Ground support for tunnels in rock by Dr.

David Oliveira

1 Key references
It is strongly recommended that for this Module 3 you complement your understanding of Rock bolting Design
by reading the Practical Rock Engineering Book by Evert Hoek, freely available at the website below.
https://www.rocscience.com/assets/resources/learning/hoek/Practical-Rock-Engineering-Full-Text.pdf
The following chapters are associated with this module:
https://www.rocscience.com/assets/resources/learning/hoek/Practical-Rock-Engineering-Chapter-5-
Structurally-Controlled-Instability-in-Tunnels.pdf
https://www.rocscience.com/assets/resources/learning/hoek/Practical-Rock-Engineering-Chapter-14-Rockbolts-
and-Cables.pdf
In addition, the papers listed below haven been included in the appendix further discussing the details of some
of the topics presented in this module. They are to be used for educational purpose only and not to be re-
distributed.
Li, C. C. (2017). Principles of rockbolting design. Journal of Rock Mechanics and Geotechnical Engineering
9, pp 396-414.
Oliveira D. (2018). Design of large span tunnels and caverns: back to basics. Invited Keynote: 2018 AGS
Victorian Symposium - Geotechnics and Transport Infrastructure, Melbourne, October, pp.25-34.
Oliveira, D.A.F., Diederichs, M.S. (2017). Tunnel support for stress induced failures in Hawkesbury
Sandstone Tunnelling and Underground Space Technology 64, pp 10-23.
Oliveira, D.A.F., Paramaguru, L. (2016). Laminated rock beam design for tunnel support. Australian
Geomechanics Journal, 51 (4), pp 1-17, December.
Oliveira D. and Pells, P. (2014). Revisiting the applicability of voussoir beam theory for tunnel design in
Sydney. Australian Geomechanics Vol 49(3), pp 29-44, September
Diederichs, M S and Kaiser P K. (1999). Stability of large excavations in laminated hard rock masses: the
voussoir analogue revisited. Int. J. Rock Mech. & Min. Sci., 36, pp 97-118.
Pells, P. J. N. (2002). Developments in the design of tunnels and caverns in the Triassic rocks of the Sydney
region. International Journal of Rock Mechanics and Mining Sciences 39, pp 569-587

1
Ground support for tunnels in rock by Dr. David Oliveira

Appendix (Papers)

2
 Journal of Rock Mechanics and Geotechnical Engineering 9 (2017) 396e414

Contents lists available at ScienceDirect

Journal of Rock Mechanics and

Geotechnical Engineering
journal homepage: www.rockgeotech.org

Full Length Article

Principles of rockbolting design

Charlie C. Li
Department of Geoscience and Petroleum, The Norwegian University of Science and Technology, Norway

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

Article history: This article introduces the principles of underground rockbolting design. The items discussed include
Received 17 January 2017 underground loading conditions, natural pressure zone around an underground opening, design
Received in revised form methodologies, selection of rockbolt types, determination of bolt length and spacing, factor of safety, and
3 April 2017
compatibility between support elements. Different types of rockbolting used in engineering practise are
Accepted 4 April 2017
Available online 3 May 2017
also presented. The traditional principle of selecting strong rockbolts is valid only in conditions of low in
situ stresses in the rock mass. Energy-absorbing rockbolts are preferred in the case of high in situ
stresses. A natural pressure arch is formed in the rock at a certain distance behind the tunnel wall.
Keywords:
Rockbolting design
Rockbolts should be long enough to reach the natural pressure arch when the failure zone is small. The
Pressure arch bolt length should be at least 1 m beyond the failure zone. In the case of a vast failure zone, tightly spaced
Bolt length short rockbolts are installed to establish an artificial pressure arch within the failure zone and long cables
Bolt spacing are anchored on the natural pressure arch. In this case, the rockbolts are usually less than 3 m long in
Factor of safety mine drifts, but can be up to 7 m in large-scale rock caverns. Bolt spacing is more important than bolt
length in the case of establishing an artificial pressure arch. In addition to the factor of safety, the
maximum allowable displacement in the tunnel and the ultimate displacement capacity of rockbolts
must be also taken into account in the design. Finally, rockbolts should be compatible with other support
elements in the same support system in terms of displacement and energy absorption capacities.
Ó 2017 Institute of Rock and Soil Mechanics, Chinese Academy of Sciences. Production and hosting by
Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/
licenses/by-nc-nd/4.0/).

1. Introduction 2. Underground loading conditions

Rockbolt is the most widely used support element in support 2.1. Low in situ stress conditions
systems in underground mines and civil tunnels. Rockbolting
design is indeed mainly based on experience and it appears that Rock blocks in the roof of an underground opening are pre-
rockbolting design is simply a business of selection of rockbolt vented to fall as long as a high enough horizontal stress exists in the
types and the determination of bolt length and spacing, but, one rock mass. However, they would fall under gravity in low in situ
essentially uses, either explicitly or implicitly, a methodology in a stress conditions. In locations close to the ground surface, the rock
specific rockbolting design. Attempts are made in this article to mass often contains well-developed rock joint sets. The rock joints
summarise the design principles and methodologies hidden in sometimes are open, which is an indication that the in situ rock
rockbolting practise, which include the relationship between the in stresses are low in the rock mass. The task of rock support in low
situ stress state and rockbolt types, the concept of pressure arch, stress rock masses is to prevent rock blocks from falling. To do so,
design methodologies, determination of bolt length and spacing, the maximum load exerted on the support elements, such as
factor of safety, compatibility between support elements and rockbolts, is the deadweight of the potentially falling block (Fig. 1).
different types of rockbolts. This is a load-controlled condition.
From the point of view of mechanics, the rockbolts must be
strong enough to bear the deadweight of the loosened rock block.
Therefore, use of a factor of safety, defined by the strength of the
E-mail address: Charlie.c.li@ntnu.no. support system and the weight force of the block (i.e. the load), is
Peer review under responsibility of Institute of Rock and Soil Mechanics, appropriate for rock support design in a load-controlled condition.
Chinese Academy of Sciences.

http://dx.doi.org/10.1016/j.jrmge.2017.04.002
1674-7755 Ó 2017 Institute of Rock and Soil Mechanics, Chinese Academy of Sciences. Production and hosting by Elsevier B.V. This is an open access article under the CC BY-
NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
 C.C. Li / Journal of Rock Mechanics and Geotechnical Engineering 9 (2017) 396e414 397

load-controlled condition as shown in Fig. 1, the strength of the

rockbolts is the most important parameter for the selection of
rockbolt type. The basic requirement is that the strength of the
rockbolts must be higher than the load on the bolts. The appro-
priate types of rockbolts under load-controlled conditions are fully
encapsulated rebar bolts, threadbar bolts and cablebolts.
In overstressed weak and soft rock, the excessive deformation
needs to be accommodated. The traditional approach to deal with
rock squeezing is to use ductile rockbolts in conjunction with other
types of ductile surface retaining elements such as mesh. Split set is
the typical rockbolt for this purpose in the mining industry. Split set
can displace significantly, but it cannot much restrain the rock
deformation because of its low load-bearing capacity. Its main
function is to avoid disintegration of the fractured rock mass. An
active measure to stabilise squeezing rock is to provide a high
support resistance to restrain the rock deformation on the one
hand, while the support elements in the support system must be
deformable on the other hand. Use of energy-absorbing rockbolts
can achieve this goal.
Rockburst is an instability issue in overstressed hard and strong
rock. The goal of rock support in such conditions is to absorb the
kinetic energy of the ejected rock. Energy-absorbing rockbolts
should be used in burst-prone rock masses. The higher the load-
bearing capacity of the energy-absorbing rockbolt is, the less the
ejected rock displaces.
Fig. 1. Gravitational load on rockbolts in low stress condition.

3. Design principles
This is essentially the design principle in structure mechanics,
which states that the load applied to a structure should not be
3.1. Natural pressure arch
higher than the strength of the structure, i.e. the strength-to-load
ratio that is called the factor of safety, should be larger than 1.
Geological exploration drilling was once carried out in a mine
This principle is valid for ground constructions where the total load
drift, excavated 5 years previously, at a depth of 1000 m. The mine
on the construction structures is usually known or easily found out.
drift was parallel with the strike of the tabular ore body and the
In shallow underground openings, this principle is also valid since
boreholes were drilled in the wall of the drift on the side of the ore
the maximum load on the rock support system is the deadweight
body that was approximately 150 m apart from the drift. The frac-
force of loosened rock blocks.
ture logging on the cores provided information on the distribution of
the secondary stresses in the rock surrounding the drift. Fig. 3 shows
2.2. High in situ stress conditions the fracture patterns in the cores taken from a horizontal borehole.
The fracture intensity in the cores varies along the borehole. The
The author observed in a deep metal mine that the number of cores are small pieces with a low value of rock quality designation
geological discontinuities in the rock mass became less and the (RQD) in the zone from the wall to a depth of 2.1 m (Zone I). The
discontinuities were less opened in depth. For instance, at a depth fracture surfaces in this zone are yellow coloured, indicating that
of 1000 m, it was observed that all of the few discontinuities they were probably created when the drift was excavated a few
exposed on an excavation face were completely closed. Therefore, it years earlier. The cores are disked in the zone from 2.1 m to 8.5 m
can be said that the rock mass quality is improved at depth because (Zone II). The fractures in this zone are fresh and perpendicular to
of the reduction in the number of geological discontinuities. the core axis. It can be said with confidence that they were created
However, the in situ rock stresses increase with depth. At depth, the during core drilling. Zone II can be further divided into two sub-
major instability issue is no longer fall of loosened rock blocks but zones. In Zone IIa, the core disking is so severe that the disks are
rock failure caused by stress. High stresses could lead to two con- tightly spaced. The disk thicknesses are obviously larger in Zone IIb
sequences in underground openings: large deformation in soft and than in Zone IIa. Zone III is from 8.5 m to the end of the borehole at
weak rock and rockburst in hard and strong rock (Fig. 2). It was the depth of approximately 180 m. The discontinuities in this zone
observed in some metal mines in Sweden that strain burst usually are believed to be mainly of geological origin. The RQD of the cores
occurred below a depth of 600 m and became intensive below in Zone III is significantly higher than the other two zones, which
1000 m. Rock failure is unavoidable in high stress conditions. The implies that Zone III is out of the disturbance distance of the drift. On
task of rock support at depth is not to equilibrate the deadweight the basis of the variation of the fracture intensity, it is inferred that
force of loosened blocks but to prevent the failed rock from disin- Zone I was the failure zone, where the rock failed either in shear or
tegration. In high stress rock masses, the support system must be in tension and the tangential stress was partially reduced, while the
not only strong but also deformable in order to deal with either tangential stress in Zone II was elevated but the rock had not yet
stress-induced rock squeezing in soft and weak rock or rockburst in fractured after excavation of the drift. Zone II was the position of the
hard and strong rock. natural pressure arch that carried the ground pressure and func-
tioned as a protection shield over the drift.
2.3. Suitable rockbolt types To illustrate the failure zone surrounding an underground
opening, numerical modelling was conducted for a horseshoe-
The suitable types of rockbolts for a given rock mass are asso- shaped tunnel of 6 m in width and 6 m in height, excavated in a
ciated with the loading condition in the rock mass. In the case of a rock mass subjected to hydrostatic in situ stresses. The in situ
398 C.C. Li / Journal of Rock Mechanics and Geotechnical Engineering 9 (2017) 396e414

Fig. 2. Loading conditions to rockbolts in high stress rock masses. (a) Rock squeezing, (b) strain burst, and (c) fault-slip burst.

Fig. 3. Cores drilled in the wall of a mine drift at a depth of 1000 m (Li, 2006a).

stresses are assumed to be s1 ¼ s2 ¼ s3 ¼ 30 MPa in the simulation underground opening, where the tangential stresses are signifi-
and the rock mass obeys the Mohr-Coulomb failure criterion with cantly elevated. This is the so-called natural pressure arch, sketched
cohesion c ¼ 5 MPa and internal friction angle f ¼ 35 . The in Fig. 5. The concept of the natural pressure arch was used for rock
constitutive model of the material is elastic and perfectly plastic, i.e. support design by, among others, Wright (1973), Krauland (1983)
the residual strength of the material is equal to the peak strength. and Li (2006b).
Fig. 4 shows the distribution of the major principal stresses that are
oriented approximately parallel with the tunnel walls and roof, i.e. 3.2. Design methodology
in the tangential directions, after excavation. The immediate sur-
rounding rock, approximately 2 m deep in the walls, fails. Beyond Rock support refers in general to any measure aiming to stabilise
that depth, the rock is still intact but the tangential stress has been rock masses by using support elements. Support elements may be
elevated somewhat, depending on the distance to the tunnel wall. It rockbolts, cables, meshes, straps, lacing, shotcrete (i.e. sprayed
reaches its maximum at a depth of about 3 m and then gradually concrete), thin liners, steel sets, shotcrete arches and cast concrete
drops to the in situ stress level (30 MPa) in locations far away from lining. A support system provides three primary functions: rein-
the tunnel. The rock portion within which the tangential stress is forcement, holding and retention (Kaiser et al., 1996). Reinforce-
significantly elevated carries the majority of the ground pressure ment refers to strengthening of the rock mass; holding to the
and forms a protection shield, i.e. a pressure arch, around the suspension of potentially loosened blocks; and retention to
tunnel. confinement of the exposed rock surfaces. Each support element
Based on the core logging shown in Fig. 3 and the numerical may perform one or more of the three primary functions. Rein-
modelling shown in Fig. 4, it can be deduced that a pressure arch (or forcement is usually achieved by installing rockbolts systematically.
ring) exists at a certain depth of the rock surrounding an The increase in the rock strength due to bolting is very limited.
 C.C. Li / Journal of Rock Mechanics and Geotechnical Engineering 9 (2017) 396e414 399

Fig. 4. Distribution of the major principal stresses in the rock surrounding a tunnel. The crosses and circles mark the zone of rock failure.

Fig. 5. A sketch illustrating the natural pressure arch surrounding an underground opening.

Assume that the load capacity of a rockbolt is 200 kN and rockbolts mainly achieved by shotcrete, mesh or other types of thin liners laid
are installed with a pattern of 1 m  1 m. The maximum confining on the rock surface in mines. In civil tunnels, the allowable rock
pressure that the rockbolts can provide is 0.2 MPa. The increase in deformation is much smaller than that in mines. Therefore, heavy
the rock strength by this confining pressure, according to the external support structures such as steel sets, concrete arches and
MohreCoulomb criterion, may be in the range of 1e2 MPa, which is even cast concrete lining are applied to restrain wall deformation.
significantly lower than the inherent strength of the rock mass. The Those structures are set up on rock surfaces, but they are similar to
essential function of bolting is to keep the fractured rock together cables installed within the rock mass in the sense that they provide
to form a pressure arch around the opened space. In other words, a holding function. A rock support system may be composed of one,
the bolts help the rock to strengthen and support itself. Rockbolts or more than one, of the following support layers, depending on the
also provide a holding function to loosened blocks and fractured loading condition and the extent of rock failure (Li, 2012):
rock. In the case of a large failure zone, rockbolts may be entirely
located within the failure zone. The use of long cablebolts is an (1) Layer 1 e Bolting: Rockbolts are installed sporadically or
option to provide an effective holding function. Surface retaining is systematically.
400 C.C. Li / Journal of Rock Mechanics and Geotechnical Engineering 9 (2017) 396e414

(2) Layer 2 e Surface retaining: Retaining elements like meshes, spaced ductile rockbolts “Split sets” or point-anchored resin bolts
straps, lacing, thin liners, shotcrete and cast concrete lining of 2.4e3 m in length. The bolt-reinforced zone is then nailed to the
are installed on the rock surface. competent strata behind the failure zone with cables (see Fig. 7).
(3) Layer 3 e Cable bolting: Single- or multi-strand cables are The tightly spaced rockbolts help the fractured rock to build up an
installed into the competent rock behind the failure zone. artificial load-bearing arch and the cablebolts integrate the arch
(4) Layer 4 e External support: Structure elements, including with the deeply located stable strata. The rock surface is retained
steel sets, concrete arches, invert, cast concrete lining and with mesh, straps and mesh shotcrete. The Agnew gold mine is
thick shotcrete liners, are set up in tunnels. located in Western Australia. Rock squeezing is the major instability
problem in the mine below the depth of 500 m because of the weak
In shallow tunnels, or in the case of a small failure zone, the task footwall rocks, ultramafic conglomerates, chlorite and talc. The
of the rock support is to hold the loosened or failed rock blocks. wallewall convergence of a 5 m  5 m mine drive reached 400 mm
Support layer 1, i.e. spot bolting or sparsely spaced pattern bolting, in 5 months. The rockbolts used in the mine were either Split sets or
may be good enough to stabilise the rock mass (Fig. 6a). The bolts point-anchored resin Posimix bolts, 2.4e3 m in length and 1 m in
should be installed into the natural pressure arch behind the failure spacing. The twin-strand cablebolts used were 6 m in length and
zone. In poor rocks, the failure zone may be so extensive that 2 m in spacing.
rockbolts cannot reach the natural pressure arch. A support system In burst-prone rock, the support system is composed of 2.4-m or
composed of layers 1 and 2 is then needed (Fig. 6b). Layer 1, in this 3-m long energy-absorbing rockbolts and meshes or fibre/mesh-
case, must be tightly spaced pattern bolting, which builds up an reinforced shotcrete. It is claimed that such a system can sustain
artificial pressure arch in the failure zone. The artificial pressure arch an ejection velocity of 5 m/s and absorb a dynamic energy of 35 kJ/m2
forms a protection shield over the opening. In extremely poor rock, (Slade and Ascott, 2007).
support layer 3, cable bolting, needs to be added to the system. The
1-2-3-layer support system is often used in deep mines and un- 3.2.2. Canadian and Scandinavian methodology
derground caverns of large span, such as underground machine The methodology of rockbolting in Canada and Scandinavian
halls in hydropower plants. This support system is characterised by countries is to integrate the failed rock with short bolts in
its flexibility in order to adapt to the prevailing rock condition. In conjunction with meshes or/and fibre/mesh-reinforced shotcrete
civil tunnels, cable bolting is less used than in mining excavations. (Fig. 8).
Instead, external support structures (layer 4), such as steel sets and In Canadian metal mines, the types of rockbolts are 2.4-m or 2.1-
concrete lining, are employed to provide the holding function. A 1- m long fully resin-encapsulated rebar and Split sets. Energy-
2-4-layer support system is preferred in civil tunnels. absorbing rockbolts are added in case of seismic rock conditions.
Pattern bolting plays a crucial role in a support system. Tightly The Craig mine, in the region of Sudbury, Ontario, is characterised
spaced bolts constrain the failed rock so that an artificial pressure by the faults going through the ore body, which creates a number of
arch is established in the failure zone. The load-bearing capacity of fault-slip seismic events in the mining operation areas below the
an artificial pressure arch was visually demonstrated by Lang depth of 1600 m. In 5-m span drives, the 1.8e2.4 m long fully resin-
(1961) in the 1960s and also recently by Hoek (2007). Li (2006b) encapsulated rebar bolts were used together with surface retaining
reported an example of applying the concept of an artificial pres- support elements (meshes, straps, shotcrete, etc.). In burst-prone
sure arch for rock support design. areas, rebar bolts and modified cone bolts were installed plus 6-
m long cables.
3.2.1. Australian methodology In Swedish mines, fully encapsulated rebar bolts with cemen-
In squeezing rock, the methodology of rockbolting in Australian titious grout are most often used. Split sets are seldom used in
mines is to reinforce the failure zone of the rock using tightly Scandinavian mines, but energy-absorbing rockbolts, such as D-

Fig. 6. Principles of rockbolting in different rock conditions of rock failure: (a) for a limited failure zone, and (b) for a vast failure zone.
 C.C. Li / Journal of Rock Mechanics and Geotechnical Engineering 9 (2017) 396e414 401

Fig. 7. The Australian methodology of rockbolting in deep metal mines.

Fig. 8. The Canadian and Scandinavian methodology of rockbolting in deep metal mines.

bolts, have been used for dynamic rock support, for instance, in the absorbing rockbolts and surface retaining elements. It is thought
iron ore mines in Sweden. The surface retaining liners are usually that the kinetic energy in a rockburst event is partially absorbed by
steel-fibre-reinforced shotcrete, but in burst-prone conditions, wire the rockbolts and partially dissipated by fragmentation of the rock
meshes are laid on the top of the steel-fibre-reinforced shotcrete. contained by the surface retaining elements. Mesh and lacing are
The bolt length is typically 2.7e3 m and the bolt spacing is 1e1.5 m. often used in South African dynamic support systems.
The Kristineberg metal mine in Sweden operates the mining ac- In South Africa, mine drifts excavated in high stress rock are
tivity in depths below 1000 m at present. The mine is subjected to typically supported with energy-absorbing rockbolts (e.g. cone
rock squeezing in zinc-and-lead ore bodies owing to the chlorite/ bolts and Durabar) and ductile bolts (e.g. split set and cables). The
talk-rich rocks, but was subjected to strain burst in the hard primary bolting method is a ring of 1.2-m long bolts and the sec-
quartzite of the gold ore bodies. The rockbolts used in the mine ondary bolting is a ring of 2.4-m long bolts plus lacing, meshing
have been 2.7-m long fully cement-grouted rebars with a bolting and 50 mm thick steel or polyester fibre shotcrete (Fig. 9). The
pattern of 1.2 m  1.2 m. bolting pattern is typically 1 m  1 m. The Mponeng gold mine in
South Africa operates its mining activity at depths below 2500 m.
3.2.3. South African methodology The dominant rock is quartzite that is burst-prone under high
In burst-prone deep mines in South Africa, the methodology of ground pressure. Rockbolt is only one of a number of ground
rockbolting is to dissipate the released kinetic energy with energy- support elements used in the mine. Split sets, 1.2 m long, are used
402 C.C. Li / Journal of Rock Mechanics and Geotechnical Engineering 9 (2017) 396e414

Fig. 9. The South African methodology of rockbolting in deep metal mines.

in mining stopes and Durabars, 2.4e3.6 m long, are used in In practise, the bolt pattern in systematic bolting is such that the
transport drives. in-row spacing and the distance between rows are equal. The bolt
spacing, s, is recommended to be in the range from 1 m to 2.5 m.
3.3. Bolt length and spacing However, rock joint spacing should be also taken into account
when the bolt spacing is determined. A rule of thumb is to set the
Determination of bolt length and spacing has been a topic for bolt spacing equal to 3e4 times the mean joint spacing in the case
discussion probably since rockbolts were first used for ground of a mean joint spacing in the range of 0.3e1 m, i.e.
support in underground excavations (e.g. Panek, 1964; Coates and
Cochrane, 1970; Lang, 1972; Barton et al., 1974; Schach et al., s ¼ ð3  4Þe (4)
1979; Farmer and Shelton, 1980; Crawford et al., 1985; Stillborg,
1994). Choquet and Hadjigeorgiou (1993) provided a review on where e represents the mean joint spacing.
this topic in their paper on the design of ground support. The In the case of a vast failure zone (Fig. 6b), the Norwegian Road
following presented are the principles for the determination of bolt Authority recommends the use of relatively short tensioned rock-
length and spacing that are used in the practise of rockbolting to bolts to establish an artificial pressure arch in the failure zone (see
date. From the point of view of operation, the bolt length should be Fig. 10b). The bolt length is still estimated using Eq. (3), but the bolt
less than half of the opening height for roof bolts and half of the spacing is recommended to be smaller than 3 times the mean joint
span for wall bolts in order to avoid installation difficulties: spacing, i.e.

Lb  0:5H ðfor roof boltsÞ (1a) s  3e (5)

For rockbolting aiming at the construction of an artificial pres-
Lb  0:5B ðfor wall boltsÞ (1b) sure arch, it is required that the rockbolts interact with each other
and an interaction zone is formed in the bolt-reinforced rock party
where Lb represents the bolt length, H is the opening height, and B (Fig. 11). Assuming that the reinforcement angle of a single rockbolt
is the opening span. The bolt length is also associated with the is 90 , the thickness of the interaction zone, t, is related to the bolt
bolting principle. In the case that the failure zone is limited to a length (Lb) and spacing (s) as follows:
relatively small depth (Fig. 6a), the bolt length should be at least
1 m longer than the depth of the failure zone, i.e. t ¼ Lb  s (6)

Lb  df þ 1 (2) The bolt length is usually short, 2e3 m, in this type of rock-
bolting. The thickness of the interaction zone is required to be at
where df is the depth of the failure zone. In the case of a vast failure least 0.5Lb in order that a strong enough artificial arch can be
zone (Fig. 6b), the bolt length is short, varying from 2 m to 3 m, but established in the broken rock. This requirement leads to a bolt
its upper limit is governed by Eq. (1). spacing that should be less than half of the bolt length, i.e.
For tunnels excavated in moderately jointed hard rock masses,
the Norwegian Road Authority proposed the following formula to s  Lb =2 (7)
determine the length of un-tensioned bolts in the central section of In the design stage of an underground rock excavation, bolt
the tunnel for the purpose of suspending the failure zone on the length and spacing are often determined with the help of empirical
natural arch (Statens vegvesen, 2000) (see Fig. 10a): methods recommended in various rock mass classification systems.
In the Q rock mass classification system (Barton et al., 1974), the bolt
Lb ¼ 1:4 þ 0:184B (3)
length and spacing can be found in a chart based on the Q-value of
 C.C. Li / Journal of Rock Mechanics and Geotechnical Engineering 9 (2017) 396e414 403

this load-controlled condition, the factor of safety (FS) for rock-

bolting is defined as

Load capacity of the bolts

FS ¼ (8)
Total load on the bolts
In this case, a safe rock support requires that the load on the bolt
is less than the strength of the bolt, i.e. FS > 1. It is required that the
factor of safety is in the range of 1.5e3 in rockbolting design.

3.4.2. Factor of safety in squeezing rock

Rock deformation can be significantly large in tunnels excavated
in highly stressed soft and weak rock because of vast rock failure.
The essential driving power for the rock deformation is the strain
energy released from the rock mass after excavation. The greater
part of the released strain energy is dissipated in rock fracturing,
which in turn brings about rock deformation. In extremely poor
rock conditions, the large rock deformation may lead to rock
collapse. The response of the rock mass during excavation is
described by the ground response curve (GRC) (see Fig. 12). Yield
rockbolts work better than stiff ones in squeezing rock. Yield
rockbolts deform together with the rock mass until the rock mass
becomes stable after a certain amount of displacement. The dashed
line in Fig. 12 represents the GRC of the rock mass after being
reinforced with yield rockbolts. In squeezing rock conditions, one
cannot find a constant load on the rock support since the support
load and the displacement are correlated. It is thus not possible to
use Eq. (8) to calculate a factor of safety. In squeezing rock, it is more
relevant to define the factor of safety with displacements rather
than load and strength. It is required, from the point of view of
stability, that the displacement of the tunnel wall at equilibrium,
ueq, has to be smaller than the critical displacement, uc, beyond
which uncontrollable rock collapse would occur. The factor of
safety for the rock support, FS, is thus defined as

uc
FS ¼ (9)
ueq
It must be pointed out that the critical displacement uc is
difficult to be quantified even with the help of numerical model-
ling. Beyond displacement uc, the rock mass becomes unstable and
calculation iterations would become non-convergent in numerical
modelling. In engineering practise, there usually exists a
maximum allowable displacement from the point of view of
operation. For example, the radial displacement of a TBM (tunnel
Fig. 10. Rockbolting methods in two different rock conditions: (a) suspension of the
failure zone to the natural arch and (b) establishment of an artificial arch within the
boring machine) tunnel usually is not allowed to be larger than
failure zone. Modified after Stillborg (1994). 150 mm in order to avoid jamming of the TBM head. Thus, there is
another factor of safety for the operation, denoted as FSop, which is
calculated as

the rock mass and a geometrical parameter called the equivalent umax
FSop ¼ (10)
dimension (Barton and Grimstad, 2014). The equivalent dimension ueq
is defined by the span of the excavation and a coefficient describing
the intended use of the excavation (road tunnel, underground where umax is the maximum allowable displacement. In addition to
station, etc.). In the rock mass rating (RMR) system by Bieniawski the factors of safety for stability and operation, it is also required
(1989), bolt length and spacing, as well as other types of support that the rock displacement at equilibrium, ueq, must be smaller
measures, are empirically recommended in a table for five classes than the ultimate displacement uult of the rockbolt to avoid pre-
of rock mass quality. mature failure of the rockbolt. To the end, the items in the criteria
for a rock support system using rockbolts are as follows:

3.4. Factor of safety

uc 9
FS ¼ >1 >
>
ueq >
>
3.4.1. Factor of safety for gravitational rock falls >
=
As mentioned in Section 2, rock blocks in the roof may become umax (11)
loosened in shallow tunnels where in situ rock stresses are low. The FSop ¼ > 1>
>
ueq >
>
loosened blocks tend to fall under gravity. The load exerted on the >
;
ueq < uult
rockbolts is equal to the deadweight force of the falling blocks. In
404 C.C. Li / Journal of Rock Mechanics and Geotechnical Engineering 9 (2017) 396e414

Fig. 11. Reinforcement interaction between rockbolts.

3.4.3. Factor of safety in burst-prone rock 9

In highly stressed rock masses, a portion of the strain energy nEab >
>
FS ¼ >1 >
>
stored in the rock mass may be released suddenly, leading to Eej >
>
=
rockburst events. Use of energy-absorbing rockbolts is one of the umax (14)
most effective means to support burst-prone rock masses. The FSop ¼ > 1>
>
ueq >
>
>
>
support principle in this case is that the energy absorption capacity ;
ueq < uult
of the rockbolts must be higher than the kinetic energy of the
ejected rock. The factor of safety for dynamic rock support needs to
be calculated with the energy absorption capacity of the rockbolts
and the energy released in the rockburst event:
3.5. Compatibility between support elements

The current methodology of rock support in civil tunnels is to

nEab install fully encapsulated stiff rockbolts in the rock and to apply
FS ¼ (12)
Eej shotcrete or cast-in concrete lining on the rock surface. Yield sup-
port elements may be imbedded in the lining in squeezing rock
where n is the number of rockbolts; Eab is the energy absorption of conditions (Schubert, 2001; Li, 2012). Rock support systems in civil
each bolt; and Eej is the kinetic energy of the ejected rock, which is tunnels are in principle composed of stiff internal elements (fully
expressed as encapsulated rebar bolts) and yield external elements (deforma-
tion-compensated concrete lining), which are conceptually
sketched in Fig. 14a. In such a support system, the stiff internal
1 elements (rockbolts) may fail after a small deformation, but the
Eej ¼ mV 2 (13) external elements (the concrete lining) can accommodate relatively
2
large rock deformation because of the embedded yield elements.
where m is the mass of the ejected rock, and V is the ejection ve- The internal and external elements in the system are thus not
locity. The ejection velocity may be estimated according to the compatible in deformation. In underground mining, yield rockbolts
horizontally dislodged distance of the ejected rock in case that the and meshes usually are used to deal with excessive rock defor-
burst occurs in walls (Kaiser et al., 1996). Rockburst, however, often mation. The support load is mainly carried by the rockbolts while
occurs in tunnel roof and also in floor (Zhang et al., 2012). In such the mesh restrains the dilation of the rock surface. Fig. 14b is a
cases it is not possible to estimate the ejection velocity by this conceptual sketch of this type of support system. In such a support
means since the horizontally dislodged distance is zero. A high system, the internal elements (rockbolts) and the external elements
ejection velocity may elevate the broken degree of the rock pile. (meshes) are compatible in deformation, but the load-bearing ca-
Thus, one may empirically estimate the ejection velocity based on pacity of the meshes is very low.
the fragmentation of the ejected rock in the case of roof rockburst. In a satisfactory rock support system, both internal and external
With a competent support system, the ejected rock will stop elements should be both strong and deformable. In other words,
moving after a displacement of ueq (Fig. 13). As mentioned above, they should be compatible both in load and deformation capacities
the displacement ueq has to be smaller than the maximum allow- in order to achieve the optimum reinforcement effect. The behav-
able displacement, umax, in order to avoid operational difficulties. iours of the internal and external support elements in such a sys-
The criteria for dynamic rockbolting design are tem are sketched in Fig. 15.
 C.C. Li / Journal of Rock Mechanics and Geotechnical Engineering 9 (2017) 396e414 405

Fig. 12. The ground response curve (GRC) and the support characteristic curve of yield rockbolts.

neglected, where fa is the apparent friction angle of the disconti-

nuity. However, the gravity of the roof wedge block does play a role
in the stability of the block. The stability of a symmetric wedge in
the roof of a tunnel has been studied by many researchers, for
instance, Crawford and Bray (1983), Shi and Goodman (1983),
Sofianos (1986), and Nomikos et al. (2002, 2006). In the
following, a simple case without considering the stiffness of rock
joints is considered to demonstrate the influence of the size of the
block on the critical dip angle when the gravity is taken into ac-
count. Consider a longitudinal wedge block, which is formed by two
discontinuities with the strikes parallel to the tunnel axis as well as
the roof surface, as an example to understand the influences of the
Fig. 13. The equilibrium displacement ueq and the maximum allowable displacement size of the block and the horizontally oriented tangential rock stress
umax related to a rockburst event.
sq (Fig. 17). The loads on the wedge block are the tangential rock
stress sq, the weight force of the block and the frictional resistance
on the sides of the block. The block tends to fall under gravity, but
4. Types of rockbolting the friction on the sides tends to prohibit the fall. The requirement
for stabilising the block is that the frictional resistance on the
4.1. Wedge block in roof discontinuity planes is higher than the downward shear force on
the planes. Taking into account all the forces exerted on the two
4.1.1. Stability analysis discontinuity planes, the stability condition for the block is ob-
Consider a horizontal overhanging rock face intersected by three tained as follows:
planar discontinuities. A tetrahedral wedge block is formed by the
three discontinuities and the rock face, as shown in Fig. 16. This 2sq  rgh
tan fa > sinð2aÞ þ cosð2aÞtan f (15)
block is kinematically feasible and tends to fall under gravity. The 2sq þ rgh
horizontal in situ rock stress is helpful in stabilising the block.
Slippage along a discontinuity is prohibited when the dip angle of where r is the density of the rock, g is the gravitational acceleration,
the discontinuity plane to the hanging rock face is larger than a h is the block height, and a is the dip angle of the discontinuities.
critical dip angle equal to (90 fa) when the weight of the block is The critical dip angle can be found by letting the two sides of Eq. (15)
406 C.C. Li / Journal of Rock Mechanics and Geotechnical Engineering 9 (2017) 396e414

Fig. 14. Sketches illustrating incompatible rock support systems: (a) in civil tunnelling and (b) in mining.

be equal. Given r ¼ 2700 kg/m3, g ¼ 10 m/s2 and f ¼ 35 , the critical reinforcement force contributed by the bolts is T ¼ St where t is the
dip angle is related to the tangential rock stress sq and the block force in a single rockbolt. All forces exerted on the block are illus-
height h, as shown in Fig. 18. The critical dip angle decreases with an trated in Fig. 19b. They are the gravitational force acting on the
increase in the tangential rock stress and approaches 55 , which is block, Wg, the total bolt force, T, the normal reaction force on the
the critical dip angle without consideration of gravity. The critical dip sliding plane, N, and the shear resistant force on the sliding plane, R.
angle increases with a decrease in the tangential rock stress, All forces are in equilibrium in all directions in the critical state at
implying that only steeply dipped wedge blocks could be stabilised which shear failure occurs along the sliding plane. By equilibrating
by the friction on the block sides in low stress conditions. The critical the forces, it is obtained that the normal force is expressed as
dip angle changes abruptly with a small change in the tangential rock
stress below 1 MPa, but it is not very sensitive to the rock stress N ¼ Wg cos j þ T sin a (17)
above 2 MPa. As shown by the three curves corresponding to
different block heights, the critical dip angle also increases with the where j is the dip angle of the sliding plane. The shear resistant
block size, particularly when the tangential rock stress is lower than force is expressed, with an assumption that the MohreCoulomb
1 MPa. The critical dip angle increases slightly with the block size criterion prevails along the sliding plane, as
when the rock stress is higher than 2 MPa. This implies that the
critical dip angle is not sensitive to block size when the tangential R ¼ cA þ ðWg cos j þ T sin aÞtan fa (18)
rock stress is high.
where A is the base area of the sliding plane. The driving shear
4.1.2. Bolting force, D, is obtained as
The load on rockbolts that are used to stabilise a rock block in
the roof is the deadweight force of the block (see Fig. 1). The D ¼ Wg sin j  T cos a (19)
number of bolts needed, Nbolt, is approximately calculated as The factor of safety of the wedge block is defined as the ratio of
the shear resistant force to the driving shear force, i.e.
Wg
Nbolt ¼ FS (16)
Pult
R cA þ ðWg cos j þ T sin aÞtan fa
FS ¼ ¼ (20)
where W is the deadweight force of the block, and Pult is the ulti- D Wg sin j  T cos a
mate load of the rockbolt. In the case of fully grouted rockbolts, the A factor of safety less than 1, i.e. FS  1, means that sliding oc-
embedment length of the bolts in the stable formation must be at curs, while the block is stable when FS > 1. A factor of safety in the
least 1 m. range of 1.5e2 is usually used for rockbolting design.
The bolt force T contributes to an increase in the normal force
4.2. Wedge block in wall and a component of the shear force. The increase in the normal
force is always positive in enhancing the frictional resistance of the
The principle of stabilising a wedge block in the wall was pre- sliding plane, but the contribution of the bolt force to the shear
sented, for instance, by Hoek and Brown (1980) and Harrison and force is either positive or negative, depending on the installation
Hudson (2000). It is demonstrated in this section through the angle, a. There exists a theoretical critical installation angle,
example illustrated in Fig. 19a. Assume that a wedge block is denoted as ac, at which the bolts most effectively reinforce the
formed in the wall and it tends to slide along the lower disconti- block. Let FS ¼ 1 in Eq. (20), representing the equilibrium state
nuity under gravity. The block is stabilised with bolts which are when shear failure is initiated along the sliding plane. The bolt force
installed with an angle of a to the discontinuity plane. The total is expressed at this moment as
 C.C. Li / Journal of Rock Mechanics and Geotechnical Engineering 9 (2017) 396e414 407

expression above with respect to a and letting the differentiation be

equal to zero, the critical installation angle is obtained as

ac ¼ fa (22)
In other words, the reinforcement effect of the rockbolts is op-
timum when they are installed at an angle of a ¼ fa.

4.3. Arching bolting

The concept of a natural pressure arch is further explained

through the arching of two blocks, as illustrated in Fig. 20. Assume
that the ceiling of an opening excavated in a laminated rock mass is
composed of two blocks formed by three transverse fractures in the
ceiling stratum. The downward movement of the two blocks at the
abutments is prohibited owing to the friction on the fracture
planes. The two blocks are then forced to rotate under gravity and
press each other at the upper part of the middle fracture plane and
the lower parts at the abutments. A pressure arch is thus formed
within the two blocks and the blocks are stabilised.
The natural pressure arch is located far from the ceiling of the
underground opening in the case of a vast failure zone around the
Fig. 15. A sketch illustrating the concept of a compatible rock support system. opening after excavation. In this case, one can consider construct-
ing an artificial pressure arch within the failure zone to prevent the
failed rock from falling. As demonstrated in physical models by
Lang (1961) and also by Hoek (2007), an artificial pressure arch may
be formed in the interaction zone of systematically installed rock-
bolts (see Fig. 21). The load-bearing capacity of such a pressure arch
can be estimated by (Krauland, 1983; Sinha, 1989):
 2
t
smax ¼ ksc (23)
B

where smax is the maximum ground pressure that the pressure arch
can bear, and sc is the uniaxial compressive strength (UCS) of the
bolt-reinforced rock party. The coefficient k is proportional to the
moment arm length in the pressure arch. Wright (1973) found that
k is approximately 0.9 based on back-calculations of his experi-
Fig. 16. A kinematic wedge bounded by three planar geological discontinuities and a mental data.
horizontal overhanging rock face.

4.4. Tieback bolting

Tieback bolting is usually used to reinforce rock pillars. With

tieback bolting, the bolts, usually equally spaced, go through the
entire width of the pillar and are pre-tensioned with a relatively
high load (see Fig. 22). It should be noted that the purpose of
tieback bolting is not to enhance the peak strength of the pillar to
avoid failure but to prevent the pillar from disintegration in the
post-failure stage. In accordance with the MohreCoulomb crite-
rion, the increase in the strength of the rock by the confining
pressure s3 is expressed as
 
f
Ds1 ¼ s3 tan2 45 þ (24)
2
Take as an example a pillar reinforced by systematic tieback
Fig. 17. A wedge block formed in the tunnel roof. bolting with a bolt spacing of 1 m. The ultimate load of the bolt is
200 kN and thus the maximum confining stress to the rock by the
bolts is s3 ¼ 0.2 MPa. Assume that the peak and residual internal
friction angles are 40 and 30 , respectively. It is obtained from Eq. (24)
sin j  cos j tan fa that the increased peak strength is 0.9 MPa and the increased
T ¼ Wg (21) residual strength is 0.6 MPa. The UCS of the rock is usually higher
cos a þ sin a tan fa
than 50 MPa. An increase of 0.9 MPa is negligible compared to the
The bolt force needed to equilibrate the other forces on the block inherent strength of the rock. However, the unconfined residual
at this moment is minimum when vT=va ¼ 0. Differentiating the strength of the rock is low in most types of rocks. Therefore, an increase
408 C.C. Li / Journal of Rock Mechanics and Geotechnical Engineering 9 (2017) 396e414

90°

Fig. 18. The critical dip angle versus the tangential rock stress in the rock for three different block sizes.

