Tunnelling and Underground Space Technology 153 (2024) 106019

Contents lists available at ScienceDirect

Tunnelling and Underground Space Technology

incorporating Trenchless Technology Research
journal homepage: www.elsevier.com/locate/tust

Considering creep in rock tunnelling with a combined support system:

Theoretical solutions and machine-learning solver
Fei Song a , Huaning Wang a, b, * , Chengshun Shang c, d , Alfonso Rodriguez-Dono c, d , Ertai Wang a ,
Jingyun Gui e , Leandro Alejano f
a
School of Aerospace Engineering and Applied Mechanics, Tongji University, Shanghai 200092, China
b
School of Civil Engineering, Suzhou University of Science and Technology, Suzhou 215009, China
c
International Centre for Numerical Methods in Engineering (CIMNE), Barcelona 08034, Spain
d
Department of Civil and Environmental Engineering, Universitat Politècnica de Catalunya (UPC), Barcelona 08034, Spain
e
School of Geology and Environment, Xi’an University of Science and Technology, Xi’an 710054, China
f
CINTECX, GESSMin Group, Department of Natural Resources and Environmental Engineering, University of Vigo, Vigo 36310, Spain

A R T I C L E I N F O A B S T R A C T

Keywords: This study provides an alternative theoretical approach to analyse the mechanical behaviour of tunnels exca-
Theoretical solutions vated in time-dependent geomaterials, considering the sequential excavation of tunnels and installation of
Supported tunnels rockbolts and elastic liner. In the theoretical analyses, the Bolted Rock Mass (BRM) is modelled as a homoge-
Rockbolts
neous material but with higher stiffness. An innovative Machine-Learning-Based solver (MLB-solver) has been
Time-dependency
Machine-learning-based solver
developed to evaluate the reinforcement ability of rockbolts for geotechnical applications. After that, the com-
plex variable method, the Laplace transformation technique, and the extension of the correspondence principle,
combined with the compatibility and boundary conditions, have been employed to obtain the analytical solutions
for stresses and displacements at the rock-support and support-support interfaces. Furthermore, the analytical
predictions have been validated by monitoring data from the Lyon-Turin Base Tunnel. Parametric analyses are
conducted to investigate the influence of rockbolt length and installation time on tunnel behaviour. Meanwhile,
the proposed theoretical solutions have been also applied to analyse tunnel stability. This study offers a new and
efficient method for the preliminary design of supported tunnels.

1. Introduction excessive convergence. If the support is installed too close to the tunnel
face, it may cause over-stress in the support system, leading to collapse.
Tunnels with a support system comprising several different struc- To avoid this stress build-up, low stiffness supports could be used, but
tures are widely used in various engineering fields, including trans- again this may cause excessive convergence (Alonso et al., 2003; Barla
portation, geological disposal, hydraulic and mining engineering (Chu et al., 2010; Song et al., 2018a, 2018b; Song and Rodriguez-Dono,
et al., 2021; Cui et al., 2021, 2022; Do et al., 2020; Guan et al., 2022; 2021). On the other hand, due to time-dependent deformations, an
Showkati et al., 2021; Sun et al., 2021; Wu et al., 2022, 2024; Xue et al., early installation of the support could lead to long-term load build-up on
2018; Zhou and Yang, 2021). The support system, typically in the form the support system (Barla et al., 2010). Therefore, the design of supports
of spray-on concrete lining, shotcrete lining, rockbolts, steel sets, or a is critical for tunnel stability.
combined support system, is usually installed to seal the host rocks and Laboratory and field tests showed that soft rocks exhibit significant
ensure the stability of underground structures (Leca and Clough, 1992; time-dependent deformations that account for a large part of the total
Oreste, 2015, 2003; Song et al., 2018a, 2018b; Song and Rodriguez- tunnel deformations (Paraskevopoulou and Diederichs, 2018; Song
Dono, 2021). et al., 2020, 2022; Barla et al., 2008; Paraskevopoulou, 2016). Creep
The art of tunnel design is to find the right type of support and/or behaviour can result in delayed deformations, which should be consid-
reinforcement and their right installation time (Alonso et al., 2003). If ered to design underground projects more accurately and safely (Alonso
the support is installed too far from the tunnel face, it may cause et al., 2003). As shown in Fig. 1, large time-dependent convergence

* Corresponding author at: School of Aerospace Engineering and Applied Mechanics, Tongji University, Shanghai 200092, China.
E-mail address: [email protected] (H. Wang).

https://doi.org/10.1016/j.tust.2024.106019
Received 11 September 2023; Received in revised form 3 June 2024; Accepted 6 August 2024
Available online 12 August 2024
0886-7798/© 2024 Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.
F. Song et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 153 (2024) 106019

develops during tunnel excavation and may last for a long period of time analytical solution model to analyse deep tunnels supported by rock-
(Barla et al., 2008; Fabre and Pellet, 2006; Sandrone and Labiouse, bolts. Furthermore, some other researchers (Bobet and Einstein, 2011)
2010). Thus, the time-dependency of rocks is a crucial topic for the developed elastoplastic analytical solutions for circular tunnels with
design of supported tunnels, particularly in soft time-dependent rockbolts under uniform in-situ stresses.
behaving rocks. For tunnels with a combined support system, Bobet (Bobet, 2009)
Tunnel design is mainly based on empirical approaches, analytically presented elastic solutions to estimate stresses and deformations for
based designs, numerical simulations, in-situ field tests or the deep tunnels. Then, analytical solutions were provided for supported
convergence-confinement method (CCM) (Bobet, 2009; Bobet and Ein- circular tunnels (Wang et al., 2019; Wu et al., 2020). Recently, Jiang
stein, 2011; Cai et al., 2015, 2004a, 2004b; Carranza-Torres and Zhao, et al. (2023) studied the reinforcement effect of rockbolts on host rocks
2009). Numerical simulations have been widely used in geotechnical using the hybrid DEM-FDM method. However, these solutions only
applications, considering more realistic geological conditions (Song apply to axisymmetric tunnels with isotropic initial stresses. Neverthe-
et al., 2022, 2023, 2024, 2025). However, numerical simulations are less, in-situ stress measurements have shown that initial stresses are
sometimes numerically unstable and generally time-consuming, espe- generally anisotropic. In summary, the solutions of supported tunnels
cially for parametric analyses (Song et al., 2018a,b, 2021; Song and were mainly focused on tunnels with a single support system or tunnels
Rodriguez-Dono, 2021). with a combined support system but for axisymmetric problems.
As an alternative, the CCM is an analytical method that can effi- Furthermore, there is a lack of solutions for supported tunnels excavated
ciently determine support forces by considering ground-support in- in time-dependent rock masses.
teractions. The CCM is widely used in tunnel design, but it meets some In the present research, we develop an alternative approach to obtain
limitations for tunnels with several support structures (Song and solutions for tunnels excavated in time-dependent rock masses, consid-
Rodriguez-Dono, 2021). Instead, in this study, a theoretical approach ering: (1) the viscoelastic behaviour models for the time-dependent
will be used. Theoretical approaches provide an efficient and quick way properties of rock masses, (2) the sequential installation of primary
to gain insights into the nature of the problem (Carranza-Torres and (rockbolts) and secondary (elastic liner) support systems, and (3) the
Fairhurst, 1999), and might be effectively utilised for preliminary design anisotropic initial stress state. First, we use a numerical-based approach
analyses (Bobet and Einstein, 2011; Zhou and Yang, 2021). to study the reinforcement mechanism of rockbolts, where the Bolted
Moreover, existing analytical solutions for supported tunnels mainly Rock Mass (BRM) is considered equivalent to the homogeneous material
focus on elastic solutions (Carranza-Torres et al., 2013; Li and Wang, but with a higher stiffness. Meanwhile, we develop a Machine-Learning-
2008; Lu et al., 2011; Wang and Li, 2009) or viscoelastic solutions with Based solver (MLB-solver) to evaluate the reinforcement of rockbolts
single support (Fahimifar et al., 2010; Lo and Yuen, 1981; Nomikos using the XGBoost algorithm and a large amount of data from numerical
et al., 2011; Pan and Dong, 1991; Sulem et al., 1987). However, in en- experiments. After that, we develop analytical solutions for tunnels
gineering applications, tunnels usually use a primary support system to excavated in time-dependent rock masses, considering a combined
seal the rock mass, and a secondary support system to ensure long-term support system (rockbolts and elastic liner). Finally, we perform sensi-
stability (Leca and Clough, 1992; Oreste, 2003; Song et al., 2018a, tivity analyses of displacements and stresses to the length and installa-
2018b; Wang et al., 2018). tion time of rockbolts, and we present an example to illustrate the
Some researchers (Wang et al., 2014, 2013) presented analytical application of our proposed method in the design of supported tunnels.
solutions for circular tunnels with a combined support system under
isotropic initial stresses, considering rock rheology, excavation process, 2. Theoretical background
and sequential installation time of liners. Then, some authors of the
present article (Song et al., 2018a, 2018b) presented analytical solutions 2.1. Basic assumptions
for tunnels with two different elastic liners under anisotropic initial
stresses, considering sequential excavation and support construction This study considers tunnels with two different support structures:
phases. Nonetheless, no support structures in the form of rockbolts were rockbolts and elastic liner. The analyses are based on the following
considered in these studies. assumptions:
However, rockbolts are an essential part of a large family of support
methods (Bobet and Einstein, 2011; Li, 2017; Zhang et al., 2014), (1) The host rock is homogeneous, exhibiting time-dependent
significantly improving the strength and stability of surrounding rock behaviour. Four different viscoelastic models are adopted to
mass (Zou et al., 2018), by generating confinement in the otherwise represent the time-dependency of rock masses: the Maxwell
unconfined surroundings of the excavation. The use of rockbolts as a model, the Kelvin model, the generalized Kelvin model and the
type of support initially emerged in the 1913s (Bobet and Einstein, 2011; Burgers model (Fig. 2).
Kovári, 2003; Maghous et al., 2012) and have been widely employed as (2) The initial stress field around the tunnel is p0 in the vertical di-
an effective and efficient reinforcement method (Wang et al., 2018; Wu rection and K0 p0 in the horizontal direction, where K0 represents
et al., 2020). Wang et al. (Wang et al., 2018) proposed a hybrid DEM-

Fig. 1. (a) Large deformation characteristics and the local collapse hazard, after Meng et al., (2013), and (b) Cross-section showing the support system between
Chainage 1400 m and 1750 m, in the Saint Martin La Porte access adit (Lyon-Turin Base Tunnel). After Barla et al. (2010).

2
F. Song et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 153 (2024) 106019

installation of rockbolts and grouting, it becomes feasible to

remedy these types of cracks or joints. Consequently, this
research assumes that after the installation of rockbolts, the
reinforced rock mass behaves as a time-independent material
with higher stiffness than the original rock mass.

