Tunnelling and Underground Space Technology 24 (2009) 689–705

Contents lists available at ScienceDirect

Tunnelling and Underground Space Technology

journal homepage: www.elsevier.com/locate/tust

Ground reaction curves for tunnels excavated in different quality rock masses
showing several types of post-failure behaviour
L.R. Alejano a,1, A. Rodriguez-Dono a,*, E. Alonso a, G. Fdez.-Manín b
a
Department of Natural Resources and Environmental Engineering, ETSI Minas. Campus Lagoas-Marcosende s/n, University of Vigo, 36280 Vigo, Spain
b
Applied Maths II Department, Campus Lagoas-Marcosende s/n, University of Vigo, 36280 Vigo, Spain

a r t i c l e i n f o a b s t r a c t

Article history: Rock mass behaviour model selection and, in particular, rock mass post-failure behaviour are key issues
Received 13 January 2009 in analysing tunnel stability, in particular in terms of the correct application of design techniques such as
Received in revised form 22 June 2009 the convergence–confinement method and also numerical modelling. Three different quality rock masses
Accepted 10 July 2009
(good, average and poor) were defined in which simulated standard tunnels were excavated. Different
Available online 8 August 2009
behaviours – including elastic perfectly plastic, elastic brittle and three strain softening behaviours –
were modelled for each type of rock mass and increasingly realistic parameters were calculated, along
Keywords:
with the corresponding ground reaction curves. The results obtained demonstrate the importance of ade-
Ground reactions curves
Post-failure behaviour
quate post-failure behaviour model selection for tunnel analysis. Also assessed are the effects of the stan-
Strain softening dard support and reinforcement.
Ó 2009 Elsevier Ltd. All rights reserved.

1. Introduction The LDP has been studied by various authors including Panet
(1995), Chern et al. (1998) and Hoek et al. (2008). Hoek et al.
Most tunnel designs are currently based on empirical methods, (2008) described a technique to obtain the LDP for different quality
but some also rely on analytical techniques such as the conver- rock masses that we used in this work in order to apply the CCM.
gence confinement method (CCM) and on 2D or 3D numerical The method for obtaining the SCC was initially proposed for differ-
models. Application of the analytical techniques requires a detailed ent types of support and reinforcement by Hoek and Brown (1980)
knowledge of rock mass behaviour in order to obtain realistic then further discussed by other authors (Hoek, 1999; Carranza-
results. Torres and Fairhurst, 2000; Oreste, 2003a,b, 2008). Finally, our
The CCM is a 2D simplified approach for resolving 3D rock-sup- main concern in this article is GRC (Fig. 1); evaluated on the basis
port interaction problems associated with the installation of sup- of rock mass behaviour, it describes the relationship between the
port near a tunnel face in underground excavations in rock. It decreasing inner pressure and increasing radial displacement of
was developed in the 1930s (Fenner, 1938), further refined by the tunnel wall.
other authors (Pacher, 1964; Salençon, 1969; Lombardi, 1975; Elastic perfectly plastic models are often used to calculate the
Hoek and Brown, 1980; Brown et al., 1983; Bouvard-Lecoanet GRC in practice (Carranza-Torres and Fairhurst, 2000; RocScience,
et al., 1988; AFTES, 1993; Brady and Brown, 1993; Panet, 1993, 2003). If failure occurs, however, these simple models do not ade-
1995; Peila and Oreste, 1995; Hoek et al., 1995) and then compre- quately represent the real stress–strain behaviour of rock masses
hensively reviewed by Carranza-Torres and Fairhurst (2000). unless the rock mass is of poor quality. For all other kinds of rock
The CCM consists of three basic components in the form of masses, strain softening (S–S) or elastic brittle models are more
graphs: the longitudinal deformation profile (LDP), which relates suitable for simulating ground behaviour correctly.
tunnel deformation to distance to the tunnel face; the support In regard to possible post-failure behaviour modes, Hoek and
characteristic curve (SCC), which represents the stress–strain rela- Brown (1997) were among the first authors to clearly reject the
tionship in the support system; and the ground reaction or re- elastic perfectly plastic assumption as inappropriate for rock
sponse curve (GRC). masses of average or high geotechnical quality (that is, with a geo-
logical strength index (GSI) of over 30). In other words – and as
illustrated in Fig. 2 – the assumption no longer holds true once
* Corresponding author. Tel.: +34 609734704; fax: +34 986812201. the maximum strength is achieved and further strain is likely. On
E-mail addresses: [email protected] (L.R. Alejano), [email protected], alfonsor-
the basis of this argument, the approach described by Carranza-
[email protected] (A. Rodriguez-Dono), [email protected] (E. Alonso), manin@d-
ma.uvigo.es (G. Fdez.-Manín). Torres and Fairhurst (2000) should not be used for rock masses
1
Tel.: +34 986812374. with a GSI > 30, for which strain softening behaviour models are

0886-7798/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.tust.2009.07.004
690 L.R. Alejano et al. / Tunnelling and Underground Space Technology 24 (2009) 689–705

Fig. 1. Main elements of the convergence–confinement method (CCM): the longitudinal deformation profile (LDP), the ground reaction curve (GRC) and the support
characteristic curve (SCC). ** Based on Carranza-Torres and Fairhurst (2000).

criterion along with other post-failure parameters (which we will

describe below) need to be estimated. Estimating the residual fail-
ure criterion is no easy task, and a number of researchers have re-
cently taken on the challenge of deepening our knowledge of rock
mass stress–strain behaviour (Crowder and Bawden, 2006). Cai
et al. (2007) have recently proposed extending the GSI system for
estimating the residual strength of a rock mass so as to adjust
the peak GSI value with respect to the residual GSIr value based
on the two major controlling factors in the GSI system (rock struc-
ture and block surface condition, both discussed below). The pro-
posed method has been validated using in situ block shear test
data from cavern construction sites.
Even after peak and residual failure criteria have been defined,
the behaviour of the rock still remains incompletely known. The
Young’s elastic modulus and Poisson’s ratio, together with the drop
modulus (usually confinement stress dependent) and the dilatancy
angle, all need to be determined to fully represent complete stress–
strain behaviour with a view to understanding and modelling exca-
Fig. 2. Different post-failure rock mass behaviour modes for rock masses with vation behaviour.
different geological strength indices (GSI). ** Based on Hoek and Brown (1997).
In recent years, the authors of this paper have worked on the
development of techniques to both calculate GRCs for tunnels
excavated in rock masses of different quality and to estimate
more suitable. Other authors (Kaiser et al., 2000; Diederichs, 2003) post-failure dilatant behaviour in rock masses (Carranza-Torres
have also indicated that a type of elastic brittle behaviour can be et al., 2002; Alonso et al., 2003; Alejano and Alonso, 2005).
anticipated for very high quality rock masses (GSI > 75), and this This study focuses on tunnels excavated in three different rock
behaviour would not fit the Hoek–Brown failure criterion masses: a poor quality rock mass (GSI = 21.4), for which elastic per-
approach. fectly plastic behaviour was anticipated; a good quality rock mass
Strain softening and elastic brittle models are difficult to char- (GSI = 64.9) for which brittle behaviour was expected (after Hoek
acterise, since both a residual failure criterion and a peak failure and Brown, 1997); and an average quality rock mass (GSI = 55).
 L.R. Alejano et al. / Tunnelling and Underground Space Technology 24 (2009) 689–705 691