Fig. 19. Use of rockbolts to stabilise a wedge block that tends to slide along the lower discontinuity plane. (a) The block and the rockbolts, and (b) the forces on the wedge block.

of 0.6 MPa in the residual strength could significantly improve the post- cavern. The pillar was subjected to extension fracture during
failure behaviour of the pillar. excavation. Cablebolts with a load capacity of 1 MN were installed
Fig. 23 shows the tieback bolting of a 6 m  8 m (width  depth) across the pillar with a spacing of 2 m. They were pre-tensioned to
pillar between two niches excavated in the wall of a hydropower 400 kN after installation. A concrete lining of 300 mm in thickness
was cast on both sides of the pillar to improve the load transfer
from the cablebolts to the pillar. Strong square plates of
Pressure arch 200 mm  200 mm were attached to the cablebolts. The maximum
confining pressure that the cablebolts can provide is approximately
1 MN/(2 m  2 m) ¼ 0.25 MPa. The increased strength of the pillar
is estimated to be 0.75 MPa, corresponding to an increase of the
load capacity of 36 MN for the 6 m  8 m pillar (assuming an in-
ternal friction angle of 30 ).

4.5. Suspension bolting

In some cases, often in coal mines, a weak layer of formation is

exposed on the roof of mine drifts (Krauland, 1983). This layer can
be secured by hanging it to the stable stratum behind with rock-
bolts (Fig. 24). The weak layer is loaded by its own weight so that
Fig. 20. Pressure arch formed in two ceiling blocks. the bolting design is simply based on the thickness of the layer and
 C.C. Li / Journal of Rock Mechanics and Geotechnical Engineering 9 (2017) 396e414 409

Fig. 21. Pressure arch formed in a bolt-reinforced roof.

the bolt spacing. The required ultimate load capacity of the 4.6. Rockbolting in large-scale caverns
rockbolt is
Rock failure may extend to a significant depth after excavation
of a large-scale underground cavern. Rockbolts and cablebolts play
Pult ¼ FSðlsC rgÞ (25)
crucial roles in the support system in this case because other
where l is the thickness of the weak layer, and C is the bolt spacing support elements, such as lining, shotcrete and mesh, can only
between rows. The minimum bolt length, Lmin, is passively respond to the rock deformation and provide very
limited effective support to the rock mass. When the rock mass
quality is poor and the in situ rock stresses are high, the size of the
Lmin ¼ l þ anchorage length (26) failure zone could be so vast that it is beyond the bolt-reinforced
The anchorage length in the stable stratum should be at least zone and the reinforced rock party continues to move toward
1 m in the case of fully encapsulated bolting. The principle is that it the opening. The principle of rockbolting in this situation is to
must be longer than the critical embedment length with a factor of reinforce the rock with tightly spaced and fully grouted rockbolts
safety of 2e4. to a relatively shallow depth (3e7 m) in combination with long
cablebolts (10e25 m). The short rockbolts aim to build up a bolt-
rock ‘shield’ surrounding the opening and the long cablebolts
suspend the ‘shield’ to competent and stable rock formations

Fig. 23. Tieback bolting in a hydropower cavern. Some of the pillar-through cablebolts
Fig. 22. Tieback bolting. are shown in the picture (Photo by J. Mierzejewski).
410 C.C. Li / Journal of Rock Mechanics and Geotechnical Engineering 9 (2017) 396e414

Fig. 24. Suspension bolting.

behind the failure zone. It is required that the cablebolts must be approximately dip toward the upstream with an angle of 30 . The
able to tolerate a significant displacement in order that they in situ horizontal stress in the rock mass is slightly higher than
function properly without premature failure. A common practise the vertical stress. The final wallewall convergence of the cavern
to enhance the deformability of a cablebolt is to de-bond its middle is expected to be 200e300 mm. The 7-m long short rockbolts,
portion from the grout with PVC pipes or other types of soft with load capacity of 300 kN, are fully grouted in the boreholes.
materials. The short bolts are spaced 1 m  1 m in the crown and
Fig. 25 shows the rockbolting design for a hydropower cavern 1.3 m  1.3 m in the walls. The load capacity of every cablebolt is
located at a depth of approximately 400 m. The cavern is 25 m in 1 MN. In every bolting profile, the cablebolts are spaced
span, 45 m in height and 120 m long. The main rock types are 4 m  4 m in the crown with the three in the middle being 15 m
sandstone (UCS of 80e150 MPa), siltstone (UCS of 40e60 MPa) long and the rest 10 m long. The cablebolts in the walls are 15 m
and mudstone (UCS of 20e50 MPa). The bedding planes long and spaced 5 m  5 m. Three 20-m cablebolts are installed in

Fig. 25. An example of rockbolting and cablebolting in a large-scale underground hydropower cavern.
 C.C. Li / Journal of Rock Mechanics and Geotechnical Engineering 9 (2017) 396e414 411

the lower portion of the wall on the downstream side because of Table 1
the risk that the rock mass there may be weakened by excavations Specification of forepoles (Bang, 1984; Ocak, 2008; Volkmann and Schubert, 2009;
DSI, 2015; Hoek, 2015).
nearby.
Type of forepoles Diameter (mm) Length (m) Spacing (m) Angle (o)

4.7. Forepoling Solid rebar spiles 20e50 4.6e6.1 0.5e1.5 6e10

Steel pipes 38e200 9e15 0.3e0.6 6e10
Self-drilling spiles 32e51 e e 6e10
In unstable ground, the crown of a tunnel can be pre-supported
by driving poles, pipes and planks into the crown ahead of the
excavation face with a small inclination angle upward. This is the
so-called forepoling reinforcement (see Fig. 26). The pole types Rockbolt
2a
range from rebar spiles, self-drilling spiles, pipe spiles and planks. 1
Rebar spiles consist of smooth or ribbed solid steel bars. They are Forepole 3
installed in pre-drilled stable boreholes with cementitious grout in
blocky and jointed rock mass against falling blocks, as well as
rammed into soft, homogenous soil to prevent loosening of the soil.
Pipe spiles consist of steel tubes. They are installed in pre-drilled
Arch
stable boreholes in blocky or jointed rock mass as well as rammed
2b
into soft, homogenous foundation soil. The hollow hole of the pipes
and the gap between the pipes and the ground can be grouted to
achieve a better load transfer.
Self-drilling spiles are simply the hollow drill rods that are left in
the strata after hole drilling. The drill bit is either left in the strata or
retrieved from the inner hole of the rod. Self-drilling spiles are 1 Pole length ahead of the face
suitable for extremely poor rock masses and consolidated but weak 2 Bearings of the poles. 2a – rockbolts, 2b – arches
soils. They are installed with conventional drill booms using rotary 3 Overlap
percussive drilling.
Forepoling planks are particularly suitable for unstable, non- Fig. 27. Longitudinal profile of forepoling.

cohesive soil. One product of such a plank is 1.25e3 m in length

and approximately 220 mm in width and is made of steel plate of
as desired (see Fig. 27). Firstly, the toe of the poles must be in a few
3e6 mm in thickness (DSI, 2015). The plank is rammed into the soil
metres ahead of the excavation face; secondly, the near-end of the
using hydraulic drifters.
poles must be supported by rockbolts, arches or both; and thirdly,
Forepoling with solid bars is usually applied after each excava-
the poles must be overlapped. The poles must have at least two
tion round with a length seldom beyond 6 m. Forepoling with
support positions, one being the face rock and the other being the
pipes, or a so-called pipe umbrella, is installed to support several
rockbolts and arches at the near end. Additional support arches are
consecutive excavation rounds. The pipe length is usually 12 m or
needed when the pole length is much longer than the advance
15 m (Volkmann and Schubert, 2009). For pre-reinforcement of a
length. Support arches can be lattice girders, steel sets, shotcrete,
12-m long tunnel excavated in a weak rock mass, Hoek (2015)
etc.
proposed the following solution: the grouted pipe forepoles
A long forepole umbrella with several support arches under-
would be 12 m long and 114 mm in diameter at a spacing of 0.3e
neath can be divided into two areas, A and B, according to the
0.6 m; the pipe forepoles would be installed every 8 m to provide a
minimum of 4 m overlap between successive umbrellas. The
specification of the most commonly used forepoles is given in
Table 1.
Forepoles are loaded laterally by the loosened ground materials
above. Three things must be done in order that the forepoles work
la la

A B

qs qs

la la

Cantilever beam Fixed-end beam

Fig. 26. Sketch of tunnelling under the protection of a forepole umbrella (Aksoy and
Onargan, 2010; Hoek, 2015). Fig. 28. Beam models for pole sections between supporting arches.
412 C.C. Li / Journal of Rock Mechanics and Geotechnical Engineering 9 (2017) 396e414

Lr Lr

(a) (b)

Fig. 29. Forepole umbrellas. (a) Without supporting arch in between and (b) with a supporting arch in the middle.

bending nature of the poles (Fig. 28). The uniformly distributed deflectable in the section close to the near end of the pole. If only
load along a pole is expressed by qs, where q is the ground pressure one arch is set up at the near end of the poles (Fig. 29a), every pole
and s is the spacing between poles. The pole section in area A can be functions as a cantilever beam. The maximum deflection of the
simplified to a cantilever beam loaded by the uniformly distributed poles is expressed, according to Eq. (27), as
load qs. The maximum deflection, dA,m, of this pole section is
expressed, according to beam theory, as
qs
dA;m0 ¼ 5:4 L4r  103 (29)
qs EI
dA;m ¼ 5:4 l4a  103 (27)
EI where Lr represents the round length. If there is an additional
supporting arch in the middle of distance Lr, the span between the
where la is the arch spacing, E is the Young’s modulus of the pole
arches becomes Lr/2. The maximum deflection of the pole section
material, and I is the bending moment of the pole. Every pole
close to the near end of the pole becomes
section between two adjacent arches in area B can be simplified by
an end-fixed beam. The maximum deflection, dB,m, of the fixed-end
beam is expressed as 1
dA;m1 ¼ d (30)
16 A;m0
qs
dB;m ¼ 2:6 l4a  103 (28) In other words, the deflection of the pole with an additional arch
EI
in the middle of the round is reduced to one sixteenth of the
It is seen by comparing the two expressions above that the deflection without the additional arch. This means that the middle
maximum deflection of the pole in area B is approximately one half arch is effective in making the poles stiffer.
of the deflection of the pole section in area A. The pole is thus most The following is an example of spiling adopted for rock support
in squeezing rock in a metal mine. The spiles are either rebars or
self-drilling rockbolts depending on the rock conditions. The
rockbolts are at least 6 m long for an advance length of 4 m and fully
grouted with cement in the boreholes. In the case of using self-
drilling bolts, the bolt must be drilled at least 1 m beyond the
advance face. No matter what type of spiles is used, they must be
longer than the advance length of blasting.
Installation of the spiles is depicted in Fig. 30. Spiling holes are
located approximately 1 m above the contour. They are drilled 10 e
15 upward with a spacing of 0.3 m.

Fig. 30. Spiling in a mine drift. Fig. 31. Systematic rockbolting in burst-prone rock conditions.
 C.C. Li / Journal of Rock Mechanics and Geotechnical Engineering 9 (2017) 396e414 413

4.8. Rockbolting in burst-prone rock Bang S. Limit analysis of spiling reinforcement system in soft ground tunneling.
Tunnel 1984;3:140e6.
Barton N, Grimstad E. Tunnel and cavern support selection in Norway, based on
Energy-absorbing rockbolts, such as the D-bolt and cone bolt, rock mass classification with the Q-system. Publication No. 23. Norwegian
should be installed in burst-prone areas in order to achieve a Tunnelling Society; 2014. p. 45e77.
satisfactory reinforcement effect. The occurrence of rockburst is Barton N, Lien R, Lunde J. Engineering classification of rock masses for the design of
tunnel support. Rock Mechanics 1974;6(4):189e239.
random both in position and time. Thus rockbolts are usually sys- Bieniawski ZT. Engineering rock mass classifications. New York: Wiley; 1989.
tematically installed, for example, with a spacing pattern s  s (see Choquet P, Hadjigeorgiou J. The design of support for underground excavations. In:
Fig. 31). Assuming that the expected ejection depth is t, the bolt Excavation, support and monitoring. Comprehensive rock engineering e prin-
ciples, practice and projects, vol. 4. Pergamon Press; 1993. p. 313e48.
length should be at least 1 m longer than the ejection depth. The Coates DF, Cochrane TS. Development of design specifications for rock bolting from
bolt spacing s is then required according to Eq. (14) to be research in Canadian mines. Research Report R224. Mining Research Centre,
Energy, Mines and Resources Canada; 1970.
1 2Eab Crawford AM, Bray JW. Influence of the in situ stress field and joint stiffness on rock
s2 ¼ (31) wedge stability in underground openings. Canadian Geotechnical Journal
FS t rV 2 1983;20(2):276e87.
Crawford AM, Ng L, Lau KC. The spacing and length of rock bolts for underground
It is crucial that the rockbolts have a strong link with surface openings in jointed rock. In: Einsenstein Z, editor. Proceedings of the 5th in-
retaining elements such as the mesh and straps so that the load on ternational conference on numerical methods in geomechanics. A.A. Balkema;
the surface support elements is transferred to the rockbolts. 1985. p. 1293e300.
DSI. Spiles and forepoling boards. Product catalog. DWIDAG-System International;
2015.
Farmer IW, Shelton PD. Factors that affect underground rockbolt reinforcement
5. Conclusions systems design. Transactions of the Institutions of Mining and Metallurgy
1980;89:68e83.
The strength of rockbolts is the key parameter for rockbolting Harrison P, Hudson J. Engineering rock mechanics. Part 2: illustrative worked ex-
amples. Pergamon; 2000.
design in low stress rock masses. Rockbolts should be deformable in Hoek E, Brown ET. Underground excavations in rock. London: Institution of Mining
addition to the requirement of high strength in high stress rock and Metallurgy; 1980.
masses. In other words, rockbolts should be energy absorbent in Hoek E. Model to demonstrate how bolts work. In: Practical rock engineering; 2007.
Hoek E. Numerical modelling for shallow tunnels in weak rock. 2015. https://www.
squeezing and burst-prone rock conditions.
rocscience.com/documents/pdfs/rocnews/Spring2003/ShallowTunnels.pdf.
There exists a natural pressure arch immediately outside of the Kaiser PK, Tannant DD, McCreath DR. Canadian rock burst support handbook.
failure zone in the rock surrounding an underground excavation. In Sudbury, Canada: Geomechanics Research Center; 1996.
Krauland N. Rockbolting and economy. In: Stephansson O, editor. Rockbolting e
the case of a shallow failure zone, the rockbolts should be long
theory and applications in mining and underground construction. Rotterdam:
enough to reach the pressure arch. In the case of a vast failure zone, A.A. Balkema; 1983. p. 499e507.
short rockbolts are tightly installed to establish an artificial pres- Lang T. Rock reinforcement. Bulletin of the International Association of Engineering
sure arch within the failure zone and long cables are anchored into Geology 1972;9:215e39.
Lang TA. Theory and practice of rockbolting. Transactions of American Institute of
the natural pressure arch. Mining, Metallurgical, and Petroleum Engineers 1961;220:333e48.
Determination of the bolt length and spacing is associated with Li CC. Design principles of rock support for underground excavations. In: Eurock
the methodology of rockbolting. In the case of the anchorage of 2012. Stockholm, Sweden; 2012.
Li CC. Evaluation of the state of stress in the vicinity of a mine drift through core
rockbolts in the natural pressure arch, the bolt length should be at logging. In: Proceedings of the 4th Asian rock mechanics symposium. New
least 1 m beyond the failure zone. In the case of establishing an Jersey, USA: World Scientific; 2006a.
artificial pressure arch, appropriate bolt lengths are approximately Li CC. Rock support design based on the concept of pressure arch. International
Journal of Rock Mechanics and Mining Sciences 2006b;43(7):1083e90.
3 m in mine drifts and up to 7 m in large-scale hydropower caverns. Nomikos PP, Sofianos AI, Tsoutrelis CE. Symmetric wedge in the roof of a tunnel
Bolt spacing is more important than bolt length in this case. The excavated in an inclined stress field. International Journal of Rock Mechanics
principle is that the bolt spacing guarantees that the rockbolts and Mining Sciences 2002;39(1):59e67.
Nomikos PP, Yiouta-Mitra PV, Sofianos AI. Stability of asymmetric roof wedge under
interact with each other. The appropriate bolt spacing is 1 m for 3-
non-symmetric loading. Rock Mechanics and Rock Engineering 2006;39(2):121e9.
m long bolts and less than 1.5 m for 7-m long bolts. Ocak I. Control of surface settlements with umbrella arch method in second stage
The rockbolting design is based on the deadweight force of excavations of Istanbul Metro. Tunnelling and Underground Space Technology
2008;23(6):674e81.
falling blocks and the strength of the rockbolt in low rock stress
Panek LA. Design for bolting stratified roof. Transactions of the Society of Mining
locations. For high rock stresses, one should take into account the Engineers 1964;229:113e9.
portion of the rock-released energy that needs to be taken care by Schach R, Garschol K, Heltzen AM. Rock bolting: a practical handbook. Oxford:
the rockbolts. The maximum allowable displacement and the ulti- Pergamon; 1979.
Schubert W. Recent experience with squeezing rock in Alpine tunnels. In: CUC e
mate displacement capacity of the rockbolt should also be taken rock support in medium to poor rock conditions; 2001.
into account. Shi G, Goodman RE. Key block bolting. In: Rockbolting e- theory and applications in
The rockbolts in a rock support system should be compatible mining and underground construction. Rotterdam: A.A. Balkema; 1983. p.143e67.
Sinha RS. Rock reinforcement. In: Underground structures e design and instru-
with other support elements with respect to displacement and mentation. Amsterdam: Elsevier; 1989. p. 129e58.
energy absorption capacities. Slade J, Ascott B. Impact of rockburst damage upon a narrow vein gold deposit in
the Easter Goldfields, West Australia. In: Challenges in deep and high stress
mining. Australian centre for geomechanics; 2007. p. 247e56.
Conflict of interest Sofianos AI. Stability of wedges in tunnel roofs. International Journal of Rock Me-
chanics and Mining Sciences & Geomechanics Abstracts 1986;23(2):119e30.
Statens vegvesen. Fjellbolting. Håndbok 215. Oslo: Trykkpartner AS; 2000 (in
The author wishes to confirm that there are no known conflicts Norwegian).
of interest associated with this publication and there has been no Stillborg B. Professional users handbook for rockbolting. Clausthal-Zellerfeld: Trans
significant financial support for this work that could have influ- Tech Publications; 1994.
Volkmann GM, Schubert W. Effects of pipe umbrella systems on the stability of the
enced its outcome. working area in weak ground tunneling. In: Sinorock e international sympo-
sium on rock characterisation, modelling and engineering design methods.
International Society for Rock Mechanics; 2009.
References Wright FD. Roof control through beam action and arching. In: SME Mining Engi-
neering Handbook, vol. 1. New York: Society of Mining Engineers; 1973. p. 80e
Aksoy CO, Onargan T. The role of umbrella arch and face bolt as deformation pre- 96. Chapter 13.
venting support system in preventing building damages. Tunnelling and Un- Zhang C, Feng XT, Zhou H, Qiu S, Wu W. Case histories of four extremely intense rock-
derground Space Technology 2010;25(5):553e9. bursts in deep tunnels. Rock Mechanics and Rock Engineering 2012;45(3):275e88.
414 C.C. Li / Journal of Rock Mechanics and Geotechnical Engineering 9 (2017) 396e414

Dr. Charlie C. Li is professor of rock mechanics for civil and out thorough studies of rock behaviour under compression in laboratory. He made
mining engineering at the Norwegian University of Science numerous field observations of rock failure and responses of rock support elements
and Technology (NTNU) in Norway. Li received his BSc in mines and other types of underground excavations. After a thorough study of the
degree in 1981 and MSc degree in 1984, both in geological performances of rockbolts, he proposed analytical models for the rockbolts currently
engineering, in Central South Institute of Mining and used in rock engineering practise, which have been acknowledged in the circle of
Metallurgy (at present Central South University), and his rock mechanics. Based on the models, Li identified the shortcomings of the conven-
PhD in mining rock mechanics at Lulea University of tional rockbolts and pointed out that rockbolts, as well as other support elements,
Technology (LUT), Sweden, in 1993. After that, he was must be not only strong and but also deformable, i.e. energy-absorbing, in high in
employed as a research associate and then associate pro- situ rock conditions. He invented a new type of energy-absorbing rockbolt, called D-
fessor at LUT until 2000. He worked then in the Kristine- Bolt in 2006. The D-Bolt is as strong as a fully encapsulated rebar bolt but its deforma-
berg mine of Boliden Mineral Ltd., Sweden, as a mining tion capacity is significantly higher than that of the rebar bolt. The D-Bolt is particu-
engineer for 4 years. He has been the professor of rock larly powerful in combating stress-induced rockburst and squeezing. The bolt has
mechanics at NTNU since 2004, in charge of the teaching been used worldwide in many deep mines and also in hydropower projects, for
and research program in the subject of rock mechanics as instance in Sweden, Canada, USA, Chile, Australia and South Africa, to combat insta-
well as the rock mechanics laboratory. He is a member of the Norwegian Academy of bility problems of rockburst. Dr. Li has practical expertise in ground support in difficult
Technological Sciences (NTVA). He is the European Vice-President of the International rock conditions (for instance, rock squeezing and rockburst), stability analysis of un-
Society for Rock Mechanics (ISRM) for the term of office 2015e2019. Prof. Li’s research derground caverns and in situ measurements and interpretation. His current research
interests are in rock failure, stability analysis of underground spaces, ground support interests are on understanding of rockburst and theories and practise of dynamic rock
and application of rock mechanic principles for underground space design. He carried support.
 Keynote Address

Design of large span tunnels and caverns: back to basics

D. Oliveira

Jacobs Engineering Group Australia and University of Wollongong; email: david.oliveira@jacobs.com

ABSTRACT

Increased demand to future-proof tunnel projects with respect to traffic has led to the proposal of some very large spans
in recent road tunnel projects in Australia. For example, four lane tunnels are currently under construction in Sydney with
mined spans of approximately 20 m and Y-junction caverns of unprecedented spans for road tunnels in Australia, all with
a requirement for 100-year design life. As these spans are unprecedented in Australian civil tunnels, a direct comparison
with local past experience is not possible and simple extrapolation of precedent designs, although potentially solving the
problem, often result in uneconomical solutions that do not necessarily target the actual failure mechanisms involved in
the excavation of such large spans. International experience could certainly be used but adequate design justification
would still have to be provided. Although there is certainly room for cutting edge innovation, robust solutions can also be
achieved by simply going back to basics. As a result, this paper intends to present and discuss how designs that focus on
first principles and the basic objectives of rock reinforcement may allow for a better understanding of the design
requirements and how to satisfy codes and standards but also provide savings with respect to ground support. The key to
the design involves understanding the failure mechanism that needs to be addressed, its relationship with the different
actions of rock bolting, i.e. suspension/anchorage and/or rock reinforcement and what could be acceptable.

Keywords: large span, rock support design, rock bolt reinforcement, suspension

1 INTRODUCTION (24 m) and both Kedron (26 m) and Lutwyche (27 m)

caverns of the Airport Link tunnels in Brisbane.
With rapid development of cities, it is crucial that the use
of the underground space is made efficiently with These large span road tunnels are currently in
projects that can cater for the needs of the population for construction for the New M5 and M4-M5 Link tunnels as
several decades. This has led to an increased demand part of the infrastructure project known as WestConnex.
to future-proof tunnel projects with respect to traffic The WestConnex project is a 33-kilometre underground
which resulted in the proposal of some very large spans motorway currently being constructed in Sydney‘s Inner
in recent road tunnel projects currently in construction in West (Figure 1).
Australia all with the requirement for 100-year design life.
As these excavation spans are unprecedented in
For example, several kilometres of four lane tunnels are Australian civil tunnelling, a direct comparison with local
currently under construction in Sydney with mined spans experience is not possible particularly considering the
of approximately 20 m. Such spans had only been semi-flat roof tunnels typically excavated in Sydney.
experienced in localized excavations in widened
sections such as breakdown bays and Y-junction Although simple extrapolation of precedent designs
caverns but not for long lengths of tunnelling. In addition, could potentially provide a solution, two risks arise: (1)
the Y—junction caverns now required for these tunnels the extrapolation based on different excavation shapes
are also unprecedented for road tunnels in Australia with that do not necessarily target the actual failure
spans reaching 31 m and exceeding experience in mechanisms involved in the excavation of such larger
Australia which include the Eastern Distributor in Sydney

M4
ZD

Figure 1. WestConnex Motorway

2018 A68 VICTORIAN SYMPOSIUM - GEOTECHNICS AND TRANSPORT INFRASTRUCTURE 25

Design of large span tunnels and caverns: back to basics D. Oliveira

y-va
—‘
spans; and (2) provide uneconomical solutions. to approximately one third of the span. On the other
International experience could certainly be used but hand, only a small number of projects with spans greater

fa
adequate design justifications and analysis would still than 20 m exist in Australia and if they are included, the
have to be provided to verify its application locally. global best fit indicates a power curve that deviates from

r“!
the linear trend. Based on such a best fit, a bolt length of
The search for solutions to new problems often target approximately 6 m would be considered a precedent
innovation. However, considering that the new challenge design for tunnels of approximately 20 m span.
described above in fact involves an “old" problem but at
a larger scale, it is considered appropriate to review the For tunnels where instability is generally controlled by
basic design assumptions to find robust solutions in geological structures, the rule of thumb for bolt length
more fundamental design principles. As a result, this approximately equal to one-third of the span is
paper intends to present and discuss how a design that somewhat related to conservative assessments of the
focus on the basic objectives of rock reinforcement may largest possible wedge assuming ubiquitous rock
allow for a better understanding of the design defects and ignoring stress effects. These assessments

7‘!
requirements and still provide savings with respect to generally result in wedge widths that span approximately
ground support. The key to the design involves the entire tunnel as illustrated in Figure 3, and wedge

1
r'1
understanding the failure mechanism that needs to be apex heights approximately equal to one third of the

K
addressed, its relationship with the different actions of span, particularly for tunnels smaller tunnels of, say,

. 1
rock bolting, i.e. suspension/anchorage and/or rock 15 m span or less.
reinforcement and what could be acceptable.
For the flat—roofed tunnels typically excavated in Sydney,

1
r.‘
2 PRECEDENT DESIGN the design approach generally was that proposed by

L
Bertuzzi and Pells (2002) as illustrated in Figure 4. The

L 4
design is no longer governed by structurally controlled

r'fi
The first step in any design is typically a comparison with
precedent experience of what has worked and what has wedges but involves selecting a roof beam thickness

A
not. Therefore, Figure 2 presents a comparison between which would in turn dictate the rock bolt length based on

"fir-“era
L
span and bolt length for several successful projects in deflections that are deemed acceptable. This deflection
Australia prior to 2014. typically varies between 10 mm and 20 mm. It is

LA
important to note that Bertuzzi and Pells (2002)
Significantly experience is observed for tunnels under suggested to add 1 m extra in bolt length to the beam
18m span. For these span tunnels, the bolt length thickness for embedment purposes.

LA
roughly follows a linear relationship with bolt length equal

PM‘
1o /

9 a a a

a /V ,_ ..
x
7 <>—————

1
7’1
y = 0.6308x° 75‘“

.
R2 = 0.8702
Bolt Length (m)

rd
U"

r‘w
4

0
0 5 10 15 20 25 30
Span (m)
——Spani4 (Lang 1958) 2 t 015 x Span (Barton at at 397-5) 14 s 0184 x Span (Barton et al 1995)
o 609a MS 9 Bondr Pump Station at Opera Hourso Car Park
at M2 Motorway A Eastem Distributor 0 Elgas Distrbutor
r Nonnsrdo Storage Tunnel - Poauna A Lane Cove Tunnel
0 Energy Australia Cabte Tunnel A ECRL x Yransgnd Canto Tunnel
x Transgnd Plant Room 0 Buranda Street 0» 80990 Rd
Coal Mnos NSB‘I’ », Brunswrck Street
NST APL Cross CrtyTunneI‘
How Southern Ralway' I Lutwycho Cavern t: Kodron Cavern
o . . o Lino 018w Flt Span'3 Linear arch (E=1 56Pa‘q=5t]<Pa den <20mm)

Figure 2. Span versus bolt length for several projects in Australia

26 2018 AGS VICTORIAN SYMPOSIUM ' GEOTECHNICS AND TRANSPORT INFRASTRUCTURE

Design of large span tunnels and caverns: back to basics D. Oliveira

STEP 1
Define rock beam thickness (i.e. bolt length) using
linear arch theory and an assummed acceptable
Roof wedge
Weight 44.4 t deflection (10-15 mm).
FS = 0
STEP 2
Assess induced stresses within pseudo<equivalent
rock beam.

STEP 3
Estimate excess shear stress within rock beam.

Left wall wedge Right wall wedge

Weight 3.3 t Weight 5.2 t STEP 4
rs = 040 FS = 0.37 , Estimate the relative horizontal movement
between the top of one bed and the bottom of
J the overlying bed.

3‘ STEP 5
Floor wedge '
Weight 181 t
Assess rock bolt contribution as reinforcement
Stable element.

Figure 4. Rock beam design steps (after Bertuzzi and

Figure 3. Example of unstable wedge formed in roof with Pells 2002).
width approximately equal to tunnel span (after Hoek,
2007) 3 WHAT CHANGES IN LARGE SPAN TUNNELS?

For a 20 m span tunnel, the required bolt length would be The precedent design above already provided some initial
approximately 6 m bolt length to achieve a mid-span indication that in large span tunnels, bolt length may not
deflection of approximately 15mm in accordance with follow similar trends to those of the smaller span tunnels.
Bertuzzi and Pells (2002) and assuming a fair quality Two main factors are considered to influence this change:
pseudo-equivalent rock beam with E =1500 MPa. This is stress effects and a change in design focus to
somewhat consistent with the best-fit of all projects shown reinforcement effects rather the suspension of unstable
in Figure 2. wedges.

The limited data beyond 18 m span indicate a distinct 3.1 Stress effects
difference from the smaller tunnels. Most of the larger
span bolt lengths deviate from both the linear-arch theory As previously discussed, for tunnels where instability is
curve and the span/3 linear trend and approximately generally controlled by geological structures, the rule of
approach the empirical rule originally devised during the thumb for bolt length is somewhat related to conservative
Snowy Mountains project in Australia by Lang (1958) for assessments of the largest possible wedge assuming
bolt lengths of span/4 (for spans greater than 18 m) and ubiquitous rock defects and ignoring stress effects. These
those of Barton et al (1974) and Barton et al (1995). The assessments generally result in wedge widths that span
exceptions are the Eastern Distributor cavern with 9 m approximately the entire tunnel and wedge apex heights
long bolts and span up to 24 m and the Lane Cove Tunnel approximately equal to one third of the span. Making
with 7 m bolts for a 22 m span excavation which are in similar assumptions in large tunnels and caverns, very
Sydney. These two projects still follow approximately the large wedges would therefore be found in the roof which
span/3 or linear arch theory curve as proposed by would require excessively long cable bolts.
Bertuzzi and Pells (2002). On the other hand, the other
projects are outside Sydney, namely the Poatina Power Figure 5 illustrates an unstable wedge, i.e. factor of safety
Station cavern in Tasmania with 7 m bolts for an FoS = 0, with an apex height of approximately 12 m within
approximately 29 m span excavation, the Kedron cavern the roof of a 40 m span cavern. Such a wedge would
in Brisbane with 7m bolts for a 26m span and the require cable bolts lengths of approximately 12—13m.
Lutwyche cavern with 6 m bolts for an approximately 26 m Firstly, the designer must question if such a large-scale
excavation. The main difference between the larger wedge is indeed reasonable based on the typical
excavations outside Sydney and the large span tunnels persistence of the controlling discontinuities. Secondly,
excavated in Sydney is that they are generally all fully an important assumption typically ignored may be
arched structures where the tunnels in Sydney are semi- reconsidered which is the positive clamping effect of
flat roof tunnels, i.e. very low arch tunnels. Stress arching horizontal stresses. If a stress state equal to the in-situ
and combined use of passive shotcrete support in arched stress is assumed, despite horizontal stress likely
excavations may allow for shorter bolts than what would increasing due to the excavation, the wedge would show
have been used in semi-flat—roofed tunnels in Sydney a stable behaviour with a FoS = 1.7.
using the Bertuzzi and Pells (2002) approach as depicted
in Figure 2. As a result, such difference must be Hoek (2007) states that this large difference in safety
considered in design. suggests a tendency for sudden failure when the in-situ
stresses are diminished for any reason and is a warning
sign that care must be taken in terms of the excavation
and support installation sequence. For this reason, many
tunnel designers consider that it is prudent to design the
tunnel support on the basis that there are no in situ

2018 AGS VICTORIAN SYMPOSIUM ' GEOTECHNICS AND TRANSPORT INFRASTRUCTURE 27

Design of large span tunnels and caverns: back to basics D. Oliveira

Without in-situ stress With in-situ stress

Roof wedge [8]
FS: 1.733
Roof wedge {8] Weight: 18.862 MN
FS: 0000 .
Apex Height: 11.79 m
Weight: 18.862 MN l
Apex Height: 11.79 m

el/zo/F/a
Figure 5. Example of the effect of in-situ stresses (40 m deep and kg = 1) on a large unstable wedge in a large span and
unsupported cavern.

stresses ensuring that, for almost all cases, the support 3.2.1 The design myth
design will be conservative.
There is significant misunderstanding and design myths
Although some level of conservatism is generally on how rock bolts provide such reinforcement effects. To
desirable, one main difference should be considered in this end, Pells (2008) stated that rock bolts are sometimes
the support of large span tunnels and caverns. In almost ascribed abilities that verge on magic. For example, they
all cases, these large span excavations will be are said to prevent stress induced failure, or said to
sequentially excavated such that large deformation that interlock a rock mass like aggregate. The latter is based
could induce wedge dislodgment would typically be on Tom Lang’s famous 1960‘s upside-down bucket
controlled and some anchorage and shear reinforcement experiment during construction of the Snowy Mountain
would already be in place prior to full exposure of the project which initiated a design myth that is often misused.
largest possible wedge, making the use of the positive Tom Lang‘s experiment intended to demonstrate to the
clamping effect of stresses more amenable. Figure 6 Snowy workers how rock bolts work.
depicts this scenario.
Evert Hoek visited the Snowy project in the 1960’s and he
An example of such large span with considerably shorter was impressed with such a demonstration (Hoek, 2007).
bolts in comparison to the excavation span is the Gjevik He then started using his own version for teaching
Olympic Hall in Norway with a record-breaking span in purposes at the University of Toronto (Hoek, 2007). In
civil tunnels of approximately 61 m which only used a Hoek’s version, he indicated that a zone of compression
maximum bolt length of 12 m (Broch et al, 1996). is induced in the region shown in red in Figure 8 and this
provides effective reinforcement to the rock mass when
3.2 Reinforcement effect the rock bolt spacing, s, is less than 3 times the average
rock piece diameter, a, and the length, L, approximately
Another important factor is the change in focus from an 28.
anchorage or suspension effect of rock bolts to the
reinforcement effect within the tunnel roof. This concept of a compression zone promoted by rock
bolts was then extended to tunnels as a rock bolt
compression ring concept around the opening by Pender
et al (1963) as depicted in Figure 9.

Figure 7. Lang’s Bucket (after Pells, 2008).

Figure 6. Example of the effect of sequential excavation
on large wedge support combined with the clamping
effect of stresses.

28 2018 AGS VICTORIAN SYMPOSIUM ' GEOTECHNICS AND TRANSPORT INFRASTRUCTURE

Design of large span tunnels and caverns: back to basics D. Oliveira

no support between washers

zone of compression

Figure 8. Hoek’s educational version of the upside-down bucket experiment (after Hoek, 2007).

3.2.2 How it really works

Despite the stress scale issue, Hoek’s experiment can still

be used to demonstrate how rock bolts work. Not
necessarily due to a confinement effect which is relevant
at the scale of the Hoek’s experiment but in fact due to its
reinforcement effect allowing stresses to arch linearly to
the abutments as shown in Figure 11. In fact, such
concept is not different to the concept of the linear arch
theory used in laminated rock beam design as depicted in
Figure 12.

Flat or semi-flat-roofed tunnels

For flat or semi-flat roofed tunnels, Oliveira and

Paramaguru (2016) presented an approach forthe design
of rock reinforcement in laminated rock beams where the
focus is on satisfying the development of the compressive
arch like the one presented in Figure 12 within a pseudo-

.szoxgzozl equivalent thicker voussoir beam. The rock bolt

reinforcement is designed to provide the necessary
capacity to overcome the excess shear stresses in the
Figure 9. Overlap of rock bolts compression zone bedding partings, thus stitching thinner beams together
forming a compression ring around the tunnels (after into an equivalent thicker rock beam that can control
Pender, 1963). deflections (Figure 13).

The above concept could be said relatively intuitive and It should be noted that the iterative design approach
simple to explain how rock bolts work. However, Pells proposed Oliveira and Paramaguru (2016) does not
(2008) demonstrated that Tom Lang's bucket experiment define the rock beam thickness and consequently rock
and consequently Hoek’s gravel table have little bolt length based on deflections defined as acceptable
relationship to the action of rock bolts around a tunnel prior to the analysis. In their approach, deflections are an
because the stress scale is all wrong. In these output, not an input, with the primary objective of the rock
experiments the confinement or compression provided by bolting design approach proposed by the authors being
the bolts is of the order of a few kilopascals which is too the development of the compressive arch within the
small, by several orders of magnitude, to have any effect “stitched” pseudo-equivalent rock beam. This is achieved
on the rock mass strength by “confinement” in a tunnel by assessing the excess shear forces developed within
scale. In addition, although Figure 9 has been reproduced the equivalent beam and comparing against the mobilised
in many text books, Pells (2008) also demonstrated via rock bolt forces estimated through methods such as those
analysis that similar scale issue is observed with this presented by Pells (2002). This approach has in generally
concept as shown in Figure 10. allowed for a reduction of bolt lengths in recent projects
when comparing to precedent designs as the effect of bolt
spacing is taken into

2018 AGS VICTORIAN SYMPOSIUM - GEOTECHNICS AND TRANSPORT INFRASTRUCTURE 29

Design of large span tunnels and caverns: back to basics D. Olivelra

r"!
rum“

"‘
mam-mm
mmnhmngcnm

“'Y
rma
rd
r‘ 1
new 1 run)

’w
= ‘rm-MAA 0»... 1mm“. w-Iw um. ”A.”

r" T"
Figure 10 Contours of major and minor principal stress induced by 2m rock bolts at 1.1m centres pretensioned to 80 kN
(after Pe/ls, 2008).

r‘fi
Likely compression arch that provides stability
Zones where tension occurs such that the plates

W1
average rock piece diameter) rockbolt spacing
(mesh/shotcrete) have an important function.

rm
l);
”@fimrfire
At this scale, the double
plates/washers do help providing

J
rockbolt L
minor confinement/interlocking (red
length

I.
zone).

zone of compression

Figure 11 Likely compression arching that provides stability in Hoek’s experiment (modified after Hoek, 2007).

("w
h
pr“,
Figure 12 Compression arching that provides stability in laminated rock beams. Note the effect of the rock bolts by the red

T‘,
stress tensors, i.e. tensile stresses (after Ollveira and Paramaguru, 2016).

account and may be used to compensate for shorter bolt Figure 15 presents both ground reaction curves together
lengths. with the countors of total displacements around the
tunnel. The unsupported case shows that collpase
Arched profiles intitiates when the excavation boundary tractions are
equivalent to approximately 15% of the original in-situ
In the case of arched profiles, the objective of the rock stress which occurs approximately 5 in behind the
reinforcement is no different, i.e. the focus remains on excavation face. At this point the crown deflection is
facilitating the stress redistribution around the tunnel. appproximately 70 mm. Full collapse is observed at a
boundary traction equivalent to 10% of the original in—situ
A discontinuum model of a 20 m span shallow excavation stresses which occurs approximately 7m behind the
in fair to poor quality siltstone rock (UCS=4-8MPa and excavation face. On the other hand, the rock bolted case
GSI = 40) using 3DEC will be adopted to illustrate the indicates that at a boundary traction equivalent to
effect of rock bolt reinforcement (Figure 14). Typical approximately 15% of the original in-situ stress or 5 m
parameters to those assumed in Sydney are used for the behind the excavation face, the crown displacement is
purpose ofthe analysis with two cases analysed: (1) afull- less than 35 mm. The tunnel reaches full stability some
face unsupported excavation and (2) a full-face 10 m behind the excavation face with a crown
excavation supported by rock bolts only spaced at 1.25 m displacement of approximately 75 mm when the
centre to centre. The Ground Reaction Curve of the two boundary traction is equivalent to 5% of the original in-situ
cases are assessed by simulating a gradual excavation stress. After this stage, the boundary traction is reduced
which is achieved through the reduction of the surface to zero in the rock bolted case simulating full excavation
tractions at the boundary of the excavation. and a distance greater than 60 m behind the excavation

30 2018 AGS VICTORIAN SYMPOSIUM ' GEOTECHNICS AND TRANSPORT INFRASTRUCTURE

Design of large span tunnels and caverns: back to basics D. Oliveira

Equivalent voussoir beam True rock beam with multiple beds

Compression arch Shear along bedding
Line of thrust

~e 6 — angle between thrust

Compression arch/thrust transferred” % force and bedding
-r e ‘ ‘ ~
through bedding ~‘.