Using the homogenisation approach, rockbolts and bolted rock form

a new composite that exhibits a stronger anti-deforming capacity
(Bernaud et al., 2009; Wu et al., 2024; Zhao et al., 2021). In this study,
the BRM (R1 < ρ < R2, t ≥ t1) can be treated as an isotropic, homoge-
neous (Fig. 3), and linear elastic material with elastic (or shear) modulus
ES (or GS) and Poisson’s ratio νS. The displacement equivalence principle
is applied to evaluate the enhancement of the original host rocks by
rockbolts, which means the displacements are consistent between tun-
nels reinforced with the BRM and those with only rockbolts. In this
study, we have developed a numerical-based approach and the corre-
sponding machine-learning-based solver to analyse the reinforcement
effect of rockbolts, which are described in detail in Section 3. Moreover,
the secondary support system (concrete liner) is assumed to be an elastic
material with shear modulus GL and Poisson’s ratio νL. In the following
derivations, the sign conventions are defined as positive for tension and
negative for compression, and the polar coordinates (ρ, θ) are employed.
Fig. 2. Conceptual one-dimensional model of four different visco-
elastic models. 2.2. Three-dimensional tunnel face effect

The analysis of the distribution of stresses and displacements near

the lateral pressure ratio, i.e., the ratio between horizontal and
the tunnel wall is a three-dimensional (3D) problem, when the tunnel-
vertical stresses (Fig. 3).
ling advance is considered (Fig. 4a). Fig. 4b shows that the support
(3) The primary support system consists of rockbolts, while the sec-
forces from the tunnel face (called the fictitious support forces), σf ,
ondary support system is a concrete liner (Fig. 3). Note that,
compared with the original rock mass, the stiffness of the Bolted decrease progressively from the initial stress σ 0 to zero as the distance to
Rock Mass (BRM) is increased due to the co-deformation of rock the tunnel face (xd) increases, until the tunnel face is at a distance where
mass and rockbolts (Bernaud et al., 2009; Chen et al., 2004; it has no influence on the considered section (Chu et al., 2019; Song and
Sakurai, 2010). Rodriguez-Dono, 2021). The 3D problem can be simplified as an
(4) The tunnel has a circular cross-section with radius R1 and is equivalent plane-strain one by applying the fictitious support forces σf ,
excavated instantaneously at time t = 0. The primary (rockbolts) to account for the tunnel face effect.
and secondary (concrete liner) support systems are sequentially The fictitious support forces are given by Eq. (1), where fR represents
installed at time t = t1 and t = t2 (t2 ≥ t1), respectively. The inner a relaxation factor calculated by tunnel convergence, which can be ob-
and outer radii of the BRM are R1 and R2, respectively, while the tained from field measurements or numerical simulations. Typically, the
inner and outer radii of the concrete liner are R3 and R2, relaxation factor fR can be written as a function of the distance to the
respectively (Fig. 3). tunnel face xd, as shown in Eq. (2).
(5) The time-dependency of rock masses can primarily be attributed [ ]
Normal : σ fρ = σ0ρ 1.0 − fR
to the propagation of cracks or joints (Hunsche and Hampel, [ ] , with 0 ≤ fR ≤ 1.0 (1)
1999; David and Helmstetter, 2006). Nonetheless, through the Tangential : σfρθ = σ0ρθ 1.0 − fR

Fig. 3. Conceptual models of tunnels supported with rockbolts and elastic liner.

3
F. Song et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 153 (2024) 106019

Fig. 4. Conceptual models of (a) the 3D effect of the tunnel face, (b) the relaxation factor fR as a function of the distance to the tunnel face (xd), and (c) the dif-
ferential element of the rock-rockbolts composite material and the reinforced rocks, in the longitudinal direction. After Chu et al. (Chu et al., 2019) and Song and
Rodriguez-Dono (Song and Rodriguez-Dono, 2021).

xd RB , is given by Eq. (4).

of rockbolts in the plane-strain model, E2D
f R = 1 − αe (2)
− β
R1
d
The two coefficients 0 ≤ α ≤ 1, β ≥ 0 characterise the curve of E2D
RB = ER + (ERB − ER ) (4)
Lz
relaxation factor fR versus xd. When α = 0 or xd → ∞, then fR=1, indi-
cating no 3D effect. When the excavation rate v is constant, then xd = vt Again, analysing the anchor’s supported tunnels in the form of the
vt plane strain model might be simple, however, it makes sense when
and, therefore, Eq. (2) becomes fR = 1 − αe R1 .
− β
involving the design of tunnels using the theoretical approach.
Generally, analysis of the anchor’s supported tunnel should be per-
formed in three-dimensional problems. Instead, in this study, following
the previous works (Bernaud et al., 2009; Bobet, 2009; Carranza-Torres 2.3. Constitutive behaviour of viscoelastic geomaterials
and Zhao, 2009; Fahimifar and Ranjbarnia, 2009; Jiang et al., 2023;
Maghous et al., 2012; Wu et al., 2024, 2020; Zhao et al., 2021; Zou et al., The constitutive model of viscoelastic geomaterials in general is
2018), the mechanical model is simplified into two-dimensional (2D) given by Eq. (5), where X is the position vector; GR (t) and KR (t) are the
plane strain problems, and the stiffness of rockbolts is reduced by the shear and bulk relaxation moduli that depend on time.
longitudinal space between rockbolts. Fig. 4c shows a differential {
sij (X, t) = 2GR (t) ∗ deij (X, t)
element of the rock-rock bolt composite and its equivalent material in (5)
σkk (X, t) = 3KR (t) ∗ dεkk (X, t)
the longitudinal direction. The equivalent radial stresses of the rein-
forced rock mass in the plane-strain model (σ2D) are approximated by In rock masses, deviatoric stress states cause deviatoric strain rates
Eq. (3), where σ R and σ RB are the radial stresses carried by the original that may lead to significant time-dependent deformations, while the
rock mass and the rockbolts, respectively, and d is the diameter of the volumetric strain is less significant (Song et al., 2020, 2021). Therefore,
rock bolt. the current research neglects the volumetric creep response. Accord-
ingly, KR (t) in Eq. (5) is treated as a constant elastic bulk modulus Ke.
σ 2D Lz = σR (Lz − d) + σ RB d (3)
Table 1 shows the expressions of GR (t) for the four viscoelastic models
After that, using the compatibility conditions of radial strains be- mentioned above, where the viscoelastic parameters, such as shear
tween the rockbolts (εRB) and the original rocks (εR), the elastic modulus modulus Gi and viscosity coefficient ηi (i = M or K) for (Song et al., 2020)

4
F. Song et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 153 (2024) 106019

Table 1
l
∑ l
∑
Shear relaxation modulus for the four different viscoelastic models.
uvx (z, t) + iuvy (z, t) = u(k)v
x (z, t) + i u(k)v
y (z, t)
Viscoelastic Maxwell Kelvin Generalized Kelvin Burgers model k=1 k=1
model model model model
l ∫ Tk
1∑
Shear G
− M t GK + G2M GM +GK
t
⎡ = I(t − τ)φ(k) (z, τ)dτ (10)
GM e ηM e 2 k=1 tbk
− ηK + t
relaxation ηK δ(t) GM + GK GM ⎣a1 e
− b
1 +
modulus GR GM GK [ ]
l ∫ Tk
(t) GM + GK ⎤ 1∑ ∂φ(k) (z, τ)
t − H(t − τ) z + ψ (k) (z, τ) dτ
a2 e b2 ⎦
−
2 k=1 tbk ∂z
b2 GM (ηM + ηK ) − b1 GK ηM b2 GK ηM − b1 GM (ηM + ηK ) [ ] [
a1 =
(b2 − b1 )A2
, a2 =
(b2 − b1 )A2
, b1 = 1 1 3 + 7G ̂ R (s) ]
H(t) = L− 1
, I(t) = L− 1
(11)
2A3 2A3 sG
̂ R (s) ̂ R (s) 3 K
sG ̂ R (s) + K̂ R (s)
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅, b2 = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅, A1 = GM GK ,A2 = GM ηM + GM ηK +
A2 + A22 − 4A1 A3 A2 − A22 − 4A1 A3
〈 ]}〉
GK ηM , A3 = ηM ηK σvx (z, t) ∑ l { [
∂φ(k) (z, t)
] [ 2 (k)
∂ φ (z, t) ∂ψ (k) (z, t)
= Re 2 ∓ z + ,
σ vy (z, t) k=1
∂z ∂z2 ∂z
each specific model can be determined experimentally. sij and eij are the 〈 ]〉
l [ 2 (k)
deviatoric components of the stress tensor (σij) and strain tensor (εij),
∑ ∂ φ (z, t) ∂ψ (k) (z, t)
v
σ xy (z, t) = Im z + .
respectively, which are given by Eq. (6). Moreover, the asterisk (*) in Eq. k=1
∂z2 ∂z
(5) denotes a convolution integral, which is defined by Eq. (7). (12)
⎧
1
⎨ sij = σ ij − δij σ kk
⎪
⎪
3 3. Analyses of the bolted rock masses
(6)
⎩ eij = εij − 1δij εkk
⎪
⎪
This study assumes that the bolted rock mass (BRM) is an isotropic
3
homogeneous elastic medium with considerably higher stiffness using
∫ t
df2 (τ) the homogenisation approach. The equivalent displacement principle,
f1 (t) ∗ df2 (t) = f1 (t)f2 (0) + f1 (t − τ) dτ (7) which is introduced in this section, allows the analysis of the rein-
0 dτ
forcement effect of rockbolts fE (fE ≥ 0), which is defined by Eq. (13),
The displacements and stresses for viscoelastic problems under where ER(t1) (or GR(t1)) is the elastic (or shear) modulus of the original
plane-strain stress boundary conditions are given by Eqs. (8) and (9), rock mass (see Eq. (5)) at time t1. The expressions of GR(t) for four
respectively, where φ(k) and ψ (k) are potentials; u(k)v
x (or σ x ) and uy (or
(k)v (k)v
common viscoelastic models (Maxwell, Kelvin, Generalized Kelvin and
y ) are the x- and y- displacements (or stresses) for the k–th load in the
σ (k)v Burgers) can be referred to in Table 1.
viscoelastic problem in the Cartesian coordinates; σ(k)v
xy is shear stress; z GS − GR (t1 ) ES − ER (t1 )
fE = = (13)
√̅̅̅̅̅̅̅ GR (t1 ) ER (t1 )
∧
= x + iy with i = − 1 is the position complex variable; L[⋅](or [⋅]) and
L - 1 [⋅] are the Laplace transform and its inverse, respectively; g is the The Poisson’s ratio νS is assumed to be the same value as the original
conjugate of the complex function g; Re[⋅] and Im[⋅] denote the real and rock mass. Section 3.1 introduces the numerical approach for assessing
imaginary components of a generic complex variable [⋅]. the reinforcement impact of rockbolts on the original host rocks. Sub-
〈 [ ̂ sequently, Section 3.2 outlines the development of a Machine-Learning-
1 3 K R (s) + 7 G
̂ R (s)
u(k)v
x (z, t) + iuy (z, t) = L
(k)v − 1
L φ(k) (z, s) Based solver (MLB-solver) that can predict the reinforcement effects
2s G
̂ R (s) 3K̂ R (s) + K
̂ R (s)
easily. Note that even the current approach has been verified that it is
]〉
∂φ(k) (z, s) available to viscoelastic cases within an acceptable range of errors,
− z − ψ (k) (z, s) (8)
∂z however, the advanced MLB-solver for viscoelastic cases can be further
developed in the future.
〈 [ (k) ] [ 2 (k) ]〉
x (z, t)
σ (k)v ∂φ (z, t) ∂ φ (z, t) ∂ψ (k) (z, t)
= Re 2 ∓ z + ,
∂z ∂z 2 ∂z 3.1. General numerical approach to obtain the reinforced factor fE
y (z, t)
(k)v
σ
(9)
〈 2 (k) 〉
∂ φ (z, t) ∂ψ (k) (z, t) In this section, a numerical-based approach is proposed to determine
xy (z, t) = Im z
σ (k)v + ,
∂z2 ∂z the reinforced factor fE, using the equivalent displacements principle.
If l different forces are applied on the structure at different times According to the algorithm presented in Fig. 5, two numerical models
before time t, i.e., the k–th force (k = 1, 2, …, l) is applied on the are required:
structure at time tbk and removed at tmk, then the total displacements can
be obtained by superposing all the quantities from the forces exerted (a) A numerical model with rockbolts, as shown in Fig. 6a. In this
before time t, and using the convolution integral property of Laplace model, the initial stresses are first applied to the whole model.
transform, as shown in Eq. (10), where Tk = min{tmk ,t}; uvx (or uvy ) is the After that, the rockbolts elements are activated to simulate their
installation.
normal x- (or y-) displacement for the viscoelastic problems; H(t) and I(t)
(b) An equivalent numerical model with Bolted Rock Mass
can be expressed as in Eq. (11) and their specific expressions for the four
(BRM), as shown in Fig. 6b. In this equivalent model, the initial
viscoelastic models are presented in Table 2. The total stresses are given
stresses are applied to the whole model, and subsequently, the
by Eq. (12), where σvx and σ vy are the normal x- and y- stresses and σvxy
tunnel cross-sections are excavated. Then, a higher value of
represents shear stresses. elastic modulus is assigned to the reinforced rock mass, to
represent the BRM.