Once the relevant parameters were estimated and the different governed by the softening parameter g. The transition is defined
models were built, the GRCs were calculated (implemented in a in such a way that an elastic regime exists as long as the softening
MATLAB code) using integration techniques for strain softening parameter is null, a softening regime occurs whenever 0 < g < g*,
behaviour as described by Alonso et al. (2003). and a residual state takes place when g > g*, with g* defined as
the value of the softening parameter controlling the transition be-
2. Rock mass behaviour models tween the softening and residual stages. Fig. 3 illustrates this type
of strength-weakening behaviour in terms of strength and strain
Hoek and Brown (1997), who suggested that rock mass post- for a confined compressive test. The slope of the softening stage –
failure behaviour varies according to rock mass quality, provided the drop modulus – is denoted by M. If this drop modulus tends
particularly relevant post-peak strength guidelines. Note, however, to infinity, behaviour is perfectly brittle, and if it tends to zero,
that their choice of residual parameters is not recommended for behaviour is perfectly plastic. Therefore, the perfectly brittle, elastic
use by engineers, given that a great deal of job-specific judgement brittle plastic and perfectly plastic behaviour models are limiting
is required. These guidelines are based on rock types: for very good cases of this strain softening model, which we think that should
quality hard rock masses, with a high GSI value (70 < GSI < 90), the be considered the most general case.
rock mass behaves in an elastic brittle manner; for heavily jointed The constitutive equation for a strain softening material can be
rock (50 < GSI < 65), moderate stress levels result in a failure of obtained in accordance with the incremental theory of plasticity.
joint systems and the rock becoming gravely; for averagely jointed The plastic strain increments can be obtained from the plastic po-
rock (40 < GSI < 50), strain softening is assumed; and for very weak tential [g(rr, rh, g)], as in the following equation:
rock (GSI < 30), elastic perfectly plastic behaviour and no dilation  @g @g
are assumed (in other words, the failure criterion is already at epr ¼ k  and eph  ¼ k  ð2Þ
@ rr @ rh
the residual stage). These notions are illustrated in Fig. 2. Strain
softening behaviour can accommodate purely brittle behaviour where k  is a plastic multiplier and unknown. Eq. (2) is the consti-
(strain softening with a drop modulus equal to infinity) and elastic tutive equation of the plastic regime and is usually called the flow
perfectly plastic behaviour (strain softening with a drop modulus rule. If the plastic potential coincides with the failure criterion, then
equal to null), so brittle and elastic perfectly plastic behaviour the rule is an associated flow rule; otherwise it is called a non-asso-
are particular cases of the strain softening behaviour. ciated flow rule.
Incremental plasticity requires consideration of a fictitious time
2.1. Strain softening variable, called s, which controls the plastic strain increments as in
the following equation:
Strain softening behaviour – also referred to as strength-weak- @ epr @ eph
ening behaviour – is grounded in the incremental theory of plastic- epr  ¼ and eph  ¼ ð3Þ
@s @s
ity (Hill, 1950; Kaliszky, 1989) that was developed to model the
process of plastic deformation. According to this theory, a material
is characterised by a failure criterion f and a plastic potential g. One 2.2. Mohr–Coulomb strain softening
of the main features of the strain softening behaviour model is that
the failure criterion and the plastic potential depend not only on For the sake of simplicity the Mohr–Coulomb rather than the
the stress tensor rij, but also on what is called the plastic or soften- Hoek–Brown failure criterion was selected for our study.
ing parameter g. The behaviour model is thus plastic strain depen- Consider the Mohr–Coulomb yield criterion:
qffiffiffiffiffiffiffiffiffiffiffiffiffi
dent. The failure criterion is defined as:
f ðrh ; rr ; gÞ ¼ rh  K / ðgÞrr  2CðgÞ K / ðgÞ ð4Þ
f ðrr ; rh ; gÞ ¼ 0 ð1Þ
a plastic potential in the form of:
Strain softening behaviour is characterised by a gradual transition
from a peak failure criterion to a residual failure criterion that is gðrh ; rr ; gÞ ¼ rh  K w rr ð5Þ

Fig. 3. Peak and residual stress envelopes and stress–strain curve for a confined test performed on a strain softening material sample.
692 L.R. Alejano et al. / Tunnelling and Underground Space Technology 24 (2009) 689–705

2.5. Dilatancy

Correctly estimating dilatancy is of paramount importance in

order to resolve certain post-failure rock mechanics issues, such
as the problem of modelling underground excavations. According
to Cai et al. (2007), constant dilation is an approximation that is
clearly not physically correct, although for practical rock engineer-
ing purposes, this assumption is adopted largely because little is
known about how the dilation of a rock mass changes past peak.
A comprehensive review of the literature and observations in
regard to published test results indicate that dilatancy is highly
Fig. 4. Cohesion and friction angle functions for the plastic parameter (g).
dependent both on the plasticity already experienced by the mate-
rial and confining stress. In a previous work (Alejano and Alonso,
2005), we made a detailed analysis of published test data with a
where Kw is the dilation coefficient or dilatancy relationship given view to proposing a sufficiently significant but conveniently simple
by: formulation for the dilatancy angle that reflects these dependen-
1 þ sin w cies and that can be readily implemented in numerical codes and
Kw ¼ ð6Þ analytical formulations.
1  sin w
In Alejano and Alonso (2005), peak dilatancy and friction angle
and piecewise linear plastic parameter functions for cohesion c(g) values recovered from a high quality series of tests were compared
and a friction angle /(g), with /peak and cpeak as the peak parameters to confinement stress in order to demonstrate that peak dilatancy
and /res and cres as the residual parameters (Fig. 4). is highly dependent on confinement stress. Peak friction angles
The elastic regime is characterised by the shear modulus G and were estimated from the slope of the corresponding Hoek–Brown
Poisson’s ratio m. failure criterion (Hoek and Brown, 1997) for the corresponding
The plastic parameter is the plastic shear strain: stress level and sample size. Peak dilatancy and friction angles cor-
relate well for low stress levels, whereas they tend to diverge for
g ¼ ep1  ep3 ¼ eph  epr ¼ cp ð7Þ
higher stress levels, with the difference becoming greater as con-
The value of g* in Fig. 4 is correlated with the drop modulus shown finement stress grows. It turns out, then, that although the
in Fig. 3. assumption wpeak = / is not erroneous for peak dilatancy at very
low stress levels, it is clearly imprecise for higher stress levels,
2.3. Peak and residual failure criteria since rock failure usually occurs in shear bands or in new disconti-
nuities – particularly in soft and medium rock.
The peak failure criterion is estimated, following the Hoek– The following equation was proposed:
Brown approach (Hoek and Brown, 1997; Hoek et al., 2002; Roc- / rci
Science, 2002), from laboratory data – for the values for rci and wpeak ¼ log10 ð8Þ
1 þ log10 rci r3 þ 0:1
m for the intact rock – and from the GSI calculated from field data.
where / (°) refers to the peak friction angle which can, moreover, be
Values for the pre-failure elastic parameters, namely Young’s elas-
calculated as the slope of the Hoek–Brown failure criterion, rci is the
tic modulus and Poisson’s ratio, are also calculated.
unconfined compressive strength of the intact rock, and r3 is the
Following Cai et al. (2004), the GSI value can be considered as
confinement stress. This equation includes scale correction via /.
depending on the description of two factors: rock structure (which
In studying decay in the dilatancy angle in line with plasticity,
can be estimated from block volume or size) and block surface con-
the first option is to assign an exponential decay function to Kw
dition (estimated from what is called the joint condition factor, JC).
(the dilatancy relationship). The decay goes from a previously esti-
JC can be considered in terms of JW (large-scale waviness), JS (small-
mated peak value to a null value corresponding to no plastic vol-
scale smoothness) or JA (joint alteration factors). The main advan-
ume increase. This null value is proposed in the light of the fact
tage of this method is that, in its extension (Cai et al., 2007), it per-
that a rock cannot dilate infinitely. Thus:
mits the residual strength envelope to be obtained for broken or
cp
plastified material from a residual value of GSI. Even if the residual 
K w ¼ 1 þ ðK w;peak  1Þe cp; ð9Þ
value of GSI is not easy to estimate, a poor estimate is preferable to
no estimate at all. where the parameter cp,*, or plasticity parameter constant, is recal-
culated for each type of rock. This formulation, which was chosen
2.4. Post-failure deformation for its simplicity, is based on Detournay (1986) but also on a study
of dilatancy decay in line with the plasticity parameter in some
The stress–strain curve of a rock or a rock mass is not fully de- reinterpreted cases.
fined only by peak and residual strength envelope values, the A series of values of cp,* was estimated for different rocks, com-
Young’s elastic modulus and Poisson’s ratio. Also necessary are mencing with various tests performed by other authors (Medhurst,
the drop modulus and the dilatancy features, which have been re- 1996; Farmer, 1983). A formula for estimating the dilatancy angle
ported to be confinement stress dependent on rock samples and for rock and rock masses was proposed in accordance with Eqs. (8)
rock masses, thereby rendering rock behaviour even more and (9). The model included confining stress dependencies and the
complex. plasticity parameter.
The post-failure drop modulus is estimated according to rock The model also includes an indirect scale effect, incorporated by
mass quality: higher for good quality rock masses and lower for means of the inclusion of / in the model. This approach has the
poor quality rock masses. As far as confining stress dependencies advantage that dilatancy is estimated by just a single parameter
are concerned, a simple formulation can be adopted: there is an in- cp,*, yet still takes scale effects and confinement stress dependency
versely proportional relationship between values for the drop into account. Modellers, therefore, need to include in their inputs a
modulus and for confinement stress (as observed in rock samples reasonable approximation of this parameter rather than a more or
in the laboratory). less unrealistic constant value for the dilatancy angle.
 L.R. Alejano et al. / Tunnelling and Underground Space Technology 24 (2009) 689–705 693

