and nonlinear models are also characterised by the same difficulty to foresee FPI: as it is
possible to deduct by the scatterplot interpretation (Figure 12.12), it is not possible to observe any
correlations between measured and predicted FPI values.
FNL Model - prediction

200
FPI predicted [(kN/cutter)/(mm/rev)]
150
100
50
0
0 50 100 150 200
Figure 12.12 Measured vs. predicted FPI, nonlinear model (FNL).

The possible causes of this predictive inability are the same discussed for the linear model FL4.
318
12.3.3 Validation of FPI models carried out for each cluster
For each cluster regression analyses have been carried out to the purpose to explain the relationship
between the geological framework defining the clusters and the FPI measured during their
excavation. Firstly, the explanatory ability of the obtained equations has been tested, then the model
obtained for cluster I has been used to forecast FPI. Equation 10.18 has been selected to predict the
index because the cluster that best represents the Domain III (micaschist) is the cluster I.
Explanatory capability
Cluster I
The regression analyses carried out about data related to cluster I allow to obtain a linear equation
that is as follows:
𝐹𝑃𝐼 = 0.576 ∗ 𝑈𝐶𝑆 + 2.991 ∗ 𝑞 + 7.651 ∗ 𝑆 − 0.065 ∗ 𝑑 − 69.901 (10.18)
The comparison between observed and calculated data is provided by Figure 12.13.
C1e Model
200
150
100
50
0
0 50 100 150 200
Figure 12.13 Measured vs. calculated FPI, cluster I.
The correlation between measured and calculated FPI is appreciable (r = 0.703; R2adj. = 0.475): the
points are aligned close to the bisector line. The model C1e is able to represent the relationship
between UCS, q, S and d and FPI. When the ground conditions described in cluster I occur, it is
319
possible to use this model to predict the field penetration index; however, due to the high
homogeneity of the geologic parameters related to the cluster, the model is highly situ-specific and
ground feature variations make the equation inapplicable.
Cluster II
The equation that represents the relationship between the geological conditions related to the cluster
II and the FPI measured during its excavation is as follows:
𝐹𝑃𝐼 = 5.522 ∗ 𝑈𝐶𝑆 0.568 + 42.618 ∗ 𝑒𝑥𝑝0.069∗𝑆 (10.21)
FPI has been calculated by means of this equation, then the observed and the evaluated data have
been compared (Figure 12.14).
C2b Model
300
250
200
150
100
50
0
0 50 100 150 200 250 300
Figure 12.14 Comparison between measured and calculated FPI, nonlinear model cluster II.
The model is not able to represent the relationship between geological variables and the response
variable; it is probably due both to the exclusion of some geological variables and to the selection of
a wrong nonlinear mathematical form.
320
Cluster III
The empirical equation describing the relation between geological parameters and FPI is as follows:
𝐹𝑃𝐼 = 17.902 ∗ 𝑈𝐶𝑆 − 5.374 ∗ 𝛼 − 0.643 ∗ 𝑑 − 3144.566 (10.22)
The corrected coefficient of determination is equal to 0.984: it means that the regression represents
very well the relationship between UCS-α-depth and FPI. It is also demonstrated by the graphical
comparison between measured and calculated FPI (Figure 12.15).
C3f Model
300
250
200
150
100
50
0
0 50 100 150 200 250 300
Figure 12.15 Comparison between the measured and calculated FPI, cluster III.
It is necessary to consider that C3f model reliability is affected by the low number of considered
elements.
321
Use of C1e model to predict FPI
The model named C1e (Equation 10.18) was developed to explain the relationship between the
geological features of cluster I and the measured values of FPI. C1e was involved to predict the
field penetration index in the tunnel tract ranging between Pk. 2 + 0.10 m and Pk. 6 + 897 m; also in
this case only the sectors with maximum depth of 1000 meters have been considered.
The values predicted by means of C1e model and the values recorded during La Maddalena
excavation have been compered. The resulting scatterplot is reported in Figure 12.16.
C1e Model - prediction

150
100
50
0
0 50 100 150
Figure 12.16 Measured vs. predicted FPI values (C1e model).

