Computers and Geotechnics 161 (2023) 105622

Contents lists available at ScienceDirect

Computers and Geotechnics

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

Failure mechanism and deformation prediction of soft rock tunnels based

on a combined finite–discrete element numerical method
Penghai Deng a, b, c, Quansheng Liu a, b, *, Bin Liu d, Haifeng Lu a, b
a
The Key Laboratory of Safety for Geotechnical and Structural Engineering of Hubei Province, School of Civil Engineering, Wuhan University, Wuhan 430072, China
b
State Key Laboratory of Water Resources and Hydropower Engineering Science, Wuhan University, Wuhan 430072, China
c
State Key Laboratory for Geomechanics and Deep Underground Engineering, China University of Mining & Technology, Beijing, Beijing 100083, China
d
State Key Laboratory of Geomechanics and Geotechnical Engineering, Institute of Rock and Soil Mechanics, Chinese Academy of Sciences, Wuhan, Hubei 430071,
China

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

Keywords: The large deformation mechanisms and predictions of soft rock tunnels have always been important but difficult
Failure mechanism to solve problems in the field of geotechnical engineering. A combined finite–discrete element numerical
Deformation prediction simulation method (FDEM) was used to study large deformation mechanism, classification and prediction. The
Combined finite–discrete element method
deformations or failure mechanisms of tunnel surrounding rock were revealed, and the failure modes and
(FDEM)
displacement value prediction of surrounding rock with different strength-stress ratios were also investigated.
Soft rock tunnel
Strength-stress ratio The following conclusions were obtained: (1) Critical hysteresis damping can be adopted to simulate the pro­
gressive large deformation process of soft rock tunnels, which can obtain the final deformation of unsupported
tunnels and avoid a dynamic response. (2) Under concentrated tangential stress, the surrounding rock undergoes
X-shaped conjugate shear fracture, and this type of fracture network continues to propagate toward the depth of
the surrounding rock until the model reaches a stable state; the reduction in tunnel cross-section is mainly caused
by the macroscopic movement and volume expansion of rock fragments, and the latter is due to the generation of
a large number of macroscopic voids. (3) As the strength-stress ratio decreases, the deformation or failure modes
of the surrounding rock can be divided into four categories: elastic–plastic deformation, closed fracturing, shear
dilation and broken expansion. (4) Finally, a prediction equation for isotropic and homogeneous soft rock un­
supported tunnel deformation with general size driven by hydrostatic in situ stress is obtained, which indicates
that with an increasing strength-stress ratio, the surrounding rock displacement decreases as an exponential
function, with a correlation coefficient of R2 = 0.997. In addition, the robustness and reliability of the prediction
equation is verified.

1. Introduction engineering investment. For example, if a soft rock tunnel could be

seriously deformed, leading to the jamming of a tunnel boring machine
A large number of tunnel engineering cases (Jimenez and Recio, (TBM), the drilling and blasting (D&B) method can be used instead, or
2011; Singh et al., 2007) have shown that large deformations of sur­ appropriate reinforcement measures can be taken in advance to prevent
rounding rock will occur during excavation in soft rock with high in situ TBM jamming.
stress. These data (Singh et al., 2007) have been mostly acquired during Therefore, researchers and engineers have performed extensive
the tunneling process, and these tunnels have been supported by linings, research on tunnel deformation prediction and obtained a large number
rockbolts or grouting. However, engineers attempt to predict the of helpful and valuable results. To eliminate the differences in tunnel
possible maximum deformations before tunneling and without any diameter, the relative deformation or strain ε is generally used, which is
support measures because the prediction results can directly affect the defined as the ratio of the maximum deformation of the tunnel surface u
selection of tunnel construction methods, support measures and to the tunnel radius R0, i.e., ε = Ru0 , and ε is typically presented as a
parameter designs and further affect the construction period, safety and

* Corresponding author at: The Key Laboratory of Safety for Geotechnical and Structural Engineering of Hubei Province, School of Civil Engineering, Wuhan
University, Wuhan 430072, China.
E-mail address: [email protected] (Q. Liu).