Then, following the iteration algorithm presented in Fig. 5, the

equivalent elastic modulus of the BRM can be obtained. Note that, the
equivalent displacements method is used in the process of determining

5
F. Song et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 153 (2024) 106019

Table 2
Specific expressions of the two functions defined in Eq. (11) for the four viscoelastic models.
H(t) I(t)
( )
Maxwell model 1 1 6 1 1
δ(t) + + δ(t) + +
GM ηM 3Ke + GM GM ηM
[ ]
6G2M 3Ke GM
Exp − t
ηM (3Ke + GM )2 ηM (3Ke + GM )
( ) ( ) ( )
Kelvin model 1 GK 6 3Ke + GK 1 GK
Exp − t Exp − t + Exp − t
ηK K η ηK ηK η ηK
( ) ( ) K
Generalized Kelvin model 1 1 GK 6 1 6G2M
δ(t) + Exp − t + δ(t) + ⋅
GM ηK ηK 3Ke + GM GM ηK (3Ke + GM )2
( ) ( )
3Ke GK + GM (3Ke + GK ) 1 GK
Exp − t + Exp − t
ηK (3Ke + GM ) ηK ηK
( ) [ ] ( )
Burgers model 1 GK 1 1 1 6 1 1 GK
Exp − t + δ(t) + + + δ(t) + Exp − t
ηK ηK GM ηM ηM (3Ke + GM ) GM ηK ηK
⎡ ( ) ⎤
N2
⎢ N1 − 2 t ⎥
3G2M ⎢ (η + η − M1 )e ⎥
⎢ M K ⎥
+ 2⎢ ( ) ⎥
ηM ηK (3Ke + GM ) ⎢ N
⎥
⎣ − N1 + 22 t ⎦
+(ηM + ηK + M1 )e
[ ]
M1 = GK ηM (3Ke + GM )(ηM − ηK ) + 3Ke GM (ηM + ηK )2 /M2 ,
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
M2 = [GM GK + 3(GM + GK )Ke ]2 η2M − 6GM Ke ηM ηK (GM GK − 3GM Ke + 3GK Ke ) + 9G2M K2e η2K ,
M2 3GK ηM Ke + GM GK ηM + 3Ke GM (ηM + ηK )
N1 = , N2 =
2(3Ke + GM )ηM ηK (3Ke + GM )ηM ηK

the equivalent elastic modulus, i.e., both models have the same dis- complex, and may not be easy to use in tunnelling design. Hence, for
placements at the tunnel periphery and the same thickness of the rein- better use in geotechnical applications, a solver is developed to estimate
forced region as the rockbolt length. the values of the reinforced factor fE based on the machine-learning
algorithm and a large number of data from numerical experiments.
3.2. Machine-learning algorithm to evaluate the reinforced factor fE More detailed information on numerical experiments can be found in
Appendix A.
Following the proposed numerical-based approach in Section 3.1,
the reinforced factor fE can be obtained by feeding the parameters of the 3.2.1. Fitting chosen variables and original data
geometrical and mechanical properties of rock mass and rockbolts. The BRM equivalence stiffness mainly depends on (1) the original
However, it should be noted that the numerical-based approach is very rock mass elastic (or shear) modulus ER (or GR), (2) the stiffness (ERB),

Fig. 5. Flow chart for determining the equivalent elastic parameters of the reinforced rock mass.

6
F. Song et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 153 (2024) 106019

Fig. 6. Conceptual model of (a) Tunnel with rockbolts, and (b) Tunnel with equivalent reinforced rock mass.

length (LRB) and number (nRB) of rockbolts, and (3) the longitudinal demonstrates a negative exponential correlation between the two
distance between two adjacent supported surfaces (Lz). For the gener- parameters.
alized analysis, the original rock mass elastic modulus (ER) is normalised Machine-learning algorithms usually require large-scale data for
by the rockbolts stiffness for plane-strain models (E2D
RB ), and the rockbolts
training. However, the dataset of 122 sets of data used in this study can
length (LRB) is normalised by the tunnel radius (R1). Therefore, fE de- be considered quite small. Therefore, it is important to select suitable
pends on ER / E2D machine-learning algorithms for application. Hence, we experimented
RB , LRB / R1 and nRB. In this study, a total of 125 sets of
numerical experiments were conducted, resulting in the acquisition of with various machine-learning algorithms, including Xtreme Gradient
125 corresponding sets of experimental data (referred to as training Boosting (XGBoost) (Chen et al., 2015), k-nearest neighbours algorithm
data) utilized for evaluating the function fE. In these experimental tests, (KNN) (Larose and Larose, 2014), Naive Bayes (Frank et al., 2000), and
a wide range of rock mass and rockbolts geometrical and mechanical Random Forests (Cutler et al., 2012). By training these different
characteristics have been adopted: the rockbolts number (nRB) varies machine-learning algorithms with the training data, several solvers have
from 4 to 20, the normalised rockbolts length (LRB / R1) varies from 0.3 been developed to determine the values of fE.
to 1.1, and the normalised rock mass elastic modulus (ER / ERB) varies
from 0.04 to 0.65, which are common engineering values. Table A.1 lists 3.2.2. Steps taken to obtain the machine-learning based solver
other input parameters of the rock mass and rockbolts in numerical First, we tested the predictive capability of the developed solver on
simulations, which are the same in all the cases. the training dataset (numerical value) as shown in Fig. 8. To quantify the
Before training the machine-learning algorithm with the dataset, it is predictive performance of different algorithms, several commonly used
important to remove data points with significant outliers (Osborne and regression evaluation indexes were employed, as defined in Eq. (14)-
Overbay, 2004; Pollet and van der Meij, 2017; Stevens, 1984). In this (17). Fig. 8 and Table 3 show that XGBoost, Naive Bayes, and Random
study, the Z-score method for outlier detection on the dataset (Ghosh Forests performed well in reproducing the original data (the training
and Vogt, 2012) has been employed. Three groups of outliers with fE data), with small absolute errors and achieving R2 values close to 1.
values of 0.0295, 0.031, and 0.03 were identified and excluded. No However, KNN exhibited slightly poorer performance in comparison.
duplicate or null values were found in the dataset. Ultimately, a total of m
1 ∑
122 sets of data were used for model training. MAE = y i )|
|(yi − ̂ (14)
m i=1
We have conducted a statistical analysis of the data to understand
their characteristics and the correlations between parameters. Fig. 7a m
shows a correlation heatmap among the parameters, revealing a sig- 1 ∑
MSE = y i )2
(yi − ̂ (15)
m i=1
nificant negative correlation between fE and ER /E2D RB . Furthermore,
Fig. 7b shows a scatter plot of fE and ER / E2D
RB , which clearly

Fig. 7. (a) Correlation heatmap of the parameters, and (b) Scatter plot of fE and ER / ERB
2D
.

7
F. Song et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 153 (2024) 106019

Fig. 8. Predictions by different machine-learning algorithms: (a) XGBoost method, (b) KNN method, (c) Naive Bayes method, and (d) Random Forests method. For
the training data.

√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
m
Table 3 1 ∑
RMSE = (yi − ̂ y i )2 (16)
Quality evaluation of the predictions for different machine-learning algorithms. m i=1
For the training data.
∑
XGBoost KNN Naive Bayes Random Forests (̂y i − yi )2
R2 = 1 − ∑i 2
(17)
i (yi − yi )
4 3 4 4
MAE 4.585 × 10− 1.625 × 10− 1.295 × 10− 2.282 × 10−
7 6 8 8
MSE 3.586 × 10− 8.654 × 10− 4.509 × 10− 9.379 × 10−
RMSE 5.989 × 10− 4
2.941 × 10− 3
2.124 × 10− 4
3.063 × 10− 4 To further verify the developed solvers from machine-learning al-
R2 9.899 × 10− 1
7.556 × 10− 1
9.987 × 10− 1
9.973 × 10− 1 gorithms, we randomly adopted 6 groups of verification data (named
test data) different from the training data. We compared the results from
the four different developed solvers and the numerical simulations.

Fig. 9. Predictions by different machine-learning algorithms: (a) XGBoost method, (b) KNN method, (c) Naive Bayes method, and (d) Random Forests method. For
the tested data.

8
F. Song et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 153 (2024) 106019

Table 4 accuracy as the numerical simulations for estimating the reinforcement

Quality evaluation of the predictions for different machine-learning algorithms. effect of rockbolts on the original rock masses.
For the tested data.
XGBoost KNN Naive Bayes Random Forests 4. Analytical solutions for supported tunnels
4 3 3 3
MAE 9.057 × 10− 1.459 × 10− 1.433 × 10− 2.265 × 10−
MSE 1.599 × 10− 6
2.847 × 10− 6
3.505 × 10− 6
1.359 × 10− 5 In this work, the construction process of the supported tunnel is
RMSE 1.265 × 10− 3
1.687 × 10− 3
1.872 × 10− 3
3.687 × 10− 3
divided into three stages:
R2 6.951 × 10− 1
4.573 × 10− 1
3.316 × 10− 1
− 1.591
(1) The excavation stage (0 ≤ t < t1): from time t = 0 to t = t1, where t1
represents the installation time of the primary support system
Table 5 (rockbolts).
Comparison between the predictions and the numerical results. The predictions (2) The primary support stage (t1 ≤ t < t2): from time t = t1 to t = t2,
are obtained by using the XGBoost model. where t2 is the installation time of the secondary support system
ER/E2D
RB
LRB / R1 nRB Calculated fE Predicted fE (elastic liner).
− 1 − 1 1 − 3 3 (3) The secondary support stage (t ≥ t2): from t = t2 onwards, where the
1.000 × 10 3.000 × 10 1.200 × 10 9.000 × 10 9.518 × 10−
1.154 × 10− 1
5.000 × 10− 1
2.000 × 101 7.333 × 10− 3
1.013 × 10− 2 tunnel is supported by both the primary (rockbolts) and second-
3.333 × 10− 1
1.100 1.600 × 101 2.800 × 10− 3
3.842 × 10− 3 ary (elastic liner) support systems.
1 1 3 3
2.800 × 10− 5.000 × 10− 8.000 4.429 × 10− 4.176 × 10−
1 1
2.000 × 10− 7.000 × 10− 1.200 × 101 5.250 × 10− 3
5.700 × 10− 3
We use the homogenisation approach (refer to sections 2 and 3) to
1 1 3 3
4.500 × 10− 9.000 × 10− 4.000 2.722 × 10− 3.095 × 10−
treat the primary support system (or BRM), which is the composite
material of rockbolts and the original rock mass, as a reinforced ho-
Fig. 9 and Table 4 show that the XGBoost model exhibits the best pre- mogenous elastic material. Both displacement and stress compatibility
dictive performance. Table 5 presents the comparison results, which are conditions must be satisfied on the interface between the original rock
quite close, with only minor fluctuations. Therefore, it can be concluded mass and the primary support system (original-reinforced interface: ρ =
that the proposed XGBoost-based prediction solver has almost the same R2), and on the interface between the primary and secondary support
systems (reinforced-liner interface: ρ = R1). Moreover, the boundary

Fig. 10. Boundary conditions at (a) the excavation stage, (b) the primary support stage, and (c) the secondary support stage.