The relationship between measured and predicted FPI values is quite appreciable, as it is possible to
observe in Figure 12.16. The correlation coefficient, r, is 0.461 and the determination coefficient is
equal to 0.212.
C1e predictive model is the most reliable one obtained in this study; it is able to reduce the
differences between the measured data and the ones calculated by means of the equation. For these
reasons, it is the most useful empirical predictive model to forecast FPI.
322
12.3.4 Validation of FPI model obtained by means of RES approach
The Rock Engineering System (Hudson, 1992) allowed to obtain a predictive model, as presented in
Chapter 11. The developed equation is as follows:
𝐹𝑃𝐼 = 8.422 ∗ 𝑒 0.043∗𝐹𝑃𝐼𝑅 (11.6)
Where FPIR is the FPI rated.
FPIR has been calculated in order to use the Equation 11.6 to predict FPI. The comparison between
measured and predicted results is shown in Figure 12.17.
RES Model
300
200
100
0
0 100 200 300
Figure 12.17 Comparison between predicted (by means of RES approach) and measured FPI.
The relationship between in situ measured FPI and predicted by means of RES approach is quite
strong (r = 0.706), as the graphical and statistical tests have confirmed.
The approach described in Chapter 11 can be used to create regression model able to estimate the
performances of a TBM.
323
12.4 Conclusion
12.4 Conclusion
Several regression analyses have been carried out with the purpose of describing the influence of
rock mass properties on TBM performances; as a result, empirical equations, in which machine
indices are the response variables and rock mass features are the independent ones, have been
obtained. These equations can be used both as a mathematical representation of the influence of
geological conditions on the machine performances and as predictive tools adopted to forecast the
TBM behavior during boring phases.
In this chapter the reliability of the obtained equations has been tested, both in terms of explanatory
and predictive capability.
The regression analyses conducted with a high level of detail are a reliable and useful tool to
explain the relationship between the geological-geotechnical properties of ground and the machine
performances observed during rock mass excavation. It was noted that the coefficients of
determination, calculated for each presented model, do not exceed the value of 0.70. It means that
they are able to explain the 70.0 % of the total variance, at maximum; the remaining part of
variance is represented by other factors which determine the performances of a TBM which usually
are not taken into account during model computing (machine specifications, cutterhead geometry,
cutter features and operational factors, as reported in Chapter 2).
When linear relationship between geological and machine parameters does not occur it is necessary
to consider nonlinear multiple regression models. It was noted that the best way to realise them is
considering the single nonlinear forms that correlate each geological parameter to the machine
index. In other words, single nonlinear regressions should be performed prior to the multiple ones in
order to select the best mathematical form able to link the independent and response variables, so
that the chosen ones have to be taken in consideration to construct the model. Nonlinear equations
performed during this study, by following this procedure, result able to explain the relation between
geological site conditions and TBM behaviour. Moreover the results of measuring-calculated
comparison are not very different from the ones obtained from the linear equations related to the
same dataset.
The geomechanical parameters collected in the tract of tunnel ranging between Pk. 2 + 010 m and
Pk. 6 + 897 m were input as independent variables into the models carry out on the basis of the data
collected in the first portion of tunnel (from Pk. 0 + 198 m to Pk. 2 + 010 m). Then, the calculated
TBM performances have been compared to the measured ones: in this way the predictive capability
of the previously described empirical formulations has been tested.
324
12.4 Conclusion
C1e model is the most reliable one, it was obtained on the basis of the cluster I data analysis and it
is able to predict FPI with appreciable accuracy, as reported in paragraph 12.3.3.
The individuated causes able to determine some difficulty in predicting TBM performances, can be
summarised as follows:
- Problems related to the empirical equations: the selected mathematical function could be
not suitable to represent the rock mass-machine interaction and/or the constant of the
equation is overestimated (or underestimated) resulting in over/under-estimated TBM
indices.
- Input data selection: model construction implies the selection of the most influencing
geological and geomechanical parameters. This selection is usually carried out on the basis
of the operator experience or, in order to be more objective, according to the statistical
significance of the parameters, determined by statistical significance tests. So that some
variables are included and some others excluded from the final model. However, it is
possible that a rock parameter is not taken into account, because in the rock mass portion
considered to develop the model it is irrelevant, but it may become more important and
seriously affect the machine behaviour in the context in which the model is used to forecast
the TBM performances. A hypothesised solution might be to consider also the not
significant parameters during model construction.
- Geology independent factors: there is a series of situations that not depend on geological
framework which can affect the performances of a TBM. For example changes in
operational and mechanical conditions or in tender contracts, which clearly cannot be
predicted by geological models.
- Machine limitations: the excavation of hard and abrasive rock masses requires high TBM
thrust, so high FPI in order to guarantee an appropriated penetration per each cutterhead
revolution. Nerveless, there is a force threshold above which TBM cannot be driven, in
relation to the rock hardness, because it is not possible to impart too high a force on very
hard rock faces to avoid machine damages and breakage. Predictive models do not consider
this limitation and evaluate FPI on the basis of geomechanical conditions, which can require
values higher than the permitted ones.
The models are developed in a specific site, characterised by specific geological and geomechanical
framework. When one or more than one variable ranges change the model is not perfectly able to
325
12.4 Conclusion
adapt itself to this variation. The situ-specificity of these models is the most important aspect that
makes them hardly applicable to different contexts from the ones in which they have been achieved.
The most situ-specific models are those developed for each cluster, given that the clustering has
divided the rock mass in very homogenous portions. If the geological framework of a rock mass
hosting a tunnel is very similar to the one characterising the cluster, the related model (the equation
carried out on the basis of cluster information) should give reliable results, as it was possible to
observe about C1e model.
A possible solution to the TBM performance prediction problems may be to create a large database
of rock mass clusters, considering several tunnels, with the concerning predictive equations
developed for each of them. Direct or indirect geological surveys, explorative tunnels and
geotechnical investigation performed prior tunnel construction, allow to recognise the cluster (or
clusters) with features similar to the ones observed in the investigated ground, so it will be possible
to consider the predictive equation performed for those conditions and apply it to predict the TBM
performances related to the planned tunnel.
326
13 CONCLUSION
The general purpose of this thesis was to perform empirical models able to predict the performances
of the TBM involved in tunnel excavation on the basis of the relationship between the geological
features of the rock mass hosting “La Maddalena” tunnel and the machine performances measured
during the excavation.
For this reason, a large database containing geological, geomechanical and structural information
collected during continuous detailed survey along the tunnel wall and also the TBM performances
registered was created. The database covers a total length of about 7000 meters.
The detailed direct surveys allowed to verify the reliability of the information obtained by means of
seismic investigation system, such as the Tunnel Seismic Prediction system. The reliability of this
system has been tested comparing the direct detected information and the indirect obtained
information. TSP was able to individuate lithological variations: it resulted useful to estimate the
fracturing degree and the orientation of principal discontinuity planes. It was noted that an
overestimation of the angles between tunnel axis and fractures, due to TSP principles of working,
occurs.
Single rock or rock mass parameters were correlated with single TBM index: it resulted of
fundamental importance to understand in which way a geological parameter affects machine
behaviour. The results of simple regression analyses have been considered in the further analyses;
in particular, it was noted that the most influencing parameters are: UCS, quartz content and
spacing, followed by joint conditions.
The information obtained by simple regressions had a primary role in the application of Rock
Engineering System to this case of study; in particular the level of correlation among variables has
been used to compile the rock interaction matrix. A predictive model was developed on the basis of
RES approach; it implied a rating of the geological parameters. It was noted that assigning ratings
on the basis of the common rock mass classification system is not suitable and the best way to
assign ratings is considering the statistical distribution of the parameters. This study allowed to
propose an approach applicable to different geological context in order to forecast the performances
of TBMs.
327
Several multiple regression analyses have been carried out with the aim of describing the combined
effects of geological features on machine performances: as a result empirical equations have been
obtained. These equations can be used both as a mathematical representation of the geology-
machine performances and as predictive tools adopted to forecast the TBM behavior during boring
phases.
The regression analyses are able to explain the relationship between the geological-geotechnical
properties of ground and the machine performances observed during rock mass excavation. It is
reasonable that the coefficients of determination of each presented model do not exceed the value of
0.70. It means that geological conditions explain the 70.0 % of the total variance, at maximum; the