Figure 13. Reinforcement effect for equivalent rock bolt stitched rock beam (after Oliveira and Paramaguru, 2016).

Geometry
Bolted rock arch — 4m bolts

Figure 14. Model of a 20 m span arched shallow tunnel in in fair to poor quality siltstone rock.

face with no further significant deflections observed in the to approximately 1 MPa. In fact, the zones with 1 MPa
tunnel crown. stresses indicate the likely line of thrust associated with
stress arching.
It should be noted that the large deflections observed in
the rock bolted case of Figure 15 would be futher reduced An analogy between conventional structural tunnel
if additional passive support such as a thick layer of support and the overall effect of rock reinforcement could
shotcrete (e.g. t> 300 mm) is applied which is generally then be made by treating the reinforced rock beam or
the case in shallow tunnels. Shotcrete would particularly reinforced arch as structural elements where loads are
control the fall-out of wedges between bolts and reduce applied with appropriate factors to test compliance to
overal rock loosening. design standards.

Based on the discussion above, the positive effect of the For example, using the previous example, a load
rock reinforcement in promoting a more stable stress equivalent to approximately 15% of the original in-situ
redistribution is evident through the comparison between stress, i.e. equivalent to the collapse initiation load, could
the ground reaction curves and equivalent deflections at be assumed and multiplied by a load factor of 1.5 in
several points behind the excavation face. accordance with AS5100. Such loads could then be
applied to a bolted rock arch modelled isolated from the
The improvement in stress arching promoted by the bolts rest of rock mass to verify its ability to promote the stress
becomes even more evident when comparing the stress arching and form the line of thrust, thus, providing support
tensor within the bolted-arch for the two cases. This is to the overlying rock mass as illustrated in Figure 17. In
illustrated in Figure 16 with the stress tensor coloured with this case, the structural capacity of the rock bolts would
respect to the magnitude of the major principal stress also have a reduction factor applied to satisfy code
(Minimum Principal in 3DEC as compression is negative). compliance.
The major principal stress indicates the ability of the rock
to redistribute or arch the stresses around the tunnel. For It is important to note, that similarly to the approach
the unsupported case, a large zone in the immediate roof adopted for rock beam design (Oliveira and Paramaguru,
of the tunnel is observed to be in tension or at zero 2016), the bolted-arch should first be allowed to deflect
stresses at a boundary traction equivalent to 10% of the elastically while maintaining a non-zero tensile strength
original in~situ stresses which confirms the loss of support within the joints. This initial elastic deflection is assumed
and associated collapse. On the other hand, the rock equivalent to the gradual excavation mechanism so that
bolted case indicates that the major principal stress in the some deformation occur before the roof is fully excavated
immediate roof of the tunnel is approximately 600 kPa at and formed.
a similar 10% boundary traction stage, locally increasing

2018 A68 VICTORIAN SYMPOSIUM - GEOTECHNICS AND TRANSPORT INFRASTRUCTURE 31

Design of large span tunnels and caverns: back to basics D. Oliveira

fa
rm
Contour 0f Displacement

,9...
Deformed Factor: 5
100 8.0000502
7,0000E-02
Percentage of in-situ stress (96)
90 Unsupported 6.0000602

rww
5.0000502
80 Rockbolts
4.0000E-02
70 3,0000E—02

7'"
2.0000E—02
60 1.0000E—02
50 0.0000E+00

W‘
40
3O

r1
20
10

o 25 50 75‘; 100 125

\
Crown vertical disptacement (mm)

rs
@fifir‘r‘roflm
Stress
Scale: 19-06
100 Minimum Prin.
0.0000E+00
Percentage of in—situ stress (%)

90 —-—-Unsuppotted -1.0000E+05
-2.0000E+05
80 Rocklnlts 6.0000905
Unsupported: 4.0000E+05
70
~5.OOOOE'I05
observe section —6.0000E+05
60
going in tension 10000905
50 3.0000905
8.0000605
40 -1.0000E*06

rm:
30
20
10

r1I
,
\
\
\

Stress tensors indicate the effect of

"1
bolts in facilitating stress redistribution
but not as compression bulbs around The line of compression/thrust can be
the bolts... seen through the 1 MPa values...

Figure 16. Ground reaction curve and representative behaviour with respect to stress tensor within “bolted-arch”.

After the initial elastic equilibrium is achieved, the joint span/4 for spans greater than 18 m as proposed by Lang
tensile strength should be set to zero and the arch allowed (1958) and the curves proposed by Barton et al (1974)
to continue deforming until either equilibrium or failure and Barton et al (1995).
occurs.
For example, long tunnels with spans of up to 20 m have
4 UPDATED EMPIRICAL CHART been recently designed with bolt lengths of 5 m in contrast
to the precedent design of 6 m long bolts. Although this
Figure 18 presents an updated empirical design chart may seem a small reduction at first, in a long tunnel, this
including recent projects where some of the concepts results in savings of several kilometres of rock bolts which
discussed above have been applied. There is an evident include drilling and grouting and the associated time
increase in points below the previous precedent design required to install a large number of them.
trends and approaching the low bound curves equal to

32 2018 AGS VICTORIAN SYMPOSIUM ° GEOTECHNICS AND TRANSPORT INFRASTRUCTURE

Design of large span tunnels and caverns: back to basics D, Oliveira

Equivalent voussoir beam True rock beam with multiple beds

HHIIIIIUIIIIHHH Comprgssinn arch Line of thrust Shear along bedding

‘‘ »
Cnmpresion arch/thrust transferreg,” %{ force and bedding
through bedding ~ e ‘4

10

8 .

: x
7 ., c
3 y = 0.5983x0766‘
g e
v
R2 = 0.875 . /
III
_: PRECEDENTED DESIGN AREA ,.o
... ., A 01
E’ 5
a)
a 2 O
. ''
-I x .x
g
o 4
03 A A

3 r " ——l
/ ,
1% 1" UNPRECEDEN TED DESIGN AREA
2 w- 7! x
1

O 5 10 15 20 25 30
Span (m)
Span/4 (Lang 1958) 2 * 015 X Span (Barton at at 1976) -—-1 4 #0184 x Span (Barton etal 1995)
0 Existing M5 9 80nd: Pump Station )1 Opera Hourse Car Park
2: M2 Motorway A Eastam Drstnbutor o Elgas Drstnbutor
1- Nonnsrde Storage Tunnel - Poatma A Lane Cove Tunnel
u Energy Austraua Cable Tunnel A ECRL x Transgnd Cable Tunnel
x Transgnd Plant Room 0 Buranda Street 4» 80990 Rd
- Coal Mines - NSBT o Brunswrck Street
NST APL Cross City‘l'unnel‘
New Southem Ralway' I Lutwycne Cavern :: Kodron Cavern
o WestConnex Stage 2 A WestConnex Stage 18 o NorthConnex
- . . - . Line of Best Fit Span/3 — Linear arch (E=l SGPa/q=50kPaJden <20mm)

Figure 18. Current precedence including recent projects in Australia.

5 CONCLUSIONS challenges of these designs, satisfy codes and standards

requirements but at the same time provide savings with
This paper presented a discussion on the design of large respect to ground support.
span tunnels and caverns with particular reference to the
Australian experience. Some precedent design was The key to the design involves understanding the key
presented and demonstrated that, although very valuable, failure mechanisms that needs to be addressed, its
past projects are not the only single source of design. relationship with the different actions of rock bolting, i.e.
suspension/anchorage and/or rock reinforcement and
Design efforts that focus on first principles and the basic what could be considered acceptable for design.
objectives of rock reinforcement can overcome the

2018 AGS VICTORIAN SYMPOSIUM - GEOTECHNICS AND TRANSPORT INFRASTRUCTURE 33

Design of large span tunnels and caverns: back to basics D. Oliveira

The compression promoted by rock bolts is often see in

text-books is an oversimplification that does not represent
a realistic case. However, rock bolts do facilitate the
development of a compression arch resulting from stress
redistribution within the rock arch or rock beam. When
properly designed, the rock bolts allow the line of thrust to
be transferred across discontinuities such that it maintains
compression but such compression it still from the “rock
formation”
Code compliance can then be verified with simplified
models based on the reinforced rock arch or beam

6 ACKNOWLEDGEMENTS

The author would like to acknowledge all his colleagues

who one way or another have contributed to some of this
work. He would also like to express his gratitude to the
organisers of the AGS Symposium for inviting him to
present this Keynote Presentation.

REFERENCES

Barton, N,R., Lien, R. and Lunde, J. (1974). Engineering

classification of rock masses for the design of tunnel
support. Rock Mech. 6(4), 189-239.
Barton, N., Grimstad, E. and Palmstrom, A. (1995). Design of
Tunnel Support. Sprayed Concrete: Properties, Design
and Application, Ed. S.A. Austin and P.J. Robins, pp 150-
170.
Bertuzzi, R. and Pells, P.J.N. (2002). Design of rock bolt and
shotcrete support of tunnel roofs in Sydney sandstone.
Australian Geomechanics, 37(3).
Broch E., Myrvang A. M. and Stjern G. (1996). Support of Large
Rock Caverns in Norway. Tunnelling and Underground
Space Technology, Vol. 11, No. 1, pp. 11-19, 1996.
Hoek, E. (2007). Practical Rock Engineering. Lecture Notes.
Lang, TA. (1958). Rock bolting speeds Snowy Mountains
project. Civil Engineering, ASCE, Vol 28. No 2, Feb 1958,
pp 40—42.
Oliveira D. and Paramaguru, L. (2016). Laminated rock beam
design for tunnel support. Australian Geomechanics Vol
51(4), pp.1—17, December.
Pells P. (2008). What happened to the mechanics in rock
mechanics and the geology in engineering geology? In: 6th
International Symposium on Ground Support in Mining and
Civil Engineering Construction, SAIMM, SANlRE and
ISRM, 30 March—3 April, Cape Town.
Pells, P. J. N. (2002). Developments in the design of tunnels and
caverns in the Triassic rocks of the Sydney region.
International Journal of Rock Mechanics and Mining
Sciences 39: 569-587.
Pender, E B, Hosking, A D and Mattner, R H (1963). Grouted
Rockbolts for Permanent Support of Major Underground
Works. Journal Institution Engineers Australia, Vol 35, No
7-8, pp 129—145.

34 2018 AGS VICTORIAN SYMPOSIUM - GEOTECHNICS AND TRANSPORT INFRASTRUCTURE

 LAMINATED ROCK BEAM DESIGN FOR TUNNEL SUPPORT

D.A.F. Oliveira1 and L. Paramaguru2

1
Principal Tunnelling/Geotechnical Engineer, Jacobs and Adjunct Principal Fellow, University of Wollongong,
2
Senior Geotechnical Engineer, Jacobs.

ABSTRACT
The design and construction of semi flat-roofed tunnels, i.e. with a high arch radius to roof span ratio, using a voussoir
beam analogy has been proven successful over time. In spite of such a success, the linear arch theory or voussoir beam
analogy has always been subjected to a certain level of scepticism due to some of its perceived limitations. Some of the
concerns are related to appropriate design methods for the design of rock bolting of multiple beds/laminations in cases
where single laminations are deemed unstable upon excavation or while addressing some adverse conditions. This paper
investigates the applicability of an analytical solution of the voussoir beam theory for the design of rock bolts in
laminated rock beams which has been confirmed with numerical analysis using DEM (Distinct Element Method)
analysis. The proposed analysis method can be easily implemented in a spreadsheet to provide rapid assessments
though it is considered only one part of the design process with other potential instability mechanisms assessed using
other analysis methods.

1 INTRODUCTION
Despite the successful design and construction of semi flat-roofed tunnels over the years in Australia, Canada and the
US, the linear arch theory or voussoir beam analogy has always been subjected to a certain level of scepticism due to
some of its perceived limitations. According to Diederich and Kaiser (1999), it generated a great deal of controversy
when it was first published by Evans (1941), even though the general notion of its successful application traces back to
ancient Rome architecture.
A significant portion of the scepticism seems to be related to perceived limitations of the conceptual model and
available analytical solutions with respect to:
§ The effect of horizontal stress
§ Span to lamination/bed thickness ratio
§ The presence of adverse geological features
§ The effect of a slightly arched roof
§ Bolting of multiple beds/laminations in cases where single laminations are unstable upon excavation
It is important to note upfront that there are two primary factors that promote a geological environment amenable to a
voussoir beam analogy in tunnel support design: (1) a horizontally bedded rock mass with no low to mid angle jointing
cross-cutting the lamination, i.e. a reasonably good quality rock mass, and (2) the existence of favourable horizontal
stresses.
This paper investigates the applicability of an analytical solution of the voussoir beam theory through comparison with
numerical modelling that focuses on the above limitations. The results illustrate that the voussoir beam analogy can be
confidently used in practice when used with reasonable engineering judgment as required in any other design method. It
adequately represents the results of more robust numerical solutions such as Discrete Element Method (DEM) when
similar governing mechanisms are considered in both models.

2 CONCEPTUAL MODEL AND ANALYTICAL SOLUTION

Rock masses dominated by parallel laminations are often encountered in underground excavations in numerous
geological environments. Fayol (1885) noted that, in these cases, the underground strata seemed to separate upon
deflection so that each laminated beam transferred its own weight to the abutments rather than loading the laminated
beam beneath. This occurred even in the cases where other discontinuities cut across the laminations at steep angles or
where reinforcement had been installed. Based on such observations, it was assumed that a compression arch could be
generated across the cross-cutting joints and within the beam upon deflection which would then transmit the beam loads
to the abutments as illustrated in Figure 1.

AUSTRALIAN GEOMECHANICS VOLUME 51 NO. 3 SEPTEMBER 2016 1

LAMINATED ROCK BEAM DESIGN FOR TUNNEL SUPPORT OLIVEIRA AND PARAMAGURU

Line of thrust

Figure 1: Voussoir beam analog

One of the most recent analytical solutions for analysis of the voussoir beam analogy, and perhaps currently the most
widely used, is that proposed by Diederichs and Kaiser (1999). The analytical solution is based on an iterative analysis
of beam deflections which is then used for estimates of maximum compressive stress, smax, developed through the
arching mechanism. Such estimates are then adopted to assess factors of safety against four main mechanisms namely
rock crushing, sliding or shear at the abutments and a Buckling Limit (BL) Index for the snap-through mechanism. Only
one minor adjustment has been made to their original proposal with abutment deflections included according to Asche
and Lechner (2003).
A full description of the solution will not be presented here as it can be found in details in Diederichs and Kaiser
(1999). However, two main aspects of the original solution are worthwhile highlighting as they will be used later in this
paper to investigate the case of multiple laminations: the horizontal stress distribution along the line of thrust within the
beam and the geometry of line of thrust itself as depicted in Figure 2.
As noted in Figure 2, Diederichs and Kaiser (1999) assumed a parabolic stress distribution with a quadratic variation at

a distance of approximately !/(2 2) from the midspan where s is the total span of the beam. This point is located
where the line of thrust crosses the centreline of the beam and it is reasonable to expect that the entire beam section is
under compression and that this stress is constant across the entire beam thickness T. As a result, the stress distribution
along the line of thrust can be represented by a two-part parabolic curve given by:

Line of thrust

Figure 2: Parabolic compressive stress variation assumed by Diederichs and Kaiser (1999) against Brady and Brown
(1993)

2 AUSTRALIAN GEOMECHANICS VOLUME 51 NO. 4 DECEMBER 2016

LAMINATED ROCK BEAM DESIGN FOR TUNNEL SUPPORT OLIVEIRA AND PARAMAGURU

where x is the distance along the rock beam and the quadratic parameters may be found from the boundary conditions
given in Figure 2 resulting in:

where fm is the maximum compressive stress is given by Diederichs and Kaiser (1999). The geometric position of the
line of thrust is given by:

where the parameters are found by:

where T is the lamination thickness and N is the depth ratio for the triangular stress block according to Diederichs and
Kaiser (1999). These two equations will play a significant role in the assessment of multiple laminations although not
directly and explicitly used in the analysis of a single lamination.

3 NUMERICAL DISCRETE ELEMENT METHOD MODELS

For comparison purposes with a more robust solution, all analyses carried out with the analytical solution are also
analysed using a Distinct Element Method (DEM) as encapsulated in UDEC V6.0 (Itasca, 2013). Similar to the
numerical models adopted by Diederichs and Kaiser (1999) as part of the development of the analytical solution, the
numerical models analysed in this paper use a two-step approach.
S or X

sj T

Figure 3: Typical boundary conditions of UDEC models.

The beams are first allowed to deflect elastically while maintaining a non-zero tensile strength within the joints. This
initial elastic deflection is assumed equivalent to a gradual excavation mechanism so that some beam deformation occur
before the roof is fully excavated and formed. After the initial elastic equilibrium is achieved, the joint tensile strength
is set to zero and the beam is allowed to continue deforming until either equilibrium or failure occurs. The joints have
no cohesion or dilation but have a frictional strength. One main difference between the model adopted in this paper and
that of Diederichs and Kaiser (1999) is that the abutment blocks are given the same deformability parameters of the
rock beam instead of a very stiff abutment. Figure 3 presents the typical boundary condition for the UDEC models.

4 EFFECT OF HORIZONTAL STRESS AND SPAN TO BED THICKNESS

The analytical solution for the voussoir beam analogy is bounded by a few limitations. Among others which will be
discussed later in this paper, Diederichs and Kaiser (1999) have pointed out that their solution was developed based on
span to lamination thickness ratio, z = S/T, greater than 10 (see Figure 3). In addition, the impact of horizontal stresses
in the rock beam is neglected.
Oliveira and Pells (2014) demonstrated some positive effect of initial horizontal stress on the overall rock beam
behaviour by means of DEM modelling which could not be captured by the Diederichs and Kaiser (1999) solution.
They suggested that increasing initial horizontal stress causes the beam to behave in a more elastic manner approaching
an uncracked rock beam, i.e. with less effect of the cross-cutting joints, and with fixed end conditions at both
abutments.

AUSTRALIAN GEOMECHANICS VOLUME 51 NO. 4 DECEMBER 2016

LAMINATED ROCK BEAM DESIGN FOR TUNNEL SUPPORT OLIVEIRA AND PARAMAGURU

Booker and Best (1990) developed a 1D finite element method which accounted for the effect of horizontal stresses
through a “cracked beam” analysis. This method was used for the design of the Sydney Opera House Carpark (Pells et
al., 1994) and seemed to capture the end effects in a more realistic manner. However, this method involves a finite
element solution that is inherently more sophisticated and less used than the Diederichs and Kaiser (1999) solution. As
a result, it will not be used or discussed in this paper.

4.1 CONFINEMENT ADJUSTMENT FACTOR

To account for some effect of the initial horizontal stress in the Diederichs and Kaiser (1999) analytical solution, a rock
mass confinement adjustment factor is proposed. Although not strictly simulating the effect of initial horizontal stresses
on the end conditions of the beam, i.e. not theoretically satisfying the gradual change in bending mechanism from a
“free end” towards a “fixed end” condition, it is considered reasonable to assume that confining stresses provide some
increase in the rock mass modulus of deformation which would in turn reduce beam deflections.
The proposed adjustment factor is based on Ramamurthy (1995). The modified horizontal rock mass Young’s modulus
incorporating the effect of confinement, Esb,modified, can be estimated based on the relationship:

where scb is the block compressive strength at the appropriate scale, a is an empirical exponent to account for the
confinement effect, sh is the initial horizontal stress and Esb is the horizontal rock mass Young’s modulus of a single
bed or lamination which is the relevant modulus for the analytical solution. This value may be estimated by:

where Eib is the modulus of the “intact” rock block at the appropriate scale, kn is the normal stiffness of the subvertical
joints at a corresponding spacing sj. In estimating Esb,modified, an upper bound value of 1.5 to 2.5Eib is proposed for
Equation (3). For most of the analyses in this paper, a limiting value of 1.5Eib will be adopted.
Based on the database used by Ramamurthy (1995), the exponent a could be said to vary between 0.06 and 0.2 (Figure
4). However, for the effect on a voussoir beam, where the block strength to confining stress ratio, i.e. scb/sh, is typically
low, a value between 0.03 and 0.06 seems to provide reasonable results compared to numerical models as demonstrated
as follows. For the low end values of a = 0.03, a lower cut-off value of 1.5Eib is recommended whereas 2.5Eib would be
recommended for the high end value of a = 0.06.
Esb/Esb,modified

a=

sci/sh

Figure 4: Confinement effect on rock mass modulus (modified from Ramamurthy, 1995).

4.2 COMPARISON WITH DEM RESULTS

The following hypothetical rock design parameters were adopted for the analytical and numerical models:
§ Intact block Young's modulus, Ei = 12000 MPa
§ Single bed with thickness T = 1.0 m

4 AUSTRALIAN GEOMECHANICS VOLUME 51 NO. 4 DECEMBER 2016

LAMINATED ROCK BEAM DESIGN FOR TUNNEL SUPPORT OLIVEIRA AND PARAMAGURU

§ Vertical joint spacing, sj = 1.875 m

§ Joint normal stiffness, kn = 10000 MPa/m
§ Bedding shear stiffness, ks = 10000 MPa/m
§ Block scale UCS, scb = 23 MPa
§ Rock unit weight, g = 26 kN/m3
§ Joint/Bedding friction angle, f = 30º
§ Bedding dip = 0º
Figure 5 shows the maximum beam deflection against the span to bedding thickness ratio (z) for varying horizontal
stresses. The modified analytical estimate of beam deflection shows a good agreement with that of the numerical model.
For example, the analytical and numerical solutions resulted in a maximum beam deflection of 21 mm for the horizontal
stress of sh = 0 MPa. It is noted that the analytical and numerical solutions show reasonable agreement for ratios z <10
and the analytical solution could still be applicable for ratios z < 10.
Figure 6 shows the maximum beam deflection against horizontal stress for different z ratios. The analytical solution
with the proposed confinement effect adjustment seems to provide reasonable agreement with the numerical model,
particularly with respect to the overall effect of initial horizontal stresses on the rock beam behaviour. The increase in
initial horizontal stress significantly reduces the beam deflection, as previously demonstrated by Oliveira and Pells
(2014). For instance, the maximum beam deflections of 21 mm and 8.5 mm were estimated for horizontal stress
sh = 0 MPa and 4 MPa respectively for a constant span to thickness ratio z = 15.

s
h
s
h
s
h
s
h
s
h
s
h
s
h
s
h
s
h
s
h

Figure 5: Maximum beam deflection versus span/bed thickness ratio.

Figure 6: Maximum beam deflection versus horizontal stress for various span-to-bed thickness ratios.

AUSTRALIAN GEOMECHANICS VOLUME 51 NO. 4 DECEMBER 2016

LAMINATED ROCK BEAM DESIGN FOR TUNNEL SUPPORT OLIVEIRA AND PARAMAGURU

It is important to note that the DEM models require small element size definition to better capture the stress arching and
appropriate deflection prediction as pointed out by Oliveira and Pells (2014).

4.2.1 Influence of the confinement factor exponent

The sensitivity of the proposed confinement adjustment factor to higher values of the empirical exponent a has been
investigated. Figure 7 shows the variation of maximum beam deflection at sh = 0.5 MPa and 1.0 MPa for two values of
a but with the same cut-off value of 1.5Eib. As expected higher values of a reduce the confinement effect in the
estimate of the deformation modulus resulting in larger beam deflections. The effect is more pronounced in large span
to bed thickness ratios (z > 10). However, as previously discussed a value of a. = 0.03 seems to better represent the
numerical model results as show in Figure 5 and Figure 6.

Figure 7: Influence of confinement exponent value on beam deflection.

5 ADVERSE GEOLOGICAL FEATURES IN LAMINATED ROCK BEAM

Like any simplified analysis tool or model, the analytical solution proposed by Diederich and Kaiser (1999) has some
limitations. For example, it relies on the assumption that the joints are rough enough to provide frictional resistance
under low to moderate confinement (i.e. no slickensides or low friction coating). Translation and sliding failure along
joints at the abutments or within the beam is not considered in the iterative solution even though an assessment of factor
of safety for such mechanism is possible. In addition, the solution is not considered valid for low to mid angle jointing
where the angle between the plane of the cross-cutting discontinuities and the normal to the discontinuities sub-parallel
to the excavation plane is more than one third to one half of the effective friction angle of these joints. As a result,
Diederichs and Kaiser (1999) did not recommend the use of the voussoir beam method for poor rock masses with low
RQD ratings (< 50) and more than three joint sets. The theory also assumes uniform rock compressive strength across
the voussoir beam, weak zones located within the compressive regions may adversely affect the stress arching.
Due to the scepticism discussed above, there have been attempts to link some of the above limitations to unacceptable
levels of sensitivity for designs based on the voussoir beam analogy. For example, Peck et al., 2013 presented a number
of adverse conditions where the voussoir beam analogy would allegedly fail to identify instability. However, besides
pointing out significant errors in the sensitivity analyses carried out by Peck et al., 2013 Oliveira and Pells (2014) also
illustrated that the linear arch analytical model definition can be adjusted in terms of geometry and loading conditions to
better capture some adverse conditions as it is also required in more sophisticated DEM models in terms of explicit
discretisation.

6 AUSTRALIAN GEOMECHANICS VOLUME 51 NO. 4 DECEMBER 2016

LAMINATED ROCK BEAM DESIGN FOR TUNNEL SUPPORT OLIVEIRA AND PARAMAGURU

Figure 8: Examples of failure mechanisms affecting the immediate tunnel roof rock beam.
Oliveira and Pells (2014) also highlighted that it is unreasonable to only contemplate design scenarios that are
favourable for the voussoir beam analogy and require “perfect knowledge” of a “perfect rock mass”. The analyses must
give cognisance to the fundamental importance of rock bolting in providing a robust design against the uncertainties in
joint directions that may be expected in strata such as the Hawkesbury Sandstone of Sydney, the Bunter Sandstone of
the UK, the Beaufort Series in South Africa and the sandstones at Poatina in Tasmania (Oliveira and Pells, 2014).
Figure 8 presents a few examples of failure mechanisms that need to be addressed as part of the rock bolting design
process. Figure 8a shows a classical problem of potentially unstable rock blocks formed by an arched shape in bedded
rock. This particular example will be further investigated later in this paper with particular attention to the applicability
of the voussoir beam analogy. Mechanisms (a) to (d) of Figure 8 illustrate potentially unstable areas (shaded in red) that
would rely on a better performance of upper rock beams. In such cases, the unstable blocks and slabs maybe considered
as dead loads for the upper rock beams as a preliminary analysis approach. Figure 8e illustrates a more conventional
issue of the voussoir beam analogy for cases where thin laminations would be unstable at a certain span when behaving
as single rock beams, therefore requiring to be bolted to the upper roof.

6 BOLTING OF MULTIPLE BEDS/LAMINATIONS

As discussed above rock bolting is of fundamental importance in designs based on a linear arch theory in order to
address some of the uncertainties in the rock mass.
The main objective of the rock bolts is to stitch together near horizontal beds of limited and variable thickness to “trick”
the rock mass into behaving as an equivalent and appropriately thicker linear arch as depicted in Figure 9. The rock
bolts are designed to reinforce the rock mass allowing the development of the compression zone across the bedding

AUSTRALIAN GEOMECHANICS VOLUME 51 NO. 4 DECEMBER 2016

LAMINATED ROCK BEAM DESIGN FOR TUNNEL SUPPORT OLIVEIRA AND PARAMAGURU

partings as illustrated in Figure 9, therefore, controlling shear and slip along the bedding partings that would otherwise
result in delamination.

Figure 9: Equivalent rock bolt stitched rock beam (sub-vertical joints not shown for clarity).
The equivalent voussoir beam compressive stresses, f, may be estimated using the same methodology described for a
single bed with the difference that the new equivalent beam thickness, Teq, corresponds to the sum of the multiple
laminations. The development of the compression arch within the equivalent beam will generate both normal and shear
forces along the bedding partings which can be estimated by the relationships suggested by Asche and Lechner (2003)
given by: Another important aspect in the rock bolt design process is to consider that the mobilisation of the majority of
the reinforcement effect generally occurs upon some rock beam deflection and associated shear displacement along the
beddings despite some bolt pretension. As a result, in order to assess the mobilised bolt forces, it is necessary to
estimate the shear displacements resulting from the shear strain induced by the compression arch within the equivalent
beam. The shear displacements are then given by:

where
f(x) is the compressive stresses estimated by Equation (1), Bt is a normal stress induced by any rock bolt pretension
before significant rock beam deflections and q(x) is the angle between the line of thrust and the horizontal bedding
parting which may be found by differentiating Equation (2) resulting in:

The available bedding parting shear resistance and consequently excess shear can now be estimated using Equations (5)
to (7). The excess shear stress is then used to assess the required rock bolting forces to “trick” the multiple laminations
into behaving as an equivalent thicker rock beam and is given by:

where fb and cb are the friction angle and apparent cohesion of the bedding partings.
An important factor in the assessment of the equivalent bolted voussoir beam behaviour is a reasonable estimate of the
effect of the bedding partings on the overall rock beam deflections. The presence of the bedding partings and associated
shear displacements upon deflection mean that the rock mass modulus can no longer be estimated by Equation (4)
which only accounts for the normal stiffness of the cross-cutting joints. This estimate could be done by a direct
assumption of a modulus ratio, Esb/Gbeq, which often varies between 10 and 20 for transversely isotropic rock masses.
However, the authors of this paper have found that the following equation provides a reasonable estimate of the
equivalent voussoir beam modulus, Ebeq:

where ks is the shear stiffness of the bedding partings, n is the rock mass Poisson’s ratio (typically n = 0.25), Tsb is the
average thickness of the individual laminations or bedding parting spacing and all other parameters as previously
defined.

8 AUSTRALIAN GEOMECHANICS VOLUME 51 NO. 4 DECEMBER 2016

LAMINATED ROCK BEAM DESIGN FOR TUNNEL SUPPORT OLIVEIRA AND PARAMAGURU

Another important aspect in the rock bolt design process is to consider that the mobilisation of the majority of the
reinforcement effect generally occurs upon some rock beam deflection and associated shear displacement along the
beddings despite some bolt pretension. As a result, in order to assess the mobilised bolt forces, it is necessary to
estimate the shear displacements resulting from the shear strain induced by the compression arch within the equivalent
beam. The shear displacements are then given by:

It is important to note that during this design process, an upper bound for the deflection of the rock bolted laminated
beam is the deflection of a single lamination. In other words, in general, multiple laminations stitched by rock bolts
cannot deflect more than a single bed as this would mean delamination of the beams.

6.1 EXAMPLE OF LAMINATED ROCK BEAM BOLTING DESIGN

An example is provided in this section to illustrate the use of the equations above. The same parameters used in
previous comparisons between the analytical solution and DEM models have been used here (Section 4.2). Rock bolts
have been initially added at an angle of 90º to horizontal as an attempt to “trick” 3 laminations of 1 m thickness into
behaving as a single equivalent 3 m thick rock beam as shown in Figure 10. The analytical solution for rock bolt shear
resistance mobilisation adopted is that proposed by Pells (2002) and as modified by Carter (2003).
15 m

3m

Rock bolts @ 1.75 m c/c at 90º and pretension of 50 kN

Figure 10: Geometry of composite rock beam and rock bolting.

The results of the analytical analysis using the equations provided above are given in Figure 11 with a predicted
deflection of approximately only 5 mm. However, it is evident that the total bolt forces mobilised upon beam deflection
(area under black solid line) is significantly less than the excess shear forces (area under the purple solid line) indicating
that delamination is possible and the compressive arch as predicted in the equivalent voussoir beam cannot form.
The low mobilization is due to the small deflections resulting in small shear displacements. For example, as anticipated,
the largest shear displacement predicted in the analysis is near the abutment with a value approximately 0.75 mm. It can
be observed on the right hand plot of Figure 11 that a joint shear displacement of 0.75 mm would only mobilise a total
shear force of approximately 70 kN which divided by the tributary area of 1.53 m2 (one bolt spacing longitudinally and
half bolt spacing transversely near the abutment) only provides an additional shear resistance of approximately 46 kPa
as shown on the left hand side plot.

Figure 11: Analytical results for 3 m thick beam with sh = 0 MPa, bolts at 90º and ks = 10000 MPa/m.
As an attempt to improve the performance of the stitched laminated beam, the bolts are now considered to be installed
at a 70º angle to horizontal as shown in Figure 12. At such an angle, the rock bolts are likely to intersect the sub-vertical
joints and consequently generate additional axial forces. The bolts may also intersect multiple beddings within the
voussoir compression arch length at slightly different locations (Figure 17). For simplicity, to account for the bolts
crossing multiple beds, the total mobilised bolt force can be multiplied by MAX[N.(nb-1),1] where N is the depth ratio
for the triangular stress block and nb the number of beds. In addition, assuming that the rock bolts intersect the sub-

AUSTRALIAN GEOMECHANICS VOLUME 51 NO. 4 DECEMBER 2016

LAMINATED ROCK BEAM DESIGN FOR TUNNEL SUPPORT OLIVEIRA AND PARAMAGURU

vertical joint near the bolt-bedding plane intersection, it can be considered reasonable to assume that this additional
axial force is similar to that developed within a bedding plane. This has been confirmed in the numerical models (Figure
17). The latter additional axial forces can then be converted to additional normal forces and consequently frictional
resistance.
The results are provided in Figure 13. The rock beam behaviour remains essentially the same as the stiffness provided
by the rock bolts added to that of the bedding has only a small influence on the beam deflections. However, the rate of
rock bolt forces mobilisation is significantly higher which results in more rock bolt forces mobilised for the same shear
displacement. For example, full bolt shear resistance is mobilised at approximately 1.2 mm. In addition, due to the rock
bolts intersecting the sub-vertical joints, additional axial forces were included in the shear resistance as discussed above.
However, the mobilised bolt shear forces/stresses are below the excess bedding shear stress which means that the
compressive arch in the equivalent beam cannot develop.

Figure 12: New rock bolting geometry.

Figure 13: Analytical results for 3 m thick beam with sh = 0 MPa, bolts at 70º and ks = 10000 MPa/m.
One could ask what would be the consequences of the compressive arch not being formed in the previous analysis. As
depicted in Figure 9, the compressive arch generates both shear and normal forces across on the bedding parting. If such
forces are not resisted by the combined bedding and rock bolting shear resistance, plastic shear displacement or slippage
along the bedding parting takes place as illustrated by the red arrows in Figure 14. Such plastic shear displacements
induce further deflection of the rock beam which in turn could mobilise additional rock bolt shear resistance. A
simplified approach to capture such a mechanism is to artificially and iteratively reduce the bedding parting shear
stiffness which in turn reduces the equivalent thicker voussoir beam modulus (Equation 9). This process, as illustrated
in Figure 14, is carried until there is equilibrium between the excess shear forces and the rock bolt mobilised forces or
until the beam behaves as single laminations, if stable, or become unstable.

If excess shear beam still exists beddings will continue to

t slip (red arrow) resulting in beam deflection

i+1 i+2

ksi+1
ksi

ksi+2

us

Figure 14: Proposed iterative approach for rock bolt mobilisation upon plastic shear displacements.

10 AUSTRALIAN GEOMECHANICS VOLUME 51 NO. 4 DECEMBER 2016

LAMINATED ROCK BEAM DESIGN FOR TUNNEL SUPPORT OLIVEIRA AND PARAMAGURU

Figure 15 presents the results of the previous analysis adopting the proposed iterative approach. Equilibrium between
excess shear force and rock bolt mobilised shear forces is reached when the bedding shear stiffness is reduced to
approximately ks = 1500 MPa/m with a corresponding beam deflection of approximately 11 mm. The rock bolt
mobilisation curve is the same is that presented in Figure 13.

Figure 15: Analytical results for 3 m thick beam with sh = 0 MPa, bolts at 70º and ks = 1500 MPa/m.
A comparison between the analytical approach proposed above and DEM models is presented in Figure 16. A single
bed analysis is presented for reference. Like the analytical solution, the DEM model also indicates that the bolts do not
mobilise enough shear resistance for the development of the thicker rock beam if installed at a 90º angle. The beam
behaviour is compatible with that predicted by the analytical approach and similar to that of a single lamination which
is the limiting deflection for a stable beam.
Figure 17 shows the stress arching in the bolt stitched 3 m beam with rock bolts installed at 70º, sh = 0 MPa and
ks = 10000 MPa/m. The resulting deflection is approximately 11 mm, similar to the analytical prediction. The
reinforcement effect of the rock bolts and tensile zones (or zero stress) developed in the beam are evident through the
stress tensors in red colour, and consistent with the conceptual model discussed above. The effect of the rock bolts on
the sub-vertical joints can also be observed.

Figure 16: Analytical versus UDEC models for 3 m thick beam with bolts at 90º and 70º.

Figure 17: Stress tensor illustrating arching in DEM model - 3 m thick beam, bolts at 70º, sh = 0 MPa and
ks = 10000 MPa/m.
In summary, both analytical and UDEC models show similar trends in all cases analysed despite the small differences
in deflection (1-4 mm) likely resulting from the simplified assumptions on the effect of initial horizontal stresses,
equivalent beam modulus and also the differences in how the rock bolts are modelled in both solutions. As a result, the
analytical approach is considered to provide a reasonably satisfactory prediction of the overall performance of the
bolted laminated rock beam.

AUSTRALIAN GEOMECHANICS VOLUME 51 NO. 4 DECEMBER 2016

LAMINATED ROCK BEAM DESIGN FOR TUNNEL SUPPORT OLIVEIRA AND PARAMAGURU

6.2 A BRIEF NOTE ON SURCHARGE LOADINGS FOR VOUSSOIR BEAM ANALYSIS

In the case of single laminations of constant thickness, where there is reasonable rock cover capable of promoting a
stress arching, the beams are generally considered to separate or delaminate upon deflection so that each rock beam
transfers its own weight to the abutments without loading the laminated beam beneath. However, it is necessary to
account for the fact that not all rock beds have the same thickness. Thinner laminations located higher in the roof would
tend to have higher deflection imposing some loading on the lower laminated beams. This becomes even more evident
in the case of multiple laminations stitched by rock bolting where the increase in the “lower beam flexural stiffness”
constrains the deflection of the upper laminations resulting in some surcharge loading on the bolted rock beam. As a
result, it is generally needed to consider some loading in the design of laminated rock beam bolting.
Similar observation was also made by Obert and Duvall (1967) who pointed out that the behaviour of rock beams in a
tunnel roof is dependent on their relative flexural stiffness. For cases where the upper beam is stiffer than the lower
beam, these two beams would behave independently and delaminate from each other (Figure 18a). Where the lower
beam is stiffer, the upper beam would load the lower beam and conversely the lower beam would support the upper
beam (Figure 18b).

Figure 18: Behaviour of multiple beams with fixed end (after Obert and Duvall, 1967).
Based on a number of assumptions but mainly that the beams would have the same deflection over their entire length,
Obert and Duvall (1967) suggested that for a simplified two beam system the additional load to be added on the lower
beam and subtracted from the upper beam, could be estimated by:

where the subscripts “u” and “l” refer to the upper and lower beams respectively, w is the beam self-weight as a
uniformly distributed load, I is the second moment of area of the beam and E the deformation modulus. Similar
estimates could be made for an assumed number of beds. However, Obert and Duvall (1967) have not considered the
effect of sub-vertical cross-cutting joints and how it would affect the values of “l”, i.e. the voussoir analogy. In addition,
the designer would still have to assume a certain number of beams in the roof.
The authors of this paper have found that, as a first guess, a parabolic surcharge equivalent to an overburden of
approximately 0.25 times the span of the beam provides a reasonable and safe first estimate when comparing to DEM
models with enough rock cover. It is important to note that upon increasing deflections, one could consider that such a
surcharge reduces up to a deflection corresponding to full delamination where the single beds are assessed stable and
sustaining their own weight. If the material above the rock beam is considered unable to arch, this load would remain
constant upon deflection. In addition, this loading percentage could increase to 1 times the span above the rock beam
for a tunnel cover in soft ground, i.e. soil.