https://doi.org/10.1016/j.compgeo.2023.105622
Received 21 May 2023; Received in revised form 15 June 2023; Accepted 25 June 2023
Available online 29 June 2023
0266-352X/© 2023 Elsevier Ltd. All rights reserved.
P. Deng et al. Computers and Geotechnics 161 (2023) 105622

percentage. Elastic-plastic mechanical analytical solutions (Aydan et al., Furthermore, Yao et al. (2009) proposed a theoretical calculation
1993; Ding et al., 2023; Liu et al., 2013; Liu et al., 2008; Zhang, 2003), equation to obtain the tunnel strain when the influences of both dila­
finite element numerical simulation methods (FEMs) (Hoek and Mar­ tation and plastic softening effects are considered, as
inos, 2000) or field monitoring data fitting methods (Chen et al., 2019) {{ [( )
1+η2
] }( )
1+η1
}
are often used to build tunnel relative deformation prediction models or ε = 2A
1
+
1 Rb
− 1
Rp
+
η1 − 1
(8)
equations, as follows: 1 + η1 1 + η2 R0 Rb 2(1 + η1 )

( )
(1) Elastic-plastic mechanical analytical solution where A = 1+E ν σRp − σ 0 , σ Rp is the radial stress at the interface between
the plastic zone and elastic zone, and σRp = 2σ1+q
0 − σc
; Rb is the radius of the
For the homogeneous and isotropic rock mass under a hydrostatic in broken zone; η1 is the parameter introduced to consider the dilatation of
situ stress state, the theoretical solution of the tunnel strain is (Liu et al., the rock mass in the plastic zone: when considering the dilatation effect
2008) of the rock mass, η1 = q, otherwise η1 = 1. η2 is the parameter introduced
{ [( ) 2sinφ ]} to consider the dilatation gradient of the rock mass in the broken zone:
1+ν c Rp 1− sinφ R2P
ε= σ0 − − 1 (1) when considering the dilatation effect of the rock mass, η2 = 1.3–1.5;
R0 ⋅E tanφ R0 R0
otherwise η2 = 1.

where E and v are the Young’s modulus and Poisson’s ratio of the sur­ (2) Numerical simulation method
rounding rock mass; c and φ are the cohesion and internal friction angle;
σ 0 is the in situ stress under the hydrostatic state; and Rp is the radius of Hoek and Marinos (2000) obtained the following empirical predic­
the plastic zone, which can be obtained by (Liu et al., 2008) tion equation by FEM numerical modeling:
[ ]1− sinφ ( ∗ )− 2
(σ 0 + c⋅cotφ)(1 − sinφ) 2sinφ σ
Rp = R 0 ⋅ (2) ε(%) = 0.2 c (9)
c⋅cotφ σ0
When the strength-stress ratio α (i.e., the ratio of the uniaxial Zhang et al. (2016) revised Eq. (9) to consider the impact of different
compressive strength of intact rock σc to the in situ stress σ 0) is larger lithologies, as
than 2.0, the surrounding rock after tunnel excavation is in an elastic ( ∗ )− 2
and intact state, and the tunnel strain can be theoretically determined as ε(%) = 0.4ξ c
σ
(10)
follows (Cai et al., 2002) σ0
1+ν where ξ is the potential squeezing ratio, as
ε= σ0 (3)
E
εp
To consider the plastic softening effect of the surrounding rock mass ξ= (11)
εp
in the plastic zone, Zhang (Zhang, 2003) proposed an improved form of
Eq. (1), as where εp is the peak strain of a special rock sample; εp is the peak strain
(
q R2P − R20
) reference value, and εp = 0.73\% .
R
ε = p2 u∗ + (4) Ding et al. (Ding et al., 2023) obtained another empirical prediction
R0 2R20
equation by finite difference method (FDM) numerical modeling as
1+sinφ
where u∗ = 1+E νRp (σ 0 sinφ + ccosφ) and q = 1− sinφ
. The radius of the ( ∗ )b ( )d
σ E
plastic zone Rp is shown in Eq. (2). ε=a c (12)
σ0 E0
In addition, to consider the influence of the differences in stress–­
strain curves between different rock types and surrounding rocks in where a, b and d are the fitting coefficients, and E0 = 1 GPa.
different regions in different strain states, Aydan et al. (Aydan et al., In addition, Shao et al. (2021) used FDM numerical modeling to
1993) proposed the following equation for calculating the tunnel strain obtain a tunnel strain prediction equation under a nonhydrostatic in situ
ε: stress state as.
⎡ ⎤qf ∗∗ +1
2
(
q− 1+α
) f +1
q− 1 α α∗
− 1
ε = 0.187e− Gn /0.104
+ 0.01 (13)
⎢q+1 ηsf − + q∗ − ⎥
(5)
q− 1 q− 1 1
ε = ηsf ⎢ ⎥
⎣ α* ⎦ where Gn is the ratio of the uniaxial compressive strength σ c to the
q* − 1
horizontal in situ stress σh, σh = λσv = λγH, λ is the lateral pressure
coefficient, σv is the vertical in situ stress, γ is the unit weight of the rock
where ηsf = , ηs = εεes , ηf = εfe , and εs, εf and εe are the strain values
ηs +ηf ε
2 mass, and H is the depth of the tunnel.
of the rock sample at different stages of the uniaxial compression
stress–strain curve, which can be obtained from a uniaxial compression (3) Case collection method
laboratory test of a special rock sample. q∗ = 1+sinφ , and φ* is the in­
∗