remaining part is represented by other factors acting on the machine behaviour which generally are
not considering during model building.
Linear and nonlinear multiple regression models have been performed It was noted that the best
way to realise them is considering the simple nonlinear regressions previously carried out.
The predictive capability of the achieved empirical models has been tested on a tract of tunnel of
about 5000 meters comparing the measured and predicted TBM indices. The models present some
difficult in forecasting machine performances, some causes have been individuated.
Some problems of over/under-estimation are imputable to the constant value calculated for each
equation.
One of the most relevant difficulties consists in the selection of input data. After this study, it is
possible to advise to include a large number of independent variables even if they do not seem to
affect the machine response, because in the context in which the model is used to forecast the TBM
performances the excluded parameter may become more important and seriously affect the machine
behaviour. Moreover, there are a series of situations that not depend on geological framework
which can affect the performances of a TBM, such as changes in operational and mechanical
conditions or in tender contracts which clearly cannot be predicted by geological models.
Machine limitations seriously affect the machine response, in particular for FPI. In fact, there is a
force threshold above which TBM cannot be driven, in relation to the rock hardness, because it is
not possible to impart a too high force on a very hard rock tunnel face to avoid machine breakage.
Predictive models do not consider this limitation and evaluate field penetration index on the basis of
geomechanical conditions, which may require values higher than the allowed ones.
328
The models are performed in a specific site so when one or more than one geological variable
ranges change the model is not able to adapt itself to this variation. The situ-specificity of these
models makes them hardly applicable to different contexts from the ones in which they have been
obtained.
The most situ-specific models are those achieved for each cluster, given that they are very
homogenous portions of rock mass. The empirical equations developed for each of them are very
reliable, so a possible solution to overcome the TBM performance prediction problems may be to
build a large database of rock mass clusters with the concerning predictive equations developed for
each of them.
Geological and geotechnical data collection performed prior tunnel construction allows the experts
to ascribe the features observed to the cluster with similar characteristics and adopt the predictive
equation typical of the recognised cluster to predict the TBM performances related to the planned
tunnel.
329
14 REFERENCES
Abdi, H., & Valentin, D. (2007). Multiple correspondence analysis. Encyclopedia of measurement and
statistics, 651-657.
Abdi, H., & Williams, L. J. (2010). Principal component analysis. Wiley interdisciplinary reviews:
computational statistics, 2(4), 433-459.
Adoko, A. C., Gokceoglu, C., & Yagiz, S. (2017). Bayesian prediction of TBM penetration rate in rock
mass. Engineering Geology, 226, 245-256.
Alber, M. (1996). Prediction of penetration and utilization for hard rock TBMs. In ISRM International
Symposium-EUROCK 96. International Society for Rock Mechanics.
Alber, M. (1998). Design of high speed TBM-drives. Proc. 8th Int. IAEG Congress, 3537-3543. Rotterdam:
A. A. Balkema.
Alber, M. (2000). Advance rates of hard rock TBMs and their effects on project economics. Tunnelling and
underground space technology, 15(1), 55-64.
Alber, M. (2008). An integrated approach to penetration, advance rates and disc cutter wear for hard rock
TBM drives. Geomechanik und Tunnelbau: Geomechanik und Tunnelbau, 1(1), 29-37.
Alber, M., Yaralı, O., Dahl, F., Bruland, A., Käsling, H., Michalakopoulos, T. N. & Özarslan, A. (2013).
ISRM suggested method for determining the abrasivity of rock by the CERCHAR abrasivity test.
In The ISRM Suggested Methods for Rock Characterization, Testing and Monitoring: 2007-2014(pp.
101-106). Springer, Cham.
Grima, M. A., Bruines, P. A., & Verhoef, P. N. W. (2000). Modeling tunnel boring machine performance by
neuro-fuzzy methods. Tunnelling and underground space technology, 15(3), 259-269.
Argand E, Blösch E, Heim A, Heim A (1911) Les nappes de recouvrement des Alpes pennines et leurs
prolongements structuraux, 31-33, In Kommission bei A. Francke (vorm. Schmid & Francke)
Armaghani, D. J., Hajihassani, M., Mohamad, E. T., Marto, A., & Noorani, S. A. (2014). Blasting-induced
flyrock and ground vibration prediction through an expert artificial neural network based on particle
swarm optimization. Arabian Journal of Geosciences, 7(12), 5383-5396.
Armaghani, D. J., Mohamad, E. T., Momeni, E., & Narayanasamy, M. S. (2015). An adaptive neuro-fuzzy
inference system for predicting unconfined compressive strength and Young’s modulus: a study on
Main Range granite. Bulletin of engineering geology and the environment, 74(4), 1301-1319.
Armaghani, D. J., Mohamad, E. T., Narayanasamy, M. S., Narita, N., & Yagiz, S. (2017). Development of
hybrid intelligent models for predicting TBM penetration rate in hard rock condition. Tunnelling and
Underground Space Technology, 63, 29-43.
Armetti, G., Migliazza, M. R., Ferrari, F., Berti, A., & Padovese, P. (2018). Geological and mechanical rock
mass conditions for TBM performance prediction. The case of “La Maddalena” exploratory tunnel,
Chiomonte (Italy). Tunnelling and Underground Space Technology, 77, 115-126.
Armetti, G., Migliazza, M., Comina, C., Padovese., P., Berti, A. (2017) Direct geostructural survey and
seismic methods: comparison for predicting geological framework of rock mass of “La Maddalena"
331
tunnel - Chiomonte, Italy. International conference AFTES 2017. Assoc. Francaise des tunnels et de
l’espace souterrain.
Atashpaz Gargari, E., Hashemzadeh, F., Rajabioun, R., & Lucas, C. (2008). Colonial competitive algorithm:
a novel approach for PID controller design in MIMO distillation column process. International
Journal of Intelligent Computing and Cybernetics, 1(3), 337-355.
Atashpaz-Gargari, E., & Lucas, C. (2007). Imperialist competitive algorithm: an algorithm for optimization
inspired by imperialistic competition. In Evolutionary computation, 2007. CEC 2007. IEEE
Congress on (pp. 4661-4667). IEEE.
Balci, C. (2009). Correlation of rock cutting tests with field performance of a TBM in a highly fractured rock
formation: A case study in Kozyatagi-Kadikoy metro tunnel, Turkey. Tunnelling and Underground
Space Technology, 24(4), 423-435.
Bamford, W. F (1984). Rock test indices are being successfully correlated with tunnel boring machine
performance. In Proceedings of the 5th Australian Tunnelling Conference, Melbourne, 2:9-22.
Barla, G., & Pelizza, S. (2000). TBM tunnelling in difficult ground conditions. In ISRM International
Symposium. International Society for Rock Mechanics.
Barton N (2000). TBM tunnelling in jointed and faulted rock. CRC Press.
Barton N, Lien R, Lunde J (1974) Engineering classification of rock masses for the design of tunnel support.
Rock Mechanics 6 (4):189-239
Barton, N. (1978). Suggested methods for the quantitative description of discontinuities in rock
masses. ISRM, International Journal of Rock Mechanics and Mining Sciences & Geomechanics
Abstracts, 15(6).
Bashir, Z. A., & El-Hawary, M. E. (2009). Applying wavelets to short-term load forecasting using PSO-
based neural networks. IEEE transactions on power systems, 24(1), 20-27.
Benardos, A. G., & Kaliampakos, D. C. (2004). Modelling TBM performance with artificial neural
networks. Tunnelling and Underground Space Technology, 19(6), 597-605.
Benzecri, J. P. (1979). Sur le calcul des taux d’inertie dans l’analyse d’un questionnaire, addendum et
erratum à [BIN. MULT.]. Cahiers de l’Analyse des Données, 4(3), 377-378.
Bieniawski, Z. T. (1989). Engineering rock mass classifications: a complete manual for engineers and
geologists in mining, civil, and petroleum engineering. John Wiley & Sons.
Bieniawski, Z. T., & Grandori, R. (2007). Predicting TBM excavability-part II. Tunnels & Tunnelling
International.
Bieniawski, Z. T., Celada, B., & Galera, J. M. (2007, June). TBM excavability: prediction and machine-rock
interaction. In Proceedings, Rapid Excavation and Tunneling Conference (pp. 1118-1130).
Bieniawski, Z. T., Tamames, B. C., Fernandez, J. M. G., & Hernandez, M. A. (2006). Rock Mass
Excavability (RME) Indicator. In Proceedings of ITA-AITES World Tunnel Congress and 32nd ITA
General Assembly.
332
Birant, D., & Kut, A. (2007). St-dbscan: An algorithm for clustering spatial–temporal data. Data &
Knowledge Engineering, 60(1), 208-221.
Blindheim, O.T. (1979) Boreability predictions for tunneling. The Norwegian Institute of Technology,
Trondheim, Norway.
Bocquet, J. (1974). Il metamorfismo prealpino nella Vanoise (Savoia) e in altri settori dello zoccolo
brianzonese. Mem. Soc. Geol Ital, 13, 271-284.
Boniface, A. (2000). Tunnel Boring Machine performance in basalts of the Lesotho formation. Tunnelling
and Underground Space Technology, 15(1), 49-54.
Bordet, C., & Comes, G. (1975). Forecasting study of the drilling properties of rocks for tunnel construction.
Example of the main arc-isere gallery. Tunnels et ouvrages souterrains, (7).
Borghi, A., & Gattiglio, M. (1997). Osservazioni geologico-petrografiche nel settore meridionale del
massiccio d’Ambin. Atti Ticinensi Sci. Terra, 5, 65-84.
Broch E, Franklin JA (1972) The point-load strength test. International Journal of Rock Mechanics and
Mining Sciences & Geomechanics Abstracts, 9 (6):669-676.
Brownlee, J. (2011). Clever algorithms: nature-inspired programming recipes. Jason Brownlee.
Bruland A (1998). Hard rock tunnel boring. Doctoral thesis, Norwegian University of Science and
Technology, Trondheim, Norway.
Callegari, E., Sacchi, R., Bovo, S., & Torassa, G. (1980). Osservazioni strutturali sul versante italiano del
massiccio di Ambin (Alpi Graie). Bollettino della Societa Geologica Italiana, 99(4), 395-404.
Casati C, Giovacchini A (1977) L’utilizzo delle immagini Landsat per indagini di neotettonica. Boll. Geod.
Sc. Affini, 36:399-410
Cassinelli, F., Cina, S., & Innaurato, N. (1983). Power consumption and metal wear in tunnel-boring
machines: analysis of tunnel-boring operation in hard rock: In: Tunneling 82, Proceedings of the 3rd
International Symposium, P73–81. Publ London: IMM, 1982. In International Journal of Rock
Mechanics and Mining Sciences & Geomechanics Abstracts (Vol. 20, No. 1, p. A25). Pergamon.
Chiaia, B. (2001). Fracture mechanisms induced in a brittle material by a hard cutting indenter. International
Journal of Solids and structures, 38(44-45), 7747-7768.
Christopher, D. M., Prabhakar, R., & Hinrich, S. C. H. Ü. T. Z. E. (2008). Introduction to information