3. Rock masses and tunnel features 3.2. Good quality rock mass

Described in this section are the main features of the rock This example of a 7 m radius tunnel, excavated to a depth of
masses where a 7 m radius tunnel is to be excavated in different 1000 m in a good quality rock mass (GSI = 64.9) composed of con-
quality rock masses (poor, good and average) and to different glomerate, sandstone and mudstone, corresponds to the CG2 do-
depths. Different post-failure behaviour models are assigned to main at the Kannagawa site (Cai et al., 2007). Laboratory testing
each kind of rock mass and all the corresponding properties are de- indicated values for rci = 162 MPa and for mi = 19.
fined. For the sake of simplicity in the application of the CCM, an For this good quality rock mass, with Vb = 303,000 cm3, JW = 1.5,
isotropic stress field is assumed for all cases. The poor and good JS = 1.5, JA = 1, JC = 2.25, a GSI value of 64.9 was obtained and the
quality rock masses are taken from literature (one of them is a real peak Mohr–Coulomb strength parameters were cpeak = 3.7 MPa
example) and the average quality rock mass was a real rock mass and /peak = 57.8°. According to the proposed method, Vbr = 10 cm3
studied by the authors. and Jcr = 0.75 for the residual rock mass. The residual GSI was
27.8, and the corresponding residual Mohr–Coulomb strength
3.1. Poor quality rock mass parameters were cres = 0.96 MPa and /res = 51°.
The peak and residual strength parameters estimated from the
Our example of a 7 m radius tunnel excavated to a depth of GSI system given by Cai et al. (2007) are shown in Table 2. Note
400 m in a poor quality rock mass was taken from Cai et al. that due to the good quality of the rock mass there were large dif-
(2007) from among examples illustrating rock mass residual ference between the peak and residual strength parameters
strength parameters for typical rock masses. Peak and residual (around 80% in cohesion and over 15% in friction).
strength parameters were calculated based on rci = 100 MPa and For this hard rock mass, an excellent quality controlled excava-
mi = 20. tion procedure that results in minimal disturbance to the confined
As mentioned earlier, the GSI value can be considered (Cai et al., rock mass surrounding the tunnel is considered (Rocscience, 2002).
2004) as depending on rock structure and block surface condition, Other parameters of interest were also obtained with the aid of the
resulting in the joint condition factor, JC, which can be considered RocLab program (Rocscience, 2002), which produced a complete
in terms of JW (large-scale waviness), JS (small-scale smoothness) set of geomechanical data for the rock mass. Some of these param-
and JA (joint alteration factors). eters and others considered in this paper are shown in Table 2.
For a very weak rock mass, whose Vb is 100 cm3, JW = 1, JS = 1,
JA = 4 and JC = 0.25, a GSI value of 21.4 was obtained. The peak 3.3. Average quality rock mass
Mohr–Coulomb strength parameters, derived from the laboratory
(assumed) data, GSI and tunnel depth were cpeak = 0.34 MPa A basaltic rock mass was chosen as representative of average
/?peak = 43°. quality rock and studied to a depth of 450 m by the authors. Based
The main advantage of this method is that its extension (Cai on laboratory tests, average unconfined compressive strength val-
et al., 2007) enables the residual strength envelope to be obtained. ues of rci = 23 MPa and mi = 10 were obtained. The GSI was esti-
Thus, according to the proposed method, Vbr = 10 cm3 and mated as having a mean value of 55. A Barton’s Q value between
Jcr = 0.1875 for the residual rock mass. The residual GSI was 15.1 1 and 5 was also estimated from field data. Taken into consider-
and the corresponding residual Mohr–Coulomb strength parame- ation in this case was very poor quality blasting and local damage
ters were cres = 0.27 MPa and /res = 40°. to the surrounding rock mass.
The peak and residual strength parameters estimated from the In this average quality rock mass estimated was a block size of
GSI system described by Cai et al. (2007) are shown in Table 1, to- around 36 cm in terms of average spacing and with JC = 1.5,
gether with other parameters of interest obtained using the RocLab roughly corresponding to the average estimated GSI. Estimated
program (Rocscience, 2002). Note that due to the poor quality of for the broken material was a block size of 10 cm in terms of aver-
the rock mass there were no major differences between the peak age spacing and with JC = 0.75, resulting in a residual GSI value of
and residual strength parameters (around 20% in cohesion and less 33.
than 10% in friction). Introducing these rock mass and laboratory test values in the
For the tunnel excavated in soft rock, appropriate is an excava- RocLab program (Rocscience, 2002) results in a complete set of
tion procedure resulting in minimal disturbance to the surround- geomechanical data referring to the material in which the tunnel
ing confined rock mass. As for dilatancy, the values is to be excavated, including rock mass classification, density, peak
recommended by Hoek and Brown (1997) were used along with and residual Mohr–Coulomb strength envelopes and elastic
the variable dilatancy model described by Alejano and Alonso parameters. This data, summarised in Table 3, is the basis for
(2005). developing the post-failure models.

Table 1 Table 2
Soft rock mass geomechanical parameters. Hard rock mass geomechanical parameters.

Parameter Unit Value Parameter Unit Value

GSIpeak 21.40 GSIpeak 64.90
Q 0.10 Q 20.00
GSIres 15.10 GSIres 27.80
rci MPa 100.00 rci MPa 162.00
mi 20.00 mi 19.00
c kN/m3 26.00 c kN/m3 26.00
E GPa 1.10 E GPa 24.00
m 0.30 m 0.25
cpeak MPa 0.34 cpeak MPa 3.70
/peak ° 43.00 /peak ° 57.80
cres MPa 0. 27 cres MPa 0.96
/res ° 40.00 /res ° 51.00
r0 MPa 10.40 r0 MPa 26.00
694 L.R. Alejano et al. / Tunnelling and Underground Space Technology 24 (2009) 689–705

Table 3 4. More complex and realistic behaviour models

Medium rock mass geomechanical parameters.