9
F. Song et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 153 (2024) 106019

Fig. 11. (a) Constraint condition, structure size and boundary stresses in numerical simulations using ANSYS, and (b) Mesh size in the vicinity of the tunnel. A, B, C,
D, E and F are the locations of study points. Far boundaries located at distances x = 100 m and y = 100 m with 19,330 quadrilateral elements.

conditions at the inner boundary of the secondary support system (ρ = Eqs. (20) and (21) show the displacements and stresses in the pri-
R3) must also be satisfied. Sections 4.1 and 4.2 present the expressions of mary support system (R1 ≤ ρ < R2), respectively, where u(m)
Sj (or σ Sk )
(m)

displacements and stresses in the whole construction process, while represents displacements (or stresses) of the primary support system
Section 4.3 shows the boundary and compatibility conditions that must
subjected to the support forces p(m) (m = 1, 2).
be satisfied. S
⎧
⎪
⎪ uex
Rj (z, t), 0 ≤ t < t1
4.1. Displacements and stresses expressions in the whole construction ⎪
⎨
uSj (z, t) = uRj (z, t1 ) + uSj (z, t), t1 ≤ t < t2
ex
(20)
(1)
process ⎪
⎪
⎪
⎩ uex Rj (z, t1 ) + uSj (z, t), t ≥ t2
(2)

In the excavation stage, only fictitious support forces are applied at

the tunnel boundary (ρ = R1), as shown in Fig. 10a. In the primary support ⎧
⃒
(1) ⃒
⎪
⎪ σ ex
Rk (z, t), 0 ≤ t < t1
stage, the interaction support forces pS ⃒ act on the original- ⎪
⎨
σ Sk (z, t) = ex
Rk (z, t1 ) + σSk (z, t), t1 ≤ t < t2 (21)
ρ=R2 (1)
σ
reinforced interface (ρ = R2), while the fictitious support forces σf are ⎪
⎪
Rk (z, t1 ) + σ Sk (z, t), t ≥ t2
⎪
⎩ ex (2)
σ
applied at the tunnel boundary (ρ = R1), as shown in Fig. 10b.
Furthermore, during the secondary support stage, the interface forces The displacements and stresses of the secondary support system (R3
⃒
(2) ⃒
pS ⃒ act on the outer boundary of the primary support system (ρ = ≤ ρ < R1) can be expressed as in Eqs. (22) and (23), respectively, where
ρ=R2
⃒ uL (or σ L ) represents the displacements (or stresses) of the secondary
(2) (2)
(2) ⃒
R2); the interaction forces pS ⃒ act on the elastic liner (along ρ = R1),
⃒ρ=R1 support system, subjected to the support forces p(2)
S .
(2) ⃒
but both interaction forces pS ⃒ and fictitious support forces σ f act on
uLr (z, t) + iuLθ (z, t) = uLr (z, t) + iuLθ (z, t), t ≥ t2 (22)
(2) (2)
ρ=R1
the first support system (along ρ = R1), as shown in Fig. 10c.
According to the superposition principle of viscoelastic theory, the σ L (z, t) = σ (2)
L (z, t), t ≥ t2 (23)
displacements and stresses of the unsupported rock mass (ρ ≥ R2) can be
expressed as in Eqs. (18) and (19), respectively, where uex ex
Rj and σ Rk are
4.2. Expressions of displacements and stresses by potentials
displacements and stresses of rock masses due to excavation considering
the effect of tunnelling advancement; uRj and σRk are displacements
(m) (m)
During the excavation stage (0 ≤ t < t1), the time-dependent exca-
and stresses of rock masses subjected to the interaction support forces vation-induced displacements are given by Eq. (24), and the total
pS (m = 1, 2). j = ρ and θ; k = ρ, θ and ρθ. stresses are expressed as in Eq. (25), where z = ρeiθ , A1 (t) =
(m)
∫t ∫t
⎧ H(t − τ)p0 fR R21 dτ, A2 (t) = 2
0 I(t − τ)p0 fR R1 dτ, and A3 (t) =
⎪
⎪ uex
Rj (z, t), 0 ≤ t < t1
∫0t 4
⎪
⎨ 0 H(t − τ )p0 fR R1 d τ . Note that, the utilized analytical solutions in the
uRj (z, t) = uRj (z, t) + u(1)
ex
Rj (z, t), t1 ≤ t < t2 (18) excavation stage, i.e., Eqs. (24) and (25), were directly obtained from the
⎪
⎪
⎪ ex
⎩ uRj (z, t) + uRj (z, t2 ) + uRj (z, t), t ≥ t2
(1) (2) previous work (Wang et al., 2015).
[
1 − iθ z 1
⎧ uex
Rρ (z, t) + iuRθ (z, t) = e
ex
A1 (t)(1 − λ) 2 − A1 (t)(1 + λ)
σex 4 z z
Rk (z, t), 0 ≤ t < t1
⎪
⎪
⎪
⎨ ]
1 1
σ Rk (z, t) = ex
Rk (z, t) + σ Rk (z, t), t1 ≤ t < t2 (19) + A2 (t)(1 − λ) − A3 (t)(1 − λ) 3 (24)
(1)
σ
⎪
⎪ ex
z z
Rk (z, t) Rk (z, t), t ≥ t2
⎪
⎩ σ +σ
(2)

10
F. Song et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 153 (2024) 106019

⎧ ⎧
σ ex [ ]
Rρ (z, t) (λ + 1)p0 (λ − 1)p0 2iθ
⎪
⎪ { [ ] [ ] }
Rρ (z, t)
⎪
⎪
⎪
⎪ = − +Re e ⎪
⎪ σ (m) R (z, t)
∂φ(m) ∂2 φ(m)
R (z, t) R (z, t)
∂ψ (m)
⎪
σ ex 2 2
⎪ = Re 2 ∓ z e2iθ
Rθ (z, t)
⎪ ⎪
⎪ +
⎪
⎪
⎪
⎪
⎪
⎨ σ (m) (z, t) ∂z ∂z2 ∂z
⎪ Rθ
⎪
⎪ 2 2 {[ ] }
⎪ 〈 − (1 − λ)Rd (t) ±[ − z (1 − λ)Rd (t) 〉 ⎪
⎪
⎪
⎪ ⎪
⎪
⎪ ∂2 φ(m)
R (z, t) R (z, t)
∂ψ (m)
⎪
⎪ z 2 z 3 ⎪
⎪ σ (m)
(z, t) = Im z + e 2iθ
⎨ +fR p0 Re
Rρθ
⎪
⎪ ⎪
⎩ ∂z2 ∂z
1 + λ R2d (t) 3(1 − λ) R4d (t) 2iθ
+ + ]e
⎪
⎪
⎪ 2 z2 2 z4 (30)
⎪
⎪
⎪ [ ]
The displacements and stresses in the primary support system sub-
⎪ σex (z, t) = Im (λ − 1)p0 e2iθ
⎪
⎪
⎪
⎪
⎪
⎪
⎪
Rρθ
2 jected to the support forces p(m) S (m = 1, 2) can be expressed as in Eqs.
⎪
⎪
⎪
⎪
⎪
〈[
(1 −
] 〉
λ)R2d (t) 1 + λ R2d (t) 3(1 − λ) R4d (t) 2iθ (31) and (32), respectively. Note that GS = GR (t1 ) + fE GR (t1 ), and fE can
⎪
⎩ +fR p0 Im z e be determined by using the numerical-based approach or the proposed
⎪
⎪ − −
z 3 2 z2 2 z 4
MLB-solver (refer to Section 3).
(25) ]
e− iθ [ (m) S (z, t)
∂φ(m)
where z = ρeiθ , A1 (t) =
∫t
p0 R21 0 fR H(t − τ)dτ,
A2 (t) = u(m)
Sρ (z, t) + iu(m)
Sθ (z, t) = κ S φ (z, t) − z − ψ (m)
(z, t) (31)
2GS S
∂z S
∫
2 t
∫t
p0 R1 0 fR I(t − τ)dτ, and A3 (t) = p0 R41 0 fR H(t − τ)dτ. Using the complex
⎧
variable method, the corresponding potentials subjected to the support { [ ] [ ] }
Sρ (z, t)
⎪
⎪
⎪ σ (m) S (z, t)
∂φ(m) ∂2 φ(m)
S (z, t) S (z, t)
∂ψ (m)
forces pS
(m)
(m = 1, 2), can be expanded as the Laurent series, as shown in
⎪
⎪
⎪ = Re 2 ∓ z + e2iθ
⎪
⎨ σ (m) (z, t) ∂z ∂z2 ∂z
Eqs. (26)-(28), where φR and ψ R are the potentials of the original rock Sθ
(m) (m)
⎪ {[ ] }
mass subjected to the support forces pS (m = 1, 2); φS and ψ S are the
(m) (m) (m) ⎪
⎪
⎪ ∂2 φ(m)
S (z, t) S (z, t)
∂ψ (m)
⎪
⎪ σSρθ (z, t) = Im z
(m)
+ e2iθ
potentials of the primary support system subjected to the support forces
⎪
⎩ ∂z2 ∂z
pS (m = 1, 2); φL and ψ L are the potentials of the secondary support
(m) (2) (2)
(32)
system subjected to the support forces pS . ak (t), bk (t), ck (t), dk (t),
(2) (m) (m) (m) (m)
Finally, the displacements and stresses of the secondary support
ek (t), fk (t), gk (t), hk (t), pk (t) and qk (t) denote time-dependent
(m) (m) (m) (m) (m) (m)
system subjected to the support forces p(2) S are given by Eqs. (33) and
undetermined coefficients. (34), respectively.
For the original rock mass (ρ ≥ R2, t ≥ t1): ]
⎧ e− iθ [ (2) L (z, t)
∂φ(2)
∑ u(2) t) iu t) t) z t) (33)
∞ (2) (2)
(z, + (z, = κ S φ (z, − − ψ (z,
⎪ φR (z, t) =
⎪
⎪
(m)
a(m)
k (t)z
− k Lρ Lθ
2GL L
∂z L
⎨ k=1
(26) ⎧
⎪
⎪ ∑
∞
{ [ ] [ ] }
R (z, t) = bk (t)z−
⎪ k
Lρ (z, t)
⎩ ψ (m) (m) ⎪
⎪
⎪ σ (2) L (z, t)
∂φ(2) ∂2 φ(2)
L (z, t) L (z, t)
∂ψ (2)
k=1
⎪
⎪
⎪ = Re 2 ∓ z + e 2iθ
⎪
⎨ σ (2) (z, t) ∂z ∂z2 ∂z
Lθ
For the primary support system (R1 ≤ ρ < R2, t ≥ t1): {[ ] }
⎪
⎪
⎪
⎪ ∂2 φ(2)
L (z, t) L (z, t)
∂ψ (2)
⎧ ∑
∞ ∑
∞ ⎪
⎪ σ Lρθ (z, t) = Im z
(2)
+ e 2iθ

⎪ φS (z, t) =
⎪
⎪
(m)
c(m)
k (t)z
− k
+ d(m)
k (t)z
k ⎪
⎩ ∂z2 ∂z
⎨ k=1 k=1
(27) (34)
⎪
⎪ ∑
∞ ∑
∞

S (z, t) = ek (t)z− k + fk (t)zk

⎪
⎩ ψ (m) (m) (m)

k=1 k=1
Note that the displacements and stresses can be addressed as long as
the coefficients of potential in Eqs. (26)-(28) have been determined. A
For the secondary support system (R3 ≤ ρ < R1, t ≥ t2): set of linear equations for these coefficients can be obtained by satisfying
⎧ ∑
∞ ∑
∞ the boundary and compatibility conditions presented in section 4.3.
⎪ φL (z, t) =
⎪
⎪
(m)
gk(m) (t)z− k + h(m)
k (t)z
k
Then, all these coefficients can be determined by solving these
⎨ k=1 k=1
(28) equations.
⎪
⎪ ∑
∞ ∑
∞
⎪
L (z, t) =
⎩ ψ (m) p(m)
k (t)z
− k
+ q(m)
k (t)z
k

k=1 k=1
4.3. Boundary and compatibility conditions
Regarding Eqs. (10) and (12), the displacements and stresses of the
original rock mass subjected to the support forces pS (m = 1, 2) can be
(m) During the primary support stage (t1 ≤ t < t2), Eqs. (35) and (36) show
expressed as in Eqs. (29) and (30), respectively. the displacements and stresses compatibility conditions on the interface
[
ρ = R2. And the boundary conditions along the tunnel wall (ρ = R1) can
{∫ t ∫t be expressed as in Eq. (37).
e− iθ ∂φ(m) (z, τ)
uRρ (z, t)+iuRθ (z, t) =
(m) (m)
I(t − τ)φR (z, τ)dτ −
(m)
H(t − τ) z R
2 tm tm ∂z { [ ] [ ] }⃒⃒
] } uex
Rρ (z,t)+iuRθ (z,t)+ uRρ (z,t)+iuRθ (z,t) −
ex (1) (1)
uex ex
Rρ (z,t1 )+iuRθ (z,t1 ) ⃒
z=R2 eiθ
{ }⃒
R (z, τ) dτ
+ ψ (m) ⃒
= uSρ (z,t)+iuSθ (z,t) ⃒
(1) (1)
iθz=R2 e

(29) (35)