retrieval. An Introduction To Information Retrieval, 151, 177.
Cook, N. G. W., Hood, M., & Tsai, F. (1984). Observations of crack growth in hard rock loaded by an
indenter. In International Journal of Rock Mechanics and Mining Sciences & Geomechanics
Abstracts (Vol. 21, No. 2, pp. 97-107). Pergamon.
Crow SC (1975) Jet tunnelling machines. A guide for design. Tunnels & Tunnelling International, 7(2)
Deere, D. U., & Miller, R. P. (1966). Engineering classification and index properties for intact rock. Illinois
Univ At Urbana Dept Of Civil Engineering.
Delisio, A., & Zhao, J. (2014). A new model for TBM performance prediction in blocky rock
conditions. Tunnelling and Underground Space Technology, 43, 440-452.
333
Delisio, A., Zhao, J., & Einstein, H. H. (2013). Analysis and prediction of TBM performance in blocky rock
conditions at the Lötschberg Base Tunnel. Tunnelling and Underground Space Technology, 33, 131-
142.
Desmons J, Mercier D (1993) Passing through the Briançon zone. In Pre-Mesozoic Geology in the Alps,
Springer Berlin Heidelberg, pp. 279-295
Desmons, J. (1992). The Briançon basement (Pennine Western Alps): mineral composition and
polymetamorphic evolution. Schweizerische Mineralogische und Petrographische
Mitteilungen, 72(1), 37-55.
Desmons, J., & Fabre, J. (1988). Contribution à la connaissance pétrographique du Mont Pourri (Savoie,
France); conséquences structurales. Géologie Alpine, 64, 13-26.
Dickmann, T. (2012). Predicting rock conditions ahead of the face. TunnelTalk.
Dickmann, T. and Krueger, D. (2013). Is geological uncertainty ahead of the face controllable? World
Tunnel Congress 2013 Geneva, Underground-the way to the future! G. Anagnostou & H. Ehrbar
(eds), CRC Press.
Dickmann, T., Tang, S. K. (2006) Application and research of seismic investigation methods to predict rock
mass conditions ahead of the face.
Dollinger GL, Handewith HJ, Breeds CD (1998) Use of the punch test for estimating TBM performance.
Tunnelling and underground space technology, 13(4):403-408
Eberhart, R., & Kennedy, J. (1995, October). A new optimizer using particle swarm theory. In Micro
Machine and Human Science, 1995. MHS'95., Proceedings of the Sixth International Symposium
on (pp. 39-43). IEEE.
Ellenberger, F. (1958). Etude géologique du pays de Vanoise. Impr. nationale.
Farmer IW, Glossop NH (1980) Mechanics of disc cutter penetration. Tunnels and Tunnelling, 12(6):22-25
Farrokh, E., Rostami, J., & Laughton, C. (2012). Study of various models for estimation of penetration rate
of hard rock TBMs. Tunnelling and Underground Space Technology, 30, 110-123.
Feng, X., & Jimenez, R. (2015). Estimation of deformation modulus of rock masses based on Bayesian
model selection and Bayesian updating approach. Engineering Geology, 199, 19-27.
Franklin, J. A. (1985). Suggested method for determining point load strength. In International Journal of
Rock Mechanics and Mining Sciences & Geomechanics Abstracts (Vol. 22, No. 2, pp. 51-60).
Pergamon.
Fudral S (1998) Etude géologique de la suture téthysienne dans les Alpes franco-italiennes nord-occidentales
de la Doire Ripaire (Italie) à la région de Bourg-Saint-Maurice(France). Doctoral dissertation.
Gay, M., Mégnien, F., & Mégnien, C. (1970). Le massif d'Ambin et son cadre de schistes lustrés (Alpes
franco-italiennes): évolution palégraphiques antéalpine. BRGM.
Gehring, K., 1995. Leistungs-und Verschleissprognosen in maschinellen Tunnelbau. Felsbau 13 (6)
Ghasemi, E., Yagiz, S., & Ataei, M. (2014). Predicting penetration rate of hard rock tunnel boring machine
using fuzzy logic. Bulletin of Engineering Geology and the Environment, 73(1), 23-35.
334
Giardino M, Polino R (1997) Le deformazioni di versante dell'alta valle di Susa: risposta pellicolare
dell'evoluzione tettonica recente. Il Quaternario, 10 (2):293-298.
Goguel, J. (1958). Présence des conglomérats à la base du “groupe d’Ambin” dans les schistes cristallins du
massif d’Ambin (Savoie). CR somm. Soc. géol. Fr, 11, 229-231.
Gong Q, Zhao J (2009) Development of a rock mass characteristics model for TBM penetration rate
prediction. International journal of Rock mechanics and mining sciences, 46(1):8-18.
Gong, Q. M., Jiao, Y. Y., & Zhao, J. (2006). Numerical modelling of the effects of joint spacing on rock
fragmentation by TBM cutters. Tunnelling and Underground Space Technology, 21(1), 46-55.
Gordan, B., Armaghani, D. J., Hajihassani, M., & Monjezi, M. (2016). Prediction of seismic slope stability
through combination of particle swarm optimization and neural network. Engineering with
Computers, 32(1), 85-97.
Graham PC (1976) Rock exploration for machine manufacturers. In: Proceedings of the Symposium on
Exploration for Rock Engineering, Johannesburg, Balkema, 1:173-180.
Greenacre, M., & Blasius, J. (Eds.). (2006). Multiple correspondence analysis and related methods. CRC
press.
Guillot, F., & Raoult, J. F. (1984). Permien et base du Trias en Vanoise septentrionale: données nouvelles et
hypothèses (zone briançonnaise interne, Alpes françaises). Ann. Soc. géol. Nord, 104(4), 183-192.
Hamidi JK, Shahriar K, Rezai B, Rostami J (2010) Performance prediction of hard rock TBM using Rock
Mass Rating (RMR) system. Tunnelling and Underground Space Technology, 25(4):333-345.
Hamilton, W. H., & Dollinger, G. L. (1979). Optimizing tunnel boring machine and cutter design for greater
boreability. RETC Proceedings, Atlanta, 1, 280-296.
Hassanpour J, Rostami J, Khamehchiyan M, Bruland A (2009) Developing new equations for TBM
performance prediction in carbonate-argillaceous rocks: a case history of Nowsood water
conveyance tunnel. Geomechanics and Geoengineering; an International Journal, 4(4):287-297.
Hassanpour J, Rostami J, Zhao J (2011) A new hard rock TBM performance prediction model for project
planning. Tunnelling and underground space technology, 26(5):595-603.
Hassanpour, J., Vanani, A. G., Rostami, J., & Cheshomi, A. (2016). Evaluation of common TBM
performance prediction models based on field data from the second lot of Zagros water conveyance
tunnel (ZWCT2). Tunnelling and Underground Space Technology, 52, 147-156.
Hoek E (1994) Strength of rock and rock masses. ISRM News Journal, 2(2):4-16.
Hoek E, Marinos P, Benissi M (1998) Applicability of the Geological Strength Index (GSI) classification
forvery weak and sheared rock masses. The case of the Athens Schist Formation. Bulletin of
Engineering Geology and Environment, 57(2): 151-160
Hoek, E. (1988). The Hoek-Brown failure criterion-a 1988 update. In Proc. 15th Canadian Rock Mech.
Symposium (pp. 31-38). Toronto, Dept. Civil Engineering, University of Toronto.
Hoek, E., Carter, T.G., Diederichs, M.S. (2013). Quantification of the Geological Strength Index
Chart. 47th US Rock Mechanics/Geomechanics Symposium, San Francisco, CA, USA.
335
Hudson, J. (1992). Rock engineering systems. Theory and practice. Pergamon.
Hudson, J. A., & Priest, S. D. (1983). Discontinuity frequency in rock masses. In International Journal of
Rock Mechanics and Mining Sciences & Geomechanics Abstracts(Vol. 20, No. 2, pp. 73-89).
Pergamon.
Hughes HM (1986). The relative cuttability of coal-measures stone. Mining Science and Technology,
3(2):95-109
Husson, F., Josse, J., & Pages, J. (2010). Principal component methods-hierarchical clustering-partitional
clustering: why would we need to choose for visualizing data. Applied Mathematics Department.
Husson, F., Lê, S., & Pagès, J. (2017). Exploratory multivariate analysis by example using R. CRC press.
Innaurato N., Mancini A., Rondena E., Zaninetti A. (1991). Forecasting and effective TBM performances in
a rapid excavation of a tunnel in Italy. In Proceedings of seventh ISRM Congress, International
Society for Rock Mechanics
Innaurato, N., Oggeri, C., Oreste, P. P., & Vinai, R. (2007). Experimental and numerical studies on rock
breaking with TBM tools under high stress confinement. Rock Mechanics and Rock
Engineering, 40(5), 429.
ISRM (1985) Suggested method for determining point load strength. International journal of rock
mechanics,mining sciences & geomechanics; abstract Vol 22, No 2, pp 51-60.