Parameter Unit Value Our study considers the significance of post-failure behaviour
GSI peak
55.00 illustrated for the three kinds of rock masses referred to in which
Q 1–5 tunnels are to be excavated: poor quality rock (GSI = 21.4), good
GSIres 33.00 quality rock (GSI = 64.9) and average quality rock (GSI = 55).
rci MPa 23.00
For each case increasingly complex models are built, as follows:
mi 10.00
c kN/m3 26.70
E GPa 3.837 (1) Elastic perfectly plastic model.
m 0.25 (2) Elastic brittle plastic model.
cpeak MPa 0.744 (3) Strain softening model with a constant drop modulus and
/peak ° 24.81
constant dilatancy.
cres MPa 0.397
/res ° 15.69 (4) Strain softening model with a variable drop modulus and
r0 MPa 12.00 constant dilatancy.

Fig. 5. Elastic perfectly plastic behaviour model for (a) a soft rock mass (b) a hard rock mass and (c) an average quality rock mass. *Note that rc represents the values for
unconfined compressive strength of the rock masses, not the intact rock.
 L.R. Alejano et al. / Tunnelling and Underground Space Technology 24 (2009) 689–705 695

(5) Strain softening model with a variable drop modulus and 4.1. Elastic perfectly plastic model (Fig. 5)
variable dilatancy.
Field and laboratory data were used to obtain the elastic param-
Below we briefly describe how each of these five models is eters and the Mohr–Coulomb peak failure criterion using standard
implemented for the three kinds of rock mass. Figs. 5–9 (represent- techniques (Rocscience, 2002). Tunnel depth was taken into ac-
ing each of the models in the order in which they are listed) depict count and an isotropic stress field was assumed. Accordingly, a
the behaviour of a cylindrical rock sample under three different set of parameters representative of the perfectly plastic behaviour
confinement pressures (0, 5 and 10 MPa). model, namely, c, /, E, m and c, could be assigned to each rock mass

Fig. 6. Elastic brittle behaviour model for (a) a soft rock mass (b) a hard rock mass and (c) an average quality rock mass.
696 L.R. Alejano et al. / Tunnelling and Underground Space Technology 24 (2009) 689–705

Fig. 7. Strain softening model with a constant drop modulus and constant dilatancy for (a) a soft rock mass (b) a hard rock mass and (c) an average quality rock mass.

where the tunnel was to be excavated. A specific weight of 26 kN/ 2c cos /

rc ¼ ð10Þ
m3 was assigned to all the rock masses and a null dilatancy value 1  sin/
was considered for this simple elastic perfectly plastic model. 1 þ sin /
According to Hoek and Brown (1997) – and as depicted in Fig. 1 – K/ ¼ ð11Þ
1  sin /
this simple model is suitable only for poor quality rock masses.
Typically applied in practice, it would theoretically be representa-
tive of the behaviour of the soft rock mass, and one of the aims of 4.2. Elastic brittle plastic model (Fig. 6)
this paper is to verify this statement.
The values for unconfined compressive strength (rc) and for the Using the same values for cpeak, /peak, E, m and c as for the elastic
passive earth pressure coefficient (K/) of the rock masses – not the perfectly plastic model, the residual failure criterion was obtained
intact rocks – were obtained as follows: by defining residual GSI values for the studied rock masses (cres,
 L.R. Alejano et al. / Tunnelling and Underground Space Technology 24 (2009) 689–705 697

Fig. 8. Strain softening model with a variable drop modulus and constant dilatancy for (a) a soft rock mass (b) a hard rock mass and (c) an average quality rock mass.

/res) and following the approach described by Cai et al. (2007); poor quality rock masses, there should be no great difference in re-
dilatancy was calculated as described by Hoek and Brown (1997). sults for the elastic perfectly plastic model and this brittle plastic
Constant dilatancy angle values were obtained as w = /peak/4 for or elastic brittle plastic rock model for poor quality rock masses.
the good quality rock mass, w = /peak/8 for the average quality rock This model would be reasonably representative of the behav-
mass and w = 0° for the poor quality rock mass. The rock masses iour of our hard rock mass, a statement that we aim to verify.
were modelled as brittle plastic or elastic brittle plastic rock. The parameter g* (the plastic parameter which marks the tran-
According to Hoek and Brown (1997), this behaviour model bet- sition to residual strength values) – as proposed by Alonso et al.
ter represents good quality rock masses. It includes peak and resid- (2003) – was calculated to fit the brittle behaviour for each rock
ual strength criteria, and, because these do not greatly differ for mass as follows:
698 L.R. Alejano et al. / Tunnelling and Underground Space Technology 24 (2009) 689–705

Fig. 9. Strain softening model with a variable drop modulus and variable dilatancy for (a) a soft rock mass (b) a hard rock mass and (c) an average quality rock mass.

 
rpeak
1 ðr3 Þ  rres
1 ðr3 Þ Kw plexity, a strain softening approach makes the sudden stress drop
g ¼ 1þ : ð12Þ associated with brittleness happen in a controlled manner. Model-
E 2
ling is by means of the drop modulus (Fig. 3).
In this initial stage of strain softening modelling, a constant
4.3. Strain softening model with a constant drop modulus and constant drop modulus was associated with each rock mass according to
dilatancy (Fig. 7) quality: M = E/5 for poor quality rock, M = E for high quality
rock and M = E/3 for average quality rock. Note that M = E/5 is
This model represents a first simple approach to modelling close to M = 0, which represents elastic perfectly plastic behaviour;
strain softening (or strength-weakening) behaviour. In this step also that the values of the drop modulus are given as functions of E,
forward towards real rock mass behaviour in terms of model com- which is a much smaller value for poor quality compared to high
 L.R. Alejano et al. / Tunnelling and Underground Space Technology 24 (2009) 689–705 699

quality rock masses. Thus, for high quality rock masses a drop rock masses, as deduced from the observations of other authors
modulus was obtained that was some hundreds of times steeper (Hoek and Brown, 1997; Cai et al., 2007) and from our own
than that for poor quality rocks. observations.
Constant dilatancy was also assumed for this model, for values This continuum approach should be considered within a broad-
the same as those used for the elastic brittle plastic model. er modelling philosophy (Starfield and Cundall, 1988) that includes
The softening parameter g* was calculated for each kind of rock empirical as well as continuous methods and an analysis of struc-
mass – soft, hard and average, respectively – to fit the strength turally controlled instabilities.
drop (Eqs. (15)–(17)). This approach represents a strength-weakening model of actual
  behaviour.
rpeak
1 ðr3 Þ  rres
1 ð r3 Þ Kw
g ¼ 6 1þ ð13Þ The dilatancy model was implemented as indicated in Eqs. (8)
E 2 and (9) above. Values of c?p,* = 0.02 = 20 m strain were assumed
peak  
r ðr3 Þ  r1 ðr3 Þres
Kw for the soft, hard and medium rock masses.
g ¼ 2 1 1þ ð14Þ
The parameters for these models are those given in Tables 1–3,
E 2
peak   along with dilatancy values as indicated in the text; plastic param-
r ð r 3 Þ  rres
1 ð r 3 Þ K w
g ¼ 4 1 1þ ð15Þ eters (g*) were calculated for each case according to the formula-
E 2
tions provided above.