6.3 DIFFERENCES BETWEEN ROCK BOLT DESIGN APPROACHES IN LAMINATED BEAMS

It is important to note the difference in the underlying philosophy of the design approach adopted in this paper and that
proposed by Bertuzzi and Pells (2002). Bertuzzi and Pells (2002) philosophy is to use the displacements derived from a
jointed rock mass analysis to design a rock bolt reinforcement that provides greater capacity than the stresses derived
from an elastic equivalent continuum analysis. The use of the continuum analysis was justified by Bertuzzi and Pells
(2002) as means to take advantage of the high in situ stresses that were ignored in the jointed beam analysis.
The current approach focuses on satisfying the development of the compressive arch of the equivalent thicker voussoir
beam analysis while embeddding the effect of high initial horitonal stresses in the jointed beam analysis, though in a
simplified manner. The rock bolt reinforcement is designed to provide the necessary capacity to overcome the excess
shear stresses in the bedding partings. For the purpose of rock bolting design in laminated beams, equivalent elastic
continuum analysis are not strictly required in the current proposal.

12 AUSTRALIAN GEOMECHANICS VOLUME 51 NO. 4 DECEMBER 2016

LAMINATED ROCK BEAM DESIGN FOR TUNNEL SUPPORT OLIVEIRA AND PARAMAGURU

However it would still be necessary to assess the stress redistribution around the excavation and associated failure
mechanisms suchs as lamination buckling when behaving as column (Euler buckling), spalling and others. In doing so,
an equivalent elastic continuum analysis could be carried out, particularly considering a transversely isotropic
behaviour where elastic anisotropy related to the differences in vertical and horizontal modulus and a ratio of E/G = 10
to 20 can be captured. This can be easily achieved using programs such as Examine2D from Rocscience. The results of
this equivalent continuum analysis can then be used as an alternative check. The designer may verify the total shear
stress induced by the tunnel excavation and the normal stresses acting on the bedding partings at a certain depth into the
roof, say 0.5 m to 1 m, therefore being able to estimate the excess shear stress. The rock bolts assessed with the current
voussoir beam analogy approach can then be verified against the excess shear stress developed in the equivalent
continuum analysis.
In doing both verifications, the designer will target two primary objectives: (1) satisfy the formation of the compressive
arch in the voussoir beam analogy, and (2) attempt to maintain the roof (laminated rock beam) behaviour as close to an
elastic behaviour as practically possible where deflections are controlled to acceptable levels.

7 EFFECT OF A SLIGHTLY ARCHED ROOF

Although the voussoir beam analogy has been successfully used in the design and construction of a number of tunnels
with flat to semi-flat roofs, with high arch radius to roof span ratio, there is certain scepticism of its application in
slightly arched-roof tunnels. This question becomes even more relevant with the recent requirements for water resisting
linings on road tunnels in NSW where some arch is required to allow the secondary lining, which is not structurally
connected to the rock bolts, to span the water loads to the tunnel abutments. As a result, this section will investigate the
applicability of the voussoir beam analogy to arched tunnels based on the initial assumption that it is the bedded nature
of the rock that dictates its behaviour.
As depicted in the failure mechanism of Figure 8a, one of the negative effects of an arched profile in a horizontally
bedded rock is the potential formation of unstable side wedges that would tend to fall-out and “square out” the
excavation. These unstable wedges need to be bolted to an upper rock beam which will now have two potential
behaviours: (1) work as single beam that is loaded by the unstable wedge or (2) work as an equivalent voussoir beam
with a portion that is removed (see grey shade in Figure 19).

Figure 19: Geometry and boundary conditions for case 1.

The example given in Figure 19 will be analysed using both analytical and numerical approaches to in order to
investigate the effect of the arch on the voussoir beam behaviour. In this case a rise to span ratio of 0.127 m/m has been
adopted resulting in a haunch depth of approximately TH = 1.58 m. The following parameters have also been adopted:
§ Intact block modulus, Ei = 4000 MPa
§ Bedding thickness, b = 1.0 m
§ Vertical joint spacing, sj = 2.075 m
§ Joint normal stiffness, kn = 8000 MPa/m
§ Bedding shear stiffness, ks = 800 MPa/m
§ Intact block UCS, sci = 12 MPa
§ Rock unit weight, g = 26 kN/m3
§ Joint friction angle, f = 40º
As an initial assumption, the potentially unstable wedges are considered simply as dead loads applied to the upper 1 m
thick rock beam. This load is applied as an equivalent distributed load that applies the same moment at the abutment

AUSTRALIAN GEOMECHANICS VOLUME 51 NO. 4 DECEMBER 2016

LAMINATED ROCK BEAM DESIGN FOR TUNNEL SUPPORT OLIVEIRA AND PARAMAGURU

contact with the rock beam (Figure 20). This is conversion is consistent with the solution of the voussoir beam targeting
a compensated moment generated at the abutment due to self-weight of the beam with a resisting moment as the beam
deflects (Diederichs and Kaiser, 1999).

Figure 20: Initial assumption for analytical model case 1.

A second assumption is to consider that the arched voussoir beam still behaves as one laminated beam. However, as
suggested by Asche and Lechner (2003), two adjustments are necessary in the voussoir algorithm. The first is that the
effective density can be reduced to take account of the removal of rock mid-span, i.e. the beam does not have a constant
Teq. Secondly, in removing the rock, the average stress must rise in the arch, and the elastic shortening which is
calculated in the algorithm increases slightly. As an adjustment factor Asche and Lechner (2003) suggested that the
effective density could be estimated by:

Assuming that some material has been removed from the beam (Figure 19), Asche and Lechner (2003) suggested that
the lowest stress, used to calculate the average stress, must be higher than that presented by Diederichs and Kaiser
(1999) as depicted in Figure 2. As a result, assuming that the force is unchanged, the new lowest stress could be
estimated by;

It is important to note that the above corrections are made within the iterative approach proposed by Diederich and
Kaiser (1999). This means that the final value of fm found with the iterative approach is already the corrected value and
no further adjustments need to be made in Equation (1).
Figure 21 shows the maximum beam deflection against the horizontal stress for the arched tunnel roof analysis. The
differences in deflections may be the result of two main assumptions: (1) the simplified approach adopted to account for
the effect of the initial horizontal stress and (2) the estimates of rock beam deformation modulus are based on average
spacing of the discontinuities, i.e. bedding partings and joints, which are no longer constant in the case analysed. The
discrete discontinuities better captured in the DEM model result in stiffer rock beams and therefore with smaller
deflections. Nevertheless, the maximum deflections predicted are considered to show a satisfactory agreement with
similar trends. The analytical solution seems to over predict the beam deflections compared to that of the numerical
solution due to the factors discussed previously. It should be noted that one rock bolt per rock wedge have been
introduced at 90º angle to provide stability to the blocks underneath the beam and these have affected to the
deformation of the beam. The analytical solution is not sensitive to initial horizontal stresses greater than 0.5 MPa
whereas the numerical model still indicates some minor reduction in deflections.
A second case was analysed considering that the tunnel roof arch creates an undercut on the first “continuous” rock
beam in the roof, leaving a beam thickness of only 0.75 m as depicted in Figure 22. Similar to “Case 1”, as an initial
assumption, the potentially unstable wedges are considered as an equivalent distributed load applied to the upper rock
beam with a corresponding value of q = 4 kPa.

14 AUSTRALIAN GEOMECHANICS VOLUME 51 NO. 4 DECEMBER 2016

LAMINATED ROCK BEAM DESIGN FOR TUNNEL SUPPORT OLIVEIRA AND PARAMAGURU

Figure 21: Maximum beam deflection for arched tunnel roof - case 1.

Figure 22: Geometry and boundary conditions for case 2.

Figure 23 shows the maximum beam deflection against the horizontal stress for arched tunnel shown in Figure 22.
Similar to the previous case, the maximum predicted deflections are in reasonable agreement with the DEM model. The
lower deflections compared to “Case 1” are a direct result of the lower equivalent surcharge applied by the smaller
wedges (4 kPa compared to 11 kPa in Case 1).

Figure 23: Maximum beam deflection for arched tunnel roof - case 2.
Again, the differences in deflections may be explained by the simplified approaches adopted for both horizontal stresses
and rock bolt mobilisation. The combined cut-off values of 1.5Eib and a = 0.03 previously adopted for the effect of
initial horizontal stresses have a particular effect which can be observed in the shape of the curve of Figure 21 and
Figure 23. To investigate this effect, additional analyses were carried out assuming an increased cut-off value of 2.5Eib
with a = 0.06. The results are shown in Figure 24. As expected, there is a better agreement with the DEM predictions,
indicating that for low stiffness rock beams such as those with low intact block modulus or with an undercut arch,
higher values of a and effect cut-off value may be more appropriate, i.e. a = 0.06 and 2.5Eib respectively.

AUSTRALIAN GEOMECHANICS VOLUME 51 NO. 4 DECEMBER 2016

LAMINATED ROCK BEAM DESIGN FOR TUNNEL SUPPORT OLIVEIRA AND PARAMAGURU

Figure 24: Maximum beam deflection for arched tunnel roof with modified confinement factors- case 2.

8 CONCLUSIONS
A significant portion of the scepticism for the use of a voussoir beam analogy seems to be related to the perceived
limitations of the conceptual model and available analytical solutions with respect to the effect of horizontal stress, span
to lamination/bed thickness ratio, the presence of adverse geological features, the effect of a slightly arched roof. In
addition, there is also some concern on appropriate design methods for the design of rock bolting of multiple
beds/laminations in cases where single laminations are deemed unstable upon excavation.
The examples adopted in this paper sought to demonstrate that the effect of initial horizontal stresses may be taken into
account through a simplified approach though not strictly capturing the theoretical mechanism. They also demonstrated
that the solution still provides acceptable results for span to lamination thickness ratio less than 10, limit suggested by
Diederich and Kaiser (1999). The effect of an undercut arch was also investigated with satisfactory results found with
the analytical solution.
The primary objective of the rock bolting design approach adopted in this paper is to promote the development of the
compressive arch within the “stitched” equivalent rock beam. This is achieved by assessing the excess shear forces
developed within the equivalent beam and comparing with the mobilised rock bolt forces estimated through methods
such as those presented by Pells (2002).
As a result, this paper illustrates with several examples and discussion of the underlying assumptions that an analytical
solution of the voussoir beam analogy can be successfully used as a simplified tool for the assessment of roof support in
strongly bedded rocks where jointing orientations and horizontal stress conditions provide a favourable environment. It
is not suggested as the only means of assessment, particularly considering the significant advances in discontinuum
rock mass numerical modelling. However, the voussoir analogy provides the designers with a simple tool that can be
programmed into a spreadsheet and used to test scenarios and conditions in a quick manner, providing some insight into
expected behaviours.

9 REFERENCES
Asche, H.R. and Lechner, M.K. (2003). Design for the Cross City Tunnel, Sydney. Proceedings of the RETC. Seattle.
pp. 122-139.
Bertuzzi, R and Pells, P J N (2002). Design of rock bolt and shotcrete support of tunnel roofs in Sydney sandstone.
Australian Geomechanics, 37(3).
Booker and Best (1990). Coffey internal communication.
Carter, J. (2003). Pells analysis of the shear behaviour of a reinforced rock joint. Report by Advanced Geomechanics,
Sydney University,
Diederichs, M S and Kaiser P K. (1999): Stability Guidelines for Excavations in Laminated Ground - The Voussoir
Analogue Revisited, Int. J. Rock Mech. & Min. Sci.; 36, pp 97-118.
Evans W H. (1941). The strength of undermined strata. Trans. Inst. Min. Metall., 50:475-500.
Fayol, M. (1885). “Sur les movements de terrain provoques par l’exploitation des mines,” Bulletin de la Société de
l’industrie Minerale, 2nd Series, Vol. 14, pp. 818.
Itasca (2014). Universal Distinct Element Code User’s Guide.
Oliveira D. and Pells, P. (2014) Revisiting the applicability of voussoir beam theory for tunnel design in Sydney.
Australian Geomechanics Vol 49(3), pp 29-44, September

16 AUSTRALIAN GEOMECHANICS VOLUME 51 NO. 4 DECEMBER 2016

LAMINATED ROCK BEAM DESIGN FOR TUNNEL SUPPORT OLIVEIRA AND PARAMAGURU

Pells, P J N, Best, R J, and Poulos, H G (1994). Design of roof support of the Sydney Opera House underground
parking station. Tunnelling and Underground Space Technology, Vol 9, No 2, pp201-207.
Pells, P. J. N. (2002). Developments in the design of tunnels and caverns in the Triassic rocks of the Sydney region.
International Journal of Rock Mechanics and Mining Sciences 39: 569-587
Peck, W A, Sainsbury, D P and Lee, M.F (2013). The importance of geology and roof shape on the stability of shallow
caverns. Australian Geomechanics, 48(3).
Ramamurthy, T. (1995). Strength and modulus response of anisotropic rocks. In Comprehensive rock engineering (ed.
J. A. Hudson), Oxford, Pergamon Press, Vol. 1, pp. 313–329.

AUSTRALIAN GEOMECHANICS VOLUME 51 NO. 4 DECEMBER 2016

 Tunnelling and Underground Space Technology 64 (2017) 10–23

Contents lists available at ScienceDirect

Tunnelling and Underground Space Technology

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

Tunnel support for stress induced failures in Hawkesbury Sandstone

David Oliveira a,⇑, Mark S. Diederichs b
a
Jacobs and University of Wollongong, New South Wales, Australia
b
Geological Sciences and Geological Engineering, Queen’s University, Kingston, Canada

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

Article history: Underlying much of Sydney, Australia is a composite rock formation known as the Hawkesbury
Received 19 December 2016 Sandstone. This unit is composed of clastic layers of variable competency including a number of thick
Accepted 9 January 2017 and strong layers. The presence of very high horizontal stresses within these layers is widely accepted.
Deep excavations such as basements or open cuts in Hawkesbury Sandstone often experience moderate
horizontal movements in excess of 1 mm per metre of rock excavation. These movements can result in
various scales of damage and excess loads on supporting elements. Stress induced failures in tunnels
and underground excavations have also been observed in a number of projects and include crushing,
spalling and or slabbing of intact rock blocks or shear failures associated with planes of weakness.
While most design approaches in ground engineering account for shear failure mechanisms, the assess-
ment of brittle failure is less common and less well understood. Conventional models and failure criteria
do not appropriately describe such behaviour and consequently the impacts on ground support may not
always be appropriately addressed. This paper presents some discussions on the modelling and assess-
ment of brittle failure in Hawkesbury Sandstone and some of the impacts of high in-situ stresses on tun-
nel support design.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction Stress induced failures in tunnels and underground excavations

have also been observed in a number of projects. These failures can
The presence of relatively high horizontal stresses in Hawkes- be sub-divided into two distinct types:
bury Sandstone is widely accepted in Sydney. A number of papers
have been published presenting in-situ stress measurements and
Brittle failures involving crushing, spalling and or slabbing of
discussing different approaches for use in practice (Enever et al., intact rock blocks, often also associated with buckling of thin
1990; Enever, 1999; Pells, 2002; Bertuzzi, 2014; Oliveira and sandstone beds in Sydney; and
Parker, 2014).
Shear failures associated with planes of weakness, either pre-
The effects of such high stresses have also been well docu- existing or induced by the excavation process, such as faults,
mented in the literature. Deep excavations such as basements or cross bedding partings, and bedding shears.
open cuts in Hawkesbury Sandstone often experience moderate
horizontal movements even in good quality sandstone. Typical Several examples of brittle failure in tunnel projects are avail-
horizontal movements between 0.5 mm/m and 1 mm/m of rock able in the Australian literature, including the Boomerang Creek,
excavation and up to 2 mm/m in more adverse cases have often Malabar Ocean Outfall, Pipehead Tunnels, Cataract Tunnel and
been observed (Pells, 1990; Hewitt et al., 1999; Walker, 2004; Cross City Tunnel (Enever et al., 1990; McQueen, 2000, 2004;
Oliveira and Wong, 2012 and Oliveira and Chan, 2016). These Bertuzzi, 2015). Shear induced failure have also been observed in
movements not only may impact neighbouring structures causing tunnel projects in Sydney such as the tunnels under the Molineaux
aesthetic damages or in more adverse cases structural damages but Point for the Sydney LPG Storage Cavern project (De Ambrosis and
also impose significant loads on supporting elements such as piles Kotze, 2004) which may also have experienced some brittle failure.
or rock bolts that need to be accounted for in design. Most design approaches in ground engineering account for
shear failure mechanisms, thus, it is a relatively straight forward
assessment. However, the assessment of brittle failure is less
common and less well understood. Conventional models and
⇑ Corresponding author. failure criteria do not appropriately describe such behaviour and
E-mail addresses: David.Oliveira@jacobs.com (D. Oliveira), diederim@queensu.ca
(M.S. Diederichs).

http://dx.doi.org/10.1016/j.tust.2017.01.003
0886-7798/Ó 2017 Elsevier Ltd. All rights reserved.
 D. Oliveira, M.S. Diederichs / Tunnelling and Underground Space Technology 64 (2017) 10–23 11

consequently the impacts on ground support may not always be exhibit long term rock strength of approximately no more than half
appropriately addressed. of the UCS value at laboratory scale near excavation boundaries.
This paper presents some discussions on the modelling and With the above understanding, it becomes evident that stan-
assessment of brittle failure in Hawkesbury Sandstone and some dard shear failure criteria based on laboratory scale parameters
of the impacts of high in-situ stresses on tunnel support design. (e.g. conventional Mohr-Coulomb or Hoek-Brown) would over-
predict the field strength in relatively intact blocks as they do
not account for the strength reduction caused by the extensional
2. Brittle failure and conceptual models type damage initiation and propagation. As a result, a number of
studies have demonstrated that, for massive to blocky brittle rock
As discussed in Kaiser et al. (2010), various studies have demon- masses (GSI > 70), the rock blocks and rock mass strength is best
strated that brittle failure processes leading to breakouts often represented by either a tri-linear or s-shaped failure envelope as
dominate rock mass behaviour near excavation boundaries in high depicted in Fig. 2.
stress conditions rather than shear failure. In the context of rock According to Diederichs et al. (2004), Diederichs and Martin
failure, the term ‘‘brittle” is used to indicate spalling-type failure (2010), damage monitoring generally indicates that the low bound
due to tensile cracking or extensile fracture propagation leading field strength threshold is less sensitive to confining stress than the
to rapid cohesion loss, and not to describe the more general peak lab envelope. As a result, the limit for major principal stress
process of plastic post-peak strain-weakening and elastic, brittle- could be approximated by r1max = CI + (1–2)r3. This threshold
plastic shear failure. only applies at low confinements.
Diederichs (2014a,b) describes the mechanisms of transition At higher confining stress the strength envelope makes a tran-
from spalling to violent energy release known as rockbursting. sition up to the envelope defined by the yield or crack damage
Diederichs et al. (2010) also noted that spalling, while brittle in threshold (CD). This is based on the understanding that, at high
nature, need not be violent. In some cases, it could also be time confinements, micro-cracks initiated at grain-scale quickly
dependent. In unsupported conditions and under an anisotropic stabilize as they propagate away from the nucleation site, and
in situ stress field, the process of spalling can form notch geome- therefore the strength will be governed by crack coalescence and
tries often confused with wedge fallout. macro-scale shear failure.
At a laboratory scale, crack initiation (CI), as formally defined by The CD threshold is defined at the point of stress-axial strain
Diederichs and Martin (2010) is typically observed at approxi- non-linearity which coincides with the point of volumetric strain
mately 40–55% of the UCS, followed by a phase of stable crack reversal in uniaxial compression as shown in Fig. 1. The CI thresh-
growth and then crack coalescence and damage (CD) with old is generally observed at the point which a systematic increase
macro-scale shear failure (Fig. 1). However, sample heterogeneity in crack emissions follows an increase in applied stress and is also
increases at larger scales, thus, increasing local tensile failure and expressed by the point of lateral strain non-linearity (Fig. 1).
unstable cracking, i.e. coalescence. This explains why brittle rocks

Fig. 1. Typical stress versus strain plot for hard rocks in uniaxial compression showing interpretation of strain and acoustic emission counts related to damage evolution
(after Martin, 1997; Martin and Christiansson, 2009)
12 D. Oliveira, M.S. Diederichs / Tunnelling and Underground Space Technology 64 (2017) 10–23

3. Brittle failure envelope for Hawkesbury Sandstone

The applicability of the mechanism and failure criteria

described above for Hawkesbury Sandstone was investigated
through a conceptual numerical model that accounts for the frac-
turing mechanism. The model was developed with the commercial
DEM program UDEC. It was intended to simulate ‘‘virtual triaxial”
tests on intact samples at typical block scale of the Hawkesbury
Sandstone observed in tunnelling projects.
The rock substance was modelled assuming a conventional
strain-softening model adopting a deformation modulus Eb = 6 -
GPa, Poisson ratio t = 0.25, friction angle / = 45° and peak cohesion
cp = 5 MPa. This provides an equivalent rock substance at lab scale
of UCS  25 MPa which is a typical value for Hawkesbury Sand-
stone. A tensile strength T = 2.5 MPa, based on a simple assumption
of one-tenth of the UCS, and a dilation angle of w = 12° were also
assumed. Friction was assumed constant upon yielding but the val-
ues of cohesion, tensile strength and dilation were reduced to
residual values of 0.5 MPa, 0.025 MPa and 0 respectively upon a
plastic strain of 0.05.
Fig. 2. Schematic of failure envelope for brittle failure, showing four zones of Heterogeneity was introduced in the samples by generating
distinct rock mass failure mechanisms: no damage, shear failure, spalling, and random defects with the Voronoi Trigon algorithm of UDEC. A
unravelling (after Diederichs, 1999, 2003). maximum defect edge of 100 mm was adopted resulting in intact
pieces that are somewhat consistent with lab scale samples. The
defects were ‘‘bonded” together with a shear strength equivalent
Based on the above, Diederichs (2007) suggested that the above to a CD threshold, assumed at 80% of the UCS at lab scale. The peak
thresholds and behaviour could be represented, in conventional cohesion was assumed equivalent to be an upper limit of the CI
strain-weakening constitutive models, with a Generalized Hoek- threshold of 60% of the UCS, resulting in cdp = 7.5 MPa. The friction
Brown (GHB) failure criterion by adopting a ‘‘peak envelope” was then back-calculated as /dp = 17° such that the peak strength
described by: at zero confinement would result in approximately 20 MPa, i.e.
the CD threshold. The tensile strength was assumed at 80% of the
asp ¼ 0:25; ssp ¼ ðCI=UCSÞ1=asp and msp ¼ ssp ðUCS=jTjÞ ð1Þ substance at Tdp = 2 MPa and dilation of 12°. Upon yielding, cohe-
sion and tension were assumed to drop to 0.1 MPa and 0.02 MPa
in a brittle manner, i.e. zero shear plastic strain. On the other hand,
where T is the tensile strength of the rock at lab scale which may be
friction was assumed to increase to /dr = 50°. The dilation angle
estimated by UCS/mi. The ‘‘residual envelope” can then be defined
was assumed to drop to zero after a defect slip greater than
as:
1 mm. The ‘‘bonded” defects were given very high values of normal
and shear stiffness.
asr ¼ 0:75; ssr ¼ 0:0001ðfor numerical stabilityÞ and msr ¼ 7to 10 Fig. 3 presents some of the fracturing patterns obtained with
ð2Þ the virtual triaxial test described above. The tensile failure at zero
confinement is evident through the axial splitting and in accor-
The above conceptual behaviour is relatively simple and dance with the behaviour of Fig. 2. The increase of crack coales-
describes that the peak envelope is primarily cohesive in nature cence is also evident for increasing confinement.
converting to a more frictional behaviour upon cracking. This is The results of the virtual triaxial tests are compared with the
based on a simple principle that friction requires movement/slip GHB failure envelopes proposed by Diederichs et al. (2010). As pre-
within the cracks to mobilise. The model above can also be approx- sented in Fig. 4, there were very good agreement between the dis-
imated using best-fit Mohr-Coulomb linear envelopes for the peak continuum fracturing samples with the simplifying assumptions
and residual strength curves above. assumed above and the GHB envelopes. The s-shaped failure envel-
As a result, the above failure criterion can also be described by a ope is evident based on the test results.
combined strain-softening/strain-hardening Mohr-Coulomb by It is interest to note that despite the higher strength of the
adopting a Cohesion-Weakening-Friction Strengthening approach intact rock (substance) and the artificial defects (heterogeneity),
(CWFS) described by Hajiabdolmajid et al. (2002). In this approach for example 25 MPa and 20 MPa at zero confinement respectively,
cohesion and friction are mobilised independently. As a starting the overall strength is still lower and closer to the CI threshold.
point, for consistency with the GHB approach above, the values The GHB envelopes were developed assuming values of
of friction and cohesion may be estimated from the above envel- CI = 0.55UCS, CD = 0.8UCS and mi = 17. These assumptions resulted
opes. The strain-softening and hardening curves are then defined in T = 1.47 MPa, asp = 0.25, ssp = 0.0915, msp = 1.555, asr = 0.75,
by the plastic strain required to affect the change in cohesion ssr = 0.0001 and msr = 12.
and friction which may be approximated by: Considering that modelling of the fracturing process becomes
costly for larger scale models and use in practice, a new set of test
epf ¼ 2CI=Eb and epc ¼ CI=Eb ð3Þ were run assuming the rock block represented by a CWFS model
with no heterogeneity or artificial defects, i.e. fully continuum.
where epf is the plastic shear strain where full friction is mobilised, In the CWFS model, the rock block modulus was also assumed to
epc is the plastic shear strain beyond which a residual cohesion be Eb = 6 GPa and the peak and residual envelopes best-fitted from
applies and Eb is the assumed intact rock block deformation the GHB envelopes presented above. This resulted in a friction
modulus. angle /cp = 20°, peak cohesion ccp = 5 MPa, tensile strength
 D. Oliveira, M.S. Diederichs / Tunnelling and Underground Space Technology 64 (2017) 10–23 13

Fig. 3. Fracturing pattern after failure of virtual triaxial test samples (only compression shown).

3.Development of brittle failure in Hawkesbury Sandstone

As previously discussed, a number of tunnels projects have

experienced some form of brittle failure in Sydney. In general, if
not in all cases, the brittle failure was also controlled by geological
features in the tunnel roof such as bedding partings or siltstone
interbeds. These features not only affect the stress redistribution
in the roof but may also induce some buckling in the sandstone
bed, particularly considering the typical semi-flat to flat-roofed
tunnels in Hawkesbury Sandstone, and therefore contributing to
the brittle failure.
In order to understand the impact of such failure mechanism in
tunnel support design, it is also important to understand how such
brittle failure develops upon excavation advance. A 3D model using
the commercial Finite Difference program FLAC3D was developed
assuming the tunnel section illustrated in Fig. 5, i.e. with a total
Fig. 4. Comparison between failure envelopes for intact block scale (Diederichs, height of approximately 7.5 m and span of 13 m which is equiva-
2007; Diederichs et al., 2010) and virtual test results. lent to a 2 lane road tunnel in Sydney future-proofed for 3-lanes.
Three perfectly horizontal sandstone bedding partings have been
Tcp = 1.47 MPa and dilation angle wcp = 12°. The residual values included in the model starting 0.5 m above the tunnel crown.
were /cr = 58°, ccr = 0.1 MPa, Tcr = 0.2 MPa and zero dilation. The These beddings were included to simulate an unfavourable feature
cohesion was set to the residual value after a shear plastic strain in the roof and also release planes as typically observed in Sydney.
of epc = 0.002 whereas friction angle after epf = 0.005 was in accor- The tunnel was modelled with a 60 m rock cover in Sandstone
dance with Eq. (3). Tensile strain was set to residual after a tensile Class I/II based on the Sydney Classification System. The CWFS
plastic strain of epT = 0.01 and dilation after a shear plastic strain model parameters previously presented were adopted. The major
epw = 0.01. principal stress was assumed horizontal and perpendicular to the
The peak strength results of the continuum CFSW samples are excavation advance following a relationship of rH = 1.2 MPa
also shown in Fig. 4 demonstrating a reasonable agreement with + 3.8rv based on the database and approach discussed in Oliveira
both GHB envelopes and the results of the discontinuum fracturing and Parker (2014) where rv is the vertical stress assuming a unit
samples. As a result, the use of a CFSW approach is considered sat- weight of c = 24 kN/m3. The minor horizontal stress (intermediate
isfactory to capture the brittle behaviour and may be used in larger principal) is orthogonal to the major and assumed as rh = 0.61rH.
scale models for assessment of tunnel excavation and support The bedding partings were modelled assuming an elastic
requirements. perfectly plastic Mohr-Coulomb criterion with / = 35°, cohesion
14 D. Oliveira, M.S. Diederichs / Tunnelling and Underground Space Technology 64 (2017) 10–23

Fig. 5. Potential spalling of a 0.5 m thick bed above crown with respect to the horizontal LDP and horizontal GRC curves.

c = 5 kPa, zero tensile strength, normal stiffness kn = 8 GPa and in Sydney. There is, however, some excavation distance between
shear stiffness ks = 0.8 GPa. The excavation was carried out in the onset of fracture initiation and well developed spalling failure.
advances of 1.5 m adopting a softening relaxation approach to sim- This is typical of spalling interaction with thinner laminations in
ulate the gradual mechanical excavation. The longitudinal dis- the roof.
placement profile was monitored on side walls and roof as well The above scenario indicates that spalling of beddings with
as the ground reaction curve. Negative values represent a distance thickness of 0.5 m or less are likely occur near the face while the
ahead of the face. No support was modelled in this example. roadheader is still operating. In fact, such a case has been observed
It is important to note that the initial condition and depth were in a number of projects such as the existing M5 tunnel. As depicted
selected such that the excavation induced stress redistribution in in Fig. 6, the roadheader pick trimming marks indicate that the
the roof would approach or slightly exceed that of the CI threshold. brittle failure of a 0.2 m thick bed in the roof occurred relatively
However, such condition can be considered relatively common for close to the excavation face.
tunnels that exceed 40–50 m depth in Sydney. In these cases, the tunnel support is installed after the spalling
Fig. 5 presents the model results as a function of the horizontal event. Some additional rock bolting outside of the pattern would
tunnel displacement ratio, i.e. the ratio between a certain displace- likely be required to support the ‘‘cantilevering” rock at the border
ment and the final total displacement, with respect to the excava- of the fractured volume. Although some significant movements
tion face (LDP). It also shows the corresponding horizontal stress along the bounding bedding parting would already have occurred,
relaxation, i.e. the ground reaction curve (GRC). monitoring for additional shear movements post support installa-
The yield indicators in the numerical model indicate that crack tion, for example using endoscopes, would be recommended. The
initiates at approximately 3 m from the excavation face and spal- durability performance of the rock bolts could be affected upon
ling likely to occur at approximately 4.5–6 m from the face. The significant shear movements along beddings. This will be further
corresponding horizontal stress relaxation factors were in excess discussed later in the paper.
of 60% and 80% respectively. Such distances are within a typical Despite the higher likelihood of failure of thinner beds close to
1-day excavation round for tunnels excavated with roadheaders the excavation face, a generalisation is not possible. Variable bed-
 D. Oliveira, M.S. Diederichs / Tunnelling and Underground Space Technology 64 (2017) 10–23 15

Fig. 6. Brittle failure observed during excavation of the existing M5 tunnel (photo
courtesy of Golder Associates).

ding thickness may have an effect on the time of failure as it will be

demonstrated. In Sydney, such variable thickness is likely consid-
ering the typical longitudinal tunnel grades and the typical bed-
ding dip angles in Sydney between 0° and 10°. In addition, the
presence of fine-grained rocks in the Hawkesbury Sandstone, such
as siltstone interbeds, may also affect the timing of the brittle fail-
ure. Such rock types are more susceptible to time-dependent
effects and deformation where an ‘‘undrained behaviour” may
result in delayed stress redistribution and failure.
Two examples are given in Figs. 7 and 8. In both cases, failure
was only observed some considerable time after excavation and Fig. 8. Time-dependent brittle failure in sandstone below siltstone interbed at the
in both cases controlled by siltstone features. A failure depth of Cataract tunnel. Failure depth up to 2 m observed after construction (after
up to 2 m was observed in the Cataract tunnel which was fully McQueen, 2000).

unsupported and initially excavated with a squared shape.

A second case was analysed under the same conditions of the

previous model. However, the minimum thickness of the immedi-
ate roof bed was increased to 1 m as illustrated in Fig. 9. The
increased thickness not only affects the stress redistribution but
reduces the rock beam mid-span deflection, thus, affecting the
potential rock ‘‘crushing” effect.
The results are presented in Fig. 9. The brittle failure in this case
develops further away behind the face, initiating at approximately
15 m and likely spalling after some further 1–3 m excavation.
These distances would be equivalent to approximately 3 excava-
tion rounds in Sydney and horizontal stress relaxation in excess
of 90%. The model also indicated significant propagation of the
brittle failure backwards, i.e. towards the already excavated zone,
as observed by the difference in the volume of ‘‘yielding elements”
between the 15 m and 18 m excavation advances. Of special inter-
est here as well is the instantaneous development of the full spal-
ling zone immediately after the onset of crack initiation. This is due
to energy stored within the thicker laminations released into the
spalling zone immediately upon initiation. This can be compared
to the case of thinner laminations in Fig. 5, where there is a delay
between initiation and full propagation, although the yield process
begins closer to the face as mentioned.
In this case, based on current tunnel practices in Sydney which
requires man-entry only under supported ground, tunnel support
consisting of rock bolts spaced at 1.5–1.75 m centre to centre
square grid and steel fibre reinforced shotcrete with thickness of
50–75 mm would likely have been installed to a distance of
approximately 6 m from the face. The shotcrete closest to the face
Fig. 7. Time-dependent brittle failure in sandstone below siltstone bedding at the
would likely be some 2 days old.
Malabar Outfall decline. Additional rock bolts installed after fallout, one year after In such a case, the rock bolting design is critical to both suspend
excavation (after McQueen, 2004). the volume of the failed rock bed on the upper rock beam as also
16 D. Oliveira, M.S. Diederichs / Tunnelling and Underground Space Technology 64 (2017) 10–23

Fig. 9. Potential spalling of a 1 m thick bed above crown with respect to the horizontal LDP and horizontal GRC curves.

discussed in Oliveira and Paramaguru (2016), but also have enough Pre-emptive design attempts to reduce some of the adverse
deformation capacity to withstand the sudden formation of the effects are not always practical due to the magnitude of the hori-
potentially dilational spalling zone. The bolts on the flans of the zontal stresses in Sydney and its unpredictability. A number of fac-
spalling zone must also have capacity for shear resistance and/or tors may affect both the occurrence and timing of the failure as
shear deformation as lateral convergence normally flows the cen- discussed above. Firstly, brittle failure cannot be avoided with rea-
tral failure of the laminations. The shotcrete also requires enough sonable practical measures. Secondly, the variability of in-situ
capacity to hold the fractured rock block that may be formed stress, quality and strength of the rock mass and presence of
between bolts. This requires sufficient energy absorption to sustain adverse features suggest that brittle failure could occur at different
some reasonable deflection. In addition, allowances for potential depths. Although it could be argued that its likelihood increases for
re-bolting and shotcrete repair would also be necessary. These con- depths in excess of 40–50 m, brittle failure of thinner bed may
siderations are further explored in the following section. occur and have been observed at shallower depths.
Nevertheless, some allowances can and have to be made as part
of the tunnel design and construction process to address some of
4. Impact on tunnel support design in Hawkesbury Sandstone the potential impacts. These can be divided into two main
categories:
The existence of relatively high horizontal stresses is often con-
sidered one of the key features for a geological environment amen-
Safety issues
able to semi-flat roofed tunnels (Pells, 1993; Oliveira and Pells,
Durability issues
2014). Based on a voussoir beam analogy, horizontal stresses typ-
ically have a positive ‘‘clamping” effect on the tunnel roof rock
beam, thus, controlling deflections and improving tunnel support 4.1. Safety concerns
(Oliveira and Paramaguru, 2016). However, adverse effects may
be expected if excessive shear displacements in the vicinity of As briefly discussed above, stress induced failures in the tunnel
the bolts and/or stress induced failure occur as discussed above. roof may result in large volumes of rock that need to be suspended
 D. Oliveira, M.S. Diederichs / Tunnelling and Underground Space Technology 64 (2017) 10–23 17

in the upper zone of the roof. In addition, some of these failures This case study illustrates the importance of considering poten-
may result in significant shear displacements that may overload tial stress-induced failure in assessing the volume of potentially
rock bolts. In the particular case of brittle failure, the shotcrete unstable rock that required to be suspended deeper into the roof.
design also needs to provide enough capacity to hold the fractured This volume also needs to be considered as a surcharge onto the
rock block that may develop between rock bolts. Some of these upper roof rock beam. Similar observation is valid for brittle fail-
aspects are discussed as follows. ures as shown later in the paper.
In addition to the impact of stress induced failures on the
anchorage mechanism of the rock bolts discussed above, another
relevant impact is on the reinforcement action mode of the bolts.
4.1.1. Rock bolts
It is known that rock bolts may be subjected to significant shear
The two ground failures observed in the Sydney LPG caverns
deformation along rock discontinuities before a structural failure.
(De Ambrosis and Kotze, 2004) provide useful insights for the
For example, Fig. 12 presents predictions of forces acting on a rock
design of the rock bolts subjected to stress induced failures. The
bolt installed at a 80° angle from the rock discontinuity plane and
tunnel was approximately 14 m wide and 11 m high and was being
subjected to shear movements. These predictions are based on the
excavated by drill and blast of a 5.5 m high top heading followed
closed-form solutions proposed by Pells (2002) and Carter (2003)
by removal of the bottom bench.
and assumed a solid bar diameter of 21.7 mm, modulus of
At the time of the first failure it was being excavated using a
200 GPa, yield strength of 600 MPa, axial rupture strain of 15%
pilot heading followed by removal of the side strips and being sup-
and a pretension of 50 kN. The bolt is assumed reinforcing a bed-
ported with 4.0 m long, hollow, high tensile steel rock bolts at a
ding plane with friction angle of 35° with zero dilation. It is
pattern of 2 m  2 m spacing (De Ambrosis and Kotze, 2004). The
observed that tensile rupture is estimated at shear displacements
bolts were initially anchored using a mechanical expanding shell
of approximately 38–47 mm.
at the top end of the bolts followed by the application of cement
Although not in sandstone, Li (2010) presented a few case stud-
grout applied through the hollow stem once the face had advanced
ies where rock bolts were subjected to high stress environments
and roof convergence had ceased.
and post-failure observation could be well documented. For exam-
The first failure occurred during or immediately after blasting of
ple, Fig. 13a illustrates a failed 20 mm rebar bolt subjected to a
a side trip when the pilot heading was some 5 m ahead of the side
shear movement of approximately 40 mm consistent with the pre-
strip removal. Considering that the rock bolts would only be
vious estimate, despite the smaller diameter.
grouted after convergence had ceased, the support was primarily
Another example is the Sydney LPG caverns (De Ambrosis and
relying on the mechanical anchors (i.e. as a friction bolt) to support
Kotze, 2004). After the first failure, the roof support was altered
wedge or rock block loads. However, the combination of blast-
to include the use of 5 m long, hollow, high tensile steel bolts fully
induced fractures and shear induced failure particularly controlled
grouted at the face on a denser grid spacing of 1.5 m (transverse) 
by cross-bedding (potentially also blast affected), resulted in a
2.0 m (longitudinal). However, a second failure of similar magni-
large volume of unstable rock (Fig. 10). The thickness of the failed
tude occurred where a number of sheared rock bolt stubs were
sandstone slab was up to 1.5 m which overloaded the mechanical
observed at the failure surface in the crown of the cavern (Fig. 13b).
anchors unable to sustain such a load. As a result, a number of rock
Inspection of the failed rock on the floor of the tunnel showed that
bolts were simply pulled-out with the failed rock slab as depicted
the bolts had been fractured at the location of a bedding plane
in Fig. 11.