Jang, J. S. R., Sun, C. T., & Mizutani, E. (1997). Neuro-fuzzy and soft computing; a computational approach
to learning and machine intelligence.
Johnson, J. B., & Omland, K. S. (2004). Model selection in ecology and evolution. Trends in ecology &
evolution, 19(2), 101-108.
Jolliffe, I. T. (2002). Principal component analysis for special types of data. Principal component analysis,
338-372
Klein, S., Schmoll, M., & Avery, T. (1995, June). TBM performance at four hard rock tunnels in California.
In Proceedings of the rapid excavation and tunneling conference (pp. 61-76). Society for mining,
metallurgy & exploration, Inc.
Koenker, R., & Portnoy, S. (1996). Quantile regression. University of Illinois at Urbana-Champaign,
College of Commerce and Business Administration. Office of Research Working Paper 97-100.
Koenker, R. (2017). Quantile regression: 40 years on. Annual Review of Economics, 9, 155-176.
Kretz, R. (1983). Symbols for rock-forming mineralsl. American mineralogist, 68, 277-279.
Latham, J. P., Van Meulen, J., & Dupray, S. (2006). Prediction of in-situ block size distributions with
reference to armourstone for breakwaters. Engineering Geology, 86(1), 18-36.
Laughton, E., & Eng, P. (2011). Value engineering the construction of bored tunnels in competent rock. Pre-
Print Society of American Value Engineers.
Liao, Z. Y., Liang, Z. Z., Yang, Y. F., Wang, Y., & Tang, C. A. (2013). Numerical simulation of
fragmentation process of jointed rock mass induced by a drill bit under dynamic loading. Chin J
Geotech Eng, 35(6), 1147-1155.
336
Liu, H. Y., Kou, S. Q., Lindqvist, P. A., & Tang, C. A. (2002). Numerical simulation of the rock
fragmentation process induced by indenters. International Journal of Rock Mechanics and Mining
Sciences, 39(4), 491-505.
Lunn, D., Jackson, C., Best, N., Spiegelhalter, D., & Thomas, A. (2012). The BUGS book: A practical
introduction to Bayesian analysis. Chapman and Hall/CRC.
Macias, F. J., Jakobsen, P. D., Seo, Y., & Bruland, A. (2014). Influence of rock mass fracturing on the net
penetration rates of hard rock TBMs. Tunnelling and Underground Space Technology, 44, 108-120.
Mahdevari, S., Shahriar, K., Yagiz, S., & Shirazi, M. A. (2014). A support vector regression model for
predicting tunnel boring machine penetration rates. International Journal of Rock Mechanics and
Mining Sciences, 72, 214-229.
Maidl, B., Schmid, L., Ritz, W., & Herrenknecht, M. (2008). Hardrock tunnel boring machines. John Wiley
& Sons.
Malusa, M. G. (2004). Post-metamorphic evolution of the Western Alps: kinematic constraints from a
multidisciplinary approach (geological mapping, mesostructural analysis, fission-track dating, fluid
inclusion analysis).
Marinos P, Hoek E (2000) GSI: a geologically friendly tool for rock mass strength estimation. In
Proceedings of ISRM International Symposium, International Society for Rock Mechanics.
Marinos, P., & Hoek, E. (2001). Estimating the geotechnical properties of heterogeneous rock masses such
as flysch. Bulletin of engineering geology and the environment, 60(2), 85-92.
Marto, A., Hajihassani, M., Jahed Armaghani, D., Tonnizam Mohamad, E., & Makhtar, A. M. (2014). A
novel approach for blast-induced flyrock prediction based on imperialist competitive algorithm and
artificial neural network. The Scientific World Journal, 2014.
Mazzoccola, D. F., & Hudson, J. A. (1996). A comprehensive method of rock mass characterization for
indicating natural slope instability. Quarterly Journal of Engineering Geology and
Hydrogeology, 29(1), 37-56.
McFeat-Smith I (1999) Mechanised tunnelling for Asia. In Workshop Manual, Organized by IMS Tunnel
Consultancy Ltd.
Momeni, E., Nazir, R., Armaghani, D. J., & Maizir, H. (2014). Prediction of pile bearing capacity using a
hybrid genetic algorithm-based ANN. Measurement, 57, 122-131.
Monie, P. (1990). Preservation of Hercynian 40Ar/39Ar ages through high-pressure low-temperature Alpine
metamorphism in the Western Alps. European Journal of Mineralogy, 343-361.
Montgomery, D. C., & Peck, E. A. (1982). Introduction to linear regression analysis, 1992.
Mooney, M. A., Grasmick, J., Kenneally, B., & Fang, Y. (2016). The role of slurry TBM parameters on
ground deformation: Field results and computational modelling. Tunnelling and Underground Space
Technology, 57, 257-264.
Nash, J. C., & Walker-Smith, M. (1987). Nonlinear parameter estimation. New York: Marcel Decker.
337
Nelson PP, Al-Jalil YA, Laughton C (1999) Improved strategies for TBM performance prediction and
project management. In Proceedings of the Rapid Excavation and Tunneling Conference, pp. 963-
980
Nelson PP, O’Rourke TD (1983) Tunnel boring machine performance in sedimentary rocks. Report to
Goldberg-Zoino Associates of New York, P.C., School of Civil and Environmental of Civil
Engineering. Cornell University, Ithaca, NY, pp. 438
Nelson, P. P., Ingraffea, A. R., & O'Rourke, T. D. (1985). TBM performance prediction using rock fracture
parameters. In International Journal of Rock Mechanics and Mining Sciences & Geomechanics
Abstracts, vol 22(3); 189-192.
Nielsen B, Ozdemir L.(1993) Hard rock tunnel boring prediction and field performance. In: Proceedings of
the RECT. Boston, Littleton, Society for Mining, Metallurgy, and Exploration, p. 833–52.
Ozdemir L, Miller RJ, Wang FD (1978) Mechanical Tunnel Boring Prediction and Machine Design. NSF
APR7307776-A03. Colorado School of Mines. Golden, Colorado, USA
Ozdemir, L. (1977). Development of theoretical equations for predicting tunnel boreability (Doctoral
dissertation, Colorado School of Mines).
Ozdemir, L., (1992). Mechanical excavation techniques in underground construction. Short course notebook,
vol 1. Istanbul Technical University, Istanbul, pp 1-49.
Padovese, P., Berti, A., Baldovin, E., De Paola, A., Morelli, G.L., Buraschi, L., Ascari, G., (2017).
Innovazioni delle tecnologie di scavo meccanizzato in ammassi rocciosi caratterizzati da roccia dura
con distacchi in calotta e alte coperture. In: Proceedings of MIR 2017 - 16°Ciclo di conferenze di
Meccanica ed Ingegneria delle Rocce ‘‘Innovazione nella progettazione, realizzazione e gestione
delle opere in sotterraneo’’. Celid, Torino, pp 33-53.
Palmström, A. (1982). The volumetric joint count-a useful and simple measure of the degree of rock
jointing. Proc. 4th Int. Cong. Int. Assoc. Eng. Geol, 5, 221-228.
Palmstrom, A. (1995). RMi-a rock mass characterization system for rock engineering purposes.
Phillips, H. R., & Bilgin, N. (1977). Correlation of rock properties with the measured performance of disc
cutters. In Proceedings of a Conference on Rock Engineering, University of Newcastle Upon Tyne,
UK (pp. 181-196).
Pognante, U. (1984). Eclogitic versus blueschist metamorphism in the internal Western Alps along the Susa
valley traverse. Sciences Géologiques, bulletins et mémoires, 37(1), 29-36.
Polino, R., Dela Pierre, F., Fioraso, G., Giardino, M., & Gattiglio, M. (2002). Foglio 132-152-153
“Bardonecchia” Carta Geologica d’Italia, scala 1: 50,000. Servizio Geologico d’Italia.
Price, N. J., & Cosgrove, J. W. (1990). Analysis of geological structures. Cambridge University Press.
Ramezanzadeh A, Rostami J, Tadic D (2008) Impact of rock mass characteristics on hard rock TBM
performance. 13th Australian tunneling conference, Melbourne, Australia, pp 213–222.
Ribacchi R, Lembo-Fazio A (2005) Influence of rock mass parameters on the performance of a TBM in a
Gneissic formation (Varzo tunnel). Rock Mech Rock Eng 38(2):105–127.
338
Rispoli, A., Ferrero, A. M., Cardu, M., Brino, L., & Farinetti, A. (2016). Hard rock TBM performance:
preliminary study based on an exploratory tunnel in the Alps. Rock mechanics and rock engineering:
from the past to the future. Taylor & Francis Group, London, 469-474.
Rispoli A. (2018). Hard rock TBM excavation: performance analysis and prediction. Thesis for the degree of
Doctor of Philosophy. University of Study of Turin.
Roby, J., Sandell, T., Kocab, J., & Lindbergh, L. (2008). The current state of disc cutter design and
development directions. In Proceeding of 2008 North American Tunneling Conference, SME C (Vol.
4, pp. 36-45).