4.4. Strain softening model with a variable drop modulus and constant 4.6. Guidelines for the estimate of the critical softening parameter
dilatancy (Fig. 8)
In the last years, we have been working in studying existing
This next strain softening modelling stage accounted for the ob- techniques and developing new approaches to include strain soft-
served confinement stress dependent nature of the drop modulus. ening behaviour in underground excavation analysis. We have
Thus, for each rock mass studied, the drop modulus considered in found that there existed no techniques, only sometimes very rough
the previous modelling stage was assigned for a confinement stress guidelines, to estimate all the parameters needed to reflect the
of 5 MPa. complete stress–strain behaviour of strain softening rock masses.
This model is a more complex strength-weakening model than In previous works (Alonso et al., 2003) and in this paper, we are
the previous model. In an endeavour to represent the trends ob- trying to provide reasonable values to define and estimate all the
served in large size rock tests, the value of the post-failure drop parameters needed to model strain softening behaviour, which is
modulus was decreased in line with increasing values for confining obviously a difficult task.
stress. In this way, we have observed that the value of the parameter
Following Starfield and Cundall (1988), the main aim of these g* tends to a low figure for very brittle rock (GSI > 75) and that it
models is not to exactly fit actual behaviour or to give a mirror image tends to infinity for very weak rock (GSI < 20). For elastic perfectly
of reality but rather to single out and make available for investigat- plastic materials, we need not a value of g*, since the peak and
ing important elements that are not usually studied. In this simple residual criterion are the same. It is obvious that the identification
formulation, the drop modulus varies in proportion to r3. Bearing of the softening parameter g* is important in this research and, as
in mind that the drop modulus is lower for softer rock masses, differ- it has been observed, neither the selection of the constants 6, 2, 4 in
ent drop moduli were assumed for the three rock masses. Thus, the Eqs. (13)–(15), nor the values in Eqs. (16)–(18) have been justified.
drop modulus for soft rock masses varied from M = E/4 to M = E/6 A reasonable approach to estimate the value of this parameter,
(between r3 = 0 and r3 = 10 MPa); for hard rock masses, from may present the following general form:
M = 2E to M = E/2 (between r 3 = 0 and r3 = 10 MPa); and for  
average rock masses, from M = E/2 to M = E/4. ½rpeak
1 ðr3 Þ  rres
1 ðr3 Þ Kw
g ¼ fg ðr3 ; GSIÞ  1þ ð19Þ
For soft rock masses at high confinements, this variable drop E 2
modulus makes this model very similar to the elastic perfectly
plastic model (Section 4.1 above), and, for hard rock masses at where fn (r3, GSI) is a function:
low confinements, very similar to the elastic brittle plastic model
E
(Section 4.2 above). fg ðr3 ; GSIÞ ¼ 1  ð20Þ
M
The same constant dilatancy values as in the previous model
were assumed. Since it has been observed that the elastic modulus E and the drop
As with all the previous models, the transition value of the plas- modulus M depend on the rock mass quality and on the confining
tic parameter, g*, was calculated to fit the model, according to stress, the value of the function also depends on these parameters.
equations for soft, hard and medium rock masses, respectively, as Remark that if dilatancy is considered to be variable, this depen-
follows: dence should also be included in the function.
r  rpeak ðr Þ  rres ðr Þ  Kw
 We have been using, for relatively simple strain softening mod-
3 1 3 1 3
g ¼ þ5 1þ ð16Þ els with constant drop modulus, a value of fn (r3, GSI) of 2, 4 and 6
5 E 2
  peak   for good, average and bad quality rock masses. These tentative val-
3r3 5 r1 ðr3 Þ  r1 ðr3 Þres
Kw ues can be equated as:
g ¼ þ 1þ ð17Þ
20 4 E 2
  peak   fg ðr3 ; GSIÞ ¼ 8  0; 08  GSI for 25 < GSI < 75 ð21Þ
7r3 25 r1 ðr3 Þ  r1 ðr3 Þ res
Kw
g ¼ þ 1þ ð18Þ This equation is suggested for strain softening models with constant
40 8 E 2
drop modulus, as a first approach.
In Section 4.3 of our paper we have been using a more complex
4.5. Strain softening model with a variable drop modulus and variable approach, including the effect of rci:
dilatancy (Fig. 9)  
rci ðMPaÞ
fg ðr3 ; GSIÞ ¼ 8; 66  0; 0812  GSI þ for 20 < GSI < 75
Finally, a variable dilatancy model – as defined by Alejano and 10
Alonso (2005) – represents a fairly realistic behaviour model for ð22Þ
700 L.R. Alejano et al. / Tunnelling and Underground Space Technology 24 (2009) 689–705

We are actually seeking information and data to test these ap- The final values have been rounded off.
proaches. Think also that, in fact, this value is variable in laboratory More work is obviously needed to fine-tune these approaches,
tests (certainly also in large-scale actual rock masses), since differ- and special attention will be placed in the future to compare re-
ent drop moduli are observed for varying confinements. However, sults with values from engineering practice.
we have included it here, due to the fact that these tentative ap-
proaches are needed if, as recommended by Starfield and Cundall
(1988), we follow heuristic approaches to study, analyze and under-
stand rock mechanics problems.
If one uses more complex strain softening models with confine-
ment stress dependent drop modulus, the following equation is
suggested as a first approach to estimate the confinement stress
dependent critical plastic parameter:
   
225  GSI 55  0:6GSI
fg ðr3 ; GSIÞ ¼  r3 þ
1000 8
for 25 < GSI < 75 ð23Þ
Again this approach is tentative. However, in our lab we have
tested, in a servo-controlled press, samples of two different moder-
ately weathered granites (rci around 50 MPa), which due to their
clay content behave as good quality rock masses (Alejano et al., in
pressa). We have measured the elastic and the drop modulus in
more than twenty tests for each of the two different granites
(named Xavier and Castro) at different confining stress levels vary-
ing between 0 and 8 MPa. The results of these tests are plot over the
lines representing Eqs. (17)–(19) in g*, r3 axes in the Fig. 10.
This is an empirical indication that, if we consider confinement
stresses between 0 and 8 MPa, the results of Eq. (23) can be a rea-
sonable starting point to model post-failure behaviour of strain
softening Mohr–Coulomb rock masses.
Observe also that for a value of confinement stress of 5 MPa, the
value of fg(r3, GSI) for bad (GSI = 25), average (GSI = 50) and good
(GSI = 75) quality rock masses obtained will be 6, 4 and 2, respec-
tively, as suggested for the model with constant drop modulus.
In Section 5.4 of our paper we have been using values roughly
coinciding with those obtained according to Eq. (23). However,
the used values were slightly changed using a more complex ap-
proach, including the effect of rci as in the following formulation.
2  3
8; 66  0; 0812  GSI þ rci ðMPaÞ
10
F g ðr3 ; GSIÞ ¼ 4 5
8  0; 08  GSI
   
225  GSI 55  0:6 GSI
  r3 þ
1000 8
for 20 < GSI < 75 ð24Þ

Fig. 10. Representation of fg(r3, GSI) as a function of GSI, for the proposed values Fig. 11. Tunnel ground reaction curves for the elastic perfectly plastic and elastic
for rock masses and for laboratory testing results in samples of different moderately brittle models for (a) a soft rock mass (b) a hard rock mass and (c) an average
weathered granites. quality rock mass.
 L.R. Alejano et al. / Tunnelling and Underground Space Technology 24 (2009) 689–705 701