Fig. 10. Mapping of first roof collapse in the Sydney LPG caverns (modified after De Ambrosis and Kotze, 2004).
18 D. Oliveira, M.S. Diederichs / Tunnelling and Underground Space Technology 64 (2017) 10–23

Similar concept can be applied to the assessment of the capacity

required to support the volume of fractured rock upon a brittle fail-
ure. However, considering the intrinsic and inter-dependent rela-
tionship between rock spalling and stress, it is considered
reasonable to account for the effect of any residual horizontal
stresses in the kinematic stability of the fractured rock block above
an existing shotcrete lining.
The residual horizontal stress acting on the fractured block may
be estimated taking into account the confinement effect provided
by the shotcrete. Such confinement is associated with the typical
dilation and displacements that take place upon rock fracturing
and spalling. These movements induce a certain deflection on the
shotcrete and associated reaction forces, thus, providing some
minor confinement to the fractured rock block.
The failure mode is illustrated in Fig. 14 with a trapezoidal frac-
tured rock block bounded by an upper bedding plane. The side
angles assume com competent rock compression considering an
arching mechanism close to the rock bolts. The weight of the frac-
tured block can then be estimated by:

W ¼ cðs  H sin aÞHs ð4Þ

where c is the unit weight of the broken rock, H is the spalling/frac-

turing depth above the tunnel crown and s is the rock bolt spacing
assuming a square grid.
The force R represents the resisting forces along the contacts of
Fig. 11. Debris of first roof collapse in the Sydney LPG cavern showing pulled-out
rock bolts (after De Ambrosis and Kotze, 2004). the fractured block and, therefore, the overall factor of safety can
then be estimated as FoS = R/W.
Considering a stress relaxation mechanism and associated dis-
involving both bending and reduction of the cross-sectional area of placements along the fractured rock block contact, the following
the bolt, i.e. ‘‘necking”, which is indicative of shear movements limit equilibrium relationship can be derived for R (further details
similar to Fig. 13a. In addition, visual observation in an open rock in Brady and Brown, 2004).
bolt hole in the roof above the failure area indicated a lateral dis-
2r1 Hs sin a sinð/  aÞ
placement of 5–10 mm on a ‘‘non-failed” bedding plane at a mea- R¼ ð5Þ
sured height of 1.0 m above the failure surface. sin a
The above case study illustrates the importance of considering where r1 is the residual horizontal stress acting on the fractured
the potential stress-induced failures and associated loading block, / is the friction angle of the fractured block and a is the angle
imposed on the rock bolts. The rock bolts need to be designed with of competent compression in the vicinity of the bolts. It is important
adequate shear capacity and number. For example, hollow bars as to note that the above equation is manly valid for values of contact
adopted in the Sydney LPG caverns may not provide enough shear normal stiffness that are significantly greater than the contact shear
capacity unless compensated by a more closely spaced number of stiffness, i.e. kn >> ks.
rock bolts. The limiting value of r1 is dependent on the confinement pro-
vided by the shotcrete and the friction angle of the fractured block.
4.1.2. Shotcrete It can be estimate by:
One of the primary objectives of shotcrete in rock bolt rein- 1 þ sin /
forced roofs is to support potential unstable rock blocks and r1 ¼ r3 ð6Þ
1  sin /
wedges that may form between rock bolts. A common design prac-
tice in Sydney is that proposed by Barrett and McCreath (1995) where r3 is the limiting shotcrete confinement. The maximum
where four shotcrete failure modes are investigated including confinement is dependent on the shotcrete capacity associated with
adhesive, direct shear, flexural and punching failures. In this a particular failure mode, i.e. adhesive and direct shear or flexural
method, the effect of horizontal stress is conservatively ignored. and punching. For example, assuming that adhesive shotcrete

Fig. 12. Forces acting on a rock bolt subjected to shear movements.

 D. Oliveira, M.S. Diederichs / Tunnelling and Underground Space Technology 64 (2017) 10–23 19

Fig. 13. Examples of structural rock bolt failure: (a) Metattliferous mine in Sweden due to combined tension-shear displacements (Li, 2010). (b) Sheared rock bolt at second
roof collapse in the Sydney LPG caverns (after De Ambrosis and Kotze, 2004).

Fig. 14. Shotcrete failure mechanism for trapezoidal fractured rock block.

failure is the governing mechanism, the limiting equivalent shot- required to be sustained by the shotcrete could be estimated using
crete confinement could be estimated according to Barrett and empirical evidence of bulking based on Kaiser et al., 1996 and
McCreath (1995) as: shown in Fig. 15. In this figure, the degree of bulking (volume
expansion during brittle failure) is a very sensitive function of con-
ra a
r3 ¼ 4 ð7Þ fining pressure (support pressure). Bulking factor is the average
s
percentage of volume increase within the yielded zone.
where ra is the shotcrete adhesive/bond stress, a is an assumed An initial estimate of the yielded zone may be carried out using
adhesive bond length and s is the rock bolt spacing. empirical relationships as that proposed by Diederichs and Martin
For example, the limiting shotcrete confinement is estimated at (2010) and depicted in Fig. 16.
approximately r3 = 7 kPa assuming a = 30 mm and ra = 0.1 MPa for
an early age shotcrete with a rock bolt spacing s = 1.75 m. Such
confinement is equivalent to an approximate load of 12 kN/m over
the rock bolt spacing. The maximum residual horizontal stress
associated with such a confinement is approximately r1 = 74 kPa
for a value of / = 56° for the fractured rock.
Assuming a reasonably conservative value of a = 30° with a typ-
ical maximum spalling depth of H = 0.5 m, the residual horizontal
stress provides a resisting force R = 57 kN/m. The block weight is
W = 31.5 kN/m for an assumed c = 24 kN/m3 which results in an
overall factor of safety FoS = 1.8. If the effect of confinement on
the internal stability of the wedge is ignored, the factor of safety
is estimated as FoS = r3/(W/s2) = 0.7, i.e. failure, which highlights
the positive effect of the very small confinement provided by the
shotcrete.
It is important to note that the above confinement effect relies
on the shotcrete being able to withstand the deflection imposed by
the effects of spalling and therefore with sufficient energy absorp- Fig. 15. Sensitivity to support pressure of (empirical) bulking factor as a result of
tion to maintain a certain residual strength. The likely deflections brittle failure.
20 D. Oliveira, M.S. Diederichs / Tunnelling and Underground Space Technology 64 (2017) 10–23

For example, using a 2D approximation, if the average depth of guidance. In addition, the energy should also be corrected for shot-
damage/yield is 0.5 m above the roof, the estimated expansion into crete thickness and span (e.g. Bernard, 2013).
the tunnel would be 25 mm based on 5% bulking factor for bolts at This is consistent with experience in Sydney where lateral
1.75  1.75 m spacing (neglecting deformation tangential to or deformation upon spalling is not significant. It is worthwhile not-
parallel to the tunnel). This average expansion will be resisted by ing that such a deflection is associated with early ages of shotcrete
both the rock bolt and the shotcrete layer. While the total deforma- (i.e. <7 days) when it better performs upon large deformations
tion of the shotcrete layer will be more than this average, the rel- (Bernard, 2008). Nevertheless such a large deflection highlights
ative deflection between the bolt plates will be less. the need for repair.
Based on the restoring force required to be provided by the It is important to note however that other shotcrete failure
shotcrete discussed above, such a deflection would be equivalent mechanisms such as pure compressive failure and buckling also
to an energy absorption requirement of approximately 300 J at need to be investigated which is more appropriately carried out
the upper bound of 25 mm deflection (relative deflection of shot- through numerical modelling.
crete would be less than this, as discussed). This energy absorption
can generally be achieved with the use of fibre reinforcement (NGI,
4.1.3. Performance of the combined support
2013) alone without steel meshes. However, it should be
In order to investigate the above mechanisms, a case similar to
recognised that such energy requirement is not directly related
that analysed in the previous 3D analysis is re-analysed using a full
to the values achieved in test panels which therefore are used as
discontinuum model in UDEC.

Fig. 16. Empirical estimate of spalling depth (after Diederichs and Martin, 2010).

Fig. 17. Spalling post-tunnel support installation.

 D. Oliveira, M.S. Diederichs / Tunnelling and Underground Space Technology 64 (2017) 10–23 21

The Sandstone Class I/II is modelled with variable bedding strength, i.e. tensile strength, of 0.2 MPa, cohesion of 0.14 MPa,
thickness between 600 mm and 1.5 m and sub-vertical joint spac- friction of 35°, kn = 1 GPa and shear stiffness ks = 0.1 GPa.
ing varying from 2 to 4 m as depicted in Figs. 17 and 18. The rock After a relaxation of 95% the properties are increased to an
block was modelled assuming the CWFS parameters described in equivalent 7-day strength where E = 10 GPa, compressive strength
the previous 3D analysis. The bedding parting also follow the prop- of 24 MPa, tensile strength of 2.5 MPa and residual of 1.1 MPa. The
erties used in the 3D model whereas the joint deformation proper- shotcrete interface increases to bond strength of 0.3 MPa, cohesion
ties were adopted as kn = 10 GPa and shear stiffness ks = 1 GPa but of 0.21 MPa, friction of 35°, kn = 10 GPa and shear stiffness
with similar strength. ks = 1 GPa.
The tunnel is assumed to be excavated in full face with 3D The results are presented in Figs. 17 and 18. Brittle failure
effects simulated using a stress relaxation approach which controls occurs after installation of the tunnel support in agreement with
the excavation boundary forces. The tunnel support consists of the 3D model, considering that the immediate roof bed is
3.5 m long bolts at a spacing of 1.75 m by 1.75 m and 50 mm fibre approximately 0.5 m.
reinforced shotcrete which are installed after a stress relaxation of The effect of brittle failure in the overall behaviour of the tunnel
60%, i.e. stresses of approximately 40% of in-situ at time of support is reflected in the relatively large vertical movements which may
activation. be considered as a trigger level during monitoring of the excava-
The rock bolts are modelled using cable bolts in UDEC with the tion. The bulging, i.e. dilation, of the fractured rock and imposed
same parameters used to generate Fig. 12 including the allowance
of a rupture strain of 15% beyond which the capacity of the bolt
drops to zero. An ultimate bond stress of 2 MPa is assumed. The
dowel effect ignored in cable bolt elements is included with rein-
forcement elements adopting a shear stiffness, Ks, and shear capac-
ity, Ps,ult estimated by:

" #3=4
2Eg pffiffiffiffiffiffiffiffiffiffiffiffiffi
F bu rci
2
K s ¼ Eb I and Ps;ult ¼ 0:67d1 ð8Þ
4Eb Iðdd21  1Þ

where d1 is the rock bolt diameter (d1 = 21.7 mm), Gg is the grout
shear modulus (Gg = 12 GPa), Eb is the Young’s modulus of the rock
bolt (Eb = 200 GPa), d2 is the rock bolt hole diameter (d2 = 48 mm),
Eg is the grout Young’s modulus (Eg = 30 GPa), I is the rock bolt sec-
ond moment of area, Fby is the ultimate stress of the rock bolts
(Fbu = 920 MPa) and rci is the uniaxial compressive strength of the
intact rock (rci = 25 MPa). The axial effect of the reinforcement ele-
ments are set to zero.
Shotcrete properties were developed to account for the early
age. Between a stress relaxation of 60% and 95% it is modelled
adopting a Youngs modulus E = 6 GPa, Poison ratio t = 0.15, com-
pressive strength of 10 MPa, tensile strength of 1.2 MPa and resid-
ual of 1 MPa. The shotcrete interface is modelled assuming a bond Fig. 19. Benefits of split heading on tunnel support.

Fig. 18. Effect on support design and construction.

22 D. Oliveira, M.S. Diederichs / Tunnelling and Underground Space Technology 64 (2017) 10–23

Fig. 20. Damage on cable bolt plastic sheathing after 15 mm and 18 mm direct shear movement (after Bertuzzi, 2004).

deflection on the shotcrete is also evident with the exaggerated cases. Large volumes of rock may be generated as a result of stress
displacements. The small confinement acting on the failure planes induced failure.
within the fractured rock is also confirmed although not shown When rock and rock mass fails in a brittle and more tensile
here. manner, the ground behaves different close to the excavation
The example confirms the expected performance of the adopted boundary than far from it. Rock fails by spalling rather than by
rock bolts where structural capacity is satisfied. However, the shot- shear. As a consequence, conventional modelling techniques and
crete and three of the central rock bolts experience significant constitutive models may be inadequate to design tunnels and their
shear displacement that may affect their long term durability as support systems, and may mislead designers leading to undesired
discussed in the following section. Due to the separation (i.e. construction delays and unexpected costs. The use of brittle mod-
deflection) of the immediate roof bed, the shear displacements els that account for cracking initiation and development, thus, dif-
within the central portion of the roof are no longer available for ferent mobilisation of cohesion and friction more appropriately
plotting. represent the failure mode.
It is important to note that a full face was modelled as a critical It has been demonstrated that the use of more appropriate
case. However, considering the large span there are some benefits models allows for a better understating of the support
in adopting a split heading excavation as illustrated in Fig. 19. requirements. However, the several triggering factors required
for the initiation of the brittle failure mean that where and when
4.2. Impact on durability it may occur is relatively unpredictable. As a result, the use of
generalised pre-emptive changes in the support may be onerous
Infrastructure tunnel projects typically require a minimum and unnecessary.
100 year design life for the permanent support which often A reasonable approach is to focus on safety concerns during
consists of rock bolts and shotcrete in rock tunnels. excavation with respect to the effects of stress induced failures
The typical approach for the design of rock bolts in Australia is such as large volumes of rock that need to be suspended and the
to address durability concerns by developing redudant corrosion bulking effect caused by spalling, both to be controlled with ade-
protection measures. It is generally accepted that a double corro- quate rock bolt spacing and shotcrete performance. The primary
sion protection system using plastic sheating as one of the protec- objective is to achieve a safe temporary excavation with construc-
tion provides the required durability. tion measures that allow repairs and reinforcements where and
As previously discussed, rock bolts may be subjected to signifi- when necessary upon an observational approach, particularly to
cant shear deformation along rock discontinuities before a struc- address durability concerns.
tural failure. On the other hand, direct shear tests on cable bolts
grouted within concrete blocks carried out by Bertuzzi (2004) indi- References
cated that damage on the plastic sheating protection may initiate
at shear displacements of less than 15 mm (Fig. 20). If such shear Barrett, S.V.L., McCreath, D.R., 1995. Shotcrete support design in blocky ground:
Towards a deterministic approach. Tunn. Undergr. Space Technol. 10 (1), 79–89.
displacment is observed along a bedding or joint that intersects a
Bernard, E.S., 2008. Embrittlement of Fiber Reinforced Shotcrete, Shotcrete, vol. 10,
rock bolt, the 100-year design life could be compromised. No. 3. American Shotcrete Association, pp. 16–21.
Similar issues are applicable to the shotcrete. Excessive move- Bernard, E.S., 2013. Development of a 1200-mm-diameter round panel test for post-
ment and deflection after application of the shotcrete could result crack assessment of fiber-reinforced concrete. Adv. Civil Eng. Mater. 2 (1), 457–
471.
in crack widths that lead to significant corrosion and reduction of Bertuzzi, R., 2004. 100-Year design life of rock bolts and shotcrete. In: Proceedings
the energy absorption capacity of the steel fibre reinforced of the 5th International Symposium on Ground Support, 28-30 September,
shotcrete. Perth, Western Australia.
Bertuzzi, R., 2014. Sydney sandstone and shale parameters for tunnel design. Aust.
As a result, allowance for both rock re-bolting and shotcrete Geomech. 49 (1).
repair and crack injection may be required in case excessive shear Bertuzzi, R., 2015. Forming a view of the rock mass strength of Hawkesbury
movements are observed during construction. Sandstone from tunnelling case histories. In: 13th International Congress of
Rock Mechanics, 10-13 May 2015, Montreal, Canada.
Brady, B.H.G., Brown, E.T., 2004. Rock Mechanics for Underground Mining. Springer,
5. Conclusions Dordrecht.
Carter, J., 2003. Pells Analysis of the Shear Behaviour of a Reinforced Rock Joint
Report by Advanced Geomechanics. Sydney University.
As demonstrated above, stress induced failures in Hawkesbury De Ambrosis, L.P., Kotze, G.P., 2004. Stress induced roof collapses during
Sandstone have been observed on a number of projects in Sydney. construction of the Sydney LPG storage cavern. In: Proc. 9th ANZ Conf. on
Such failures have an impact on tunnel support designs that need Geomechanics, Auckland, 8-11 February, pp. 159–165.
Diederichs, M.S., 1999. Instability of Hard Rockmasses: The Role of Tensile Damage
to be addressed. For instance, rock bolting is not only dependent on and Relaxation Ph.D. thesis. Dept. of Civil Engineering, University of Waterloo,
the existing block sizes as typically considered in gravity fall-out Waterloo, Canada.
 D. Oliveira, M.S. Diederichs / Tunnelling and Underground Space Technology 64 (2017) 10–23 23

Diederichs, M.S., 2007. CGS Geocolloquium Award Lecture: Damage and spalling Martin, C.D., 1997. Seventeenth Canadian geotechnical colloquium: the effect of
prediction criteria for deep tunnelling. Can. Geotech. J. 44 (9), 1082–1116. cohesion loss and stress path on brittle rock strength. Can. Geotech. J. 34 (5),
Diederichs, M.S., Kaiser, P.K., Eberhardt, E., 2004. Damage initiation and propagation 698–725.
in hard rock tunnelling and the influence of near-face stress rotation. Int. J. Rock Martin, C.D., Christiansson, R., 2009. Estimating the potential for spalling around a
Mech. Mining Sci. 41 (5), 785–812. deep nuclear waste repository in crystalline rock. Int. J. Rock Mech. Min. Sci. 46
Diederichs, M.S., 2003. Rock fracture and collapse under low confinement (2), 219–228.
conditions. Rock Mech. Rock Eng. 36 (5), 339–381. McQueen, L.B., 2000. Stress relief effects in Sandstone in Sydney underground and
Diederichs, M.S., Carter, T., Martin, C.D., 2010. Practical rock spall prediction in deep excavations. In: McNally, G.H., Franklin, B.J. (Eds.), Proceedings of the
tunnel. In: Proceedings of World Tunnelling Congress ’10 – Vancouver, p. 8. ‘Sandstone City’ symposium at the 15th Australian Geological Convention.
Diederichs, M.S., Martin, C.D., 2010. Measurement of spalling parameters from Geological Society of Australia (Monograph No. 5).
laboratory testing. In: Proceedings of Eurock 2010, Lausanne, p. 4. McQueen, L.B., 2004. In situ rock stress and its effect in tunnels and deep
Diederichs, M.S., 2014a. Managing rockburst risk in D&B tunnels. In: Tunnelling excavations in Sydney. Aust. Geomech. 39 (3), 43–58.
Association of Canada Conference. Vancouver, p. 8. Oliveira, D.A.F., Wong, P.K., 2012. Selection of rock mass design parameters for
Diederichs, M.S. 2014b. When does brittle failure become violent? Spalling and assessing excavation induced movements in the Sydney CBD. In: Proceedings:
rockburst characterization for deep tunneling projects. In: Proceedings of the ANZ 2012 Conference, Melbourne, 15–18 July, pp. 789–795.
World Tunnel Congress 2014, Iguazu, Brazil, p. 10. Oliveira, D., Parker, C., 2014. An alternative approach for assessing in-situ stresses in
Enever, J.R, Walton, R.J., Windsor 1990. Stress regime in the Sydney basin and its Sydney. In: Proceedings: Australasian Tunnelling Conference 2014, Sydney, pp.
implications for excavation design and construction. Inst. Eng. Aust. VII 189–194. September.
Tunnelling Conf. Sydney, pp. 49–59. Oliveira, D., Pells, P., 2014. Revisiting the applicability of voussoir beam theory for
Enever, J.R., 1999. Near surface in situ stress and its counterpart at depth in the tunnel design in Sydney. Aust. Geomech. 49 (3), 29–44. September.
Sydney metropolitan area. In: Proc. 8th ANZ Conf. on Geomechanics, Hobart, pp. Oliveira, D., Chan, G., 2016. Ground control for a deep basement excavation in
591–597. Sydney’s GPO fault zone. In: Proceedings: Australian Geomechanics Society –
Hajiabdolmajid, V., Kaiser, P.K., Martin, C.D., 2002. Modelling brittle failure of rock. Sydney [Chapter 2016 Symposium, November].
Int. J. Rock Mech. Min. Sci. 39 (6), 731–741. Oliveira, D., Paramaguru, L., 2016. Laminated rock beam design for tunnel support.
Hewitt, P.H., McQueen, L.B., Davies, P.R., 1999. Genting Centre, Sydney – Deep Aust. Geomech. 51 (3).
Excavation Adjacent to Railway Tunnels. In: Proceedings of the 8th Australia NGI (Norwegian Geotechnical Institute) 2013. Rock Mass Classification and Support
New Zealand Conference on Geomechanics, Hobart,. Australian Geomechanics Design Handbook. NGI (<www.ngi.no>). p. 57.
Society, pp. 611–617. Pells, P.J.N., 1990. Stressed and displacements around deep basement in the Sydney
Kaiser, P.K., Amann, F., Steiner, W., 2010. How highly stressed brittle rock failure area. Seventh Australian Tunnelling Conference, Sydney, pp. 241–249.
impacts tunnel design. In: Zhao, J. et al. (Eds.), Proceedings of European rock Pells, P.J.N., 1993. Rock mechanics and engineering geology in the design of
mechanics symposium, Lausanne, pp. 27–38. underground works. EH Davis Mem. Lect., Aust. Geomech. 25. August.
Kaiser, P.K., McCreath, D., Tannant, D., 1996. Canadian Rockburst Support Pells, P.J.N., 2002. Developments in the design of tunnels and caverns in Triassic
Handbook. Geomechanics Research Centre and CMIRO Sudbury, Ontario. rocks of the Sydney region. Int. J. Rock Mech. Mining Sci. 39, 569–587.
Li, C.C., 2010. Field observations of rock bolts in high stress rock masses. Rock Mech. Walker, B., 2004. Stress relief on hillsides and hillside excavations. Aust. Geomech.
Rock Eng. 43, 491–496. 39 (3), 59–72.
 REVISITING THE APPLICABILITY OF VOUSSOIR BEAM THEORY FOR
TUNNEL DESIGN IN SYDNEY

D.A.F. Oliveira1 and P.J.N. Pells2

1
Associate Geotechnical Engineer, Coffey, Sydney and Research Fellow, University of Wollongong, Australia,
2
Senior Geotechnical Consultant, Pells Consulting, Sydney, Australia.

ABSTRACT
The design of semi flat-roofed tunnels, i.e. with a high arch radius to roof span ratio, in Sydney has been proven
successful over time. The horizontally bedded nature of Sydney’s Hawkesbury Sandstone draws designers to the
voussoir beam theory. Such analogy and the associated method of analysis can be easily implemented in computer
spreadsheets, which significantly facilitates the design of semi flat-roofed tunnels in geological conditions such as
Hawkesbury Sandstone. Like any other engineering simplified model or theory, the voussoir beam theory has some
limitations. However, it seems that some of these limitations are not well understood and often ignored and/or
misinterpreted. Such lack of understanding of the theory and its limitations often raises question about the applicability
of the analytical solution in practice, misleading engineers to believe that the only reliable and comprehensive design
method is through numerical analysis such as Distinct Element Method (DEM). This paper investigates the applicability
of the analytical or closed-form solution of the voussoir beam theory through comparison with numerical modelling,
focusing on some of the perceived limitations and their impact on the design of tunnels in Sydney. The results illustrate
that the voussoir beam theory can be confidently used in practice if its limitations are well understood and good
engineering judgment is applied to take the local geology into account. In addition, the results also demonstrate that
some of the limitations can be on the conservative side. For example, the potentially positive effect of high horizontal
stresses ignored in the voussoir beam theory may explain why some of the unfavourable conditions are less pronounced
in practice.

1 INTRODUCTION
According to Diederichs and Kaiser (1999), the notion of a linear arch theory often referred as voussoir beam was first
proposed by Evans (1941) specifically to explain the stability of a jointed or cracked beam. After some initial
controversy when first published, the voussoir beam analogue has been generally accepted and has since been revisited
by several authors (Pells et al., 1994; Sofianos, 1996; Diederichs and Kaiser, 1999) and presented as a simplified tool
for stability analysis of excavations in civil construction and in mining.
The horizontally bedded nature of the Hawkesbury Sandstone in Sydney draws designers to the voussoir beam or linear
arch theory. The theory describes the conditions for the stability of unsupported flat roof spans by means of a rock beam
analogy with joints sub-orthogonal to the tunnel span. It accounts for the arching effect of the vertical stresses and loads
upon deflection of the tunnel roof which in turn promotes compressive zones across the joints and consequently the
stability of the rock beam. Diederichs and Kaiser (1999) present the four main failure mechanism associated with a
voussoir beam as illustrated in Figure 1.

(c)
(a) (b)

(d) (e)

Figure 1: (a) Voussoir beam analogue with compression arch shown; Failure modes of the voussoir beam: (b) snap-
through; (c) crushing; (d) sliding and (e) diagonal cracking (modified after Diederichs and Kaiser, 1999).
The analytical solution is based on an iterative analysis of beam deflections which is then used for estimates of
maximum compressive stress, σmax, developed through the arching mechanism. Such estimates are then adopted to
assess factors of safety against crushing, sliding or shear at the abutments and a Buckling Limit (BL) Index for the

Australian Geomechanics Vol 49 No 3 September 2014 29

REVISITING THE APPLICABILITY OF VOUSSOIR BEAM THEORY FOR TUNNEL DESIGN IN SYDNEY
OLIVEIRA AND PELLS

snap-through mechanism. Diederichs and Kaiser (1999) state that a BL index of 35% indicates yielding of the voussoir
beam and 80% collapse.
The voussoir beam theory has limitations like any other simplified engineering model or theory. However, it seems that
some of these limitations are not well understood and often ignored and/or misinterpreted which may raise questions
about its applicability in practice.

2 LIMITATIONS OF THE VOUSSOIR BEAM THEORY

Diederichs and Kaiser (1999) pointed out a few limitations of the voussoir beam theory and their analytical solution.
Some that are considered more relevant for the design of civil tunnels in Sydney are discussed as follows.
The analytical solution is valid for span to thickness ratios greater than 10 and relies on the assumption that the joints
are rough enough to provide frictional resistance under low to moderate confinement (i.e. no slickensides or low friction
coating). Sliding failure along joints at the abutments or within the beam is not considered in the iterative solution even
though an assessment of factor of safety for such mechanism is possible.
The solution is mostly valid for no low to mid angle jointing where the angle between the plane of the cross-cutting
discontinuities and the normal to the discontinuities sub-parallel to the excavation plane is recommended less than one
third to one half of the effective friction angle of these joints. As a result, Diederichs and Kaiser (1999) did not
recommend the use of the voussoir beam method for poor rock masses with low RQD ratings (< 50) and more than
three joint sets. In addition, as the theory assumes uniform rock compressive strength across the voussoir beam, weak
zones located within the compressive regions may adversely affect the stress arching.
Peck et al. (2013) seek to illustrate unacceptable sensitivity of designs based on voussoir beam theory by means of
simple numerical models which are shown in this paper to be misleading. In addition, their discussion also seems to
imply that the only reliable and comprehensive design method is through numerical analysis such as Distinct Element
Method (DEM), and analytical solutions are not capable of dealing with such geological variations. Figure 2 is a
reproduction of those models, each of which will be addressed in this paper. We note, up front, the following important
points:
1. The analyses presented by Peck et al. (2013) do not include the impact of horizontal virgin stresses in the
beam, let alone additional horizontal stresses that are induced by the excavation when the virgin horizontal
stresses are equal to or greater than overburden pressure. The existence of relatively high horizontal stresses
are the second key feature in determining a geological environment amenable to flat or semi-flat roof designs
(Pells, 1993).
2. The models are for a 1 m thick bed spanning 15 m, and no sane rock mechanics engineer would contemplate
such a scenario, at least not without expecting some sort of roof instability, as it requires perfect knowledge of
a perfect rock mass. Peck et al. (2013) may well have chosen this extreme scenario to illustrate points of
applied mechanics, but the case is so extreme that even when their calculations were valid they are considered
irrelevant.
3. The Peck et al. (2013) analyses give no cognisance to the fundamental importance of rock bolting in designs
based on linear arch theory. As shown by Pells et al. (1994) and detailed in analyses presented below, it is
these rock bolts that stitch together near horizontal beds of limited and variable thickness, to ‘trick’ the rock
mass into behaving as an appropriately thick linear arch, robust against the uncertainties in joint directions that
may be expected in strata such as the Hawkesbury Sandstone of Sydney, the Bunter Sandstone of the UK, the
Beaufort Series in South Africa, and the sandstones at Poatina in Tasmania.
This paper revisits the applicability of the analytical or closed-form solution of the voussoir beam theory through
comparison with numerical modelling focusing on some of the above-perceived limitations and their impact on the
design of tunnels in Sydney. In addition, other potentially positive factors such as the effect of high horizontal stresses
ignored in the analytical solution, are investigated as it may explain why some of the more unfavourable conditions are
less pronounced in practice.

30 Australian Geomechanics Vol 49 No 3 September 2014

REVISITING THE APPLICABILITY OF VOUSSOIR BEAM THEORY FOR TUNNEL DESIGN IN SYDNEY
OLIVEIRA AND PELLS

Figure 2: Sensitivity of the voussoir beam analogue to relatively minor geotechnical variations (after Peck et al., 2013

3 ANALYTICAL AND NUMERICAL SOLUTIONS

One of the most recent analytical solutions for analysis of voussoir beam, perhaps currently the most widely used, is
that proposed by Diederichs and Kaiser (1999).
In order to make a proper assessment of the concerns expressed by Peck et al. (2013), analyses have been undertaken
using an extension of the analytical methodology presented by Diederichs and Kaiser (1999) and a Distinct Element
Model as encapsulated in UDEC V4.0 (Itasca, 2006). Abutment deflections have been included, according to Asche and
Lechner (2003). The analytical model code has also been checked against the 1D finite element method developed by
Booker and Best (1990) as used for the design of the Sydney Opera House Carpark – Pells et al. (1994). The Booker
and Best (1990) method accounts for the effect of horizontal stresses on a “cracked beam” analysis. The cracked beam
model is not discussed in this paper but Figure 3 shows a comparison between the analytical code used herein, with one
of the calculations for the Opera House Carpark. The agreement is considered satisfactory.

Figure 3: Comparison of the analytical solution adopted in this paper and Pells et al. (1994).

Australian Geomechanics Vol 49 No 3 September 2014 31

REVISITING THE APPLICABILITY OF VOUSSOIR BEAM THEORY FOR TUNNEL DESIGN IN SYDNEY
OLIVEIRA AND PELLS

4 ASSESSMENT OF THE LIMITATIONS POSTULATED BY PECK ET AL. (2013)

4.1 PROCEDURE
In the analyses presented below, the same parameters as given by Peck et al. (2013) in Figure 2 have been used with the
unit weight given in Figure 4. The intact rock was represented by an elastic-plastic behaviour with a Mohr Coulomb
failure criterion assuming φ = 34º, c = 6 MPa (which combined results in an UCS = 23 MPa) and σt = 3 MPa. In
addition, in order to model brittle failure, the intact rock was assigned a strain softening behaviour whereby the
cohesion is gradually reduced upon development of plastic strain to a minimum value of 10% of the initial cohesion at a
5% plastic strain. For the analytical solution, it is important to convert the discontinuum parameters into an equivalent
continuum. The horizontal rock mass Young’s modulus, Eh,rm, which is the relevant modulus for the analytical solution
is estimated by:
Ei k n s (1)
Eh ,rm =
Ei + k n s
where Ei is the intact rock Young’s modulus (Ei = 12 GPa), kn is the normal stiffness of the joints (kn = 10 GPa/m) and s
is the joint spacing (s = 1.875 m). The resulting horizontal rock mass modulus is Eh,rm = 7.3 GPa.
Similar to the numerical model adopted by Diederichs and Kaiser (1999) during their development of the analytical
solution, the beam was first allowed to deflect elastically while maintaining a non-zero tensile strength within the joints.
This initial elastic deflection is assumed equivalent to a gradual excavation mechanism such that some beam
deformation occurs before the roof is fully excavated and formed. After the initial elastic equilibrium was achieved, the
joint tensile strength was set to zero and the beam was allowed to continue deforming until either equilibrium or failure
occurred. The joints had no cohesion or dilation but had a frictional strength. One main difference between the model
adopted in this paper and that of Diederichs and Kaiser (1999) is that the abutment blocks are given the same
deformability parameters of the rock beam instead of a very stiff abutment. Figure 4 presents the typical boundary
condition for the UDEC models.

15 m

1m
γ = 26 kN/m3

Figure 4: Typical boundary conditions of UDEC models.

4.2 BASE CASE

The base-case (Figure 2a) has perfectly vertical joints (dip 90º) consistent with the voussoir model presented by
Diederichs and Kaiser (1999). Peck et al. (2013) numerical model and analytical solution resulted in a stable rock beam
with deflection predictions of 13 mm and 13.5 mm respectively
The analytical results of the base case run by the writers of this paper are presented in Figure 5 and Table 1. In Figure 5,
results are shown for different beam thicknesses other than one metre to illustrate the thickness where the beam would
start to experience instability (i.e. yielding) and potential collapse under such “geological condition”. As depicted, a 1 m
thick beam is assessed stable with predicted mid-span deflection of approximately 21 mm in contrast to the 13.5 mm
predicted by Peck et al. (2013). A deflection of 13 mm is only predicted if the intact rock modulus (i.e. Ei = 12 GPa) is
used in the analytical solution in contrast to a value of Eh,rm = 7.3 GPa which includes the effect of the vertical joints.
Peck et al. (2013) present other models in their paper where a value of kn = 100 GPa/m was used. If there was a typo
and the latter normal stiffness was used in their examples, the equivalent horizontal rock mass modulus would indeed
be close to the intact rock modulus. However, such a normal stiffness is unrealisticaly high for an intact rock with Ei =
12 GPa and the level of stresses in the current analyses.

32 Australian Geomechanics Vol 49 No 3 September 2014

 REVISITING THE APPLICABILITY OF VOUSSOIR BEAM THEORY FOR TUNNEL DESIGN IN SYDNEY
OLIVEIRA AND PELLS

1.6 1.6 1.8

1.4 1.4 1.6

Collapse
Yielding

Yielding
Collapse
1.4

Beam  thickness  (m)

Beam  thickness  (m)

Beam  thickness  (m)

1.2 1.2

Yielding

Collapse
1.2
1.0 1.0
1.0
0.8 0.8
0.8
0.6 0.6
0.6
0.4 0.4 0.4
0.2 0.2 0.2
0.0 0.0 0.0
0 50 100 150 0% 20% 40% 60% 80% 100% 0 5 10 15 20
Midspan  deflection  (mm) Buckling  limit  index  (BL) σmax (MPa)

(a) (b) (c)

Figure 5: Analytical voussoir beam solution for different beam thicknesses. (a) Mid-span deflection; (b) Buckling limit
index (BL); (c) Maximum compressive stress.
Table 1: Factors of safety predicted for the base case with analytical solution.
Failure Mode Snap-through Crushing Sliding/Shear
Factor of Safety 2% (Buckling Limit Index) 3.7 4.2
The base case UDEC model (Figure 6) also indicates a deflection of approximately 19 mm which is similar to the
analytical prediction but also in contrast to the 13 mm predicted by Peck et al. (2013). As illustrated in Figure 7, it is
possible that Peck et al. (2013) have used coarse finite difference zones which resulted in a beam behaviour that is too
stiff and consequently with smaller deflections.
The effect of increasing initial horizontal stress causes the beam to behave in a more elastic manner (Figure 6) which
tends to approach an uncracked beam fixed at both ends (abutments). The deflection of such elastic beam behaviour can
be estimated by:

ql 4 (2)
δ=
384EI
where q is the uniformly distributed load acting on the beam (in the base case only the self-weight), E is the relevant
beam Young’s modulus (i.e Eh,rm = 7.3 GPa) and I is the uncraked second moment of area of the beam.
STABLE

(a) Deflection (mm)

0 10 20 30 40 50

STABLE

(b)

(c)
Figure 6: Deflection contours for base case. (a) σh = 0 MPa; (b) σh = 4 MPa; (c) effect of initial horizontal stress.
It is important to note that the effect of horizontal stress has been modelled as an initial value, i.e. an in situ stress prior
to any excavation which changes only as a function of the rock beam deflection but not as a result of stress
redistribution of the excavation. In other words, it does not fully represent the stress redistribution and potential increase
induced by the tunnel excavation. However, the range of initial horizontal stresses adopted in the analyses is considered
representative of the expected horizontal stress values and roof beam behaviour for tunnels depths varying from 5 m to
40 m in Sydney’s Hawkesbury Sandstone.

Australian Geomechanics Vol 49 No 3 September 2014 33

REVISITING THE APPLICABILITY OF VOUSSOIR BEAM THEORY FOR TUNNEL DESIGN IN SYDNEY
OLIVEIRA AND PELLS

STABLE

(a)
Deflection (mm)
0 4 8 12 16 20

STABLE

(b)

Figure 7: Deflection contours for base case showing finite difference zone sizes: (a) fine; (b) coarse.

4.3 SINGLE STEEPLY DIPPING JOINT

Peck et al. (2013) presented an example of a single steeply dipping joint (Figure 2b) where the dip angle was stated to
be 70º. According to them, the single steeply dipping joint prevented the formation of the compressive arch causing the
beam to collapse. This could be considered a significant issue for the design of flat-roofed tunnels in Sydney as the
typical dip angle of the NNE and ESE sub-vertical joint sets observed in the Hawkesbury Sandstone varies between 70º
and 90º (Bertuzzi and Pells, 2002a).
As discussed above, the analytical solution was developed for sub-vertical joints where the angle between the plane of
the cross-cutting discontinuities and the normal to discontinuities sub-parallel to the excavation plane is less than one
third to one half of the effective friction angle of these joints. As a result, considering that such an angle in this case is
20º, equal to two thirds of the friction angle of the joints, the analytical solution would be deemed not applicable and if
used would provide similar results to that given in Figure 5 and Table 1. In other words, it cannot predict the collapse
described by Peck et al. (2013). However, the UDEC model (Figure 8) surprisingly indicates stability with a behaviour
similar to that of Figure 6a. This seems to indicate that either the dip angle of the single joint analysed by them is in fact
shallower and there was a typo in Peck et al. (2013) or perhaps they have not allowed for a gradual loading of the beam
simulating the gradual excavation discussed above. Another factor playing a role is that the base case of Peck et al.
(2013) already predicted smaller deflections resulting in lower compressive stresses across the sub-vertical joint,
consequently lower frictional resistance potentially causing the collapse.
STABLE

Deflection (mm)
0 10 20 30 40 50

Figure 8: Deflection contours for single steeply dipping joint (dip 70º) with σh = 0 MPa.
A second case was run by the writers in UDEC considering a dip angle of 60º which could also be argued possible in
Hawkesbury Sandstone even though less frequent. Under such conditions, i.e. with an angle between the plane of the
cross-cutting discontinuity and the normal to discontinuities sub-parallel to the excavation plane equal to the friction
angle of the joints (30º), the UDEC model (Figure 9) predicts full collapse of the beams even at higher normal stresses.
However, it is important to recognize that due to current tunnelling standards in Sydney, it is normal practice to have a
minimum rock bolting pattern even in massive sandstone considered self-supporting. Such rock bolting pattern may
reduce such risk by reinforcing the single joint triggering the instability. The effect of rock bolts is investigated later in
this paper.

34 Australian Geomechanics Vol 49 No 3 September 2014

 REVISITING THE APPLICABILITY OF VOUSSOIR BEAM THEORY FOR TUNNEL DESIGN IN SYDNEY
OLIVEIRA AND PELLS

COLLAPSED

(a)
Deflection (mm)
0 10 20 30 40 50

COLLAPSED

(b)

Figure 9: Deflection contours for single steeply dipping joint (dip 60º): (a) σh = 0 MPa; (b) σh = 4 MPa.

4.4 SINGLE NON-PERSISTENT SHALLOW ANGLE JOINT OR CROSS BEDDING PARTING

Peck et al. (2013) presented the case of a single non-persistent shallow angle joint as another limitation of the voussoir
beam theory. In their example, slip along the sub-horizontal joint allowed excessive rotation at the point of maximum
deflection initiating collapse of the roof beam.
Similar to the previous case and as discussed above, the analytical solution is unable to account for low to mid angle
joints. Therefore, if a 1 m thick rock beam was assessed with the analytical solution such instability would not be
predicted. However, it is evident that such a shallow angle joint will not mobilise enough normal stress and shear
strength that promotes the stress arching mechanism across the entire beam thickness. As a result, the correct beam
thickness to be used in the analytical solution is 0.5 m which, based on the results given in Figure 5 and Table 1, already
indicated a meta-stability, i.e. proximity to a collapse. However, in the current case it is also necessary to account for
the additional uniformly distributed load imparted by the upper 0.5 m of the beam that is not contributing to the arching
effect, i.e. a load equivalent to 13 kPa. Figure 10 presents the results of such analysis with different beam thickness for
comparison purposes. It is then demonstrated that the analytical solution also predicts the collapse of a 0.5 m thick
beam with yielding initiated at a beam thicknesses of 0.75 m.

1.2 1.2 1.2

Yielding

1.0 1.0 1.0

Collapse

Collapse
Yielding
Beam  thickness  (m)
Beam  thickness  (m)

Beam  thickness  (m)

Yielding

Collapse

0.8 0.8 0.8

0.6 0.6 0.6

0.4 0.4 0.4

0.2 0.2 0.2

0.0 0.0 0.0

0 100 200 300 400 500 0% 20% 40% 60% 80% 100% 0 10 20 30
Midspan  deflection  (mm) Buckling  limit  index  (BL) σmax (MPa)

(a) (b) (c)

Figure 10: Analytical voussoir beam solution for different beam thicknesses with 13 kPa uniformly distributed. (a) Mid-
span deflection; (b) Buckling limit index (BL); (c) Maximum compressive stress.
The UDEC model, with no applied surcharge, (Figure 11) confirms the predictions of the analytical solution for low
values of horizontal stresses. However, stability is observed for horizontal stresses exceeding 1.75 MPa. In addition,
once stability is achieved upon increase of the initial horizontal stress, the model predicts deflections that are smaller
than that predicted with a 0.5 m thick elastic beam fixed at both end. This indicates that, although the defect is affecting
the stress arching, the upper 0.5 m of the beam starts to carry some compressive stresses limited by the shear strength
mobilised along the shallow angle joint, thus reducing the overall beam deflection. In other words, the horizontal
stresses promote an initial normal stress along the joint which improves its shear capacity.