Rostami J (1997) Development of a force estimation model for rock fragmentation with disc cutters through
theoretical modeling and physical measurement of crushed zone pressure. Doctoral dissertation,
Colorado School of Mines.
Rostami J, Ozdemir L (1993) A new model for performance prediction of hard rock TBMs. In Proceedings
of the Rapid Excavation and Tunneling Conference, pp. 793-793
Rostami, J. (2008). Hard rock TBM cutterhead modeling for design and performance
prediction. Geomechanik und Tunnelbau: Geomechanik und Tunnelbau, 1(1), 18-28.
Rostami, J. (2016). Performance prediction of hard rock Tunnel Boring Machines (TBMs) in difficult
ground. Tunnelling and Underground Space Technology, 57, 173-182.
Roxborough F.F., Phillips H.R. (1975). Rock excavation by disc cutter. International Journal of Rock
Mechanics and Mining Sciences & Geomechanics Abstracts, 12 (12): 361-366.
Salimi, A., Rostami, J., & Moormann, C. (2016). TBM performance prediction in Basalt and Pyroclastic
rocks of Deacn traps, A case study of Maroshi-Ruparel water supply tunnel. Rock Mechanics and
Rock Engineering: From the Past to the Future, Taylor & Francis Group, London, 975-980.
Sanio HP (1985). Prediction of the performance of disc cutters in anisotropic rock. International Journal of
Rock Mechanics and Mining Sciences & Geomechanics Abstracts, 22(3):153-161
Sapigni M, Berti M, Bethaz E, Busillo A, Cardone G (2002). TBM performance estimation using rock mass
classifications. International Journal of Rock Mechanics and Mining Sciences, 39(6):771-788
Sato, K., Gong, F., & Itakura, K. (1991). Prediction of disc cutter performance using a circular rock cutting
ring. In Proceedings 1st international mine mechanization and automation symposium.
Snowdon RA, Ryley MD, Temporal J (1982) A study of disc cutting in selected British rocks. International
Journal of Rock Mechanics and Mining Sciences & Geomechanics Abstracts, 19(3):107-121.
Spalla, M. I., Zanoni, D., Marotta, A. M., Rebay, G., Roda, M., Zucali, M., & Gosso, G. (2014). The
transition from Variscan collision to continental break-up in the Alps: insights from the comparison
between natural data and numerical model predictions. Geological Society, London, Special
Publications, 405, SP405. 11.
Spiegelhalter, D. J., Best, N. G., Carlin, B. P., & Van Der Linde, A. (2002). Bayesian measures of model
complexity and fit. Journal of the Royal Statistical Society: Series B (Statistical
Methodology), 64(4), 583-639.
339
Stampfli, G. M., & Marthaler, M. (1990). Divergent and convergent margins in the North-Western Alps
confrontation to actualistic models. Geodinamica Acta, 4(3), 159-184.
Staub R (1937) Gedanken zum bau den Westalpen Zwischen Bernina und Mittelmeer. Buchdruckerei Gebr.
Fretz.
Stevenson, G. W. (1999). Empirical estimates of TBM performance in hard rock. In Proceedings of the rapid
excavation and tunneling conference (pp. 993-1010).
Sue C, Delacou B, Champagnac JD, Allanic C, Tricart P, Burkhard M (2007). Extensional neotectonics
around the bend of the Western/Central Alps: an overview. International Journal of Earth Sciences
96(6):1101-1129.
Sun, J. S., Chen, M., Chen, B. G., Lu, W. B., & Zhou, C. B. (2011). Numerical simulation of influence
factors for rock fragmentation by TBM cutters. Rock and Soil Mechanics, 32(6), 1891-1897.
Sundaram NM, Rafek AG, Komoo I (1998).The influence of rock mass properties in the assessment of TBM
performance. In Proceedings of the 8th IAEG Congress, Vancouver, British Columbia, Canada, pp.
3553-3559.
Sundin NO, Wänstedt S (1994) A boreability model for TBM's. In Proceedings of first North American Rock
Mechanics Symposium, American Rock Mechanics Association.
Tarkoy PJ (1973). Predicting TBM penetration rates in selected rock types. In Proceedings of Ninth
Canadian Rock Mechanics Symposium, Montreal.
Tarkoy, P. J. (1975). Rock Hardness Index Properties and Geotechnical Parameters for Predicting Tunnel
Boring Machine Performance (Doctoral dissertation, University of Illinois at Urbana-Champaign).
Tarkoy, P. J., & Marconi, M. (1991). Difficult rock comminution and associated geological conditions.
In Proceedings of the 6th international symposium: Tunnelling (Vol. 91, pp. 195-207).
Thélin P, Sartori M, Lengeler R, Schaerer JP (1990). Eclogites of Paleozoic or early Alpine age in the
basement of the Penninic Siviez-Mischabel nappe, Wallis, Switzerland. Lithos, 25(1):71-88
Torabi, S. A., Sahebjamnia, N., Mansouri, S. A., & Bajestani, M. A. (2013). A particle swarm optimization
for a fuzzy multi-objective unrelated parallel machines scheduling problem. Applied Soft
Computing, 13(12), 4750-4762.
Tuncdemir, H., Bilgin, N., Copur, H., & Balci, C. (2008). Control of rock cutting efficiency by muck
size. International Journal of Rock Mechanics and Mining Sciences, 2(45), 278-288.
Valantin A. (1973). Examen des diffe´rents proce´de´s classiques de´termination de la nocivite´ des roches
vis-a`-vis de l’abattage me´canique. Expose Presente´ Aux Journe´es d’information Techniques de
Creusement, Luxemburg, pp 133–140.
Wang, L., Kang, Y., Zhao, X., & Zhang, Q. (2015). Disc cutter wear prediction for a hard rock TBM
cutterhead based on energy analysis. Tunnelling and Underground Space Technology, 50, 324-333.
Wickham GE, Tiedemann HR, Skinner EH (1972). Support determinations based on geologic predictions. In
Proceedings of North America Rapid Excavation & Tunnelling, 1.
Wijk G (1992) A model of tunnel boring machine performance. Geotechnical and Geological Engineering,
10(1):19-40.
340
Wyllie, D. C. (1999) Foundations on Rock, 2nd Edition, Taylor and Francis, London, UK, 401.
Wyllie, D. C., & Mah, C. (2014). Rock slope engineering. CRC Press.
Xia, Y., Zhang, K., & Liu, J. (2015). Design Optimization of TBM Disc Cutters for Different Geological
Conditions.
Yagiz S (2002) Development of rock fracture and brittleness indices to quantify the effects of rock mass
features and toughness in the CSM Model basic penetration for hard rock tunneling machines.
Doctoral dissertation, Colorado School of Mines.
Yagiz, S (2006). TBM performance prediction based on rock properties. Proceedings of Multiphysics
Coupling and Long Term Behaviour in Rock Mechanics, EUROCK, 6, 663-670.
Yagiz S (2008). Utilizing rock mass properties for predicting TBM performance in hard rock condition.
Tunnelling and Underground Space Technology, 23(3):326-339
Yagiz, S., & Karahan, H. (2011). Prediction of hard rock TBM penetration rate using particle swarm
optimization. International Journal of Rock Mechanics and Mining Sciences, 48(3), 427-433.
Yagiz, S., Gokceoglu, C., Sezer, E., & Iplikci, S. (2009). Application of two non-linear prediction tools to
the estimation of tunnel boring machine performance. Engineering Applications of Artificial
Intelligence, 22(4-5), 808-814.
Yang, X. S., & Deb, S. (2010). Engineering optimisation by cuckoo search. arXiv preprint arXiv:1005.2908.
Yavari, M., & Mahdavi, S. (2005). Prediction of penetration rate of TBM using ANN. In National mining
conference (pp. 1-3).
Yin, L., Miao, C., He, G., Dai, F., & Gong, Q. (2016). Study on the influence of joint spacing on rock
fragmentation under TBM cutter by linear cutting test. Tunnelling and Underground Space
Technology, 57, 137-144.
Zhang, L., & Einstein, H. H. (2000). Estimating the intensity of rock discontinuities. International Journal of
Rock Mechanics and Mining Sciences, 37(5), 819-837.
Zhao Z, Gong Q, Zhang Y, Zhao J (2007) Prediction model of tunnel boring machine performance by
ensemble neural networks. Geomechanics and Geoengineering: An International Journal, 2(2):123-
128
Zou, D. (2017). Mechanical Underground Excavation in Rock. In Theory and Technology of Rock
Excavation for Civil Engineering (pp. 435-472). Springer, Singapore.
Zurada, J. M. (1992). Introduction to artificial neural systems(Vol. 8). St. Paul: West publishing company.
341
www. googlemaps
www.europeancommission.com
www.herrenknecht.com
www.igm.org.my
www.ltf
www.palmierigroup.com
www.R-project.org
www.robbinscompany
www.telt-sas.com
www.therobbinscompany.com
www.tunneltalk.com
342
APPENDIX I
Schemes used during geological surveys