11
F. Song et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 153 (2024) 106019

{ }⃒ { }⃒
⃒ ⃒ verification.
Rρ (z, t) − σ Sρ (z, t) ⃒
σ(1) = 0, Rρθ (z, t) − σ Sρθ (z, t) ⃒ =0
(1)
σ(1) (1)
z=R2 eiθ z=R2 eiθ The numerical model built in ANSYS is consistent with the hypoth-
(36) esis made in the analytical solutions, and both models are calculated
⃒ ⃒ under plane-strain conditions with small deformations. Only a quarter of
⃒ ⃒
Sρ (z, t)⃒
σ (1) = σfρ (R1 , t), Sρθ (z, t)⃒
σ(1) = σ fρθ (R1 , t) (37) the tunnel structure is analysed in the numerical model (see Fig. 11a)
z=R1 eiθ z=R1 eiθ
because of the double symmetry of the geometry and the boundary
Then, according to the extension of the correspondence principle, conditions on both the x and y axes. Moreover, the normal displace-
and the Laplace transform of the boundary conditions, a set of linear ments along the bottom (y = 0) and the left (x = 0) boundaries are
algebraic equations can be established for the undetermined coefficients restrained. Fig. 11b shows the mesh size in the vicinity of the tunnel. A
with respect to the coefficients in Eqs. (26)-(28) in the Laplace domain, mesh of 19,330 quadrilateral elements is adopted, with smaller elements
as shown in Appendix B. After that, substituting the expressions of H(t) near the excavation zone. Table 6 shows the input parameters of rock
and I(t) for the utilized viscoelastic models (see Table 2) into these masses and support systems, including the construction process, geom-
equations, the explicit coefficients are obtained. Then, by applying the etry, and material properties.
inverse Laplace transform, substituting the determined coefficients into In the numerical modelling, two different surfaces I and II were built
Eqs. (24)-(34), the displacements and stresses of the surrounding rocks in the area of the primary support system (6 m ≤ ρ ≤ 7 m) to represent
and the primary support system in the primary support stage are obtained. the installed progress of rockbolts. The viscoelastic and elastic models
After installing the secondary support system (t ≥ t2), the compati- are assigned to surfaces I and II, respectively, to characterise the prop-
bility conditions of displacements and stresses at the original-reinforced erties of the original rock mass and the BRM. Note that, surface II is
interface (ρ = R2) are given by Eqs. (38) and (39). Moreover, the dis- deactivated at the beginning of numerical calculations and, therefore,
placements and stresses compatibility conditions at the reinforced-liner before installing the primary support system, the model properties in the
interface (ρ = R1) are presented in Eqs. (40) and (41). Eq. (42) shows the area of 6 m ≤ ρ ≤ 7 m are the same as the original rock mass. At the
stress boundary conditions at the inner boundary of the secondary installation time of the primary support system, i.e. t = t1, surface I is
support system (ρ = R3). deactivated while surface II is active. After that, the properties of this
⎧ [ ] ⎫ ⃒⃒ area are characterised by elastic behaviour.
⎨ uex
⎪
Rρ (z, t) + iuRθ (z, t) + uRρ (z, t2 ) + iuRθ (z, t2 ) +
ex (1) (1) ⎪
⎬⃒ At the beginning of the numerical simulations, the model is run for a
⃒
[ ] [
⎪ u (z, t) + iu (z, t) − uex (z, t ) + iuex (z, t ) ⎭
(2) (2) ] ⎪⃒
⃒ sufficiently long time before excavation (100 days in this example), to
⎩ 1 1 ⃒
Rρ Rθ Rρ Rθ
z=R2 eiθ reach the equilibrium stress–strain conditions before starting the exca-
{ }⃒
⃒ vation, ensuring that we only consider the incremental deformations
= uSρ (z, t) + iuSθ (z, t) ⃒
(2) (2)
(38)
iθ z=R2 e induced by excavation. Subsequently, the tunnel cross-section is
{ }⃒ { }⃒ instantaneously excavated. That particular time is considered t = 0 day
⃒
Rρ (z, t) − σ Sρ (z, t) ⃒
σ(2) (2)
= 0, Rρθ (z, t) − σ Sρθ (z, t) ⃒
σ(2) (2) ⃒
=0 in this study. Five days later, the constitutive model in the area of the
z=R2 eiθ z=R2 eiθ
primary support system is changed from the viscoelastic model to the
(39) elastic model, to simulate the installation of the rockbolts. Then, after
[ ] [ ]⃒
⃒ five more days, i.e., t = 10 days in analytical solutions, the new elements
u(2)
Sρ (z, t) + iuSθ (z, t) −
(2)
u(1)
Sρ (z, t2 ) + iuSθ (z, t2 ) ⃒
(1)
in the region of the secondary support system (5.9 m ≤ ρ ≤ 6 m) are
z=R1 eiθ
⃒
⃒ activated to simulate the installation of the elastic liner. Fig. 12 shows
= uLρ (z, t) + iuLθ (z, t)⃒
(2) (2)
iθ
(40) the comparisons of the incremental displacements and total stresses
z=R1 e
between analytical and numerical results. A good agreement is
{ }⃒ { }⃒
⃒ ⃒ observed, serving as a verification for the developed analytical solutions.
Sρ (z, t) − σ Lρ (z, t) ⃒
σ(2) = 0, Sρθ (z, t) − σ Lρθ (z, t) ⃒ =0
(2)
σ(2) (2)
z=R1 eiθ z=R1 eiθ

(41) 5.2. Validation of the analytical solutions

⃒ ⃒
⃒ ⃒
Lρ (z, t)⃒
σ (2) = 0, σ(2)
Lρθ (z, t)⃒ =0 (42) To validate the applicability of the developed analytical solutions in
z=R3 eiθ z=R3 eiθ
predicting tunnelling convergence, a comparison is performed between
Similar to the solving procedure in the primary support stage, ac- field test data, numerical predictions and analytical solutions for the
cording to the extension of the correspondence principle, and the Lap- Saint Martin La Porte access adit (Chainage 1406 m). Fig. 13a shows the
lace transform of the boundary conditions, a set of linear algebraic geological profiles, and Fig. 13b shows the tunnel cross-section with the
equations can be established for the undetermined coefficients with geometry and support system information.
respect to the coefficients in Eqs. (26)-(28) in the Laplace domain, as In this section, numerical and analytical results are compared with
shown in Appendix B. After substituting the expressions of H(t) and I(t) on-site tunnel convergence observations. Numerical simulations have
into these equations, the explicit coefficients are obtained. Then, by been performed utilizing the Finite Difference Method software FLAC
applying the inverse Laplace transform, substituting the determined (Barla et al., 2010). In their investigations (Barla et al., 2010), they used
coefficients into Eqs. (24)-(34), the displacements and stresses of the a circular cross-section with an equivalent radius of 6 m, and they
surrounding rocks and the support systems in the secondary support stage simulated the primary support system by applying a uniform pressure of
are obtained. 0.1 MPa at the tunnel boundary. In the numerical calculations, the
ground was initially elastic-perfectly-plastic and the time-dependency
5. Verification and validation of the proposed analytical was simulated by running the CVISC model after installing the second
solutions support system (after 23 days).
For the analytical solutions, the Burgers viscoelastic constitutive
5.1. Verification of the analytical solutions model is used, and Table 7 shows the input parameters for the con-
struction process and material properties. Tunnelling advancement rate
To verify the developed analytical solutions in Section 4, numerical is set as a constant value: 1.5 m/day in the analytical predictions. It is
simulations of tunnelling problems are carried out in the finite element important to note that some assumptions are inevitable when using
method software ANSYS. The generalized Kelvin model is used to numerical calculations or analytical solutions to analyse tunnelling
characterise the properties of host rocks. In this section, the obtained behaviour. However, these assumptions need to be carefully assessed
numerical results are compared with the analytical solutions for when using analytical solutions in geotechnical applications. Hence,

12
F. Song et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 153 (2024) 106019

Table 6
Construction information, geometrical and mechanical properties for rock mass and support systems in the numerical model (ANSYS).
Vertical initial Lateral pressure Radius of the tunnel wall, The radius of the outer boundary of the primary support The radius of
stress, ratio, R1 [m] system, R2 [m] the inner
p0 [ MPa] K0 boundary of the
secondary
support system,
R3 [m]

15 0.65 6 8 5.85

Parameters for the Generalized Kelvin model (the original rock mass) Parameters for
the primary
support system

GM [MPa] GK [MPa] ηK [MPa day] Bulk modulus of the GS νS

original rock mass, Ke [MPa]

2000 1000 10,000 ∞ (volume incompressible) 7003.5 0.5

Parameters for the secondary support Installation time of the primary support system, Installation time of the secondary support system, t2 [Day] Speed of
system t1 [Day] longitudinal
excavation, v
GL [MPa] νL
[m/day]

12,000 0.2 5.0 10.0 ∞

analytical solutions should be used as an approximation to tunnelling field test. The analytical results, on the other hand, capture the gradual
convergence. decrease in the rate of convergence, and reach a stable condition when
Fig. 13c shows that the field data of convergence along array 1–5 time is large enough, consistent with the on-site observations. Therefore,
agree well with the numerical results during the primary support stage (0 – we can conclude that the analytical solutions can provide a good
23 days). However, in the secondary support stage (after 23 days), the approximation to tunnelling deformation.
numerical predictions deviate from the observed deformations from

Fig. 12. Comparison between analytical predictions and numerical results for (a) The incremental displacements versus time at points A, B, and C, (b) The in-
cremental displacements versus time at points D, E, F, (c) The total stresses versus time at points A, B, and C, and (d) the total stresses versus time at points D, E, F.
The locations of these points are plotted in Fig. 11b.

13
F. Song et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 153 (2024) 106019

Fig. 13. (a) Geological conditions of the Saint Martin La Porte access adit, (b) tunnel cross-section showing the support system between chainage 1400 m and 1750
m, and (c) comparisons between field test data, numerical predictions and analytical results. Based on the work of Barla et al. (Barla et al., 2010).

6. Analyses and discussions ⃒ ⃒

qL = ⃒σLθ − σLr ⃒ (45)

In this section, we consider the viscosity of rock mass as described by If no additional information is provided, we use the following pa-
the generalized Kelvin model, which is commonly used for geomaterials rameters for the examples in this section: the normalized length of
with good mechanical properties or subjected to low-stress conditions. rockbolts is LRB / R1 = 0.6, the number of rockbolts is nRB=20, the
The Generalized Kelvin model is shown in Fig. 2, where the constitutive normalised elastic modulus of rockbolts is ER /E2D
RB = 0.3. The normalized
parameters are GM / GK=2.0, ηK / GK=10.0. We also assume that the rock shear modulus and the normalized thickness of the secondary support
mass is an incompressible material (i.e. Ke → ∞, νR=0.5). To simplify the system (elastic liner) are GL / GK=1000 and dL / R1 = 0.03, respectively,
analysis, we use dimensionless variables throughout this section. The which implies that R3 / R1 = 0.97. The normalized installation time of
shear moduli, geometry parameters and stresses are normalised by the the primary and secondary support systems are t1 / TK=0.5 and t2 /
reference values GK, R1 and p0, respectively. The generic time is nor- TK=1.0, respectively. The Poisson’s ratio of the primary support system
malised by the retardation time TK, which is defined as TK=ηK / GK for is νS=0.5, while the Poisson’s ratio of the secondary support system is
the Generalized Kelvin model. νL=0.2. The lateral pressure ratio is K0 = 0.65, and we assume a suffi-
Tunnel convergences are normalised by Δu eρ given in Eq. (43), which ciently large tunnelling advancement rate.
represents the excavation-induced elastic convergence for the unsup-
ported circular tunnel subjected to isotropic initial stress p0, where Gp = 6.1. Influence of the length of rockbolts and constitutive models selection
GM GK / (GM+GK) is the permanent stiffness of the generalized Kelvin
model. The critical stresses qi (i = S, L) of the primary (S) and secondary We examine the effect of the length of rockbolts (LRB /R1) on the
(L) support systems are defined in Eqs. (44) and (45), respectively. The mechanical responses of tunnels by considering four options: LRB /R1 =
failure criterion is that the critical stresses should not exceed the 0.4, 0.6, 0.8 and 1.0, with all the other parameters fixed. Using the
allowable stress of the geomaterials. machine-learning-based solver, we obtain the reinforcement factors fE as
p0 R1 0.375 % for the case of LRB /R1 = 0.4 and fE = 0.364 % for the other three
Δueρ = (43) cases with LRB /R1 = 0.6, 0.8 and 1.0.
2Gp
The host rocks exhibit time-dependent behaviour, which leads to
⃒ ⃒
qS = ⃒σ Sθ − σSr ⃒ (44) tunnel convergences that stabilize after a long time. Fig. 14 shows the
distributions of stable displacements and critical stresses at the tunnel

14
F. Song et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 153 (2024) 106019

Table 7
Input parameters in the validation example,
Initial stress Geometry Burgers model

p0 [MPa] K0 R1 [m] R2 [m] R3 [m] GM [MPa] GK [MPa] ηM [MPa⋅day] ηK [MPa⋅day]