Australian Geomechanics Vol 49 No 3 September 2014 35

REVISITING THE APPLICABILITY OF VOUSSOIR BEAM THEORY FOR TUNNEL DESIGN IN SYDNEY
OLIVEIRA AND PELLS

COLLAPSED

70
(a) Deflection (mm) 60

Beam  deflection  (mm)

0 10 20 30 40 50
50

COLLAPSED 40
30 Collapse UDEC
(b)
20 Analytical
META-STABLE / YIELDING Elastic  beam
10

(c) 0
0 1 2 3 4 5
STABLE Horizontal  stress  (MPa)

(e)
(d)

Figure 11: Deflection contours for single non-persistent shallow angle joint. (a) σh = 0 MPa (b) σh = 1 MPa (c) σh =
1.75 MPa (d) σh = 4 MPa (e) effect of initial horizontal stress.
A relatively similar condition in Hawkesbury Sandstone could be argued to be the effect of cross-bedding partings on
composite beams, i.e. where low angle partings typically with dips of 15º to 30º (Bertuzzi and Pells, 2002a) exist above
a more massive bed and both are clamped together by rock bolts forming a composite thicker voussoir beam. However,
it is important to emphasize that this refers to the effect of cross-bedding partings, i.e. open cross beddings, and not
distinct intact cross bedding often observed in the fresher Hawkesbury Sandstone.
It is important to note that the effect of horizontal stresses is not necessarily always positive and high values of stress
may cause compressive buckling of thin beams, which may be assessed approximately by Euler’s buckling failure
mechanism given by:

π 2 Et 2 (3)
σ cr = 2
12lef
where σcr is the critical compressive buckling stress, E is the relevant beam Young’s modulus (i.e Eh,rm = 7.3 GPa), t is
the beam thickness, and lef is the effective length, equal to the length of the beam for pinned ends and half for fixed end
cases. Based on Equation (3), the critical buckling stress in this case would be between 6.7 MPa for a beam with pinned
ends and 26 MPa for fixed ends. As discussed above, it could be expected that the rock beam would behave more like a
fixed end beam under the effect of high horizontal stresses, therefore, tending towards the high bound value of the
buckling stress. However, it is important to consider that in this case, the effect of the sub-vertical joints may have an
increased effect particularly considering that the beam is under a combined flexural-compressive load. Such combined
effect potentially reduces the critical stress that causes pure compressive buckling. Therefore, it seems appropriate to
adopt an estimate between these two values, and further assessed by numerical analyses if the approximate analysis
results indicate buckling may be an issue. Compressive buckling issues will be further illustrated in the next case. An
additional mechanism that should also be considered is the effect of stress induced failure. It is generally accepted that,
for unconfined conditions, significant but stable micro-cracking starts at approximately 40% of the UCS of the intact
rock with unstable cracking typically initiating at 70% to 80% of the UCS. For instance, Figure 10c demonstrates that
for a thickness of 0.6 m the maximum compressive stress developed through arching is approaching the rock UCS.
Such cracking may deteriorate the rock modulus causing further beam deflections which could promote a transition of
the beam behaviour from a meta-stable/yielding condition to a collapse.

4.5 WEAK ABUTMENT CONTACT

As discussed above, the analytical solution does not directly consider shear displacements and failure along joints but
can use the maximum stress developed through the stress arching mechanism to estimate the factor of safety against
sliding at the abutments. Therefore, the analytical results for the case of a weak abutment (Figure 2d) are similar to that
of the base case given in Figure 5 and Table 1. The exception is the factor of safety against sliding or shear at the
abutment joints which for a friction angle φ = 30º was FoS = 4.2 but when replaced by φ = 18º reduces to FoS=2.41.

36 Australian Geomechanics Vol 49 No 3 September 2014

REVISITING THE APPLICABILITY OF VOUSSOIR BEAM THEORY FOR TUNNEL DESIGN IN SYDNEY
OLIVEIRA AND PELLS

STABLE
140
120 UDEC
(a) Deflection (mm) Analytical

Beam  deflection  (mm)

φ = 18º, kn/ks = 1 GPa/m 0 10 20 30 40 50 100
80
STABLE Collapse
60
40
(b) φ = 12º, kn/ks = 1 GPa/m 20

COLLAPSE 0
1 2 3
Shear  strength  reduction  factor
(c) φ = 7.8º, kn/ks = 1 GPa/m (d)
Figure 12: Deflection contours for weak abutment contact (a) FoS = 2.31 MPa (b) FoS = 1.5 MPa (c) FoS = 1 (d)
Impact of shear strength reduction.
The UDEC model presented in Figure 12a indicates a similar factor of safety to that of the analytical solution, i.e. a
stable condition. Figure 12d illustrates the development of the voussoir beam maximum deflection upon a shear
strength reduction up to a value of FoS=2.31 where collapse occurs (Figure 12c) at the abutment. Note that the
maximum beam deflection in Figure 12a is approximately only 10% higher than that depicted in Figure 6a indicating
that for FoS ≥ 2, the effect of a single weak abutment is less pronounced and beam deflections assessed with the
analytical solution are acceptable. Such results contradict the analysis carried out by Peck et al. (2013) and confirm the
ability of the analytical solution to provide a meaningful assessment for the case of a single weaker abutment, with a
difference in FoS of less than 5%. The effect of the horizontal stress case was not assessed as it was evident that it
would yield similar results to that of Figure 6c.

4.6 WEAK ZONE WITHIN INTACT ROCK

The effect of a small weak zone within the central intact rock block was investigated by Peck et al. (2013) as depicted
in Figure 2e. They stated that in such a case, crushing failure within the weak zone is predicted which then allows
excessive rotation at the point of maximum deflection and consequent collapse.
As mentioned above, the voussoir analytical solution assumes a uniform compressive strength and stiffness across the
beam and therefore, such geological condition cannot be directly assessed with the theory. However, such an
assessment can be carried out indirectly similarly to the weak abutment contact. For example, Figure 5c indicates that
the maximum compressive strength developed for a 1 m thick beam is approximately 4 MPa which would not cause
crushing of the weak zone with an UCS = 5 MPa even though potentially causing unstable cracking.
The UDEC model depicted in Figure 13 confirms the analytical assessment above with a 20% increase in deflections,
i.e. the mid-span deflection increases to 23 mm while the base case was only 19 mm for the UDEC model. Such
increase in deflection is expected considering that the weaker zone has also an associated lower stiffness. In addition,
the model indicates minor yielding of the weak zones that take place during development of the stress arching but not
enough to cause full crushing. Such results are again in contradiction to the assessment of Peck et al. (2013). Similar to
the base case, horizontal stress promotes a positive effect, reducing deformations up to a limit of approximately 4 MPa
beyond which there is an increase in deflection. At a horizontal stress of 16 MPa compressive buckling becomes an
issue causing shear failure of the beam and collapse consequently.

Australian Geomechanics Vol 49 No 3 September 2014 37

REVISITING THE APPLICABILITY OF VOUSSOIR BEAM THEORY FOR TUNNEL DESIGN IN SYDNEY
OLIVEIRA AND PELLS

STABLE
50

UDEC
(a) 40

Beam  deflection  (mm)

Deflection (mm) Analytical

Collapse
0 10 20 30 40 50 Elastic  beam
30

STABLE 20

10
(b)
0
META-STABLE / YIELDING 0 5 10 15 20
Horizontal  stress  (MPa)

(c) (d)

Figure 13: Deflection contours weak zone within intact rock (a) σh = 0 MPa; (b) σh = 8 MPa; (c) σh = 16MPa; (d) Effect
of initial horizontal stress.

5 ROCK BOLT REINFORCEMENT IN VOUSSOIR (OR LINEAR ARCH) DESIGN

5.1 BOLTING OF MULTIPLE BEDS
Flat-roofed tunnels in Sydney sandstone often involves the design of a composite voussoir beam where multiple beds,
typically spaced at 1 m to 3 m, are clamped together by a pattern of rock bolts. Such rock bolts could be said to have
two primary objectives: (1) create a safe work zone for tunnel workers and (2) improve the overall behaviour of the roof
beam by artifically increasing its thickness.

5.2 EFFECT OF WEAK INTERBED SURFACES

Weathered, clay rich, interbed surfaces, clay seams, and occasional thin, laminite beds occur within the Hawkesbury
Sandstone. It would be expected that such weaker seam materials have some impact on the overall behaviour of the
composite beam. Figure 14 depicts a simple composite beam model used to investigate the effect of weak seams. The
beam contains weak seams within the bedding partings equivalent to approximately 3.3% of its overall thickness. A
rock bolting pattern of 1.75 m x 1.75 m centre to centre installed at a 90º angle to horizontal was adopted as
representative of typical tunnel support used for good quality sandstone roof in Sydney (Class I/II Sandstone) with
similar span.
15 m

3m

Rock bolts @ 1.75 m c/c

50 mm seam, E= 200 MPa, φ = 26º, c = 20 kPa

Figure 14: Geometry of composite rock beam with seams and rock bolting
Figure 15 presents the beam deflections of two comparison models: with and without weak seams. The weak seams
increase the beam deflections by approximately 50% for all levels of stress but both are stable. It is important to note
that Figure 15c also demonstrates that, in this case, the rock bolts are not fully effective in reducing the overall beam
deflection with the composite beam behaviour very similar to that of a single 1 m thick bed. This is mainly due to both
rock joints and rock bolts being modelled at an angle of 90º to horizontal which results in small mobilisation of axial
and shear forces along the bolts for the current level of deflections. The impact and performance of rock bolts will be
further investigated in the next section.

38 Australian Geomechanics Vol 49 No 3 September 2014

REVISITING THE APPLICABILITY OF VOUSSOIR BEAM THEORY FOR TUNNEL DESIGN IN SYDNEY
OLIVEIRA AND PELLS

STABLE
40
UDEC  -­‐  3  x  1m  bed  with  seams
UDEC  -­‐  3  x  1m  bed  without  seams

Beam  deflection  (mm)

30 UDEC  -­‐1m  bed
Analytical:  1m  bed
(a) Deflection (mm)
20
0 10 20 30 40 50

STABLE
10

0
0 5 10
(b) Horizontal  stress  (MPa)

(c)
Figure 15: Deflection contours for rock bolted beds (a) σh = 0 MPa without seams (b) σh = 0 MPa with seams (c) Effect
of seams and initial horizontal stress.
The rock bolts were modelled in UDEC using reinforcement elements which offer both axial and pure shear resistance.
The axial and shear stiffness, Ka and Ks respectively, were estimated by Equation 4:
3/ 4
⎡ ⎤
⎢ ⎥
G g Eb ⎢ 2Eg ⎥ (4)
K a = πd1 K s = Eb I
⎛ d ⎞ ⎢ ⎛ d 2 ⎞ ⎥
2⎜⎜ 2 − 1⎟⎟ ⎢ 4 Eb I ⎜⎜ − 1⎟⎟ ⎥
⎝ d1 ⎠ ⎣ ⎝ d1 ⎠ ⎦
where d1 is the rock bolt diameter (d1 = 21.7 mm), Gg is the grout shear modulus (Gg = 12 GPa), Eb is the Young’s
modulus ot the rock bolt (Eb = 210 GPa), d2 is the rock bolt hole diameter (d2 = 54 mm), Eg is the grout Young’s
modulus (Eg = 30 GPa) and I is the rock bolt second moment of area. The rock bolts shear capacity, Ps,ult were estimated
by:
2
Ps,ult = 0.67d1 Fbyσ ci (5)
where Fby is the ultimate stress of the rock bolts (Fby = 920 MPa) and σci is the uniaxial compressive strength of the
intact rock (σci = 23 MPa). The axial capacity was assumed Pa,ult = 340 kN or limited by a bond strength of 1 MPa. In
addition, an ultimate axial strain of 15% was set to the rock bolts beyond which both axial and shear capacity drops to
zero.

5.3 ROCK BOLT PERFORMANCE

As discussed above, one of the main objectives of the rock bolts in a flat-roof design is generally to “clamp” the
multiple beds together in order to achieve a composite thicker voussoir beam. The underlying philosophy for the design
of flat-roofed tunnels in Sydney is to use the displacements derived from a jointed rock mass analysis to design a rock
bolt reinforcement that provides greater capacity than the stresses derived from an elastic continuum analysis (Bertuzzi
and Pells, 2002b). In doing so, each bolt is designed to provide a shear resistance greater than the joint excess shear
stress over its area of influence. Such bolt shear resistance is mobilised by the shear movement that would occur along
the bedding plane at that point where the bolt crosses the bedding plane.
In order to understand the negligible impact of the rock bolts in promoting the composite beam behaviour in Figure 15c,
it is important to verify the shear forces mobilised at the locations the bolts cross the bedding planes and their inter-
relationship with the bedding plane shear displacements. Figure 16 presents the rock bolt forces and shear
displacements of the same model used in Figure 15a (i.e. without the seams). The bolt forces in this figure are divided
by the out-of-plane bolt spacing (i.e. 1.75 m) which is necessary in UDEC to account for the plane strain condition. The
average mobilised bolt shear force is approximately 23 kN upon a shear displacement of approximately 3 mm. Such
shear force is only 10% of the estimated shear capacity of the rock bolts which seems to explain the typical small
impact of the rock bolts.

Australian Geomechanics Vol 49 No 3 September 2014 39

REVISITING THE APPLICABILITY OF VOUSSOIR BEAM THEORY FOR TUNNEL DESIGN IN SYDNEY
OLIVEIRA AND PELLS

STABLE

(a)

Figure 16: Shear displacements along bedding partings and forces for rock bolts at 90º, σh = 0 MPa and no surcharge.
The low mobilisation of bolt shear forces observed in Figure 16 is related to the small beam deflections and consequent
small shear displacements mobilised along the bedding partings. In order to further investigate the performance of the
rock bolts, a triangular surcharge ranging from zero to q = 195 kPa mid-span is applied at the top of the model used in
Figure 16. This triangular surcharge is equivalent to a 7.5 m overburden above roof beam which is considered
reasonable for a span of 15 m even for depths of more than 10 m due to stress arching effects. As expected, greater
beam deflections promotes additional shear displacements which further highlights the positive effect of the rock bolts
at low values of horizontal stress (Figure 17).
Figure 17 also illustrates the significant improved effect rock bolts at a shallower angle, in this case 70º (Figure 20).
The impact of installing rock bolts at an angle can be clearly illustrated (Figure 18) through the analytical solution for
rock bolt design proposed by Pells (2002). For bolts at 90º, the analytical solution predicts a shear force of
approximately 45 kN mobilised at a shear displacement of about 3 mm whereas 300 kN is mobilised with less than 2
mm for an angle of 70º, and clearly explaining the significantly lower deflection of Figure 17c. It is important to note
that the UDEC reinforcement element only predicts 50% of the analytical solution bolt shear force at 90º. However,
similar results can be achieved if the UDEC parameters, particularly the shear and axial stiffnesses, are calibrated with
the analytical solution or perhaps based on laboratory bolt shear tests.
STABLE
60
UDEC  -­‐  3  x  1m  bed  without  bolts
50 UDEC  -­‐  3  x  1m  bed  with  bolts  @  90º
Beam  deflection  (mm)

Deflection (mm)
(a) UDEC  -­‐  3  x  1m  bed  with  bolts  @  70º
0 10 20 30 40 50 40

30
STABLE
20

10
(b)
0
STABLE
0 5 10
Horizontal  stress  (MPa)

(d)
(c)

Figure 17: Deflection contours for bedded beam with triangular surcharge q = 195 kPa (a) σh = 0 MPa without rock
bolts (b) σh = 0 MPa with rock bolts at 90º (c) σh = 0 MPa with rock bolts at 70º (d) Effect of initial horizontal stress.

40 Australian Geomechanics Vol 49 No 3 September 2014

REVISITING THE APPLICABILITY OF VOUSSOIR BEAM THEORY FOR TUNNEL DESIGN IN SYDNEY
OLIVEIRA AND PELLS

(a) (b)
Figure 18: Analytical solution for rock bolts (a) at 90º angle to bedding (c) at 70º.

6 BOLTED BEAM WITH THROUGH-GOING INCLINED JOINT

As discussed above, the presence of steeply dipping joints may significantly affect the stability of a rock beam,
particularly for dip angles greater than half of the friction angle of joints, and in the case of a single joint 2/3 of the
friction angle. Figure 9 illustrates the collapse of a 1 m thick beam due to the a single joint with a dip of 60º. As a result,
considering the possibility of such dip angles in Hawkesbury Sandstone and that the current tunnelling practice in
Sydney is to have supported roof, it seems appropriate to investigate the performance of rock bolts in such composite
beams under such jointing condition.
Figure 19 presents the behaviour of a composite beam with a single steeply dipping joint (dip 60º) and a rock bolt
pattern at 1.75 m x 1.75 m centres installed at a 90º angle similar to Figure 14. For such pattern, only one rock bolt
crosses the steeply dipping joint. This single bolt experiences axial strains exceeding 15% causing it to rupture and
consequently beam collase.
COLLAPSED

(a)
Deflection (mm)
0 10 20 30 40 50

Figure 19: Deflection contours for single steeply dipping joint (dip 60º) with rock bolts at 90º and σh = 0 MPa.
A second rock bolt pattern is investigated. The locations of the rock bolts are exactly the same as in Figure 19 but
installed at an angle 70º. Such bolt pattern not only improves the performance of the rock bolts as discussed above but
also increases the chances of more bolts crossing steeply dipping joints. In this example, the number of bolts crossing
the joints increases to two as depicted in Figure 20. Note that to achieve the same number with a 90º angle, the bolts
spacing would have to be less than 1 m and they would still not have the same shear performance. Spot bolting is not a
practical option as such steep feature may not always be visibly evident in the roof during excavation.
15 m

3m

Rock bolts @ 70º and 1.75 m c/c

Figure 20: Geometry of composite rock beam with single steeply dipping joint.
Figure 21 presents the results of the UDEC model with rock bolts installed at a 70º angle. As expected, there is
significant improvement in the rock beam behaviour which becomes stable with a maximum deflection of

Australian Geomechanics Vol 49 No 3 September 2014 41

REVISITING THE APPLICABILITY OF VOUSSOIR BEAM THEORY FOR TUNNEL DESIGN IN SYDNEY
OLIVEIRA AND PELLS

approximately 8 mm. Such smaller deflection results in lower axial strains in the bolts now able to sustain the applied
loads.
STABLE

(a) Deflection (mm)

0 10 20 30 40 50

3.5
10
UDEC 3.0

Yielding

Collapse
Beam  thickness  (m)
8
Beam  deflection  (mm)

Analytical:  1  x  3m  bed 2.5

Elastic  beam
6 2.0

1.5
4
1.0
2 0.5
(b)
(c)
0 0.0
0 50 100 150 200
0 1 2 3 4 5 6 7 8 9 10
Horizontal  stress  (MPa) Midspan  deflection  (mm)

Figure 21: (a) Deflection contours for single steeply dipping joint (dip 60º) with rock bolts at 70º and σh = 0 MPa (b)
Effect of initial horizontal stress (c) Analytical voussoir beam.
It is important to note that the analytical solution does not account for multiple beds, thus, not directly applicable for
composite beams. However, analysis is often carried out ignoring the bedding partings, thus, with a beam thickness
assumed to be 3 m. In this case, the analytical solution predicts a mid-span deflection in the order of 2.5 mm (Figure
21b and Figure 21c). The major difference in this case, i.e. 2.5 mm compared with the 8 mm of the UDEC model, could
be attributed to the single steeply dipping joint. However, as discussed above, the roof span to beam thickness is 5, i.e.
less than 10, which again invalidates the direct applicability of the analytical solution. In fact, if a 3 m beam is modelled
in UDEC without the bedding partings, the model predicts collapse of the beam due to sliding at the abutments which
seems to indicate that for low roof span to thickness ratios, the Diederichs and Kaiser (1999) parabolic function used to
assess the stress arching mechanism is not applicable. Interestingly, in this case, if the joints at the abutment are made
elastic in shear (infinite cohesion) but allowing for tensile failure (zero tensile strength), the UDEC beam deflection
matches the analytical solution.

7 BRIEF NOTES ON POOR QUALITY ROCK MASSES

As previously discussed, no one would argue the important geology. It is evident that in poor rock masses, such as those
with closely spaced laminations and a high frequency of steeply dipping joints, the application of a voussoir analogue
should be assessed with great care. It is obvious that under such conditions, a flat-roofed tunnel would require
significantly heavier roof support and perhaps alternate roof shapes considered to manage risk.
For example, it is well accepted that there is a higher frequency of steeply dipping to mid angle joints in the Ashfield
Shale. As a result, flat-roofs in such formation are typically less used or only smaller spans.
In order to illustrate the above discussion, Figure 22 presents the simulation of a flat-roofed tunnel design in Class II
Shale where one set of joints has a dip of 45º. A stress ratio of ko = 1.5 was assumed for the Class IV Shale, ko = 2 for
the Class II Shale and ko = 4.5 for the Class II Sandstone. The roof support consists of 4 m long rock bolts installed at a
similar spacing to what would be adopted for a Class II/III Sandstone roof with such span, i.e. 1.5 m c/c. A FoS = 1.5
was applied to the rock bolts axial and shear capacities which means that this model is to some extent an ultimate limit
state analysis, therefore the absolute values of deformation are less relevant. The large deformations in Figure 22a are
shown just for illustration purposes as the model did not fully converge which clearly indicates roof collapse due to
inadequate factor of safety. Figure22b again reflects the positive effect of the bolts installed at 70º. Although the model
converged and collapse is not evident, the pattern of deformation indicates a high-risk condition particularly considering
that the factor of safety was only applied to the rock bolts and best-estimated intact rock or discontinuities strength
parameters were adopted without reduction factors. It would seem reasonable to increase the tunnel support or the arch
radius to roof span ratio, i.e. a more arched roof, to reduce such risks.

42 Australian Geomechanics Vol 49 No 3 September 2014

REVISITING THE APPLICABILITY OF VOUSSOIR BEAM THEORY FOR TUNNEL DESIGN IN SYDNEY
OLIVEIRA AND PELLS

Vert. Disp. (mm) Vert. Disp. (mm)

0 40 80 120 160 200 0 20 40 60 80 100

12 m 7m

(a) (b)
Figure 22: Example of flat-roofed tunnel in Ashfield Shale (residual soil overlying Class IV Shale, Class II Shale and
Class II Sandstone): (a) bolts at 90º; (b) bolts at 70º.

8 CONCLUSIONS
The design of semi flat-roofed tunnels in Sydney using a voussoir beam analogue has been proven successful over time,
despite the theory limitations. However, if the geological conditions are clearly understood and good engineering
judgement is applied to take these conditions into account, the voussoir beam analytical solution can be confidently
used in practice to yield similar results to that of more comprehensive discontinuum numerical models. In other words,
if the appropriate mechanisms are well understood and considered in the analytical solution, as they should also be
accounted for in the numerical models, the predicted behaviours are in most cases similar.
As detailed in analyses presented above, rock bolts are typically used to stitch together near horizontal beds of limited
and variable thickness, to ‘trick’ the rock mass into behaving as an appropriately thick linear arch, robust against the
uncertainties in joint directions that may be expected in strata such as the Hawkesbury Sandstone of Sydney. In
addition, the existence of relatively high horizontal stresses are the second key feature in determining a geological
environment amenable to flat or semi-flat roof designs.

9 ACKNOWLEDGMENTS
The authors of this paper would like to acknowledge the contribution of Patrick Wong who reviewed the paper
providing useful comments.

10 REFERENCES
Asche, H.R. and Lechner, M.K. (2003). Design for the Cross City Tunnel, Sydney. Proccedings of the RETC. Seattle.
pp. 122-139.
Bertuzzi, R and Pells, P J N (2002a). Geotechnical parameters of Sydney Sandstone and Shale. Australian
Geomechanics, 37(5).
Bertuzzi, R and Pells, P J N (2002b). Design of rock bolt and shotcrete support of tunnel roofs in Sydney sandstone.
Australian Geomechanics, 37(3).
Booker and Best (1990). Coffey internal communication.
Diederichs, M S and Kaiser P K. (1999): Stability Guidelines for Excavations in Laminated Ground - The Voussoir
Analogue Revisited, Int. J. Rock Mech. & Min. Sci.; 36, pp 97-118.
Evans W H. (1941). The strength of undermined strata. Trans. Inst. Min. Metall., 50:475-500.
Itasca (2006). Universal Distinct Element Code User’s Guide.
Pells, P J N, Best, R J, and Poulos, H G (1994). Design of roof support of the Sydney Opera House underground
parking station. Tunnelling and Underground Space Technology, Vol 9, No 2, pp201-207.
Pells (1993) Rock Mechanics and Engineering geology in the Design of Underground Works. EH Davis Memorial
Lecture, Australian Geomechanics.
Pells, P. J. N. (2002). Developments in the design of tunnels and caverns in the Triassic rocks of the Sydney region.
International Journal of Rock Mechanics and Mining Sciences 39: 569-587

Australian Geomechanics Vol 49 No 3 September 2014 43

REVISITING THE APPLICABILITY OF VOUSSOIR BEAM THEORY FOR TUNNEL DESIGN IN SYDNEY
OLIVEIRA AND PELLS

Peck, W A, Sainsbury, D P and Lee, M.F (2013). The importance of geology and roof shape on the stability of shallow
caverns. Australian Geomechanics, 48(3).
Sofianos A I (1996). Analysis and Design of an Underground Hard Rock Voussoir Beam Roof. Int. J. Rock Mech. &
Min. Sci., 33 (2), pp 153-166.

44 Australian Geomechanics Vol 49 No 3 September 2014

PERGAMON International Journal of Rock Mechanics and Mining Sciences 36 (1999) 97±117

Stability of large excavations in laminated hard rock masses: the

voussoir analogue revisited
M.S. Diederichs *, P.K. Kaiser
Geomechanics Research Centre, F217 Laurentian University, Sudbury, Ont., Canada P3E 2C6
Accepted 8 November 1998

Abstract

The voussoir beam analogue has provided a useful stability assessment tool for more than 55 years and has seen numerous
improvements and revisions over the years. In this paper, a simpli®ed and robust iterative algorithm is presented for this model.
This approach includes an improved assumption for internal compression arch geometry, simpli®ed displacement determination,
support pressure and surcharge analysis and a corrected stabilizing moment in the two dimensional case. A discrete element
simulation is used to verify these enhancements and to con®rm traditional assumptions inherent in the model. In the case of
beam snap-through failure, dominant in hard rock excavations of moderately large span, design criteria are traditionally based
on a stability limit which represents an upper bound for stable span estimates. A de¯ection threshold has been identi®ed and
veri®ed through ®eld evidence, which corresponds to the onset of non-linear deformation behaviour and therefore, of initial
instability. This threshold is proposed as a more reasonable stability limit for this failure mode in rockmasses and particularly
for data limited cases. Design charts, based on this linearity limit for unsupported stability of jointed rock beams, are presented
here summarizing critical span±thickness±modulus relationships. # 1999 Elsevier Science Ltd. All rights reserved.

1. Introduction stress calculations [7, 8]. It is more common, however,

to encounter other joint sets cutting through the lami-
Rockmass behaviour dominated by parallel lami- nations. These joints reduce and, in the extreme, elim-
nations is often encountered in underground exca- inate the ability of the rockmass to sustain boundary-
vations in numerous geological environments. These parallel tensile stresses such as those assumed in con-
laminations can be the result of sedimentary layering, ventional beam analysis. However, where these joints
extensile jointing, fabric created through metamorphic cut across the laminations at steep angles or where re-
or igneous ¯ow processes or through excavation-paral- inforcement has been installed, it is possible to assume
lel stress fracturing of massive ground. This structure that a compression arch can be generated within the
can be the dominant factor controlling the stability of beam which will transmit the beam loads to the abut-
roofs in large civil excavations, in coal or other hori- ments (Fig. 1).
zontal mining stopes [1, 2] and can also dominate the It was noted by Fayol [9] that underground strata
stability of inclined open stope hangingwalls such as seemed to separate upon de¯ection such that each
those encountered in hard rock mining in Canada [3, 4], laminated beam transferred its own weight to the abut-
in Australia [5, 6] and elsewhere. ments rather than loading the beam beneath. Stability
of an excavation in this situation, it was concluded,
In rare circumstances, the lamination partings rep-
could be determined by analyzing the stability of a
resent the only joint set present in the rockmass. Roof
single beam de¯ecting under its own weight.
stability and de¯ection in this instance can be assessed
Conventional beam analysis, however, signi®cantly
using conventional elastic beam de¯ection and lateral
underestimated the inherent stability of such beams.
Even intact laminations would crack at midspan as
* Corresponding author. 105 William St. W., Waterloo, Ont.,
predicted, but would, after additional deformation,
Canada N2L IJ8. Tel.: +1-519-578-5327; e-mail: mdiederi@nickel. become stable again. The notion of the voussoir, while
laurentian.ca. traceable back to the architecture of ancient Rome [10],

0148-9062/99/$ - see front matter # 1999 Elsevier Science Ltd. All rights reserved.
PII: S 0 1 4 8 - 9 0 6 2 ( 9 8 ) 0 0 1 8 0 - 6
98 M.S. Diederichs, P.K. Kaiser / International Journal of Rock Mechanics and Mining Sciences 36 (1999) 97±117

Fig. 1. (a) Jointed rock beams; (b) voussoir beam analogue.

was ®rst proposed by Evans [11] speci®cally to explain The primary modes of failure assumed in the model
the stability of a jointed or cracked beam. After gener- and veri®ed in laboratory tests by Sterling [15] are
ating a great deal of controversy when ®rst published, buckling or snap-through failure, lateral compressive
the voussoir beam analogue has been generally failure (crushing) at the midspan and abutments, abut-
accepted and has since been reworked and presented ment slip and diagonal fracturing (Fig. 3). Shear fail-
as a simpli®ed tool for stability analysis of excavations ure (Fig. 3c) is observed at low span-to-thickness
in civil construction and in mining [10, 12±14]. Fig. 2 ratios (thick beams), while crushing (Fig. 3b) and
illustrates two example cases where the voussoir beam snap-through failure (Fig. 3a) is observed at higher
analogue can be invoked to explain the inherent stab- span-to-thickness ratios (thin beams). An examination
ility of a laminated hangingwall (Fig. 2a) and cross of the model data presented by Ran et al. [16] shows
jointed back (Fig. 2b) in hard rock environments. that if the angle between the plane of the cross-cutting

Fig. 2. Voussoir beams encountered at (a) Winston Lake Mine, Ontario and (b) mine access, Sudbury.
 M.S. Diederichs, P.K. Kaiser / International Journal of Rock Mechanics and Mining Sciences 36 (1999) 97±117 99

joints and premature shear failure of the beam is

likely [16].
Stimpson and Ahmed [17] also showed by physical
modelling that for thick beams, external loading (sur-
charge) can produce diagonal tensile ruptures (Fig. 3d)
extending from the upper midspan to the lower abut-
ments (parallel to the compression arch). While this
failure mode is partially the result of the loading con-
®guration, it may also be an important mechanism
where weak or broken material exists above the beam
or where internal rockmass damage and bulking due
to stress overloads the surface beam. This paper deals
with thin laminations (span-to-thickness ratio greater
than 10) under the in¯uence of self-weight or moderate
surcharge loading. Therefore, only crushing and snap-
through failure are hereafter considered. Sliding stab-
ility is included in the analyses but does not control
limiting dimensions for thin beams.
The following is a summary of the voussoir analysis
procedure based on the iterative scheme proposed by
Brady and Brown [13], including a number of improve-
ments and corrections by the authors. Most impor-
tantly, a more realistic yield threshold is introduced
for snap-through failure, to replace the ultimate rup-
ture limit originally proposed [11, 12]. This procedure
is summarized in several design charts and is veri®ed
using a discrete element simulation.

Fig. 3. Failure modes of the voussoir beam: (a) snap-through; (b)

crushing; (c) sliding and (d) diagonal cracking. 2. The voussoir model

Consider a laminated rock beam above an exca-

joints and the normal to the lamination plane (and the vation with a horizontal span given by S. The normal
normal to the excavation surface) is less than one third thickness of the single layer under analysis is T. For
to one half of the e€ective friction angle of these an elastic beam, with no joints and with constant cross
joints, then it is valid to apply the voussoir beam sol- section, a distribution of compression and tension,
ution. For shallower cross-jointing, slip along these symmetrical about the horizontal centreline of the

Fig. 4. Elastic beam with (a) ®xed ends and (b) simple (pin) supports.
100 M.S. Diederichs, P.K. Kaiser / International Journal of Rock Mechanics and Mining Sciences 36 (1999) 97±117

beam, is found across all plane sections within the encountered in newly developed underground spans at
beam (Fig. 4a). The solution (Eqs. (1) and (2)) for the low to moderate depth. This initial elastic phase fol-
maximum stress values, at the abutments, for com- lowed by progressive fracture and deformation of
pression (bottom of beam) or tension (top of beam), laminated hangingwalls has been observed and is
smax, as well as the maximum beam de¯ection, d, can described in detail by Milne [18].
be easily calculated using closed form beam It is highly unlikely that any thinly laminated roof
equations [7]: structure will remain in a completely continuous and
elastic state after excavation. The transition from con-
gS 2
smax ˆ 1† tinuous elastic beam to voussoir beam is normally
2T
assured and assumed in most cases. Failure is not
inevitable in this situation, however. When the tensile
gS 4 cracking, described above, or lamination-normal joints
dˆ 2†
32ET 2 crosscut the beam and render it incapable of sustaining
where E is the Young's modulus of the rock and g is tensile stresses, a compression arch develops within the
the speci®c weight. The maximum stress at the mid- beam, rising from the abutments to a high point at
span is one half of the maximum stress at the abut- midspan. For a half-span, this arch generates a
ments. Therefore, for such a beam with ®xed ends and moment between the reaction force at midspan and at
distributed loading, yield is assumed when the maxi- the abutments (Fig. 5) which acts to resist the moment
mum tensile stress, in the upper part of the beam at imposed by self-weight. Horizontal stress symmetry
the abutments, exceeds the tensile strength of the rock. within the beam is lost, making a closed form solution
Vertical tensile fractures form at the abutments and impossible without assuming an average thickness NT
the beam becomes simply supported (assuming no slip for this arch. This is the case for the solution intro-
at the abutments), as shown in Fig. 4(b), with a maxi- duced by Evans [11] and later by Beer and Meek [12].
mum tensile stress at the midspan given by In both cases N is assumed to be 0.5. Numerical exper-
imentation by Evans showed that this was an incorrect
2gS 2 assumption and that the equilibrium value was closer
smax ˆ 3†
3T to 0.7. Nevertheless, Evans chose the value, N = 0.5,
This stress is now higher than the previous abutment to simplify the practical solution which he was present-
stress, and therefore higher than the rock tensile ing. Investigations by the authors of this paper have
strength. This leads to subsequent fracturing centred shown that N is variable. While N is closer to 0.75 for
about the midspan as shown by Stimpson and stable beams at equilibrium, N drops to below 0.5 as
Ahmed [17]. This process of progressive cracking at the critical (unstable) beam geometry is approached.
the abutments, followed by cracking at the midspan In this formulation, the average thickness, NT, of
and other parts of the beam can be responsible for a the compression arch within the beam is initially
¯urry of low-level seismic emissions (rock noise), often unknown, as is the ultimate moment arm, Z0, between

Fig. 5. Voussoir beam (half-span shown) and nomenclature.

 M.S. Diederichs, P.K. Kaiser / International Journal of Rock Mechanics and Mining Sciences 36 (1999) 97±117 101

the horizontal reaction forces, F, at the midpsan and

at the abutments. Considering the half beam (Fig. 5)
and building on the algorithm introduced by Brady
and Brown [13] the solution procedure begins with an
assumption of the initial moment arm prior to de¯ec-
tion, Z0:
 
2
Z0 ˆ T 1 ÿ N 4†
3

The length of the reaction arch (horizontal reaction

force locus) is given by
8 2
LˆS‡ Z 5†
3S 0
This equation is based on the assumption of a para-
bolic compression arch. Numerical [16] and
physical [17] experiments con®rm this assumption.
Note that this equation is correctly reported by Beer
and Meek [12] but is in error in the summary of Brady
and Brown [13].
The solution proceeds on the assumption that for
stability to be obtained, the moment generated at the
abutment due to self-weight of the beam, given by Fig. 6. External beam loading due to (a) uniformly distributed sup-
port pressure; (b) linearly varying support pressure; (c) parabolically
gTS 2 varying support pressure and (d) parabolically varying surcharge
MW ˆ 6† loading.
8
must be exactly compensated, as the beam de¯ects a
for the case of a uniformly applied support pressure,
distance d at the midspan, by an opposite or resisting
p, illustrated in Fig. 6(a).
moment, MR. This moment is generated by the oppos-
If the support pressure is applied in a triangular dis-
ing horizontal reaction forces in the beam centre and
tribution varying from zero at the abutments to p at
at the abutments separated by a distance Z = Z0 ÿ d:
the midspan (Fig. 6b) as in the case of passive rock-
fmax NTZ bolts, then use
MR ˆ FZ ˆ 7†
2 2p
ge* ˆ ge ÿ 10†
where fmax is the maximum stress acting in the beam 3T
(at the bottom edge of the abutment and at the top
A parabolic distribution of support pressure (Fig. 6c)
edge of the midpsan section).
can be applied as
In Eq. (6), the speci®c weight, g, of the rockmass
can be replaced with an e€ective speci®c weight, ge, 7p
g*e ˆ ge ÿ 11†
given by 9T
ge ˆ g cos a 8† To introduce a distributed surcharge, q, replace (+p)
with (ÿq) in Eqs. (9)±(11) as appropriate. For example,
where a is the dip (angle from the horizontal) of the in the case of a distributed surcharge loading ranging
lamination plane. This analogue considers only the parabolically, as in Fig. 6(d), from 0 to q (due to bro-
component, of the driving weight, oriented normal to ken rock above the beam for example) the modi®ed
the beam. For inclined beams, therefore, the e€ect of e€ective speci®c weight becomes
beam-parallel loads due to settling is not considered.
In this model, as in reality, an inclined roof is more 7q
ge* ˆ ge ‡ 12†
stable under gravity loading than a horizontal roof. 9T
In addition, active support pressure can be con-
Assuming a triangular horizontal stress distribution
sidered in the equation by further adjustment of the
within the compression arch (of thickness NT) at the
e€ective speci®c weight:
abutments and at the midspan as illustrated in Fig. 5,
p the following equilibrium equation is obtained for
gep ˆ ge ÿ 9†
T fmax:
102 M.S. Diederichs, P.K. Kaiser / International Journal of Rock Mechanics and Mining Sciences 36 (1999) 97±117

Elastic shortening of the arch can then be obtained

from
 
fav L 2 N
DL ˆ L ˆ fmax ‡ 16†
E E 9 3

where E is the rockmass modulus in the direction par-

allel to the beam.
This shortening leads to a downward displacement
of the arch midspan and a new reaction moment arm
given by
s
 
3S 8 2
Zˆ Z 0 ÿ DL 17†
8 3S

The de¯ection, d, at midspan is given by (Z ÿ Z0) and

a negative value for the term under the square root
sign in Eq. (17) indicates that the critical beam de¯ec-
tion has been exceeded. In other words, as de¯ection
increases, the resisting moment, MR, a product of
Fig. 7. Conventional assumptions for compressive stress variation
inside the beam compared with the parabolic variation proposed in increasing reaction force, F, and decreasing moment
this paper. arm, Z, passes through its maximum without achieving
equilibrium with the weight moment MW. In this case,
snap-through failure would occur for the speci®ed
ge* †S 2 value of compression arch thickness, NT. If there are
fmax ˆ 13† no values for N (between 0 and 1) for which a stable
4NZ
solution can be obtained, ultimate collapse of the
In order to calculate the elastic shortening of the arch beam is assumed to occur.
(and thereby calculate the central de¯ection) an In order to ®nd a solution for the equilibrium pos-
assumption must be made about the internal distri- ition of the beam, Evans [11] chose to maximize the
bution of compressive stress within the beam. Previous product of the resisting moment and the moment arm,
authors [11±13] have based their calculation on the Z, although by his own admission, this choice was
assumption of a quasi-linear variation of stress (Fig. 7) somewhat arbitrary. In the procedure of Brady and
along a constant arch section NT, yielding an average Brown [13] a two-variable relaxation technique is
stress along the reaction line, fav, given by employed to solve for N, the arch thickness and Z, the
  ®nal moment arm. The equilibrium solution corre-
fmax 2 N sponds to the unique pair (N, Z) which results in ana-
fav ˆ ‡ 14†
2 3 2 lytical equality through the sequence of equations
summarized in the forgoing discussion. This approach
Careful examination of this assumption shows it to be was found by the authors to be highly unstable and
in error. It is reasonable to expect that at some point convergence was often dicult. Fortunately it can be
the entire beam section must be under compression seen through examining the results of this process that
and that at the point where the reaction line crosses equilibrium solution also corresponds with the mini-
the centreline of the beam, this stress is constant across mization of fmax, the maximum stress at the abutments
the entire beam section T. It is also reasonable to and midspan.
assume that the variation of stress along the reaction In the approach presented here, N is varied in incre-
line is not linear. Numerical experimentation by the ments (e.g. 0.01) over its ®nite range (0 to 1). Z is
authors of this work and examination of numerical modi®ed in an iterative fashion and a convergent sol-
results obtained by others [16, 19] con®rms that the ution is thereby obtained with only a few steps for
distribution of stress along the centreline of the arch is each value of N. For a stable beam with a span well
in fact parabolic (Fig. 7). This yields the following cor- below the critical limit for the given geometry and
rected equation for average stress in the beam: rockmass properties, a solution for Z is possible for all
  values of N. Equilibrium corresponds to the minimum
fmax 2 value of fmax. As the critical span is approached, the
fav ˆ ‡N 15†
3 3 percentage of N values which yield rational results for
 M.S. Diederichs, P.K. Kaiser / International Journal of Rock Mechanics and Mining Sciences 36 (1999) 97±117 103

Fig. 8. Flow chart for the determination of stability and de¯ection of a voussoir beam.