GENERAL INFORMATION
Date:………………… Survey N°:…………………
Pk. Initial: …………m; Pk. Final:………….……..m
Tunnel axis direction:……… Tunnel depth:……………m
GEOLOGICAL DESCRIPTION OF INTACT ROCK
(global aspect, colour, forming minerals, texture, weathering, lithologies, evidences of instability
etc)
…………………………………………………………………………………………………………
…………………………………………………………………………………………………………
…………………………………………………………………………………………………………
…………………………………………………………………………………………………………
GRAPHICAL REPRESENTATION OF ROOF AND WALL STRUCTURES
Sinistral wall 
Dextral wall 
Roof
DISCONTINUITIES DESCRIPTION
Types of discontinuities: joints, faults, veins, ….. Number of joint sets: …………...
Schistosity: dip/direction …………; dip ………..
HORIZONTAL INTERCEPT [cm]
VERTICAL INTERCEPT [cm]
ORIENTATION OF DISCONTINUITIES
Dip Direction Dip Dip Direction Dip Dip Direction Dip

SCHEME FOR PRINCIPAL SET DESCRIPTION
ID set: ………………………. Type of discontinuity: …………………..
Estimation of mean orientation:……….Dip Direction;……….Dip. ...…………. β: ……………...…
SPACING [cm]
PERSISTENCE [m]
<1 □ 1–3□ 3 – 10 □ 10 – 20□ >20 □
APERTURE [mm]
< 0.5 □ 0.5 – 10.0 □ > 10.5 □
FILLING [mm]
Absent □ Hard, width < 5 □ Hard, width > 5 □ soft width < 5 □ soft width > 5 □
ROUGHNESS
Very rough □ Rough □ Slightly rough □ Smooth □ Slickenside □
WEATHERING DEGREE
Unweathered □ Slightly w. □ Moderately w. □ Highly w. □ Decomposed □
GSI:………………….
OBSERVATIONS:
…………………………………………………………………………………………………………
…………………………………………………………………………………………………………
APPENDIX II
Histograms of frequency of the principal geological and geotechnical variables