5. Ground reaction curves for the tunnel point out that the support and reinforcement effect is highly depen-
dent on the installation, usually controlled by means of distance to
5.1. Initial considerations the face and maximum displacement (in turn controlled by means
of the LDP). Nevertheless, the actual behaviour of the tunnel must,
GRCs for tunnels can be rigorously calculated for the different in all cases, lie somewhere in between the two extremes repre-
behaviour models described by applying the approach described sented by the elastic perfectly plastic and elastic brittle models.
in Alonso et al. (2003). Note that very few changes had to be intro- For the low geotechnical quality rock mass, the final displace-
duced in the original MATLAB code to account for the particulari- ments change from around 21 cm for the elastic perfectly plastic
ties of these models. case to around 28 cm for the elastic brittle case. For the hard rock
An alternative method for obtaining GRCs was recently devel- mass, the final displacements change from around 10 mm for the
oped by Lee and Pietruszczak (2008), based on a simplified numer- elastic perfectly plastic case to around 16 mm for the elastic brittle
ical method (with a suitable degree of accuracy) for strain case. Final differences should not be very high, however, if we take
softening rock masses like those described here. proper account of natural variability in the rock parameters. For
the average quality rock mass, the final displacements change from
5.2. GRC analysis for elastic perfectly plastic and elastic brittle models around 18 cm for the elastic perfectly plastic case to around
230 cm for the purely elastic brittle case, a clear indication that
Analytical solutions also exist for the elastic perfectly plastic adequate selection of the post-failure behaviour model is essential.
and the elastic brittle behaviours, such as those described by Panet Following the work by Oreste (2003), we now obtain the SCCs
(1995) and Carranza-Torres (1998). The GRCs representing these corresponding to support and reinforcement as determined using
two models are depicted in Fig. 11, where both analytical (Panet, Barton’s Q classification system.
1995 and Carranza-Torres, 1998) and numerical (Alonso et al., For the soft rock mass (Fig. 12), we assumed reinforcement and
2003) solutions are represented, showing similar results. support adequate to support the tunnel to be 25 cm of fibre shot-
A significant difference between the final displacements for the crete and Swellex rockbolts with spacing of 1  1 m. The shotcrete
elastic perfectly plastic and elastic brittle models for each kind of reinforcement and the Swellex support were assumed to be in-
rock mass is a clear indication of the error that might arise if model stalled at a distance of 1 and 1.5 m, respectively, from the tunnel
selection is not based on suitable criteria. It is also important to face.

Fig. 12. Tunnel ground reaction curve and support for a soft rock mass using (a) the elastic perfectly plastic model and (b) the elastic brittle model.

Fig. 13. Tunnel ground reaction curve and support for a hard rock mass using (a) the elastic perfectly plastic model and (b) the elastic brittle model.
702 L.R. Alejano et al. / Tunnelling and Underground Space Technology 24 (2009) 689–705

For the hard rock mass (Fig. 13), it is assumed that adequate behaviour by accounting for a drop modulus that decreases as con-
support for the tunnel will be Swellex rockbolts spaced at intervals finement increases (as has been observed in actual rock masses);
of 2 m. The Swellex support is assumed to be installed at a distance and finally, the third strain softening model represents further
of 4 m from the tunnel face. development by accounting for variable dilatancy. To our knowl-
For the medium quality rock mass, the necessary reinforcement edge, this third strain softening model should suitably represent
and support for the tunnel is assumed to be 15 cm of fibre shot- actual rock mass behaviour. The corresponding GRCs for the differ-
crete and Swellex rockbolts with spacing of 1 x 1 m. Support and ent rock masses are shown in Fig. 15.
reinforcement are installed very close to the tunnel face (at dis- As can be observed in this figure, for the soft rock mass the
tances of 2 and 3 m, respectively). The results are shown in GRC corresponding to the third strain softening model (with var-
Fig. 14a (elastic perfectly plastic case) and Fig. 14b (elastic brittle iable drop modulus and variable dilatancy) is close to those corre-
case). It is clear that the support will be able to support the tunnel sponding to all the other models with the exception of the elastic
in the first case, but not in the second case, where displacements brittle model, that is, the results are, from a practical engineering
will not be tolerated. perspective, essentially identical. It can thus be concluded that
The values for the most significant rock mass quality parame- the elastic perfectly plastic model is reasonably representative
ters and proposed support and reinforcement (derived from Bar- of continuous rock mass behaviour in soft rock (GSI < 35) unless
ton’s Q classification) are shown in Table 4. Note that support squeezing conditions occur (see Hoek and Marinos (2000) and
and reinforcement are only indicative, given that the classification Hoek et al. (2008) for further details on the validity of the CCM
system does not account specifically for depth. Note also that the for very low quality rock masses). For soft rock masses, the sup-
rock mass with GSI = 64.9 requires reinforcement with shotcrete port and reinforcement curves would be identical (from an engi-
and does not require support. neering point of view) for the strain softening model with variable
drop modulus and variable dilatancy and the elastic perfectly
5.3. GRC analysis for the strain softening behaviour models plastic model.
A tunnel excavated in the soft rock mass would thus be stable,
The differences between final displacements in the elastic per- with an equilibrium pressure of around 0.9 MPa, and with an initial
fectly plastic and elastic brittle behaviour models (extreme mod- displacement of around 40 mm after support and reinforcement
els) have been demonstrated. Consequently, for the GRCs installation and a final displacement of around 100 mm. According
corresponding to the strain softening models, the final displace- to the Hoek (1999), the safety factor would be around 1.1, rising to
ments would obviously have to lie between the final displacements around 1.3 according to Oreste (2003a).
corresponding to these two extreme models. For the hard rock mass (Fig. 15b), the GRC corresponding to the
The first of the three strain softening models has a constant strain softening model with variable drop modulus and variable
drop modulus; the second represents a further step towards actual dilatancy is close to those corresponding to all the other models

Fig. 14. Tunnel ground reaction curve and support for a medium rock mass using (a) the elastic perfectly plastic model and (b) the elastic brittle model.

Table 4
Rock mass classification (RMR) estimates for poor, good and medium quality rock masses, the main parameters calculated from the corresponding ground reaction curves,
support and reinforcement recommendations along with distances to the tunnel face.*

GSI = 21.4 GSI = 64.9 GSI = 55

RMR 21.4 64.9 55
Q 0.1 20 1–5
Support – fibre shotcrete (cm) 25 No 15
Reinforcement – rockbolt (spacing) 1m1m 2m2m 1m1m
L (unsupported distance to face according to Barton’s Q) (m) 1.2 8.2 5.3
Average distance to face for support installation 1 N/A 2
Average distance to face for reinforcement installation 1.5 4 3
*
Based on a calculation of longitudinal deformation profiles obtained according to the method proposed by Vlachopoulos and Diederichs (2009) and excerpted in Hoek et al.
(2008).
 L.R. Alejano et al. / Tunnelling and Underground Space Technology 24 (2009) 689–705 703