Z decrease, converging to a single rational solution The factor of safety with respect to crushing (com-
pair (N, Z) at the absolute critical limit. pressive failure) at the abutments and at midspan is
Finally. taking Z and Z0 (as determined using the given by the ratio of uncon®ned compressive strength
equilibrium value of N), the de¯ection, d, at the mid- of the rock with respect to the maximum compressive
stress calculated in the model:
span is simply Z ÿ Z0. The complete procedure for the
determination of stability and equilibrium de¯ection is UCS
F:S:crush ˆ 18†
summarized in Fig. 8. fmax
104 M.S. Diederichs, P.K. Kaiser / International Journal of Rock Mechanics and Mining Sciences 36 (1999) 97±117

The factor of safety with respect to vertical sliding, of In order to obtain a reasonable yield threshold for
an unsupported beam under self-weight, along joints the buckling limit, the relationship between midspan
at the abutments is given by displacement, d, and thickness, T, was considered.
Fig. 10 illustrates summary results for several beam
fmax N sti€nesses. The de¯ection/thickness relationships are
F:S:slide ˆ tan f 19† log-linear for beam geometries with ample thickness.
ge S
As the thickness is reduced (or the span is increased)
the relationship becomes non-linear and eventually
A numerical buckling limit, B.L., is introduced here becomes unde®ned at ultimate failure. Ultimate failure
which is the percentage of values of N within the occurs at a buckling limit of 100%, corresponding to a
range of 0 to 1 for which a solution (i.e. a real value displacement equivalent to approximately 0.25 T.
of Z) cannot be obtained. Fig. 9 illustrates the The onset of non-linearity consistently occurs at a dis-
decrease in normalized equilibrium arch thickness, N, placement equivalent to approximately 0.1 T. This
and the increase in buckling limit, B.L., with increasing `yield' point also consistently corresponds to a buck-
span/thickness ratio. N consistently drops below 0.5 as ling limit, B.L., of 35%.
the critical span/thickness ratio is approached and The yield threshold determined in Fig. 10 more clo-
reaches a limit of 0.35 immediately prior to failure sely corresponds to the critical displacement limit of
(snap-through). Ultimate collapse and the critical span approximately 0.15 T obtained by Mottahed and
versus thickness relationships proposed by Evans [11] Ran [19] through numerical modelling. Beyond this
and by Beer and Meek [12] correspond to a buckling displacement the jointed beam model used in their ex-
limit of 100%. In other words, stability is impossible if perimentation began to exhibit unstable behaviour and
there is no arch thickness which yields an equilibrium failure became imminent. This limit was independent
solution. For a more conservative approach, however, of span and the modulus.
a threshold can be speci®ed for the buckling limit in The choice of buckling limit a€ects the calculated
order to obtain design dimensions for the beam. critical span or the critical thickness (limiting case for

Fig. 9. Variation of N and buckling limit with span to thickness ratio.

 M.S. Diederichs, P.K. Kaiser / International Journal of Rock Mechanics and Mining Sciences 36 (1999) 97±117 105

Fig. 10. Example determination of beam `yield' limit (example span = 20 m).

beam stability) as illustrated by the example in Fig. 11. the discussion which follows. Typically thin hard rock
These limits are obtained by including the analysis out- beams (signi®cant compressive strength) under their
lined in Fig. 8 inside a simple iterative bisection algor- own self-weight will tend to snap-through before they
ithm to ®nd the critical value of a speci®ed input crush (Fig. 12). If an otherwise stable beam is sub-
parameter (in this case, thickness) which gives a value jected to surcharge loading, however, as is the case in
(span) straddling the interface between stability and many of the laboratory experiments described in the
failure. literature, then crushing failure will result (as may
The displacement limit of 0.1 T (B.L. = 35%) is abutment sliding or diagonal cracking). The in¯uence
used forthwith, as a yield limit, due to its convenience of surcharge loading on the stability of the beam as
and to account conservatively for uncertainties in the well as on the failure mode is demonstrated in Fig. 13.
approach. This limit is important for monitoring appli- The transition line (snap-thru/crushing) indicates
cations since it is independent of the span and rep- which failure mode controls the critical limits as
resents a universal rule of thumb for determining the shown.
signi®cance of measured displacements. Displacements
of less than 10% of the e€ective lamination thickness
(determined using a borehole camera or from mapping 3. Field evidence of snap-through limit
data) can be con®dently assessed as being within the
elastic limit of the voussoir beam, independent of rock 3.1. Mount Isa Mine
modulus or e€ective span.
Fig. 11 indicates critical limits for snap-through fail- A limiting displacement (for linear voussoir beha-
ure of the beam. Crushing failure is also considered in viour) of 0.1 to 0.15 times the mapped bedding thick-
106 M.S. Diederichs, P.K. Kaiser / International Journal of Rock Mechanics and Mining Sciences 36 (1999) 97±117

Fig. 11. E€ect of buckling limit on critical beam geometry for `snap-thru' failure of a horizontal beam (Erm = 10 GPa, g = 0.03 MN/m3).

ness is observed in the data, presented by Milne [18],

from numerous extensometers installed in the stope
hangingwalls at Mount Isa Mine in Australia. The dis-
placement data from one of the extensometers is sum-
marized in Fig. 14. In this case, the extensometer was
installed in the hangingwall, slightly above the stope
midspan and prior to mining of a 50 m high stope at a
depth of 870 m. The hangingwall consists of laminated
shale with bedding plane partings observed every 20 to
25 cm. The stope was mined via full face longitudinal
retreat. Measurements were recorded as the stope was
progressively mined past and beyond the location of
the extensometer.
The radius factor shown on the horizontal axis of
Fig. 14 is a scale and shape index developed by
Milne [18]. The details of this index are summarized
elsewhere [20]. For illustrative purposes, the radius fac-
tor describes the harmonic average of distance, from a
point in a excavation surface or face, to all abutments.
The radius factor, R.F., of the hangingwall at the
extensometer location increases as the stope is widened
along strike (constant height and thickness). The maxi-
mum R.F. prior to back®lling was approximately 14.
Milne [18] identi®es four displacement zones with
respect to increasing span including:

1. Elastic rock behaviour.

Fig. 12. Typical critical limits for the stability of slender beams (axis 2. Stable voussoir beam behaviour.
scales are identical for both plots). 3. Unstable voussoir beam behaviour
 M.S. Diederichs, P.K. Kaiser / International Journal of Rock Mechanics and Mining Sciences 36 (1999) 97±117 107

Fig. 13. In¯uence of surcharge loading on beam stability and failure mode.

4. Ultimate failure. beam thickness. A further transition to rapidly

increasing displacement occurs at around 55 to 60
Stage (1) is not presented in Fig. 14 due to the mm. This is consistent with the ultimate collapse
scale. Stages (2) through (4) are clearly apparent in limit of 0.25 T. No extensometer anchor displace-
Fig. 14. The lamination thickness, T, is determined to ments are recorded beyond 80 mm although it is
be relatively consistent and equal to 0.2 to 0.25 assumed that had mining and recording continued
m [18]. The onset of non-linear or unstable voussoir beyond a R.F. of 14, without back®lling, this stope
displacement (interface between stages (2) and (3) wall would have experienced signi®cant surface fail-
occurs at a surface displacement of approximately ure. This corresponds (given uncertainties regarding
20 to 30 mm (0.08 T to 0.15 T), re¯ecting the the local parting thickness) with the predictions from
proposed yield limit for de¯ection of 10% of the the voussoir analogue.

Fig. 14. Hangingwall response to mining compared with voussoir limits. Data from Milne [18].
108 M.S. Diederichs, P.K. Kaiser / International Journal of Rock Mechanics and Mining Sciences 36 (1999) 97±117

3.2. Winston Lake Mine this study is used here to illustrate the voussoir beam
behaviour of the chert hangingwall. Borehole camera
A second case study from Winston Lake Mine in data [24] is also used here to estimate the e€ective
Ontario, Canada is used here to demonstrate the utility thickness of continuous partings parallel to the hang-
and validity of the voussoir analogue. The typical ingwall as the wall de¯ects.
stope geometry is similar to that in the previous The rockmass quality as indicated by the index
example except that the stopes are smaller (20 m high) Q [25] is approximately 7 to 28 which, using the
and less steeply dipping (458 to 608) and in this case, relationship [26]
shallower (550 m). Winston Lake Mine uses a modi®ed
Erm ˆ 25 log10 Q 20†
version of the longitudinal retreat technique called
modi®ed AVOCA, in which the stope is blasted in full indicates a rockmass modulus in the range of 20 to 35
height slices perpendicular to the strike of the ore GPa. The data set, collected from numerous sources
body. Rock ®ll is introduced by end dumping some contains a large scatter and according to variability
distance back from the active face, creating an open limits proposed by Barton [27] this rockmass quality
stope of variable open strike length. The ®ll is com- could have a modulus as low as 10 GPa and as high
pressible rock®ll and cannot be considered tight ®ll as 50 GPa. The lower modulus could apply if the rock-
until signi®cant closure has occurred. As a stability mass is relaxed due to undercutting as was the case in
analogue, therefore, the two dimensional beam is valid some of the stopes [28].
in this case. The displacement performance of the hangingwall
As previously illustrated in Fig. 2(a), the hanging- can be predicted by the voussoir analogue. The results
wall is composed of blocky chert with surface parallel for this case are shown in Fig. 16. The equilibrium dis-
jointing along foliation planes and two orthogonal sets placement is plotted as a function of lamination thick-
of cross-joints. Cablebolt support is installed from a ness and rockmass modulus. The critical displacement
remote drift to supply full coverage at the hangingwall. limits for yield (10% of the thickness) and for collapse
Fig. 15 shows the layout of drifts, stopes, cablebolts (25% of the thickness) are overlain on the plot. The
and instrumentation for an experiment investigating joints at Winston Lake were not fully continuous and
cablebolt performance in hangingwalls and the impact the e€ective lamination thickness, therefore, evolved as
of stress change [21, 22]. Extensometer data [23] from the hangingwall de¯ected and as joints propagated

Fig. 15. Plan layout and typical cross sections through study area, Winston Lake Mine.
 M.S. Diederichs, P.K. Kaiser / International Journal of Rock Mechanics and Mining Sciences 36 (1999) 97±117 109

Fig. 16. Measured displacements and observed lamination thicknesses (data points), compared with voussoir predictions, Winston Lake Mine.

along the foliation planes or connected with other ing to the stability of the hangingwall by applying an
joints. The borehole camera surveys [24] can be used e€ective support pressure.
to crudely estimate the e€ective beam thickness at any
point in space and time. The measured displacements
are plotted against the estimated lamination thickness
in Fig. 16 for speci®c points along di€erent extens- 4. Numerical veri®cation of snap thru mechanism
ometer-borehole camera clusters (Fig. 15).
The results show that the rockmass behaviour corre- A discrete element model [30] was employed to
lates well with the predicted displacements. The data further verify the analogue. Other authors have per-
with higher beam thicknesses are from earlier in the formed similar analyses with mixed results. A recent
paper by So®anos [31] describes a simulation using
mining. As more partings form, the e€ective thickness
®nite elements which appears to show that the voussoir
decreases and the points move to the left on Fig. 16. It
model presented by Brady and Brown [13], which is
is interesting that while the earlier points correlate
similar to the model described here, poorly predicts
with a rockmass with low modulus, the later data
the simulated displacements and beam stresses, primar-
seems consistent with a sti€er rockmass. This can be
ily due to an over-prediction of the e€ective thickness
linked to two possible mechanisms.
of the compression arch at the midpsan and at the
The ®rst is that as the voussoir beam de¯ects the abutments. The numerical simulations of So®anos [31]
compression in the arch increases. The rockmass mod- predict a normalized arch thickness of less than 0.3 at
ulus of a jointed rockmass is normally pressure equilibrium and less than 0.1 near failure. These are
dependent [29] and therefore, in this case displays a low compared to the values 0.75 for equilibrium and
strain-sti€ening behaviour. In addition, plain strand, 0.3 to 0.4 at failure, predicted by the iterative
cement grouted cablebolts tend to require up to approach in this paper. Numerical simulations by Ran
between 30 and 40 mm of slip to develop their peak et al. [16] indicate an arch thickness more consistent
frictional capacity [14]. If the e€ective embedment with the predictions in this paper.
length of the strand, equal to the lamination thickness, Clearly, something is amiss. The answer appears to
is less than the critical length required to break the be the numerical boundary conditions and discretiza-
steel, the cablebolt can provide this capacity over rela- tion used in the model. It was found during the course
tively large rockmass displacements. It can be seen of this research that a UDEC model [30] composed of
that late in the experiment, as mining progresses rigid blocks and elastic joints (no tension) and a model
farther past the instrumentation sites, the de¯ection composed of deformable blocks and elastic joints (no
beyond 40 mm follows a more stable trend than that tension) both seemed to exhibit similar behaviour to
predicted by the model. Cablebolts must be contribut- the model of So®anos [31] if the beam had ®xed sup-
110 M.S. Diederichs, P.K. Kaiser / International Journal of Rock Mechanics and Mining Sciences 36 (1999) 97±117

Fig. 17. Exaggerated displacement pro®le (top) for a typical UDEC voussoir beam simulation and contours of internal horizontal compressive
stress (bottom).

ports or was bounded by rigid blocks as abutments. potential in cases where lamination thickness and mod-
This creates problems with discretization which do not ulus is uniform.
seem to permit the formation of a stress triangle The beam was ®rst allowed to de¯ect elastically
(0 < 1) at the abutments. Instead, reaction forces are while maintaining a non-zero tensile strength within
concentrated at the bottom corner of the abutments the joints. If this initial elastic de¯ection is not per-
leading to over-prediction of the moment arm and in- mitted, sliding will occur before arch stresses are gener-
accurate de¯ection predictions. ated. It is assumed that excavation is a gradual process
The model ultimately used for this paper utilized de- such that beam deformations occur before the abut-
formable elastic blocks (discrete blocks with internal ments are fully liberated. After elastic equilibrium was
®nite di€erence zones), elastic joints (no tension) and achieved, the joint tensile strength was set to zero and
¯exible (elastic and internally discretized) but very sti€ the beam was allowed to continue deforming until
abutment blocks. This provided better displacement either equilibrium or failure occurred. The joints had
compatibility at the abutments. The results of this no cohesion but had a frictional strength. The initial
model correlate very well with the voussoir analogue elastic deformation stage allowed lateral stress and
as will be discussed presently. Simulations included frictional strength to accumulate in the joints, particu-
single beams only since the purpose of the model was larly at the abutments. Without this stage the beam
a comparison with the analogue model. Hatzor and would simply slip past the abutments under its own
Benary [32] have recently examined the in¯uence of weight.
multiple layers with interbed friction. The single beam Several con®gurations were tested, using di€erent
prediction represents the worst case failure initiation rock sti€nesses, Ei, di€erent joint normal sti€ness,
 M.S. Diederichs, P.K. Kaiser / International Journal of Rock Mechanics and Mining Sciences 36 (1999) 97±117 111

Fig. 18. Predicted variation of horizontal stress along arch axis (ABC) compared with UDEC results (top) and predicted and simulated stress dis-
tributions at abutment and at midspan (bottom).

JKN, and joint spacings, sj. The rockmass modulus, bolic variation introduced in this paper. Again the
Erm for comparison with voussoir predictions, was cal- match is acceptable for engineering application of the
culating using the relationship: method.
Models of unit thickness (T = 1) with several di€er-
1 1 1 ent rockmass moduli were analyzed with increasing
ˆ ‡ 21†
Erm Ei JKN†sj span, in increments of 5 m, up to snap-through failure
(blocks fall and beam disintegrates). Crushing failure
The de¯ection pro®le and the internal stress distri- was not investigated in this analysis. Fig. 19 illustrates
bution are shown in Fig. 17 for one example simu- the calculated equilibrium displacements in the UDEC
lation. The parabolic nature of the compression arch is models and the associated voussoir predictions. The
immediately apparent. Fig. 18 compares the lateral modeled de¯ections match the predicted relationships
stress distribution at the abutment and at the midspan well, with the exception of a slight over-prediction by
with the assumed triangular distribution and the pre- the voussoir model at higher sti€nesses. In addition
dicted voussoir arch thickness (0.65 T). The distri- the UDEC model failed in each case immediately
bution at midspan correlates very well, while the beyond the critical span predicted by the voussoir ana-
numerically simulated abutment stress distribution is logue.
more non-linear, resulting in a higher maximum stress The maximum stress occurred inevitably at the bot-
near the bottom edge of the beam. In addition, the lat- tom edge of the beam at the abutment. The midspan
eral stress variation along a parabolic path, ABC, in maximum stress correlated fairly well with predictions
the UDEC model is compared with the assumed para- (Fig. 20) while the abutment stress was consistently
112 M.S. Diederichs, P.K. Kaiser / International Journal of Rock Mechanics and Mining Sciences 36 (1999) 97±117

Fig. 19. Midspan displacements (data points) from UDEC beam simulations compared with voussoir predictions (T = 1 m).

Fig. 20. Maximum compressive stress in UDEC beams compared with voussoir predictions (T = 1 m).
 M.S. Diederichs, P.K. Kaiser / International Journal of Rock Mechanics and Mining Sciences 36 (1999) 97±117 113

Fig. 21. Simpli®ed relationships for rockmass modulus as a function of rock quality and con®nement.

higher. If this is real and not an artifact of the model- and sample disturbance. The results have been further
ling, it is unlikely to devalue the predicted crushing modi®ed here to provide convenient functions for
limit for the voussoir analogue, since the area over modulus based on rock quality and con®nement
which the increased stresses occur in the UDEC model (Fig. 21). These functions can be used to estimate
is quite limited and any resulting initial crushing rockmass modulus for the voussoir beam. For most
would be highly localized. These numerical results, excavations where the voussoir analogue is valid, the
therefore, give us con®dence that the voussoir model lower bound moduli should be used.
can be used to predict the de¯ection and stability of Finally the compressive strength must be speci®ed.
rock beams which form in laminated ground. Crushing failure is local in nature and therefore relates
to the strength of the intact rock. However, in accord-
ance with recommendations by Martin [34] regarding
the strength of intact rock in situ, one third to one
5. Stability guidelines half of the laboratory UCS is used for this analysis.
The summary design limits based on yield (buckling
The results of this work can be summarized into limit = 35%) are presented in Fig. 22. If a beam thick-
normalized stability charts for use in design of under- ness is speci®ed, then a critical span is obtained for
ground openings. Parametric modelling has shown snap-through failure and for crushing failure. The de-
that the snap-through stability limits for critical span sign span is the lesser of the two values. Likewise, if a
and thickness can be related to the normalized rock- span is speci®ed, the critical thickness is the greater of
mass modulus, E *, which is equal to the modulus the two.
divided by the e€ective speci®c gravity, S.G.*(E * = E/ Actual spans are rarely fully two dimensional in
S.G.*). Similarly, the limits for crushing failure can be nature. In fact the two-dimensional beam represents a
related to the normalized compressive strength lower bound model for determining critical span. An
(UCS* = UCS/S.G.*). The e€ective speci®c gravity, upper bound is obtained from the model for a square
S.G.*, for a beam with a dip of a can be obtained span. Again, Brady and Brown [13] derive a relation-
from Eq. (22). ship for a square span by assuming that the span fails
as four triangular panels. The weight of each panel
S:G:* ˆ S:G: cos a† 22†
creates a moment about the abutment edge:
In order to simplify the estimation of rockmass mod-
ulus based on rockmass classi®cation, the test data for gTS 3
MW ˆ 23†
®eld modulus presented by Barton et al. [27] and by 24
Bieniawski [33] was reexamined by Hutchinson and
Diederichs [14] to account for the e€ect of con®nement which is resisted by a reaction moment
114 M.S. Diederichs, P.K. Kaiser / International Journal of Rock Mechanics and Mining Sciences 36 (1999) 97±117

Fig. 22. Stability guidelines for jointed rock beams (tunnel spans). E€ective speci®c gravity, S.G.* = S.G. cos a.

fmax NTSZ Combining Eqs. (23) and (25) yields a modi®ed re-
MR ˆ 24†
2 lationship for fmax:
This equation, however, is in error due to the assump-
tion of a constant moment arm, Z, through all sections ge S 2
of the triangular panel. The de¯ection at the hinge fmax ˆ 26†
6N Z ‡ Z0 †
point is zero and therefore, the moment arm must
vary from Z = Z0 ÿ d, at the midspan, to Z0 at the
corners. Eq. (24) must therefore be revised to Eq. (26) is then substituted for Eq. (13) in the original
analysis. One ®nal adjustment is necessary, however.
fmax NTS Z ‡ Z0 † For the square span, Eq. (16), for the shortening of
MR ˆ 25†
4 the beam must be replaced by

Fig. 23. Stability guidelines for jointed rock plates (square spans). E€ective speci®c gravity, S.G.* = S.G. cos a.
 M.S. Diederichs, P.K. Kaiser / International Journal of Rock Mechanics and Mining Sciences 36 (1999) 97±117 115
 
1 ÿ n†L 2 N erated by the active pressure created by mechanical
DL ˆ fmax ‡ 27†
E 9 3 rockbolts) resulted in a small increase in stability,
while cohesive resistance (no slip) generated by fully
where n is the Poisson's ratio for the rockmass. grouted bolts (or rock bridges) had a signi®cant sta-
The results for the square span are summarized in bilizing e€ect.
Fig. 23. . These charts are only valid if no low to mid angle
A few limitations with respect to these charts should jointing is present. In this case or in the case of
be noted: thinly laminated ground, support is necessary in
order to create a reinforced beam. Rockbolts or
. These charts can be used with con®dence in lami- (preferably [1]) grouted rebar should be approxi-
nated ground where the lamination thickness is mately equivalent in length to the desired beam
known and where the modulus can be estimated thickness [10]. Assuming a lower-bound rockmass
with some degree of reliability.
modulus, a reinforced thickness can be estimated
. It is imprudent to rely on such an assumptive
from Figs. 22 and 23 which will ensure adequate
method for very large spans (>100 m for steeply
stability. A factor of safety of 1.5 to 2 with respect
inclined spans and >60 m for horizontal spans)
(i.e. use half the span or twice the thickness) to is
where other in¯uences, not considered here, may
adequate for most situations [14].
govern stability. Consistent excavation quality is
. This method is not suitable for poor rockmasses
also assumed. Over-blasting, or uneven excavation
surface geometry will negatively impact roof stab- with a low RQD rating (<50) and more than three
ility. joints sets.
. The assumption in these charts is that the joints are . This technique is designed to predict the onset of
rough enough to provide frictional resistance under roof instability. Thus only the ®rst (lowermost) lami-
low to moderate con®nement (i.e. no slickensides or nation is considered and not a composite beam
low friction coating) and that the span to thickness structure. This is based on the assumption that, pro-
ratios are greater than 10. Sliding failure along joints vided the beam thickness is stability of the whole
at the abutments or within the beam is not con- roof is controlled by the stability of this ®rst beam.
sidered. . The in¯uence of the gravity load component parallel
. It is assumed that there is no frictional or cohesive to inclined laminations is ignored in order to achieve
resistance along the interfaces between the lami- a tractable solution. This leads to the apparently
nations. This represents a worst case assumption reasonable conclusion that inclined layers (dip = a)
since such resistance increases the stability of the are signi®cantly more stable than horizontal layers.
beams. Snyder [1] showed that limited friction (gen- The de¯ecting beam must still achieve the same

Fig. 24. Examples of parallel weight correction to critical span for inclined beams.
116 M.S. Diederichs, P.K. Kaiser / International Journal of Rock Mechanics and Mining Sciences 36 (1999) 97±117

Fig. 25. Corrected equilibrium midspan de¯ections for stable steeply dipping beams (dip = 808), accounting for parallel weight component.

moment equilibrium (between the perpendicular 100% for near-vertical beam (dip = 808).
weight component and the compression arch reac- Undoubtedly, the in¯uence of the parallel weight com-
tion force) regardless of initial (gravitational) com- ponent is more complex than described here, leading
pression within the beam. The beam does, however, to assymetrical beam de¯ection and other compli-
settle under this imposed loading. The e€ect of this cations. A program of numerical experimentation is
additional compression can be considered implicitly warranted to investigate this in¯uence and provide
by imposing an initial shortening of the beam, DdC, more accurate stability predictions for inclined beams.
equivalent to the compression due to gravity (paral- Such a program is beyond the scope of this paper.
lel to the beam) given by Eq. (28). The beam must
®rst close this gap before additional compression
can be created during beam de¯ection. This gap is 6. Conclusions
used to calculate a new initial moment arm in
Eq. (29). This value is substituted for Z0 in Eq. (17). An improved iterative approach to a classic ana-
logue for stability assessment of laminated ground has
S 2 g sin a been presented with several improvements and correc-
DdC ˆ 28† tions including an improved assumption for lateral
E
stress distribution and arch compression, the appli-
s
  cation of support pressure and surcharge loading, sim-
3S 8 2 pli®ed displacement determination and a robust
Z *0 ˆ Z 0 ÿ DdC 29†
8 3S iteration scheme.
A linearity limit or yield limit has been identi®ed,
Examples of the in¯uence of this correction on critical corresponding to a midpsan displacement of approxi-
span are shown in Fig. 24. Given the bevy of simplify- mately 10% of the lamination thickness. This displace-
ing assumptions already inherent in the model, the ment limit appears to be independent of rockmass
impact of this correction is of limited practical signi®- modulus and is a useful guideline for performance
cance for rockmass moduli greater than one GPa and monitoring and observational design. Beyond this dis-
can be neglected in the interest of producing the sim- placement, stability cannot be assured. The yield limit
pli®ed and uni®ed charts in Figs. 22 and 23. On the and the methodology in general has been veri®ed using
other hand, according to this simpli®ed correction, ®eld evidence and numerical simulations.
non-critical equilibrium de¯ections are signi®cantly This limit is used to present stability charts for de-
a€ected as in the examples shown in Fig. 25, with sign. These design charts have been normalized with
increases, at one half of the critical span, exceeding respect to e€ective speci®c gravity which is a function
 M.S. Diederichs, P.K. Kaiser / International Journal of Rock Mechanics and Mining Sciences 36 (1999) 97±117 117

of rock density, excavation dip angle and surcharge [14] Hutchinson DJ, Diederichs MS. Cablebolting in underground
mines. Canada: Bitech, 1996. 406 pp.
loading or support pressure. Two failure modes are
[15] Sterling RL. The ultimate load behaviour of laterally con-
presented for thin laminations. These are snap-through strained rock beams. The state of the art in rock mechanics.
and crushing. Other failure modes apply to thick Proc. of the 21st US Symposium on Rock Mechanics. 1980. p.
beams and are not considered. Critical span of the 533±42.
beam is determined by the critical failure mode (mode [16] Ran JQ, Passaris EKS, Mottahed P. Shear sliding failure of the
which gives the lowest critical span). jointed roof in laminated rock mass. Rock Mech. Rock Eng.
1994;27(4):235±51.
More work is needed to account for the boundary [17] Stimpson B, Ahmed M. Failure of a linear Voussoir arch: a lab-
parallel component of weight in steeply dipping vous- oratory and numerical study. Can. Geotech. J. 1992;29:188±94.
soir beams. While geometrical limits for stability are [18] Milne D. Underground design and deformation based on sur-
marginally a€ected by the neglect of this component, face geometry. Ph.D. Thesis, Mining Department, University of
British Columbia, Canada, 1996.
equilibrium displacement predictions may be signi®-
[19] Mottahed P, Ran J. Design of the jointed roof in strati®ed rock
cantly in error for steeply dipping beams. based on the voussoir beam mechanism. CIM Bull.
1995;88(994):56±62.
[20] Milne D, Pakalnis RC, Felderer M. Surface geometry assess-
Acknowledgements ment for open stope design. Rock Mechanics: Proc. Of the 2nd
North American Rock Mechanics Symposium, vol. 1.
Rotterdam: Balkema, 1996. p. 315±22.
This research has been funded by the Natural [21] Maloney S, Fearon R, Nose J, Kaiser PK. Investigations into
Science and Engineering Research Council (Canada). the e€ect of stress change on support capacity. Rock support in
A special thanks is due to Doug Milne (currently at mining and underground construction. Rotterdam: Balkema,
the University of Saskatchewan) for supplying raw 1992. p. 367±76.
[22] Kaiser PK, Diederichs MS, Yazici S. Cable bolt performance
extensometer data for the Mount Isa case example. during mining induced stress change: three case examples. Rock
Thanks are also due to Sean Maloney of the support in mining and underground construction. Rotterdam:
Geomechanics Research Centre and to Winston Lake Balkema, 1992. p. 337±84.
Mine and Noranda Technology Centre. [23] Maloney SM, Kaiser PK. Stress change and deformation moni-
toring for mine design: a case study. Field measurements in geo-
mechanics. Rotterdam: Balkema, 1991. 481±90.
[24] Maloney SM, Kaiser PK. Field investigation of hanging wall
References support by cable bolt pre-reinforcement at Winston Lake Mine.
Research Report. Geomechanics Research Centre, Laurentian
[1] Snyder VW. Analysis of beam building using fully grouted roof University, Canada, 1993. 140 pp.
bolts. Proc. of the Int. Symp. on Rock Bolting. Rotterdam: [25] Barton NR, Lien R, Lunde J. Engineering classi®cation of rock
Balkema, 1983. p. 187±94. masses for the design of tunnel support. Rock Mech.
[2] BeÂtournay MC. A design philosophy for surface crown pillars 1974;6(4):189±239.
in hard rock mines. CIM Bull. 1987;80(90):45±61. [26] Barton NR, Lùset F, Lien R, Lunde, J. Application of Q-sys-
[3] Miller F, Choquet P. Analysis of the failure mechanism of a tem in design decisions concerning dimensions and appropriate
layered roof in long hole stopes at Mines GaspeÂ. 90th CIM support for underground installations. In: Bergmand M, editor.
AGM. Edmonton, 1988. Paper No. 120. Subsurface space, vol. 2. New York: Pergamon, 1980. p. 553±
[4] Milne DM, Pakalnis RC, Lunder PJ. Approach to the quanti®- 61.
cation of hanging-wall behaviour. Trans. Inst. Min. Metall. [27] Barton N. Application of Q-system and index tests to estimate
1995;105:A69±A74. shear strength and deformability of rockmasses. Proc. Int.
[5] Villasceusa E. Excavation design for bench stoping at Mount Symp. on Engineering Geology and Underground Construction,
Isa Mine, Queensland, Australia. Trans. Inst. Min. Metall. vol. 2. Lisbon: IAEG, 1983. p. II51±70.
1996;106:A1±A10. [28] Kaiser PK, Maloney S. The role of stress change in under-
[6] Beer G, Meek JL, Cowling R. Prediction of behaviour of shale ground mining. Eurock '92. Rotterdam: Balkema, 1992. p. 396±
hangingwalls in deep underground excavations. 5th I.S.R.M. 401.
Symposium. Melbourne, 1981. p. D45±51. [29] Barton NR, Bandis S, Bakhtar K. Strength, deformation and
[7] Obert L, Duvall WI. Rock mechanics and the design of struc- conductivity coupling of rock joints. Int. J. Rock. Mech. Min.
tures in rock. John Wiley and Sons, 1966. 649 pp. Sci. Geomech. Abstr. 1985;22:121±40.
[8] Hoek E, Brown ET. Underground excavations in rock. London: [30] Itasca Consulting Group. UDEC: Universal Distinct Element
Inst. of Min. and Metall., 1980. 527 pp. Code, Version 3.00. 1996.
[9] Fayol M. Sur les movements de terain provoques par l'exploita- [31] So®anos AI. Analysis and design of an underground hard rock
tion des mines. Bull. Soc. Indust. Min. 1885;14:818. voussoir beam roof. Rock Mech. Min. Sci. Geomech. Abstr.
[10] Corlett AV. Rock bolting in the voussoir beam: the use of rock 1996;33(2):153±66.
bolts in ground support. CIM Bull. 1956;LIX:88±92. [32] Hatzor YH, Benary R. The stability of a laminated voussoir
[11] Evans WH. The strength of undermined strata. Trans. Inst. beam: back analysis of a historic collapse using DDA. Int. J.
Min. Metall. 1941;50:475±500. Rock. Mech. Min. Sci. 1998;35(2):165±82.
[12] Beer G, Meek JL. Design curves for roofs and hanging-walls in [33] Bieniawski ZT. Engineering rock mass classi®cations. New
bedded rock based on `voussoir' beam and plate solutions. York: Wiley, 1989.
Trans. Inst. Min. Metall. 1982;91:A18±A22. [34] Martin CD. 17th Canadian Geotechnical Colloquium: the e€ect
[13] Brady BHG, Brown ET. Rock mechanics for underground of cohesion loss and stress path on brittle rock strength. Can.
mining. Chapman and Hall, 1993. 571 pp. Geotech. J. 1997;34:698±725.
 International Journal of Rock Mechanics & Mining Sciences 39 (2002) 569–587

Developments in the design of tunnels and caverns in the

Triassic rocks of the Sydney region
P.J.N. Pells*
Pells Sullivan Meynink Pty Ltd, Suite 11, 10 East Parade, Eastwood, NSW 2122, Australia
Accepted 30 April 2002

Abstract

Details are given of the analytical methods used to design rockbolt and shotcrete support for tunnels and large span caverns under
relatively low cover in the near horizontally bedded Triassic sandstones of the Sydney region.
The paper provides a concise description of the engineering geology of the Sydney sandstones because it is fundamental to tunnel
support design that a valid geological model be the basis of any analytical design. Equations are provided which allow calculations
of the lengths, density and capacity of rockbolts for support in this geological environment. The paper also discusses the design
approach adopted for support of tunnels and caverns which are at sufficient depth to generate compressive and shear failure of the
rock mass, so-called True Rock Pressure.
r 2002 Elsevier Science Ltd. All rights reserved.

1. Introduction artist is required to bring his canvas or paper. And

once that design is articulated by the engineer as
This paper traces what the writer and his co-workers artist, it must be analysed by the engineer as scientist
have learned in regard to the design and construction of in as rigorous an application of the scientific method
caverns and tunnels in the Sydney region over the past as any scientist can make.
two decades. This work is quite specific to the near
The first step in this process is to present the
horizontally bedded Triassic sandstones and shales of
geological setting, because a valid geological model is
the area, and comprises the construction of some of the
the starting point of tunnel design.
world’s widest span near surface caverns, all without
passive support. These include the donut shaped cavern
2. The geological setting
for the Sydney Opera House underground car park,
with a span of 17 m and only 6 m of rock cover (see
2.1. Regional
Figs. 1 and 2), and the 24 m wide section of the double
deck Eastern Distributor tunnel (see Figs. 3 and 4).
Rock engineering works in the Sydney region relate
This paper contains no truly new concepts but seeks
largely to the Triassic sandstones and shales that
to show how facets of the science of rock mechanics
underlie most of the metropolitan area. The Triassic
have been appropriately linked to a particular geological
stratigraphy comprises four major divisions, namely:
environment to provide a design process which is
primarily scientific and not simply educated guesswork
Wianamatta maximum thickness 300 m, consisting
(often called ‘art’). In this regard the writer subscribes
Group mainly of shales and siltstones
wholly to the following quote from Petroski [1]

The conception of a design of a new structure can Mittagong a thin horizon (o20 m) of interbedded
involve as much a leap of the imagination and as Formation shales and sandstones
much a synthesis of experience and knowledge as any
Hawkesbury maximum thickness 290 m, predomi-
*Tel.: +61-2-9874-8855; fax: +61-2-9874-8900. Sandstone nantly near-horizontally bedded sand-
E-mail address: mailbox@psmsyd.com.au (P.J.N. Pells). stone but with some laminite beds

1365-1609/02/$ - see front matterr 2002 Elsevier Science Ltd. All rights reserved.
PII: S 1 3 6 5 - 1 6 0 9 ( 0 2 ) 0 0 0 5 8 - 8
570 P.J.N. Pells / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 569–587

Fig. 1. Location plan of Sydney Opera House underground parking station.

Fig. 2. Artistic impression of Opera House underground parking station.

Narrabeen maximum thickness 700 m, comprising 2.2. Engineering geology of the Hawkesbury Sandstone
Group sandstones with claystone horizons
When viewed in vertical section the Hawkesbury
As shown in Fig. 5 the Wianamatta Group shales and Sandstone may be divided into three facies, namely:
the Hawkesbury Sandstone are the dominant near
surface rocks, and because the shales form an eroded sheet facies
cap it is in the Hawkesbury Sandstone that most of the massive facies
structures discussed in this paper are located. mudstone facies o5% of formation
 P.J.N. Pells / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 569–587 571

Fig. 3. Double deck tunnel of the 3 km long Eastern Distributor.

The sheet facies comprises sets of cross-bedded strata clay matrix 15–30%
bounded by planar near horizontal surfaces, see Fig. 6. secondary quartz 4–10%
The units range in thickness from fractions of a metre to siderite (iron carbonate) 2–4%
greater than 5 m but are typically of the order of a
The average composition of the matrix clay is 55–75%
metre. The near horizontal bedding planes give this
kaolinite, 20–30% illite and the balance mixed-layered
facies a sheet-like appearance when viewed from a
clays.
distance. The cross-beds typically dip to the northeast.
The term massive facies was coined to covey the gross 2.2.1. Strength
aspect when viewed from a distance and does not mean The sandstone comprises medium grained quartz
wholly structureless at closer inspection. Frequently, grains; quartz overgrowth of the grains frequently
sandstone bodies of this facies have a discordant provides an interlocking structure and the development
erosional lower surface and a planar concordant upper thus of crystal faces imparts a glistening effect in
surface. Mudstone (or shale) breccia commonly occurs fractured faces.
within troughs at or above the basal surface, but clasts, Substance strength properties (nominally 50 mm
and in particular mudchips and mudflakes, can occur diameter core) of fresh or slightly weathered sandstone
dispersed throughout. are typically:
Petrographic analyses indicate that typically the
Saturated unconfined compressive strength ðUCSÞ
Hawkesbury Sandstone has the following composition:
¼ 25245 MPa;
detrital quartz grains 50–75%
lithic fragments 2–4% Saturated Brazilian tensile strength ¼ 223 MPa:
572 P.J.N. Pells / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 569–587

The strength of the material oven-dried is about 1.5–

2.0 times the saturated strength. Intact Young’s
Modulus values range from about 2.5 to 8 GPa,
indicating a low to average modulus ratio.
An important question is what substance strength to
use at the ‘‘tunnel scale’’ when evaluating the potential
for stress induced failures. The approach taken by the
writer is based on the view that around excavations the
rock volume potentially subject to high induced
compressive stresses is of the order of bedding plane
and joint spacings, i.e. 0.5–2 m. Based on the data on
size effects on rock strength summarised by Hoek and
Brown [2] a reduction factor of about 0.6 is adopted
with respect to 50 mm diameter core strength. This
means that in practice a field scale strength of about
20 MPa is adopted for the fresh Hawkesbury Sandstone.

2.2.2. Bedding discontinuities

Two main forms of bedding discontinuities are
present (see Fig. 6).
Facies bedding. Major depositional horizons repre-
sented by near horizontal, undulating, bedding disconti-
nuities which have a typical spacing of between about 1
and 2.5 m. These may be continuous for hundreds of
metres and are marked by continuous partings, clay
seams or petrographic changes. Local increase in the dip
of the facies bedding occurs where sand has been
deposited in channel structures.
Clay seams, typically between 5 and 25 mm thick, are
Fig. 4. The Eastern Distributor double deck tunnel during
construction. very common within the sequence. The origin of these

Fig. 5. Triassic geology of the Sydney region.