APPENDIX III
Longitudinal geological section of “La Maddalena” tunnel

LONGITUDINAL GEOLOGICAL SECTION OF “LA MADDALENA” TUNNEL
N315
1500 m
925 m
access
1000 m
1000 m
LEGEND
Micaschist – Clarea complex Minuti Gneiss
Gneiss Micaschist alternation – Ambin complex Drill and Blast Tunnelling
Aplitic Gneiss – Ambin complex Schistosity
Albitic Gneiss Principal fracture Tunnel

APPENDIX IV
The generic underground excavation matrix; Hudson, 1992

PHD Unimi R11266

Uploaded by

Copyright:

Available Formats

PHD Unimi R11266

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

PHD Unimi R11266

Uploaded by

Copyright:

Available Formats

UNIVERSITÀ DEGLI STUDI DI MILANO

Dottorato di Ricerca in Scienze della Terra

Geomechanical characterization of rock masses

Tutor Academic Year Coordinator

1.1 General purpose 1

1.2 Work organisation and thesis structure 2

2.2 Types of Tunnel Boring Machines 6

2.2.1 Open type TBM 7

2.2.2 Shield type TBM 9

2.2.3 Closed mode TBM 11

2.3 Rock breakage mechanism 12

2.3.1 Disc cutters 12

2.3.2 Rock failure 13

2.4 TBM performance indices 16

2.4.1 Rate of penetration 16

2.4.2 Advance rate 16

2.4.3 Coefficient of utilisation 17

2.4.4 Penetration rate per revolution 17

2.4.5 Field penetration index 17

2.4.6 Specific penetration 18

2.4.7 Specific excavation rate 18

2.5 Factors influencing the TBM performances 19

2.5.2 Mechanical factors 20

2.5.3 Operational factors 21

3 LA MADDALENA EXPLORATORY TUNNEL PROJECT 23

3.1.1 Mont Cenis Base Tunnel 25

3.2 La Maddalena exploratory tunnel 28

3.2.1 General features and purpose 28

3.2.2 Geographical location 28

3.2.3 Tunnel route 29

3.2.4 Encountered rock mass lithologies 30

3.2.5 Rock supports and reinforcements 31

3.2.6 The Tunnel Boring Machine “Gea” 34

4.2 Penninic domain 40

4.2.1 Grand Saint Bernard nappe 40

4.3 Tectono-stratigraphic unit of Ambin 42

4.3.1 Clarea complex 44

4.3.2 Ambin complex 46

4.4 Structural evolution of Ambin 48

4.4.1 Pre-Alpine evolution 48

4.4.2 Ductile Alpine evolution 48

5 ROCK MASS CHARACTERIZATION 55

5.2 Intact rock property determination 57

5.2.1 Lithological description 57

5.2.2 Micro-petrographic description 58

5.2.3 Uniaxial compressive strength 61

5.2.4 Determination of rock abrasivity 69

5.2.5 Determination of rock hardness 73

5.3 Quantitative description of discontinuities 76

5.3.1 Types of discontinuity 80

5.3.3 Spacing between discontinuities 89

5.3.4 Volumetric joint count 94

5.3.5 Rock Quality Designation index, RQD 96

5.3.6 Rock block volume 98

5.3.7 Groundwater conditions 100

5.3.8 Joint conditions 101

5.4 Rock mass classifications 110

5.4.1 Rock Mass Rating (RMR) system 110

5.4.2 Geological Strength Index (GSI) 117