26 1.0 6 14 5.8 566 498.1 10212.7 1554.9

Parameters of support systems

GS [MPa] GL [MPa] νS νL fE t1 [day] t2 [day]

(1 + fE) GR(t1) 8333.3 0.5 0.2 0.5 % 10 23

adapted from the previous works (Barla et al., 2011, 2010).

wall (ρ = R1, t → ∞), for the four lengths of rockbolts. Fig. 14a shows maximum difference in the critical stresses due to different rockbolts
that the radial displacements increase gradually with the angle θ; while lengths is nearly 6.5 % of the largest critical stress.
Fig. 14b shows that the circumferential displacements peak at θ = 45◦ Furthermore, Fig. 15 presents the evolutions of normalised radial
and then decrease to zero at θ = 90◦ . and circumferential displacements, as well as the evolution of the nor-
The BRM has a higher stiffness than the original rock masses, so it malised critical stresses at the study point ρ = R1, θ = 45◦ . The dis-
can be regarded as a support system that reduces the tunnelling de- placements and stresses increase with time, until they reach stable
formations and enhances the tunnel stability. Therefore, longer rock- values, which reflects the creep behaviour of the Generalized Kelvin
bolts result in smaller convergences. In these examples, the maximum model (Fig. 2). Moreover, the displacements and stresses are the same
difference in the radial and circumferential displacements due to for different LRB /R1 cases at the time of rockbolt installation, but they
different rockbolt lengths is about 9 % and 7 % of the largest values, diverge as time progresses until they stabilise.
respectively. Additionally, to further investigate the effect of constitutive models
Higher critical stresses indicate a more dangerous state for the tun- selection on the resulting tunnelling responses, four different cases are
nels. Fig. 14c shows that the normalised critical stresses decrease considered: (a) the Generalized Kelvin model, with GM / GK=2.0, ηK /
gradually with the angle θ, which agrees with Song et al.’s conclusion GK=10.0, (b) the Burgers model, with GM / GK=2.0, ηM / GK=10.0, ηK /
that the point with θ = 0◦ for the case of λ < 1 is the most vulnerable GK=10.0, (c) the Burgers model, with GM / GK=2.0, ηM / GK=20.0, ηK /
point to fail first (Song et al., 2018a, 2018b). In this example, the GK=10.0, and (d) the Burgers model, with GM / GK=2.0, ηM / GK=30.0,

Fig. 14. Distributions of: (a) Normalised radial displacements, (b) normalised circumferential displacements, and (c) normalised critical stress.

15
F. Song et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 153 (2024) 106019

Fig. 15. Evolutions of: (a) Normalised radial displacements, (b) normalised circumferential displacements, and (c) normalised critical stress.

ηK / GK=10.0. Note that, in all cases, the normalized length of rockbolts

is set as LRB / R1 = 0.6. As can be observed in Fig. 16, tunnelling de-
formations obtained by Burgers viscoelastic model are obviously bigger
than the ones obtained by the generalized Kelvin model due to the
contribution of the viscous dashpot of the Maxwell model part. As ex-
pected, for cases with bigger ηM / GK values, smaller deformations will
be developed by the viscous dashpot of the Maxwell model, in turn,
results are closer to the ones obtained by the Generalized Kelvin model.
It is the reason why the Burgers model can be simplified to the gener-
alized Kelvin model for cases of ηM → ∞ (Song et al., 2020).

6.2. Influence of installation time of rockbolts

The installation time of support systems is an important factor in

tunnel design. In this section, four different cases of t1 / TK are consid-
ered: t1 / TK=0.1, 0.4, 0.7 and 1.0. For simplicity, fE is set as 0.364 % in
this example. Fig. 17 shows the distributions of normalised final dis-
placements (radial and circumferential) and normalised final critical
stresses along the tunnel wall (ρ = R1, t → ∝) for the four cases of t1 / TK.
Earlier installation of rockbolts leads to smaller deformations at the
tunnel wall because less creep displacements occur. However, it also
leads to higher critical stresses in the BRM, which makes the primary
support system (i.e., the BRM) more vulnerable. Different installation Fig. 16. Evolutions of normalised radial displacements. Four different cases are
times of rockbolts affect the support forces at the original-reinforced considered: generalized Kelvin model and three different Burgers model.
interface (ρ = R2) and, therefore, the mechanical responses. In these
cases, the maximum difference of radial displacements, circumferential
displacements, and critical stresses between different cases of t1 / TK is
about 46 %, 14 % and 18 % of the corresponding largest values,

16
F. Song et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 153 (2024) 106019

Fig. 17. Distribution of: (a) Normalised radial displacements, (b) normalised circumferential displacements, and (c) normalised critical stress.

respectively. ⃒ ⃒
Fig. 18 shows the evolutions of displacements and critical stresses at FL = σ Lc − ⃒σ Sθ − σSr ⃒ (46)
the study point ρ = R1, θ = 45◦ . Due to the creep behaviour of host rocks, The first point to fail would be at the inner boundary of the support
the mechanical responses are time-dependent: the displacements and system with θ = 0◦ for K0 < 1, or with θ = 90◦ for K0 > 1. Then, the safety
critical stresses rise initially and then stabilise after a sufficiently long factor of the secondary support system (F Lsafety ) is defined as shown in Eq.
time. On the other hand, earlier installation of rockbolts helps the tunnel
(47).
achieve stable states sooner. ⎧ ⃒
⎪ σ cL ⃒
⎪
⎪ ⃒ ,K < 1
6.3. Reliability analyses and application in tunnelling design ⎨ |σθL − σ rL | ⃒r=R ,θ=0◦ 0
⎪
3
Fsafety
L
= ⃒ (47)
⎪
⎪ σ cL ⃒
The support forces applied on the outer boundary of the secondary
⎪
⎪
⎩ |σ − σ | ⃒
⃒ , K 0 > 1
θL rL r=R3 ,θ=90◦
support system are significantly affected by the installation time and the
stiffness of the support structures. An early installation of the secondary In the example, the secondary support system is an elastic liner.
support system leads to larger final support forces acting on the sec- Table 8 lists the parameters employed in the example. Fig. 19 shows how
ondary support system, which increases the risk of failure. Moreover, the installation time (t2 / TK) and the thickness (d2 / R1) of the secondary
larger thickness of supports improves the tunnel stability, but also in- support system affect the safety factor of the secondary support system
creases the cost of the support. In this section, we present an example of (F Lsafety ). The safety factor increases with larger values of d2 / R1 or t2 /
using the proposed solutions to perform the reliability analyses of the TK.
secondary support system, providing suggestions for the tunnelling Fig. 19a and 19b show the safety factor (F Lsafety ) as a function of t2 /
design in terms of the installation time (t2 / TK) and the thickness (d2 / L
TK and d2 / R1, respectively, where the dashed lines represent F safety =
R1) of the secondary support system.
1.0. In Fig. 19a, the minimum allowable installation time of the sec-
The secondary support system is usually designed to ensure the long-
ondary support system can be obtained when the safety factor is equal to
term stability of tunnels, and it is not expected to fail. Therefore, the
1.0. Note that the bigger the value of d2 / R1, the smaller will be the
stability (or the safety factor) of the secondary support system is a key
recommended values of t2 / TK. Moreover, in Fig. 19b, for a fixed
factor in tunnel design. To conduct the reliability analyses of the sup-
installation time of the secondary support system, increasing the support
ported tunnels, the critical function of the secondary support system (FL)
thickness improves the tunnel stability, and the economic limit for the
is defined as shown in Eq. (46). In this study, we assume that the support
support thickness can be determined by the intersection of the dashed
fails when FL≤0, where σ Lc represents the compressive strength of the
line (F Lsafety = 1.0) and the safety factor curves. Therefore, the proposed
secondary support system

17
F. Song et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 153 (2024) 106019

Fig. 18. Evolutions of: (a) Normalised radial displacements, (b) normalised circumferential displacements, and (c) normalised critical stress.

Table 8
Input parameters in the application example.
Initial stress Geometry Generalized Kelvin model

p0 [MPa] K0 R1 [m] R2 [m] GM [MPa] GK [MPa] ηK [MPa⋅day] Poisson’s ratio

15 0.65 6 9.6 2000 1000 10,000 0.5

Parameters of support systems

GS [MPa] GL [MPa] νS νL σcL [MPa] t1 [day] fE

(1 + fE) Grock(t1) 10,000 0.5 0.2 14.7 1 0.364 %

solutions can help to optimise both the installation time and the thick- Turin Base Tunnel. Furthermore, for practical application in tunnel
ness of the secondary support system. design, a general numerical approach and a machine-learning-based
solver have been developed to evaluate the reinforced ability of rock-
7. Conclusions bolts to the original rock mass.
In the parametric analyses, the influence of the length and installa-
This study has developed analytical solutions for displacements and tion time of rockbolts on displacements and stresses of tunnels have been
stresses in the whole process of tunnelling and construction of support discussed. The radial displacements, circumferential displacements and
systems, considering time-dependent properties of rock mass, aniso- critical stresses increase over time and then eventually stabilise.
tropic initial stress state, and sequential installation of the primary Increasing the length of rockbolts improves the tunnel stability. An
(rockbolts) and secondary (elastic liner) support systems. The three- earlier installation time of rockbolts reduces both radial and circum-
dimensional effect can be also included. The obtained analytical solu- ferential displacements. Finally, the proposed analytical solutions have
tions have been verified by numerical results, showing a good agree- been used to analyse supported tunnels and optimise the design of the
ment. In addition, analytical predictions have been validated by secondary support system. In conclusion, the proposed approach pro-
comparing tunnelling convergence and monitoring data from the Lyon- vides an alternative method for the preliminary design of tunnels.

18
F. Song et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 153 (2024) 106019

L
Fig. 19. The safety factor of the secondary support system, F safety , versus: (a) Normalized installation time of the secondary support system, and (b) normalized
thickness of the secondary support system.

Note that, even if the proposed method in this article can analyse Writing – review & editing. Ertai Wang: Software. Jingyun Gui:
different cases of supported tunnels constructed in time-dependent rock Writing – review & editing. Leandro Alejano: Supervision, Writing –
masses, it still has some limitations. For instance, in the theoretical review & editing.
analyses, the Bolted Rock Mass (BRM) is modelled as the homogeneous
material and geomaterials exhibit viscoelasticity without considering Declaration of competing interest
plasticity. In addition, address the role of rock bolts in soft rocks
exhibiting soil-like mechanical behaviour will be further studied. The authors declare that they have no known competing financial
interests or personal relationships that could have appeared to influence
8. Codes availability the work reported in this paper.

The original data, programs and the machine-learning-based solver Data availability
are available upon request. Please contact [email protected] for
access. Data will be made available on request.

CRediT authorship contribution statement Acknowledgements

Fei Song: Conceptualization, Software, Validation, Visualization, This study was financially supported by the National Natural Science
Writing – original draft. Huaning Wang: Conceptualization, Method- Foundation of China (Grant No. 12272274); the State Key Laboratory of
ology, Visualization, Writing – review & editing, Funding acquisition, Disaster Reduction in Civil Engineering (SLDRCE19-A-06). Their sup-
Supervision. Chengshun Shang: Investigation, Software, Writing – port is greatly appreciated. The authors thank the contribution of Hao-
original draft. Alfonso Rodriguez-Dono: Investigation, Supervision, guang Liang (Tongji University) on simulations by using FLAC.

Appendix A. Numerical-based approach to evaluate fE

In the numerical experiments, the Finite Difference Method software FLAC 2D is used to perform two-dimensional plane-strain numerical models.
Fig. A.1 presents the constraint condition, structure size and boundary conditions of the numerical models, and two different geometry models are
built for (a) the numerical model with rockbolts, and (b) the numerical model with Bolted Rock Mass (BRM). Due to the double symmetry of the x- and
y- axes of geometry and boundary conditions, only a quarter of the tunnel structure is analysed. The left (x = 0) and the bottom (y = 0) boundaries are
fixed in normal displacements, and normal stresses are applied to the remaining boundaries. The rock masses are simulated by the elastic constitutive
model, and the behaviour of rockbolts is simulated by the cable units. Table A.1 lists the input parameters employed in the simulations. The diameter
of rockbolts is 50 mm, which is a common value. Fig. A.1 shows the mesh size in the vicinity of the tunnel.