with the exception of the elastic perfectly plastic model. Thus, from iour in good quality rock masses (60 < GSI < 70). The upper limit of
a practical engineering perspective, the four results are essentially validity, however, will depend on the spalling behaviour of the
the same. It can be concluded, therefore, that the elastic brittle rock (Kaiser et al., 2000; Diederichs, 2003).
model is reasonably representative of continuous rock mass behav- A tunnel excavated in a hard rock mass could thus be consid-
ered to be very stable, with an equilibrium pressure around
6 kPa, and with around 10 mm of initial displacement after support
and reinforcement installation and a final displacement of around
15 mm. This means, in practice, that a final operative radius equal
to the original radius would be obtained.
According to Hoek (1999), the safety factor would be around
6.5. However, it would rise to around 7.8 if the strain-based safety
factor were defined as the ratio between the final and the actual
strain on the support system (as per Oreste, 2003a). Obviously,
the reinforcement and support systems are over-dimensioned for
rock mass confinement purposes, but they are appropriate if struc-
turally controlled instability mechanisms are taken into account.
Finally, for the medium rock mass (Fig. 15c), the corresponding
to the strain softening model with variable drop modulus and var-
iable dilatancy is very dissimilar to both the elastic perfectly plastic
and elastic brittle models. This means that using either of these
two extreme models for average quality rock masses
(40 < GSI < 60) is likely to result in significant error. To achieve
an acceptable degree of accuracy, more realistic models that ac-
count for strain softening behaviour (at least), if not variable dilat-
ancy, should be used.
Such a model is the strain softening model with variable drop
modulus and variable dilatancy, whose GRC (for the medium rock
mass) yields a final displacement of around 400 mm (Fig. 16); in
our experience, this seems to be a more acceptable range for these
excavations. Using this GRC for our CCM analysis, the tunnel exca-
vated in the medium rock mass would have an equilibrium pres-
sure of around 0.6 MPa, with an initial displacement of around
60 mm after support and reinforcement installation and a final dis-
placement of around 400 mm. The other two strain softening mod-
els (constant drop modulus/constant dilatancy and variable drop
modulus/constant dilatancy) would have produced very similar
results.

Fig. 15. Tunnel ground reaction curves for five different behaviour models* for (a) a
soft rock mass (b) a hard rock mass and (c) an average quality rock mass. *The five
models are the elastic perfectly plastic and elastic brittle plastic models and three
strain softening models (with a constant drop modulus and constant dilatancy, with Fig. 16. Tunnel ground reaction curve and support for a medium quality rock mass
a variable drop modulus and constant dilatancy, and with a variable drop modulus resulting from the strain softening model with a variable drop modulus and
and variable dilatancy). variable dilatancy.
704 L.R. Alejano et al. / Tunnelling and Underground Space Technology 24 (2009) 689–705

Table 5 acceptable degree of error. Some such models have been described
Parameters (internal pressure (Peq) and displacement (Ueq) at the equilibrium point above; the most accurate of these models, in our opinion, would be
and safety factors) for three behaviour models representing a medium rock mass.
capable of accommodating strain softening, confinement stress
Elastic Elastic Strain softening model dependent drop modulus and variable dilatancy.
perfectly brittle with variable drop We demonstrated the extent of possible error with average qual-
plastic model modulus and dilatancy
model
ity rock masses if more realistic models – like the strain softening
model with a variable drop modulus and variable dilatancy – are
Ueq (mm) 110 800 400
Peq (MPa) 0.63 0.68 0.64
not used. Note that the CCM represents a continuum approach to
Safety factor Hoek (1999) 1.05 0.97 1.02 rock masses, so, in all cases, the role of discontinuities and structur-
Safety factor Oreste (2003) 1.14 0.99 1.02 ally controlled instabilities also need to be considered – a topic, how-
ever, that is beyond the scope of this paper. As for the range of
applications of the CCM, an article on this subject is shortly to be
published (Alejano et al., in pressb).
According to both Hoek (1999) and Oreste (2003a), the safety A concluding comment is that, for the analysed cases, the vari-
factor would be around 1.02, indicating that the tunnel would able dilatancy model does not produce significantly different re-
not be very stable. sults from the approach described by Hoek and Brown (1997).
However, Fig. 13 indicates that if we use the elastic perfectly Nonetheless, we are of the opinion that it is more realistic and
plastic model (remember that GRCs often apply this kind of model accurate to model dilatancy as variable rather than constant, even
in practice), we would obtain an equilibrium pressure of around if parameter estimation is more difficult. Application would be
0.6 MPa with a much smaller final displacement of around very appropriate, for example, for studies of large rock mass zones
110 mm, resulting in a safety factor of 1.05 (Hoek, 1999) or 1.14 with yielding, such as, for example, coal pillar yielding or rock mass
(Oreste 2003a). If, however, we use the elastic brittle model, we caving.
would obtain an equilibrium pressure of around 0.7 MPa with a
much greater final displacement of around 800 mm, resulting in Acknowledgements
a safety factor of less than 1 according to both Hoek (1999) and
Oreste (2003a). We can thus see how unrealistic values in both fi- The authors thank the Spanish Ministry of Science and Technol-
nal displacements and safety factors would be obtained (Table 5 ogy for funding under contract reference number BIA2006-14244
shows a comparison of these results). for the research project entitled ‘Analysis of rock mass post-failure
behaviour’.
6. Conclusions Ailish M.J. Maher provided assistance with English usage in a
version of the manuscript.
We describe a case study in which tunnels were excavated in
different kinds of rock masses, with all significant parameters cal-
culated. GRCs were obtained for increasing levels of model com- References
plexity, reflecting elastic perfectly plastic, elastic brittle and
AFTES, 1993. Groupe de travail n. 7 – Soutenement et revetement, Emploi de la
strain softening models. Post-failure parameters were calculated méthode convergence–confinement. Tunnels et Ouvrages Souterrains, Suppl.
by means of recently developed techniques (Cai et al., 2007). Final- 117, pp. 118–205.
ly, a strain softening model was built that accounted for confining Alejano, L.R., Alonso, E., 2005. Considerations of the dilatancy angle in rocks and
rock masses. Int. J. Rock Mech. Min. Sci. 42 (4), 481–507.
stress and plastic strain dependent dilatancy. The effects of stan- Alejano, L.R., Posada, D., Rodriguez-Dono, A., in pressa. Servo-controlled strength
dard support and reinforcement were also taken into account. tests on moderately weathered granite. EUROCK’2009. Dubrovnik-Catvat,
Our main conclusion is that the application of CCM varies Croatia.
Alejano, L.R., Alonso, E., Rodriguez-Dono, A., Fdez.-Manin, G., in pressb. Towards a
according to the quality of the rock mass. Note that, for each tunnel reliable application of the convergence–confinement method for tunnels
described, the plastic zone remains constant in all the models; con- excavated in rock masses exhibiting Hoek–Brown strain-softening behaviour.
sequently, the variability observed in final displacements can only Int. J. Rock Mech. Min. Sci.
Alonso, E., Alejano, L.R., Varas, F., Fdez.-Manin, G., Carranza-Torres, C., 2003. Ground
be due to the post-failure strain behaviour. reaction curves for rock masses exhibiting strain-softening behaviour. Int. J.
For softer rock masses (GSI < 40), actual behaviour is so similar Num. Anal. Meth. Geomech. 27, 1153–1185.
to elastic perfectly plastic behaviour that negligible error would be Bouvard-Lecoanet, A., Colombet, G., Esteulle, F., 1988. Ouvrages Souterrains:
Conception, Realisation, Entretien. Presses de L’école Nationale des Ponts et
attained if this model were used in practice to obtain the GRCs or
Chaussées, Paris.
to implement numerical models. If squeezing conditions occur, the Brady B.H.G., Brown E.T., 1993. Rock Mechanics for Underground Mining, second ed.
continuum approach must be considered within a wider modelling Chapman and Hall.
Brown, E.T., Bray, J.W., Ladanyi, B., Hoek, E., 1983. Ground response curves for rock
philosophy (Starfield and Cundall, 1988) that takes account of 3D
tunnels. J. Geotech. Eng. 109, 15–39.
numerical models (see Hoek et al. (2008) and Hoek and Marinos Cai, M., Kaiser, P.K., Uno, H., Tasaka, Y., Minamic, M., 2004. Estimation of rock mass
(2000)). deformation modulus and strength of jointed rock masses using the GSI system.
A similar conclusion can be drawn for hard rock masses (GSI in Int. J. Rock Mech. Min. Sci. 41 (1), 3–19.
Cai, M., Kaiser, P.K., Tasaka, Y., Minamic, M., 2007. Determination of residual
the range 60–75). Negligible error would be attained if the elastic strength parameters of jointed rock masses using the GSI system. Int. J. Rock
brittle model was used in practice to obtain the GRCs or to imple- Mech. Min. Sci. 44 (2), 247–265.
ment numerical models. Again, however, this continuum approach Carranza-Torres, C., 1998. Self similarity analysis of the elastoplastic response of
underground openings in rock and effects of practical variables. Ph.D. Thesis.
must also be considered within a wider modelling philosophy University of Minnesota.
(Starfield and Cundall, 1988) – in this case, one that includes an Carranza-Torres, C., Fairhurst, C., 2000. Application of convergence–confinement
analysis of structurally controlled instabilities and spalling control method of tunnel design to rock masses that satisfy the Hoek–Brown failure
criterion. Tunn. Undergr. Space Technol. 15 (2), 187–213.
approaches (Kaiser et al., 2000; Diederichs, 2003). Carranza-Torres, C., Alonso, E., Alejano, L.R., Varas, F., Fdez.-Manin, G., 2002. Elasto-
Finally, for the case of average quality rock masses (GSI in the plastic analysis of deep tunnels in brittle rock using a scaled form of the Mohr–
range 40–60), behaviour is not similar to either of the models rep- Coulomb failure criterion. In: Hammah et al. (Ed.), Proc. 5th North Am. Rock
Mech. Symp. – Tunn. Assoc. Can., Toronto, pp. 283–293.
resenting the extremes (that is, the elastic perfectly plastic and
Chern, J.C., Shiao, F.Y., Yu, C.W., 1998. An empirical safety criterion for tunnel
elastic brittle plastic models), so more accurate models are re- construction. In: Proceedings of the Regional Symposium on Sedimentary Rock
quired to describe the behaviour of these rock masses with an Engineering. Balkema, Taipei, Taiwan, pp. 222–227.
 L.R. Alejano et al. / Tunnelling and Underground Space Technology 24 (2009) 689–705 705