 P.J.N. Pells / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 569–587 573

Fig. 6. Typical exposure of Hawkesbury Sandstone.

seams is not clearly understood but they provide the RQD values in the fresh or slightly weathered rock
major weakness within the Hawkesbury Sandstone typically lie in the range of 75–100%.
sequence.
Cross bedding. Cross bedding (also termed current 2.2.4. Faulting
bedding) is an ubiquitous feature of the sheet facies. Faulting in the Hawkesbury Sandstone is relatively
Cross bedding planes are often marked by the deposi- rare [3] but there are at least three 20–40 m wide zones
tion of flakes of mica, graphite, and carbonaceous comprising high angle normal faulting, oriented NNE,
matter. The cross bedding usually does not represent running through the central business district. The zones
planes of weakness in fresh or slightly weathered are spaced at between 300 and 600 m. In addition, low
sandstone. However, in moderately to highly weathered angle thrust faults, substantially confined to bedding
sandstone the cross beds can form surfaces of relatively horizons, are also found.
low tensile and shear strength (f0 E302351). The above structures play little part in the rock
mechanics discussed in this paper but serve as a warning
that the Hawkesbury Sandstone is not always a benign
2.2.3. Jointing medium.
The dominant joint set strikes NNE with dips ranging
from 651 to 901 east or west, depending on location
across the city. A secondary, orthogonal, set comprises 2.2.5. Regional stress field
near vertical joints. Measurements of the natural stress field have been
The joints have substantial horizontal continuity made on many projects in the Sydney region using
(typically greater than 20 m). Their vertical continuity hydrofracture, rock slotter and strain cell overcoring
is variable. Many of the joints terminate on bedding techniques. The results of these measurements have been
horizons and may have vertical continuities of the order presented in several papers including Enever et al. [4],
of 5 m. However, about 30% of the joints transgress Enever [5] and McQueen [6]. Unfortunately, many of
several bedding horizons and have vertical continuities the stress measurement compilations combine results
of between 10 and 30 m. Spacings of the dominant NNE from different geological units and include results
joint set range between less than 0.3 m to about 5 m, obviously affected by topography.
with an average of about 1.5 m. A pattern frequently The writer considers that, within the Hawkesbury
observed within the Sydney area is that the joints may Sandstone to a depth of about 150 m, the following
occur in swarms. This means that anywhere between equations represent an appropriate expression of the
three and ten joints may occur over a distance of a few natural total stress field:
metres, with there then being a substantial gap before s1 ¼ sNS ¼ 1:5 þ 1:2sv to 2:0sv MPa; ð1Þ
encountering further joints of the same set.
The orthogonal joints (ESE) have similar continuities
to the main set but with spacings in the range 6–20 m. s2 ¼ sWE ¼ 0:5s1 to 0:7s1 MPa; ð2Þ
574 P.J.N. Pells / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 569–587

s3 ¼ sv ¼ 0:024H MPa; ð3Þ

* increase in the frequency and thickness of near
horizontal seams of extremely weathered material
where sv is the vertical stress, sNS the horizontal stress along facies bedding horizons.
oriented approximately 201 East of true north, sWE the
However, it should be noted that in some zones,
orthogonal to sNS :
weathering causes strength increase due to the conver-
Enever [5] has proposed a step function in the stress
sion of siderite (iron carbonate) to iron oxide. Therefore
field at a depth of 20 m. This does not make sense from a
geological categorisation of weathering is not always an
geological viewpoint and is not adopted by the writer
indicator of engineering properties.
and his co-workers.

2.2.6. Hydrogeological parameters 3. Classification systems and their limitations

The substance permeability of the sandstone is in the
range of 10
9–10
11 m/s. Mass permeability is governed 3.1. The Sydney system for sandstone and shale
by joints and bedding horizons. Analysis of the results
of some 2240 m of Lugeon permeability tests for the Before discussing experience with the Q-system and
Ocean Outfall Tunnels, the Eastern Distributor, the RMR system in the Sydney environment, it is appro-
Chatswood–Parramatta rail tunnels, the Cross City priate to describe a classification system used in Sydney
Tunnel and four electric cable tunnels gives the since 1978 [7,8]. This system was developed specifically
following results: for assessing design parameters for heavily loaded
foundations on the sandstones and shales. It was never
Lugeon value Cumulative length Percentage of total intended for use in tunnel design. However, with the
of borehole length (%) passage of time it has proven to be a very valuable tool
for rapid communication between investigators, de-
o0.1 620 28 signers and contractors of information on mass quality
0.1–1 800 36 of the sandstones and shales. The widespread adoption
1–3 277 12 of the system within the Sydney area has demonstrated
3–10 208 9 that it encapsulates the key features which affect
10–25 200 9 engineering performance of the rocks. It is a five-class
25–100 78 3 system with Class I being the best quality. The system is
>100 53 2 set out in Table 1 and is based on:
* substance unconfined compressive strength,
The log mean permeability from these data is 0.48 * degree of fracturing, usually assessed from core, and
Lugeon (i.e. k ¼ 5  10
8 m/s). Experience from mon- * the percentage cumulative thickness of sub-horizon-
itoring seepages of tunnels and deep basements below tal clay seams within the zone being assessed.
the water table is that average mass permeability is in
the range 3  10
7–5  10
8 m/s. Occasional open joints The lowest rating of any factor defines the class. Thus,
produce moderate seepage flows (0.2–1 l/s), but typically for example, an 8 m horizon of sandstone selected as
only for a few days because of the low storage being reasonably consistent, with the parameters
characteristics of the mass sandstone. * substance UCS=10 MPa,
The regional groundwater table follows a muted * slightly fractured (equivalent to RQD>75),
reflection of the topography, being at sea level along the * cumulative 50 mm of clay seams over 8 m (i.e. 0.6%)
harbour and at a depth of 5–8 m below the ridges.
would classify as Class III because the UCS controls. If
2.2.7. Effects of weathering the UCS were >24 MPa the zone would classify as Class
The parameters presented in Sections 2.2.1–2.2.6 II because the degree of fracturing would control.
apply primarily to fresh or slightly weathered sandstone.
The horizon of extremely (EW) to moderately weathered 3.2. Classification using Bieniawski’s RMR system
(MW) rock typically extends to a depth of between 5
and 15 m. The following is typically apparent with While every site is different, one can obtain a good
increased degree of weathering: idea of how the range of Sydney sandstones classify in
the RMR system [9] by analysing each of the major
* UCS decrease to o20 MPa for MW, o10 MPa for classes given in Table 1. The results of this process are
HW and o2 MPa for EW, given in Table 2 for a north–south oriented tunnel at a
* RQD decrease to 40–70% for MW, 10–40% for HW depth of between 20 and 50 m (equivalent to the Eastern
and o10% for EW, and Distributor tunnels).
 P.J.N. Pells / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 569–587 575

The data in Table 2 show that: 3.3. Classification using the Q-system

* the RMR system is very insensitive to the intact In accord with the approach set out in Section 3.2 for
strength, a parameter which is very important in the the RMR system, Table 3 gives the Q-system classifica-
engineering behaviour of the Sydney sandstone, tion [10,11] for the range of Sydney sandstone, again
* the best sandstone classifies at the bottom of assuming a north–south oriented tunnel at a depth of
Bieniawski’s Class 2, indicating a fair to good rock 20–50 m.
mass—which is very conservative given the specta- Unlike the RMR system, the Q-value classification
cular unsupported vertical cuttings and unsupported provides good discrimination of the range of mass
openings made in this rock, and properties of the sandstones.
* the system does not discriminate well between the
sandstone grades encountered in the Sydney Basin. 3.4. Limitations for support design

Table 1 A 1997 review [12] of the adopted support in five

Engineering classification of shales and sandstones in the Sydney major tunnelling projects in the Hawkesbury Sandstone,
region—a summary guide compared with that indicated by the RMR and Q-
Class Unconfined Defect spacing Allowable
systems, reached the following conclusions:
compressive (mm) seams (%)a
1. The RMR system provides poor discrimination of the
strength qu
(MPa) range in rock mass quality, and only provides
support guidelines for nominally 10 m span tunnels
Classification for sandstone at shallow depth; hence the system has little practical
I >24 >600 o1.5
II >12 >600 o3
value for tunnel design in the Sydney region.
III >7 >200 o5 2. The Q-system predicts a significantly lesser level of
IV >2 >60 o10 support than actually adopted in the five cases studied.
V >1 NA NA
At the time of the 1997 study all that could be said
Classification for shale about the Q-system guidelines was that either the
I >16 >600 o2 designs adopted in the Hawkesbury Sandstone had
II >7 >200 o4
been conservative or the guidelines were potentially
III >2 >60 o8
IV >1 >20 o25 dangerous. Since then information has become available
V >1 NA NA from crown collapses in a storage cavern project, which
a
suggests the latter to be true. For legal reasons full
Allowable seams: seams include clay, fragmented rock and highly
to extremely weathered zones, usually sub-horizontal. The limits
details of these collapses cannot be given at this time.
suggested in the tables relate to a defined zone of influence. For pad Suffice to say that the span was about 13 m, overburden
footings, the zone of influence is defined as 1.5 times the least footing cover about 125 m and the Q-value designated from the
dimension. For socketed footings, the zone includes the length of the site investigation boreholes was 24. The original Q-
socket plus a further depth equal to the width of the footing. For system recommendations [10], for ESR=0.8, would
tunnel or excavation assessment purposes the defects are assessed over
a length of core of similar characteristics.
place this structure in Support Category 14 and would
The classification system is based on rock strength and defects using require tensioned bolts at 1.5–2.0 m spacings at lengths
three parameters as set out below. The lowest rating of any one factor of 3, 5 and 7 m plus chain link mesh. The updated (1993/
defines the class. 1996) recommendations [11] would require systematic

Table 2
Classification of typical range of Sydney sandstone using RMR

Item Parameter Sandstone class according to Sydney system

I II III IV V

1 Intact strength 2 2 2 1 0
2 RQD 18 17 13 5 3
3 Discontinuity spacing 15 15 12 10 8
4 Discontinuity conditiona 25 22 20 10 10
5 Groundwater 10 8 8 8 8
6 Adjustment for orientation
5
5
5
5
5
Total RMR 65 59 50 29 24
Class 2 3 3 4 4
Description Good Fair Fair Poor Poor
a
This is difficult to assess because of the need to define the relative priorities of continuity, roughness, joint infill and joint wall condition.
576 P.J.N. Pells / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 569–587

Table 3
Typical classification of range of Sydney Sandstones using NGI Q-system (o50 m depth)

Item Parameter Sandstone class according to Sydney system

I II III IV V

1 RQD 90 80 65 25 5
2 Jn 2 4 4 6 12
3 Jr 3 3 1.5 1 1
4 Ja 1 1.0 2.0 3.0 6.0
5 Jw 0.8 0.8 0.8 0.66 0.66
6 SRFa 2.5 2.5 5.0 5.0 5.0

Q-value 43 19 2 0.18 0.009

Description Very good Good Fair Very poor Exceptionally poor
a
Practitioners in Sydney vary in their assessment of this parameter.

bolting at 2.6 m spacing with bolt lengths of 4.5 m (no CSIR to work on tunnel design, the most striking
mesh, no shotcrete). The actual support was similar to personal discovery was that there appeared to be no
the original recommendations and heavier than the appropriate analytical design methods for rock tunnel
updated guidelines. The first collapse comprised about support, certainly not as understood in the fields of
300 tonne of crown, the second about 20 tonne. After the structural engineering, hydraulic engineering and even
collapses the support was increased to substantially soil mechanics. There was a lot of talk about the ‘art’ of
greater density and capacity than would be indicated by tunnel design, but on close examination much of this
the Q-system. seemed to be educated guesswork. This is probably why
It may be argued that the Q-value adopted from the those working in rock tunnel support design enthusias-
site investigation boreholes was wrong. The writer has tically, and somewhat uncritically, adopted the RMR
reviewed this matter using logging of the actual and Q classification systems when they appeared in 1973
excavation. This suggests a Q-value of about 13 (i.e. and 1974.
half the original value). However, the support recom- The writer accepts that tunnel design is different from
mendations would be the same. many other engineering design processes. However, it
In early 2000 a major system of TBM driven tunnels can be performed on a scientific basis using an intimate
was completed on the north side of Sydney Harbour to blend of engineering geology, precedent, structural
store peak sewage flows. The scheme included about analysis and the observational method during construc-
6.5 km of 3.8 m diameter tunnel, 3.7 km of 6.0 m tion.
diameter tunnel and 3.5 km of 6.3 m diameter tunnel. Fortunately, the relatively simple geology of the
All the tunnels were in Hawkesbury Sandstone, with Hawkesbury Sandstone has facilitated the development
depths of cover ranging from about 20 m to about 80 m. of analytical design methods over the past two decades,
The initial primary support, comprising rockbolts and methods which have been demonstrated to work well in
mesh, was designed using the Q-system. The actual rail, road and water tunnels, and major caverns. Before
density of rock bolting (bolt per metre) which proved proceeding to discuss the analytical methods currently in
necessary to install, following inadequate performance use it is critically important to reiterate Lauffer’s [13]
of the initial design, ranged between 5 and 9 times the categorisation of tunnel support loadings, namely:
initial design densities.
(i) Loosening pressure: forces or pressures caused by
Based on the information given above it is the practice
the weight of blocks or prisms of loosened or
of the writer, and co-workers, not to rely on the
potentially loosened rock in the crown and side-
recommendations of the Q and RMR systems for
walls of a tunnel.
support design in the Hawkesbury Sandstone. The
(ii) Swelling pressure: caused by volumetric increases
approaches which are used are set out in the remainder
of clays, claystones or other rocks due to exposure
of this paper.
to the atmosphere under altered stress conditions.
(iii) True rock pressure: arising when compressive
stresses generated in the rock around the tunnel
4. Adopted support design methodology
are sufficient to cause compressive yielding and
fracture.
4.1. Basis of design
Swelling pressures are not an issue in Sydney’s
Some 30 years ago when, as a young civil engineer, the Triassic rocks. Loosening pressures are the prime
writer joined Bieniawski’s rock mechanics group at the consideration at depths of less than about 50 m, and
 P.J.N. Pells / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 569–587 577

are discussed in Section 4.2. True rock pressure creates a [22]. This requires a decision to be made on what is the
much more difficult design problem and is discussed in allowable crown sag, and cognisance must be taken of:
Section 4.3. * the mass modulus of the crown strata and the
effective abutment stiffness, and
4.2. Design for loosening pressures * the natural stress field.

Traditionally, loosening pressures have been calcu-

Because, these variables are usually not known
lated using the Terzaghi [14] support loadings, or some
precisely it is appropriate to do a parametric study to
modification thereof. These have no meaning where
develop an understanding of what factors are critical to
support in jointed rock is provided by anchors, dowels,
design. Fig. 8 shows the typical results of such calcula-
mesh and shotcrete. With these forms of support the key
tions for a bolted crown beam in Class II Hawkesbury.
concept is to retain the integrity of the rock mass, which
It can be seen that for a 15 m span (the example given in
will then support itself. In applying this concept to the
Fig. 9) an effective linear arch thickness of 5 m is
near horizontally bedded, steeply jointed Hawkesbury
required if crown sag is to be less than 20 mm.
Sandstones it is appropriate to invoke the concepts of
linear arch theory published by Evans [15].
4.2.1. Bedding plane shear
Cutting an arch-shaped crown in this type of rock
In linear arch theory the steeply dipping joints play no
mass is counterproductive because this creates unneces-
substantial role in a crown stability. If shear along near
sary ‘cantilevers’ of rock and negates the positive aspects
horizontal bedding planes can be limited to approxi-
of a relatively high horizontal stress field (see Fig. 7).
mately that which would occur in a thicker elastic beam
Linear arch theory clearly shows that spans in excess of
then the effective thickness of the linear arch is not
15 m can easily be created in sandstone of Class III or
controlled by physical bedding plane spacings. Hence
better (see Table 1) provided the effective bedding
the fundamental tenet of the analytical design process is
thickness is greater than 5 m (see Fig. 8). The problem is
to limit shear displacements on bedding horizons within
that real bedding spacings are between 0.5 and 2 m (see
the desired thickness of the linear arch.
Fig. 6). The solution of this problem is to provide
In order to implement this procedure two initial sets
internal reinforcing to the rock so as to artificially create
of calculations have to be made, namely:
the requisite effective bed thickness. The analytical
process whereby this is achieved is set out below. In 1. Calculation of the probable bedding plane shear
essence it is an extension of the theory of reinforcement displacement that would occur at an acceptable
of a laminated crown beam by friction effects as set out maximum crown sag, if the crown rock were
in Chapter 20.3 of Obert and Duval [16]. unreinforced. This can be done using a jointed finite
The requisite thickness of the linear arch can be element model. The results of such an analysis are
determined using the modified version of the original given in Figs. 9b and c for the example tunnel shown
Evans linear arch theory as given by Sofianos [17], or by in Fig. 9a. If only horizontal bedding discontinuities
a more sophisticated version as developed by Booker are considered then a closed form solution can be

Fig. 7. Geometric disadvantages arising from cutting arch crown in Hawkesbury Sandstone.
578 P.J.N. Pells / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 569–587

Fig. 8. Theoretical design curves from linear arch analysis with examples from Sydney projects.

used to estimate these shear displacements, as used for local support of isolated loosened blocks of
discussed by Bertuzzi and Pells [24]. rock.
2. Calculate the shear stresses which would occur at the Fig. 10 shows the general case of a single rockbolt
locations of physical bedding horizons if behaviour crossing a discontinuity. The reinforcement acts to
were purely elastic. This can be done using the same increase the shear resistance of the joint by the following
finite element model but with elastic bedding plane mechanisms:
behaviour. The results for the simple example are
1. an increase in shear resistance due to the lateral
given in Fig. 9d. Again, if only horizontal bedding
resistance developed by the rockbolt via ‘‘dowel
discontinuities are considered, a closed form solution
action’’—force R1 ;
can be used for these shear stresses [24].
2. an increase in normal stress as a result of prestressing
Once the process of calculating the bedding plane of the rockbolt—force R2 ;
shear displacements and shear stresses is completed as 3. an increase in normal stress as a result of axial force
per (i) and (ii) above, attention can be turned to developed in the rockbolt from dilatancy of the
calculating the rock bolt capacities, orientations and joint—force R3 ; and
distributions required to create the effective linear arch. 4. an increase in normal stress as a result of axial force
This process is set out in Section 4.2.2. developed in the rockbolt from lateral extension—
force R4 :
4.2.2. Calculating rock bolting requirements The first component can be considered as increasing
At the outset it should be noted that consideration is the cohesion of the joint, while the other three
given here only to fully grouted rockbolts. These are components increase the frictional component of inter-
typically so far superior to end anchored bolts in their face strength by increasing the effective normal stress on
influence on rock mass behaviour that the latter are only the interface. If the rockbolts are at a spacing S; so that
 P.J.N. Pells / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 569–587 579

each anchor affects an area, S 2 ; the equivalent increases

in cohesion, Dc; and normal effective stress, Dsn ; are as
follows:
R1 þ R5
Dc ¼ ; ð4Þ
S2
R2 þ R3 þ R4
Dsn ¼ : ð5Þ
S2
As a result of the anchors, the equivalent strength of
the joint, sj ; will be as follows:
sj ¼ ðcj þ DcÞ þ ðsn0 þ Dsn Þtan fj ; ð6Þ
where cj is the effective cohesion of joint, fj the effective
friction angle of joint, sn0 the initial effective normal
stress on joint, Dc the equivalent increase in effective
cohesion (Eq. (4)), Dsn the equivalent increase in
effective normal stress (Eq. (5)).
Methods of calculating forces R1 ; R3 and R4 as a
function of joint shear displacement are set out in
Sections 4.2.2.1 and 4.2.2.2. Force R2 is created by the
initial bolt pretension.

4.2.2.1. Calculation of dowel action: force R1. Calcula-

tions of dowel action is based on laboratory test data
and theoretical analyses presented by Dight [18]. The
experimental data showed that:
* plastic hinges formed in the fully grouted rockbolts at
small shear displacements (typically o1.5 mm); these
plastic hinges occur only a short distance on either
side of the joint.
* crushing of the grout, or rock (whichever was the
weaker) occurred at similar small displacements.
Based on his experiments, on plastic bending theory,
Fig. 9. (a) Example tunnel, (b) shear displacements, (c) normal and Ladanyi’s [19] expanding cylinder theory, Dight
stresses, (d) elastic shear stresses. developed equations for calculating the ‘dowel’ force R1 :
For the simplified assumptions of grout strength equal
to or less than the rock, and for the joint having no infill,
the equations are
 
D2 pffiffiffiffiffiffiffiffiffiffiffi R2 2
R1 ¼ 1:7sy Pu p 1
; ð7Þ
4 Ty
where
 A=2
d
Pu ¼ ð8Þ
KðpD þ 2dÞ
and
   
1
n2 sc sc
K ¼ sc ln þ
E 2P0
st ð2P0
st Þ
 
2nðP0
st Þ
st
 ð9Þ
E
and
2 sin f
A¼ ; ð10Þ
Fig. 10. Single full column grouted rockbolt across a joint. 1 þ sin f
580 P.J.N. Pells / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 569–587

where st is the tensile strength of the rock, sc the They appear to give similar predicted load versus shear
compressive strength of the rock, E the modulus of displacement curves to those obtained using the
the rock, f the friction angle of the crushed rock, Ty the equations presented in this paper. The difficulty is that
yield strength of the rockbolt, n the Poisson’s ratio of they include a ‘‘correction constant’’ which appears to
the rock, R2 the initial tension in the bolt, P0 the initial have no objective method of quantification.
stress in the rock in the plane of the joint (assumed equal
all round the bolt), d the shear displacement on the joint. 4.2.2.3. Rock bolt length. The bolt length is usually
Eqs. (7)–(10) can be solved, using a MathCAD routine taken as the required linear arch thickness plus 1 m. This
or a spreadsheet, to give a relationship between joint presumes there to be a physical bedding plane at the
shear displacement and the dowel action resistance R1 : upper surface of the nominated linear arch and is
intended to allow sufficient bond length for mobilization
4.2.2.2. Calculation of axial forces due to dilation and of bolt capacity at this postulated plane.
lateral rockbolt extension: forces R3 and R4. Calculation
of forces R3 and R4 is based on observations in 4.2.2.4. Bolt density. The design process is iterative
experimental tests by Dight [18], Pells [20] and Pellet because of the following variables in regard to the bolts
and Egger [21] that, for fully grouted rockbolts, axial alone:
bolt yield is attained at a joint after very small shear or * bolt capacity—a function of diameter and bolting
joint opening movements (typically o1.5 mm). material (typically either 400 MPa reinforcing steel or
The distance to the plastic hinges from the joint nominally 950 MPa steel associated with Macalloy/
surface is given approximately by the equation VSL/Diwidag bars),
sffiffiffiffiffiffiffiffiffiffiffiffiffiffi 
D 1:7psy R2 2
* bolt inclination,
L¼ 1
: ð11Þ * bolt spacing across and along the tunnel.
4 Pu Ty
Typically, for tunnels of spans up to about 12 m use is
If the assumption is made that, under small shearing made of standard rockbolt steel (nominally 400 MPa).
displacements, axial strain in the rockbolt is dominantly For larger spans some, or all, of the bolts comprise high-
between the two plastic hinges, then from the geometry grade steel.
given in Fig. 10 the equations for R3 and R4 are, as a It is advantageous to incline bolts across the bedding
function of shear displacement: planes provided one is certain as to the direction of
 
d tan i pD2 shearing. However, as shown in the example presented
R3 ¼ Es ; ð12Þ
2L 4 in Fig. 9 such shearing is not always symmetrical about
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1 the tunnel centreline. Bolts inclined across bedding
0
L
L2
ðd=2Þ2 pD2 against the direction of shearing can be largely
R4 ¼ Es @ A ; ð13Þ ineffective. Therefore given the uncertainty in this
L 4 regard it is considered appropriate that only those bolts
located over the tunnel abutments should be inclined,
where Es is the modulus of the rockbolt material (steel), the central bolts are installed vertically. As an example,
i the joint dilatancy angle. Fig. 12 shows the support used for the wide span section
If the bolt is inclined at an angle (p) to the joint such of the Eastern Distributor.
that forces in the bolt resist shear movement, then the Having made the above decisions regarding bolt
increased shear resistance due to R3 and R4 is lengths and inclinations the process of bolt density
Ds ¼ ðR3 þ R4 Þ sinðp þ d=LÞ tanðfj þ iÞ þ ðR3 þ R4 Þ computation proceeds, in principle, as set out below.
 cosðp þ d=LÞ; ð14Þ
Step 1 The tunnel crown is divided into patches at each
where p is the angle of bolt to shear surface (radians). bedding horizon, as illustrated in Fig. 9, with
Eqs. (7)–(10) and (14) can be used to calculate the each patch intended to cover one rock bolt. It
total shear resistance provided by any rockbolt at any should be noted that the first major bedding
angle to a joint. The equations give good agreement with horizon above the crown usually controls design.
laboratory tests performed by Dight [18] and Pellet and Step 2 From the jointed finite element analysis (see
Egger [21], and with calculations made by Poulos [22] Fig 9b) the average shear displacement and the
using the analogy of a laterally loaded pile. Fig. 11 normal stress (see Fig. 9c) within each patch
shows the calculated combined forces R1 ; R2 ; R3 and R4 are calculated.
using the equations presented above, and similar Step 3 A rockbolt type (diameter, material, inclina-
predictions made by Poulos. tion) is selected for a patch and the forces
Equations to calculate the effect of fully anchored R1 2R4 are calculated as per the equations
rockbolts are also provided by Pellet and Egger [21]. given in Section 4.2.2.
 P.J.N. Pells / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 569–587 581

Fig. 11. Calculated resistance forces R1 2R4 for typical full column grouted rockbolts.

Step 5 The average shear stresses (tapplied ) in the same

patch is computed from the elastic finite
element analyses (see Fig. 9d).
Step 6 The ‘‘factor of safety’’ against shearing within
each patch is defined as
FOS ¼ tstrength =tapplied :
It is required that each patch have a FOSX1.2
although it may be found that one or two
patches on some joints may have lower factors
of safety.

4.2.3. Calculating shotcrete requirements

4.2.3.1. Loading. The basic principle behind the design
Fig. 12. Twenty-four metre span section of double decker Eastern of shotcrete, in the loosening pressure environment, is
Distributor. that it is to support and contain the rock between the
rockbolts. The size of the rock blocks which potentially
have to be supported (the ‘‘design block’’) have to be
Step 4 Using the values of R1 –R4 ; and the normal assessed on a probability basis from the known geology.
stress from Step 2, the shear strength of the However, the point should be noted that there is no way
bolted patch is calculated (tstrength ). of knowing, in advance of excavation, where exactly
582 P.J.N. Pells / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 569–587

these blocks will be located. In reality they will occur at Steel fibres Dramix 30 or 40 mm at about 50–
only a few locations in the crown of the tunnel, but 60 kg/m3
because the shotcrete must be applied in a pre-planned, Aggregate grading EFNARC [25]
systematic manner, and because safety requirements Microsilica about 5% of cement content
dictate that not even a brick size piece of rock may be
unsupported, it is necessary to assume that the ‘design Shotcrete designed for adhesion
block’ can occur anywhere. It comprises a patch of
gravity load on the shotcrete. UCS 35–45 MPa
Steel fibres Dramix 30 mm at 20 kg/m3
4.2.3.2. Shotcrete action. The shotcrete may be de- Aggregate grading EFNARC [25] or finer
signed to act in one of the two ways: Microsilica About 8–10% of cement content
(i) as a membrane spanning between bolts, or
4.2.4. Role of the observational method
(ii) by adhesion to the rock immediately around the
The design process described above cannot be done
design block.
for each metre of tunnel, and the practice is to develop
The two mechanisms are illustrated in Fig. 13. designs for the Typical, Adverse and Special conditions
Under membrane action the flexural strength of the expected in each Region of the geotechnical model
shotcrete is the key material parameter and steel fibres, prepared for the project, as described by Pells and Best
or mesh, are essential. In addition considerable attention [26]. Mapping and monitoring is essential during
has to be given to detailing a good structural connection construction to check that geotechnical conditions are
between shotcrete and rockbolt heads. Spider plates, as as expected and that crown deflections are consistent
illustrated in Fig. 14, were used for this purpose on the with design expectations.
Eastern Distributor project. The Opera House Carpark cavern crown, with its 6 m
The adhesion mechanism involves the shotcrete acting of rock cover, was designed using the principles
in shear, [23], and therefore flexural strength (and high described above and crown sag was carefully monitored
dosage of steel fibres) is of little relevance. The key as the 17.5 m span was progressively created by stripping
factor for ‘adhesion’ behaviour is that the rock surface is from an initial 6 m heading. Fig. 15 shows how crown
very clean. centreline sag increased as the span was progressively
Current practice in the Hawkesbury Sandstone is to increased, compared with theoretical predictions made
use the adhesion mechanism when designing in Class I at design stage.
or Class II sandstone, and to use the membrane concept Maximum crown sag measurements from four differ-
in poorer quality sandstone and shales. ent projects are shown in Fig. 8 compared with
Details of the methods of calculation are given by theoretical ‘‘linear arch’’ predictions.
Bertuzzi and Pells [24] and Barrett and McCreath [23] An important observation which has been made from
and are not repeated here. extensometers installed in the crowns of the Eastern
Distributor and M5 tunnels is that it is impractical to
4.2.3.3. Typical shotcrete specifications. Current design install rockbolts sufficiently close to the advancing faces
specifications in the Sydney tunnels are typically as of initial headings to prevent significant opening of
summarised below. bedding planes which exist less than about 0.5 m above
Shotcrete designed for membrane action the crown. With opening of such bedding planes close to
the crown it is not possible to generate the shear
UCS 40–50 MPa response equivalent to a pseudo-clastic beam, as
Residual flexural 2 MPa at 2 mm required by the design method given in Section 4.2.2.
strength (ASTM) To allow for this reality it is recommended that the

Fig. 13. Alternative structural actions of shotcrete.

 P.J.N. Pells / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 569–587 583

Fig. 14. Spider rockbolt plate used on Eastern Distributor to ensure adequate connection between shotcrete and rockbolts.

Fig. 15. Prediction & measurement of roof sag of the Opera House Carpark cavern.

lower 0.5 m of crown be treated as dead load (15 kN/m) manifests as loading imposed on crown support by
and not be included as part of the design linear arch compressive, shear and tensile failure of the near
thickness. horizontally bedded rock.
Design for true rock pressure is more difficult than for
4.3. Design for true rock pressure loosening pressure because a coherent analytical design
method is yet to be developed. However, the principles
Given the relatively high horizontal stress in the of design are quite clear, and safe designs for deep
Sydney Basin (see Section 2.2.5), true rock pressure tunnels and caverns in the Sydney Basin rocks
584 P.J.N. Pells / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 569–587

can be developed as outlined below. The important steps periphery exceed the mass strength of the sandstone
are: (typically about 20 MPa as discussed in Section 2.2.1).
However, good evidence has been presented by Stacey
(i) Assess the likely locations and extents of rock mass
[27] and Martin et al. [28] that the onset of, and the
yielding.
extent of, brittle rock mass failure can better be estimated
(ii) Take steps to reduce the extent of the problem by
using a tensile strain criterion. The writer’s current
altering the position and shape of the excavation.
practice is to use a stress versus strength criterion for
(iii) Recognise that no reasonable amount of support
assessment of failure likelihood, and for selecting an
can prevent the rock mass yielding and that
appropriate excavation shape. The tensile strain criterion
support is there to contain the failed material and
is used to assess the likely volume of failure once it is
maintain the geometric integrity.
known that true rock pressure cannot be avoided. In
(iv) Make use of the observational method by appro-
Class I and Class 2 Hawkesbury Sandstone it is reason-
priate instrumentation, and implement the princi-
able to adopt a critical tensile strain of about 0.0005.
ples of NATM.
As discussed by Pells [29], crown stress concentrations
Steps (i) and (iii) are dealt with in more detail in the under a high horizontal stress field are substantially
following sub-sections. Step (iv) is dealt with in many increased by the presence of low shear strength and/or
current texts and there are no special features applicable low stiffness near-horizontal bedding discontinuities.
to the Sydney Basin rocks. This is illustrated in Fig. 16 for the cases of a circular
TBM tunnel.
4.3.1. Location and extent of crown yielding If, at excavation level, the natural stresses are
Usual practice is to assume that crown yielding
initiates when the peak induced stresses at the tunnel sh ¼ s1 ¼ K0 sv ¼ K0 gd;

Fig. 16. Effect on low stiffness bedding horizons on stress concentrations around TBM tunnel.
 P.J.N. Pells / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 569–587 585

where sh is the horizontal stress, s1 the major principal Clearly, particular site conditions affecting rock
stress, sv the overburden pressure, K0 the constant strength, bedding plane defects and topographic effects
which is calculated from Eqs. (1)–(3), d the overburden on the stress field must be taken into account in a
depth, g the unit weight. detailed assessment, and it should be noted that lengths
Then, in general, the maximum compressive crown of tunnel below 100 m in massive sandstone have
stresses can be expressed as shown no signs of failure. For flat-crown tunnels the
stress concentrations may be less than for a circular
s0 ¼ ðSf K0
1Þgd; tunnel [29].
where Sf is a constant depending on the shape of the
4.3.2. Bolting and structure requirements
tunnel and the anisotropy of the rock.
The characteristics of rockbolts appropriate for true
For a circular tunnel in isotropic rock, the value of
rock pressure are in some ways quite different from
Sf ¼ 3: Therefore, for g ¼ 0:024 MN/m3 the induced
those required for loosening pressures. As already
stresses (s0 ) would exceed a rock strength of 20 MPa at
stated, there is no way bolting can prevent yielding of
an overburden depth of 166 m. Actual experience in the
the rock. Therefore bolts must be able to accommodate
Hawkesbury Sandstone is that crown and invert failure
potentially large shear movements along joints and new
has developed around circular tunnels at depths of as
fractures, while at the same time providing high shear
little as 60 m. Invariably such failures are observed to be
resistance. Again full column grouted bolts are far
associated with low shear strength bedding surfaces just
superior to end anchored bolts.
above crown or just below invert (see Fig. 17). There is
The following types of bolts are ineffective or, at best,
no doubt that failure at such relatively shallow depth is
inefficient:
due to the stress considerations being similar to those
indicated by Fig. 16. * glass fibre reinforced plastic—because the long-term
For rapid ‘‘back-of-the-envelope’’ assessments the strength of this material is only about 30% of the
writer uses a value of Sf ¼ 6 for TBM tunnels, where short-term strength, and high local shear strains can
low-strength bedding planes are likely to be present, easily result in the long-term strength being exceeded
indicating that significant true rock pressure problems without the designer having any control of this
may have to be dealt with below a depth of about 75 m. matter.

Fig. 17. Typical stress induced failure in TBM tunnel of North Side Storage Project.
586 P.J.N. Pells / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 569–587

* thin wall hollow steel bolts—because they provide standing of the important facets which must be covered
much less shear resistance than solid bolts. by a design, and the paper provides guidelines in this
* high tensile strength steel bolts—because strain to regard for rock masses similar to the Hawkesbury
failure may be as low as 5%. Sandstone.

Solid steel bolts composed of material with at least

20% strain to failure are appropriate. Furthermore,
tests performed by Pells [20] and Dight [18] suggest that References
smooth sided bolts may perform better than rib profile
[1] Petroski H. To engineer is human. New York: St Martins Press,
bolts.
1982.
Bolt lengths are chosen so as to extend about 1 m [2] Hoek E, Brown ET. Underground excavations in rock. London:
beyond the yielded volume of rock calculated using the Institution of Mining and Metallurgy, 1980.
tensile strain criterion as discussed in Section 4.3.1. [3] Branagan DF. An overview of the geology of the Sydney region.
Total bolt capacity is preferably 1.5 to 2.0 times the In: Pells P, editor. Engineering geology of the Sydney region.
Rotterdam: Balkema, 1985. p. 3–46.
weight of the estimated yielded volume of rock. This
[4] Enever JR, Walton RJ, Windsor CR. Stress regime in the Sydney
factor of safety reflects the uncertainty associated with basin and its implications for excavation design and construction.
predicting the volume of fractured rock. Proceedings of the Seventh Australian Tunnelling Conference,
One final point to make in regard to rockbolting for Sydney, 1990. p. 49–59.
true pressure is that long term design life (greater than [5] Enever JR. Near surface in-situ stress and its counterpart at depth
in the Sydney metropolitan area. Aust Geomech J 1999;34(2):
25 years) is difficult to attain because the degree of local
65–76.
distortion of the bolts at joints and new fractures is such [6] McQueen LB. Stress relief effects in sandstones in Sydney
that the continuity of most corrosion protection underground and deep excavations. Sandstone City, Monograph
measures (galvanising, epoxy coating and HDPE No. 5, Environmental, Engineering and Hydrogeology Specialist
sheathing) cannot be assured. Group, Geological Society of Australia, 2000.
[7] Pells PJN, Douglas DJ, Rodway B, Thorne CP, McMahon BK.
Shotcrete is designed on the basis of membrane action
Design loadings for shales and sandstones in the Sydney region.
because adhesion cannot be assumed when large rock Aust Geomech J 1978;G8:31–9.
mass deformations occur. The practice is to adopt a [8] Pells PJN, Mostyn G, Walker BF. Foundations on sandstone and
uniformly distributed load on the shotcrete, which is shale in the Sydney region. Aust Geomech J 1988;33(3):17–9.
considered to be point supported by the bolts. [9] Bieniawski ZT. Engineering classification of jointed rock masses.
Civ Eng S Afr 1973;15(12):335–43.
[10] Barton N, Lien R, Lunde J. Engineering classification of rock
masses for the design of tunnel support. Rock Mech 1974;6(4):
5. Conclusions 189–236.
[11] Barton N. Investigation, design and support of major road
tunnels in jointed rock using NMT principles. Proceedings of
In the rocks of the Sydney Basin existing rock mass the Ninth Australian Tunnelling Conference, Sydney, 1996.
classification systems have value in rapid communica- p. 145–60.
tion between professionals but are considered to have [12] Pells PJN. Classification systems—good for communication but
limited value in assessing support requirements. not always for design. Tunnelling Under Difficult Conditions,
International Conference, Basel, ITC, 1997. p. 99–118.
Within the Hawkesbury Sandstone, with its near
[13] Lauffer H. Gebirgklassifizierung fur der Stollembau. Geol
horizontal bedding and vertical jointing, it has been Bauwes 1958;24(1):46–51.
possible, using a combination of old and new rock [14] Terzaghi K. Rock defects and loads on tunnel support. In:
mechanics concepts, to develop analytical methods for Proctor RV, White T, editors. Rock tunnelling with steel support.
designing rock reinforcing for ‘loosening pressures’. The Youngstown, OH: Commercial Shearing and Stamping Co.,
methodology set out in the paper has been used Youngstown, OH, 1946.
[15] Evans WH. The strength of undermined strata. Trans Inst Min
successfully in several major projects where flat crown Metall 1941;50:475–532.
excavations with spans up to 24 m have no passive [16] Obert L, Duvall WI. Rock mechanics and the design of structures
lining. Support comprises permanent rockbolts and a in rock. New York: Wiley, 1966.
thin layer of steel fibre reinforced shotcrete. Monitoring [17] Sofianos AI. Analysis and design of underground hard rock
of crown deflections and rock bolt loads has shown voussoir beam crown. Int J Rock Mech Min Sci 1996;33(2):
153–66.
good agreement between theory and reality. [18] Dight PM. Improvements to the stability of rock walls in open pit
It is believed that the methodology proven in the mines. PhD thesis, Monash University, Australia, 1982.
Hawkesbury Sandstone can find application in similar [19] Landanyi B. Expansion of cavities in brittle media. Int J Rock
rock masses elsewhere in the world. Mech Min Sci 1967;4:301–38.
[20] Pells PJN. The behaviour of fully bonded rockbolts. Proceedings
At depths where rock mass yielding occurs, and
of the Third Congress ISRM, vol. 2B, Denver, 1974. p. 1212–71.
designs have to deal with ‘true rock pressure’, the stage [21] Pallet F, Egger P. Analytical method for the mechanical
has not been reached that an integrated analytical design behaviour of bolted joints subjected to shearing. Rock Mech
method is in place. However, there is good under- Rock Eng 1996;29(2):73–97.
 P.J.N. Pells / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 569–587 587

[22] Pells PJN, Best RJ, Poulos HG. Design of crown support of the [26] Pells PJN, Best RJ. Aspects of primary support design for tunnels
Sydney Opera House underground parking station. Tunnelling in the Sydney area. Trans Inst Eng Aust 1991;CE33(2):57–66.
Underground Space Technol 1994;9(2):201–7. [27] Stacey TR. A simple extensions strain criterion for fracture of
[23] Barrett SVL, McCreath DR. Shotcrete support design in blocky brittle rock. Int J Rock Mech Min Sci 1981;18:469–74.
ground: towards a deterministic approach. Tunnelling Under- [28] Martin CD, Kaiser PK, McCreath DR. Hoek–Brown parameters
ground Space Technol 1995;10(1):79. for predicting the depth of brittle failure around tunnels. Can
[24] Bertuzzi R, Pells PJN. Design of rock bolts in tunnels in Sydney Geotech J 1999;36:136–51.
Sandstone. Sydney: ITA World Tunnel Congress, 2002. [29] Pells PJN. Geometric design of underground openings for high
[25] EFNARC. The European specification for sprayed concrete, horizontal stress fields. Proceedings of the Third Australian–NZ
Hampshire, UK, 1996. Geomechanics Conference, Wellington, NZ, 1980. p. 2–183, 2–188.