Table A.1
Input parameters of loads, geometrical and mechanical properties of rock mass and rockbolts in FLAC 2D numerical simulations.

Initial stress, Lateral pressure ratio, K0 Parameters of the original rock mass
p0 [MPa]
Poisson’s ratio, νR Compression strength [MPa] Tension strength [MPa] Weight [t⋅m− 3]

2.0 1.0 0.4 6 0.33 2.7

Parameters of the rockbolts

Diameter, DRB [m] Tension strength [MPa] Shear strength of grouting [Pa] Cohesion of grouting [MPa] Friction angle of grouting [◦ ] Length of grouting [m]

0.05 335 109 1.5 20 0.314

19
F. Song et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 153 (2024) 106019

Fig. A.1. Mesh size in the vicinity of the tunnel, for (a) The numerical model with rockbolts, and (b) The equivalent numerical model with the BRM.

Appendix B. Derivation of the coeffocients in the potentials

Non-zero coefficents in the Laplace domain can be calculated by solving the following euations to obtain solutions in the primary support stage:
̂e 1 (s)
2̂
d 1 (s) + =0 (B.1)
R21

̂c 1 (s)R−1 2 + 3 ̂
d 3 (s)R21 + ̂f 1 (s) = 0 (B.2)

− ̂c 1 (s)R−1 2 + ̂
d 3 (s)R21 + ̂e 3 (s)R−1 4 = 0 (B.3)

b 1 (s)
̂ ̂e 1 (s)
= 2̂
d 1 (s)R2 + (B.4)
R2 R1

a 1 (s)
̂ b 3 (s)
̂ ̂c 1 (s) ̂e 3 (s)
− 4 + 3 4 = − 4 2 + 3 4 − ̂f 1 (s) (B.5)
R22 R2 R2 R2

a 1 (s)
̂ b 3 (s)
̂ ̂c 1 (s) ̂e 3 (s)
− 2 + 3 4 = − 2 2 + 6̂ d 3 (s)R22 + 3 4 + ̂f 1 (s) (B.6)
R2 2
R2 R2 R2
{ }
1+λ ̂ A1 (t1 ) 1̂ 1 [ ]
− A1 (s) − − H(s)R2 b 1 (s) =
− 1̂
(κS − 1) ̂
d 1 (s)R2 − ̂e 1 (s)R−2 1 (B.7)
4R2 s 2 2GS
{ ̂3 (s) A3 (t1 )} 1 [ ]
1− λ ̂ A1 (t1 ) A a 1 (s) ̂
̂ b 3 (s) 1 [ ̂ ]
A1 (s) − − + 2 + H(s)
̂ − = κS d 3 (s)R32 + ̂c 1 (s)R−2 1 − ̂e 3 (s)R−2 3 (B.8)
4R2 s R22
R2 s 2 R2 R2 3
2GS
{ }
1− λ ̂ A2 (t1 ) 1 1 [ ]
A2 (s) − + ̂I(s)R−2 1 ̂
a 1 (s) = κS ̂c 1 (s)R−2 1 − 3 ̂
d 3 (s)R32 − ̂f 1 (s)R2 (B.9)
4R2 s 2 2GS
Furthermore, non-zero coefficents in the Laplace domain can be calculated by solving the following euations to obtain solutions in the secondary
support stage:
[ ∫ t2 ]
1 1 1 1 ̂ ̂(2)
− L[A1 (t) − A1 (t1 )](1 + λ) − L H(t − τ)b1 (τ)R−1 1 dτ − H(s) b 1 (s)R−2 1
(1)
4 R2 2 t1 2
(B.10)
1 [ (2) ]
(κS − 1) ̂
d 1 (s)R2 − ̂e 1 (s)R−2 1
(2)
=
2GS1

1 1 p0 1
L[A1 (t) − A1 (t1 )](1 − λ) − L[A3 (t) − A3 (t1 )](1 − λ) 3 +
4 R2 4 R2
[ ∫ t2 ] [ ∫ t2 ]
1 1 1̂
L H(t − τ)a(1)
1 (τ)R2 dτ −
− 1
⋅L 3 (τ)R2 dτ + H(s) a 1 (s)⋅
H(t − τ)b(1) − 3
̂ (2) (B.11)
2 t1 2 t1 2
1 ̂ ̂(2) 1 [ ̂ (2) ]
R−2 1 − H(s) b 3 (s)R−2 3 = κS d 3 (s)R32 + ̂c 1 (s)R−2 1 − ̂e 3 (s)R−2 3
(2) (2)
2 2GS

20
F. Song et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 153 (2024) 106019

[ ∫ t2 ]
p0 1 1
L[A2 (t) − A2 (t1 )](1 − λ) + L I(t2 − τ)a(1)
1 (τ )R − 1
d τ
4 R2 2 t1
2
(B.12)
1 1 [ (2) (2) ]
+ ̂I(s) ̂
a 1 (s)R−2 1 = κS ̂c 1 (s)R−2 1 + 2 ̂
d 3 (s)R32 − ̂f 1 (s)R2
(2) (2)
2 2GS
(2) (2)
b 1 (s)R−2 1 = 2 ̂
̂ d 1 (s)R2 + ̂e 1 (s)R−2 1 (B.13)
(2)

(2) (2)
− k̂
a 1 (s)R−2 1 + ̂
b 3 (s)R−2 (3) = ̂
d 3 (s)R32 − k̂c 1 (s)R−2 1 + ̂e 3 (s)R−2 (3) (B.14)
(2) (2) (2)

(2) (2)
a 1 (s)R−2 k = ̂c 1 (s)R−2 1 + (k + 2) ̂
d 3 (s)R32 + ̂f 1 (s)R2 (B.15)
(2) (2)
̂

1 [ (2) (1) ]
(κS − 1) ̂
d 1 (s)R1 − ̂e 1 (s)R−1 1 − (κS − 1) ̂ d 1 (s)R1 + ̂e 1 (s)R−1 1
(2) (1)
2GS
(B.16)
1 [ (2) ]
(κL − 1) ̂
h 1 (s)R1 − ̂p 1 (s)R−1 1
(2)
=
2GL

1 [ ̂ (2) (1) ]
κS d 3 (s)R31 + ̂c 1 (s)R−1 1 − ̂e 3 (s)R−1 3 − κS ̂
d 3 (s)R31 − ̂c 1 (s)R−1 1 + ̂e 3 (s)R−1 3
(2) (2) (1) (1)
2GS
(B.17)
1 [ ̂ (2) ]
κL h 3 (s)R31 + ̂ g 1 (s)R−1 1 − ̂
p 3 (s)R−1 3
(2) (2)
=
2GS2

1 (2) (2) (1)

[κS ̂c 1 (s)R−1 1 − 3 ̂
d 3 (s)R31 − ̂f 1 (s)R1 − κS1 ̂c 1 (s)R−1 1 + 3 ̂
d 3 (s)R31
(2) (1)
2GS
(B.18)
(1) 1 [ (2) (2) ]
+̂f 1 (s)R1 ] = g 1 (s)R−1 1 − 3 ̂
h 3 (s)R31 − ̂
q 1 (s)R1
(2)
κL ̂
2GL
(2) (2)
2̂
d 1 (s)R1 + ̂e 1 (s)R−1 1 = 2 ̂
h 1 (s)R1 + ̂
p 1 (s)R−1 1 (B.19)
(2) (2)

̂c (2) ̂ (2) (2)

1 (s)R1 + (k + 2) d k+2 (s)R1 + ̂f k (s)Rk1 =
− 1 k+2

(2)
(B.20)
g k (s)R−1 k + (k + 2) ̂
h k+2 (s)Rk+2 q k (s)Rk1
(2) (2)
̂ 1 +̂

p 1 (s)
(2)
(2) ̂
2̂
h 1 (s) + =0 (B.21)
R23

(2) (2)
h 2 (s)R3 − 2 ̂
2̂ h 2 (s)R3 = 0 (B.22)

(2)
g k (s)R−3 k− 1 + (k + 2) ̂
h k+2 (s)Rk+1 q k (s)Rk−3 1
=0 (B.23)
(2) (2)
̂ 3 +̂

− k̂ h k+2 (s)Rk+1
g k (s)R−3 k− 1 + ̂ 3 p k+2 (s)R−3 k−
+̂ 3
=0 (B.24)

References Cai, Y., Jiang, Y., Djamaluddin, I., Iura, T., Esaki, T., 2015. An analytical model
considering interaction behavior of grouted rock bolts for convergence-confinement
method in tunneling design. International Journal of Rock Mechanics and Mining
Alonso, E., Alejano, L.R., Varas, F., Fdez-Mańin, G., Carranza-Torres, C., 2003. Ground
Sciences. 76, 112–126.
response curves for rock masses exhibiting strain-softening behaviour. International
Carranza-Torres, C., Fairhurst, C., 1999. The elasto-plastic response of underground
Journal for Numerical and Analytical Methods in Geomechanics. 27, 1153–1185.
excavations in rock masses that satisfy the Hoek-Brown failure criterion.
Barla, G., Bonini, M., Debernardi, D., 2008. Time dependent deformations in squeezing
International Journal of Rock Mechanics and Mining Sciences 36 (6), 777–809.
tunnels. In: The 12th International Conference of International Association for
Carranza-Torres, C., Rysdahl, B., Kasim, M., 2013. On the elastic analysis of a circular
Computer Methods and Advances in Geomechanics (IACMAG), pp. 1-11.
lined tunnel considering the delayed installation of the support. International
Barla, G., Bonini, M., Debernardi, D., 2010. Time dependent deformations in squeezing
Journal of Rock Mechanics and Mining Sciences. 61, 57–85.
tunnels. ISSMGE International Journal of Geoengineering Case Histories 2, 40–65.
Carranza-Torres, C., Zhao, J., 2009. Analytical and numerical study of the effect of water
Barla, G., Zhao, K., Janutolo, M., 2011. 3D advanced modelling of TBM excavation in
pressure on the mechanical response of cylindrical lined tunnels in elastic and elasto-
squeezing rock condition. In: First Asian and 9th Iranian Tunneling Symposium,
plastic porous media. International Journal of Rock Mechanics and Mining Sciences.
Underground Space for Sustainable Development.
46 (3), 531–547.
Bernaud, D., Maghous, S., de Buhan, P., Couto, E., 2009. A numerical approach for design
Chen, T., He, T., Benesty, M., Khotilovich, V., Tang, Y., Cho, H., Chen, K., Mitchell, R.,
of bolt-supported tunnels regarded as homogenized structures. Tunnelling
Cano, I., Zhou, T., 2015. Xgboost: extreme gradient boosting. R Package Version 0.4-
Underground Space Technol. 24 (5), 533–546.
2 1, 1–4.
Bobet, A., 2009. Elastic solution for deep tunnels. Application to excavation damage zone
Chen, S., Qiang, S., Chen, S., Egger, P., 2004. Composite element model of the fully
and rockbolt support. Rock Mechanics and Rock Engineering. 42, 147–174.
grouted rock bolt. Rock Mechanics and Rock Engineering. 37, 193–212.
Bobet, A., Einstein, H.H., 2011. Tunnel reinforcement with rockbolts. Tunnelling
Chu, Z., Wu, Z., Liu, B., Liu, Q., 2019. Coupled analytical solutions for deep-buried
Underground Space Technol. 26 (1), 100–123.
circular lined tunnels considering tunnel face advancement and soft rock rheology
Cai, Y., Esaki, T., Jiang, Y., 2004a. An analytical model to predict axial load in grouted
effects. Tunnelling and Underground Space Technology. 94, 103111.
rock bolt for soft rock tunnelling. Tunnelling Underground Space Technol. 19,
Chu, Z., Wu, Z., Liu, Q., Liu, B., Sun, J., 2021. Analytical solution for lined circular
607–618. https://doi.org/10.1016/j.tust.2004.02.129.
tunnels in deep viscoelastic burgers rock considering the longitudinal discontinuous
Cai, Y., Esaki, T., Jiang, Y., 2004b. A rock bolt and rock mass interaction model.
excavation and sequential installation of liners. Journal of Engineering Mechanics.
International Journal of Rock Mechanics and Mining Sciences 41, 1055–1067.
147 (4), 04021009.
https://doi.org/10.1016/j.ijrmms.2004.04.005.