Crowder, J.J., Bawden, W.F., 2006. The Estimation of Post-Peak Rock Mass Lee, Y.-K., Pietruszczak, S., 2008. A new numerical procedure for elasto-plastic
Properties: Numerical Back Analysis Calibrated using In Situ Instrumentation analysis of a circular opening excavated in a strain-softening rock mass. Tunn.
Data. Rocnews. Available from: <http://www.rocscience.com/library/rocnews>. Undergr. Space Technol. 23, 588–599.
Detournay, E., 1986. Elasto-plastic model of a deep tunnel for a rock with variable Lombardi, G., 1975. Qualche aspetto particolare della statica delle cavitá
dilatancy. Rock Mech. Rock Eng. 19, 99–108. sotterranee. Rivis. Italiana Geotec. 9, 187–206.
Diederichs, M.S., 2003. Rock fracture and collapse under low confinement Medhurst, T.P., 1996. Estimation of the in situ strength and deformability of coal for
conditions. Rock Mech. Rock Eng. 36 (5), 339–381. engineering design. Ph.D. Thesis. University of Queensland, Australia.
Farmer, I.W., 1983. Engineering Behaviour of Rocks, second ed. Chapman Hall. Oreste, P., 2003a. Analysis of structural interaction in tunnels using the covergence–
Fenner, R., 1938. Untersuchungen zur Erkenntnis des Gebirgsdruckes. Glückauf 74, confinement approach. Tunn. Undergr. Space Technol. 18, 347–363.
681–695 [705–715]. Oreste, P., 2003b. A procedure for determining the reaction curve of shotcrete lining
Hill, R., 1950. The Mathematical Theory of Plasticity. Oxford University Press, New considering transient conditions. Rock Mech. Rock Eng. 36 (3), 209–236.
York. Oreste, P., 2008. Distinct analysis of fully grouted bolts around a circular tunnel
Hoek, E., 1999. Support for very weak rock associated with faults and shear zones. considering the congruence of displacements between the bar and the rock. Int.
In: Villaescusa, Windsor, Thompson (Ed.), Rock Support and Reinforcement J. Rock Mech. Min. Sci. 45 (5), 1052–1067.
Practice in Mining. Balkema, Kalgoorlie, Australia, pp. 19–34. Pacher, F., 1964. Deformationsmessungen in Versuchsstollen als Mittel zur
Hoek, E., Brown, E.T., 1980. Underground Excavations in Rock. The Institution of Erforschung des Gebirgsverhaltens und zur Bemessung des Ausbaus.
Mining and Metallurgy, London. p. 527. Felsmech. u. Ing. Geol. 1, 149–161.
Hoek, E., Brown, E.T., 1997. Practical estimates of rock mass strength. Int. J. Rock Panet, M., 1993. In: Hudson (Ed.), Understanding Deformations in Tunnels.
Mech. Sci. Geom. Abstr. 34 (8), 1165–1187. Comprehensive Rock Engineering, vol. 1. Pergamon Press, pp. 663–690.
Hoek, E., Marinos, P., 2000. Predicting tunnel squeezing problems in weak Panet, M., 1995. Le calcul des tunnels par la méthode des curves convergence–
heterogeneous rock masses. Tunn. Tunnell. Int. Part 1, 32(11), 45–51; Part 2, confinement. Presses de ĺÉcole Nationale des Ponts et Chaussées, Paris, France.
32(12), 33–36. Peila, D., Oreste, P., 1995. Axisymmetrical Analysis of Ground Reinforcing in
Hoek, E., Kaiser, P.K., Bawden, W.F., 1995. Support of Underground Excavations in Tunnelling Design, Computer and Geotechnics, vol. 17. Elsevier Science Ltd.,
Hard Rock. Balkema, Rotterdam. London, UK. pp. 253–274.
Hoek, E., Carranza-Torres, C., Corkum, B., 2002. Hoek–Brown failure criterion – 2002 RocScience, RocLab, 2002. Rocscience Inc., Toronto, Canada.
ed. In: Proceedings of the NARMS-TAC 2002, Mining Innovation and RocScience, RocSupport, 2003. Rocscience Inc., Toronto, Canada.
Technology. Toronto, Canada, pp. 267–273. Salençon, J., 1969. Contraction quasistatique d’une cavité à symètrie sphérique ou
Hoek, E., Carranza-Torres, C., Diederichs, M.S., Corkum, B., 2008. Kersten Lecture cylindrique dans un milieu elastoplastique. Annales des Ponts et Chaussées 4,
Integration of geotechnical and structural design in tunnelling. In: 56th Annual 213–236.
Geotechnical Engineering Conference. University of Minnesota. Starfield, A.M., Cundall, P.A., 1988. Towards a methodology for rock mechanics
Kaiser, P.K., Diederichs, M.S., Martin, D., Sharpe, J., Steiner, W., 2000. Underground modelling. Int. J. Rock Mech. Min. Sci. Geom. Abstr. 25 (3), 99–106.
works in hard rock tunnelling and mining. In: Proceedings, GeoEng, Melbourne. Vlachopoulos, N., Diederichs, M.S., 2009. Improved longitudinal displacement
Technomic Publishing, Lancaster, pp. 841–926. profiles for convergence confinement analysis of deep tunnels. Rock Mech.
Kaliszky, S., 1989. Plasticity: Theory and Engineering Applications. Amsterdam, Rock Eng. 42, 131–146.
Elsevier.