21
F. Song et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 153 (2024) 106019

Cutler, A., Cutler, D.R., Stevens, J.R., 2012. Random forests. Ensemble machine learning: Song, F., Wang, H.N., Jiang, M.J., 2018a. Analytically-based simplified formulas for
Methods and applications. circular tunnels with two liners in viscoelastic rock under anisotropic initial stresses.
Cui, L., Sheng, Q., Dong, Y., Xie, M., 2022. Unified elasto-plastic analysis of rock mass Construction and Building Materials 175, 746–767. https://doi.org/10.1016/j.
supported with fully grouted bolts for deep tunnels. International Journal for conbuildmat.2018.04.079.
Numerical and Analytical Methods in Geomechanics. 46 (2), 247–271. Song, F., Wang, H.N., Jiang, M.J., 2018b. Analytical solutions for lined circular tunnels in
Cui, L., Sheng, Q., Dong, Y., Ruan, B., Xu, D., 2021. A quantitative analysis of the effect of viscoelastic rock considering various interface conditions. Applied Mathematical
end plate of fully - grouted bolts on the global stability of tunnel. Tunnelling and Modelling 55, 109–130. https://doi.org/10.1016/j.apm.2017.10.031.
Underground Space Technology. 114, 104010. Song, F., Rodriguez-Dono, A., Farfan, P., 2022. Modelling underground excavations in
David, A., Helmstetter, A., 2006. Brittle creep, damage, and time to failure in rocks. rock masses with anisotropic time-dependent behaviour. Geomechanics and
Journal of Geophysical Research Solid Earth. 111 (B11), 1–17. Geophysics for Geo-energy and. Geo-resources 8, 1–25.
Do, D., Tran, N., Mai, V., Hoxha, D., Vu, M., 2020. Time-Dependent Reliability Analysis Song, F., Rodriguez-Dono, A., Olivella, S., 2021. Hydro-mechanical modelling and
of Deep Tunnel in the Viscoelastic Burger Rock with Sequential Installation of Liners. analysis of multi-stage tunnel excavations using a smoothed excavation method.
Rock Mechanics and Rock Engineering. 53, 1259–1285. Computers and Geotechnics 135, 104150.
Fabre, G., Pellet, F., 2006. Creep and time-dependent damage in argillaceous rocks. Song, F., Rodriguez-Dono, A., Olivella, S., Gens, A., 2021. Coupled solid-fluid response of
International Journal of Rock Mechanics and Mining Sciences 43 (6), 950–960. deep tunnels excavated in saturated rock masses with a time-dependent plastic
Fahimifar, A., Ranjbarnia, M., 2009. Analytical approach for the design of active grouted behaviour. Applied Mathematical Modelling 100, 508–535.
rockbolts in tunnel stability based on convergence-confinement method. Tunnelling Song, F., Rodriguez-Dono, A., Olivella, S., Zhong, Z., 2020. Analysis and modelling of
and Underground Space Technology. 24(4), 363–375. longitudinal deformation profiles of tunnels excavated in strain-softening time-
Fahimifar, A., Tehrani, F., Hedayat, A., Vakilzadeh, A., 2010. Analytical solution for the dependent rock masses. Computers and Geotechnics. 125, 103643.
excavation of circular tunnels in a visco-elastic Burger’s material under hydrostatic Song, F., Gens, A., Collico, S., Grgic, D., 2025. Fully coupled THM constitutive model for
stress field. Tunnelling and Underground Space Technology. 25(4), 297–304. clay rocks: formulation and application to laboratory tests. Journal of Rock
Frank, E., Trigg, L., Holmes, G., Witten, I., 2000. Naive Bayes for regression. Machine Mechanics and Geotechnical Engineering.
Learning. Song, F., Gens, A., Collico, S., Plúa, C., Armand, G., Wang, H., 2024. Analysis of
Ghosh, D., Vogt, A., 2012. Outliers: An evaluation of methodologies. Section on Survey thermally-induced fracture of Callovo-Oxfordian claystone: from lab tests to field
Research Methods – JSM 2012. 3455–3460. scale. Geomechanics for Energy and the Environment.
Guan, K., Zhu, W., Yu, Q., Cui, L., Song, F., 2022. A plastic-damage approach to the Song, F., González-Fernándeza, M., Rodriguez-Dono, A., Alejano, L., 2023. Numerical
excavation response of a circular opening in weak rock. Tunnelling and analysis of anisotropic stiffness and strength for geomaterials. Journal of Rock
Underground Space Technology 126, 104538. Mechanics and Geotechnical Engineering 15 (2), 323–338.
Hunsche, U., Hampel, A., 1999. Rock salt — the mechanical properties of the host rock Song, F., Rodriguez-Dono, A., 2021. Numerical solutions for tunnels excavated in strain-
material for a radioactive waste repository. Engineering Geology 52 (3–4), 271–291. softening rock masses considering a combined support system. Applied
Jiang, M., Bo, Y., Wang, H., Che, N., 2023. Study on the reinforcement mechanism of Mathematical Modelling. 92, 905–930.
grouted bolts with or without prestress via the hybrid DEM-FDM method. Stevens, J., 1984. Outliers and influential data points in regression analysis.
Transportation Geotechnics. 40, 100967. Psychological Bulletin. 95(2), 334–344.
Kovári, K., 2003. History of the sprayed concrete lining method-part I: Milestones up to Sulem, J., Panet, M., Guenot, A., 1987. An analytical solution for time-dependent
the 1960s. Tunnelling and Underground Space Technology 18 (1), 57–69. displacements in a circular tunnel. International Journal of Rock Mechanics and
Larose, D.T., Larose, C.D., 2014. k-nearest neighbor algorithm. Mining Sciences & Geomechanics Abstracts. 24(3), 155–164.
Leca, E., Clough, G., 1992. Preliminary design for NATM tunnel support in soil. Journal Sun, Z., Zhang, D., Fang, Q., Dui, G., Tai, Q., Sun, F., 2021. Analysis of the interaction
of Geotechnical Engineering. 118(4), 558–575. between tunnel support and surrounding rock considering pre-reinforcement.
Li, C., 2017. Principles of rockbolting design. Journal of Rock Mechanics and Tunnelling and Underground Space Technology. 115, 104074.
Geotechnical Engineering. 9(3), 396–414. Wang, G., Han, W., Jiang, Y., Luan, H., Wang, K., 2019. Coupling analysis for rock mass
Li, S., Wang, M., 2008. An elastic stress-displacement solution for a lined tunnel at great supported with CMC or CFC rockbolts based on viscoelastic method. Rock Mechanics
depth. International Journal of Rock Mechanics and Mining Sciences. 45(4), and Rock Engineering 52, 4565–4588.
486–494. Wang, M., Li, S., 2009. A complex variable solution for stress and displacement field
Lo, K., Yuen, C., 1981. Design of Tunnel Lining in Rock for Long Term Time Effects. around a lined circular tunnel at great depthInternational. Journal for Numerical and
Canadian Geotechnical Journal. 18(1), 24–39. Analytical Methods in Geomechanics 33, 939–951.
Lu, A., Zhang, L., Zhang, N., 2011. Analytic stress solutions for a circular pressure tunnel Wang, H., Li, Y., Ni, Q., Utili, S., Jiang, M., Liu, F., 2013. Analytical solutions for the
at pressure and great depth including support delay. International Journal of Rock construction of deeply buried circular tunnels with two liners in rheological rock.
Mechanics and Mining Sciences 48 (3), 514–519. Rock Mechanics and Rock Engineering. 46, 1481–1498.
Maghous, S., Bernaud, D., Couto, E., 2012. Three-dimensional numerical simulation of Wang, H., Utili, S., Jiang, M., 2014. An analytical approach for the sequential excavation
rock deformation in bolt-supported tunnels: A homogenization approach. Tunnelling of axisymmetric lined tunnels in viscoelastic rock. International Journal of Rock
and Underground Space Technology. 31, 68–79. Mechanics and Mining Sciences 68, 85–106.
Nomikos, P., Rahmannejad, R., Sofianos, A., 2011. Supported axisymmetric tunnels Wang, H., Utili, S., Jiang, M., He, P., 2015. Analytical Solutions for Tunnels of Elliptical
within linear viscoelastic burgers rocks. Rock Mechanics and Rock Engineering. 44, Cross-Section in Rheological Rock Accounting for Sequential Excavation. Rock
553–564. Mechanics and Rock Engineering. 48, 1997–2029.
Oreste, P., 2003. Analysis of structural interaction in tunnels using the covergence- Wang, H., Xiao, G., Jiang, M., Crosta, G., 2018. Investigation of rock bolting for deeply
confinement approach. Tunnelling and Underground Space Technology 18 (4), buried tunnels via a new efficient hybrid DEM-Analytical model. Tunnelling
347–363. Underground Space Technol. 82, 366–379.
Oreste, P., 2015. Analysis of the interaction between the lining of a TBM tunnel and the Wu, K., Shao, Z., Li, C., Qin, S., 2020. Theoretical Investigation to the Effect of Bolt
ground using the convergence-Confinement method. American Journal of Applied Reinforcement on Tunnel Viscoelastic Behavior. Arabian Journal for Science and
Sciences 12 (4), 276–283. Engineering. 45, 3707–3718.
Osborne, J., Overbay, A., 2004. The power of outliers (and why researchers should Wu, K., Shao, Z., Sharifzadeh, M., Chu, Z., Qin, S., 2022. Analytical approach to
always check for them). Practical Assessment, Research and Evaluation. estimating the influence of shotcrete hardening property on tunnel response. Journal
Pan, Y., Dong, J., 1991. Time-dependent tunnel convergence-I. Formulation of the of Engineering Mechanics. 148 (1), 04021127.
model, in. International Journal of Rock Mechanics and Mining Sciences & Wu, K., Wang, Y., Zheng, X., Zhao, N., 2024. A better understanding of tunnel
Geomechanics Abstracts 28 (6), 469–475. deformable supports: from analytical model to engineering application. Archives of
Paraskevopoulou, C., Diederichs, M., 2018. Analysis of time-dependent deformation in Civil and Mechanical Engineering. 24, 114.
tunnels using the Convergence-Confinement Method. Tunnelling and Underground Xue, Y., Zhang, X., Li, S., Qiu, D., Su, M., Li, L., Li, Z., Tao, Y., 2018. Analysis of factors
Space Technology. 71, 62–80. influencing tunnel deformation in loess deposits by data mining: a deformation
Paraskevopoulou, C., 2016. Time-dependency of Rocks and Implications Associated with prediction model. Engineering Geology. 232, 94–103.
Tunnelling. PhD thesis. Department of Geological Sciences and Geological Zhang, K., Zhang, G., Hou, R., Wu, Y., Zhou, H., 2014. Stress Evolution in Roadway Rock
Engineering, Queen’s University. Bolts During Mining in a Fully Mechanized Longwall Face, and an Evaluation of Rock
Pollet, T., van der Meij, L., 2017. To remove or not to remove: the impact of outlier Bolt Support Design. Rock Mechanics and Rock Engineering. 48, 333–344.
handling on significance testing in testosterone data. Adaptive Human Behavior and Zhao, N., Shao, Z., Wu, K., Chu, Z., Qin, S., 2021. Time-Dependent Solutions for Lined
Physiology. 3, 43–60. Circular Tunnels Considering Rockbolts Reinforcement and Face Advancement
Sakurai, S., 2010. Modeling strategy for jointed rock masses reinforced by rock bolts in Effects. International Journal of Geomechanics. 21 (10), 04021179.
tunneling practice. Acta Geotechnica 5, 121–126. Zhou, J., Yang, X., 2021. An analysis of the support loads on composite lining of deep-
Sandrone, F., Labiouse, V., 2010. Analysis of the evolution of road tunnels equilibrium buried tunnels based on the Hoek-Brown strength criterion. Tunnelling and
conditions with a convergence-confinement approach. Rock Mechanics and Rock Underground Space Technology. 118, 104174.
Engineering. 43, 201–218. Zou, J., Chen, K., Pan, Q., 2018. An improved numerical approach in surrounding rock
Showkati, A., Salari-rad, H., Aghchai, M., 2021. Predicting long-term stability of tunnels incorporating rockbolt effectiveness and seepage force. Acta Geotechnica. 13,
considering rock mass weathering and deterioration of primary support. Tunnelling 707–727.
and Underground Space Technology. 107, 103670.

22