1− sinφ∗
ternal friction angle of broken rock; f and f* are the test constants. α = Meng et al. (2022) obtained the following empirical fitting equation
of tunnel deformation by collecting more than 200 typical soft rock
σ0 ,
α∗ = σσ0c and σ ∗c is the uniaxial compressive strength of the rock mass,
∗
σc

tunnel cases:
and theoretically, σc and σ ∗c are
( )− 0.714
σc
σc =
2c⋅cosφ
(6) u = 204.31⋅ (14)
1 − sinφ σ0

2c∗ ⋅cosφ∗ where u is the tunnel deformation at the tunnel boundary, and the unit is
σ ∗c = (7) cm.
1 − sinφ∗
According to deformation monitoring data of 81 tunnel cases, Chen
where c* represents the cohesion of broken rock. (Chen et al., 2019) proposed a tunnel strain prediction equation based

2
P. Deng et al. Computers and Geotechnics 161 (2023) 105622

on the modified [BQ] value as. will undoubtedly improve the fitting accuracy and the application
( )1.40 scope of the prediction model.
H
ε = 0.50 + 0.10 (15)
[BQ] Therefore, a combined finite–discrete element numerical method
15+0.24RMR
(FDEM) is employed to reveal the deformation or failure mechanism of
where [BQ] = 170ln 5.7− 0.06RMR, RMR = 9lnQ + 44, and Q is the quality soft surrounding rock after tunnel excavation, and the empirical fitting
of the rock mass. equation of tunnel strain with strength-stress ratio can be obtained.
Furthermore, according to the deformation characteristics of the Different influence factors are investigated, including FDEM element
Wushaoling soft rock tunnel, Liu et al. (2008) obtained a fitting curve size, excavation diameter, Young’s modulus, Poisson’s ratio and internal
between the tunnel deformation and dimensionless comprehensive co­ friction coefficient of the surrounding rock. In addition, the reliability of
1+λ E σ c
efficient κ (κ = 1+2λ ⋅σv ⋅σv ); Chen et al. (2018) obtained a fitting curve the prediction equation is verified by comparison with engineering
between the tunnel deformation and strength-stress ratio based on 93 monitoring data. Compared with theoretical analytical solutions or FEM
tunnel deformation monitoring results; Yang et al. (2022) collected the numerical simulations, the FDEM can not only simulate the elastic–­
deformation data of a soft rock tunnel with similar lithology located in plastic deformation of surrounding rock but also simulate the fracture
the Jiuzhaigou–Mianyang Expressway of China and thus obtained a initiation, propagation and connection process of surrounding rock and
fitting curve between the tunnel deformation and the strength-stress the resulting volume expansion effect (Deng et al., 2021c; Deng et al.,
ratio. However, no specific fitting equations are given for the above 2023).
three fitting curves. In fact, the feasibility of FDEM numerical simulation and the reli­
The above models have made important contributions to the pre­ ability of simulation results have been verified by many previous studies
diction of tunnel deformation. However, previous prediction models (Deng and Liu, 2020; Deng et al., 2021c; Deng et al., 2020; Lisjak et al.,
have the following limitations: 2015). Although the FDEM has been widely used in tunnel excavation
simulations, including the failure mechanism of isotropic (Deng and Liu,
• The elastic–plastic mechanical analytical solution and FEM numeri­ 2020; Deng et al., 2020; Lisjak et al., 2014a), anisotropic (Deng et al.,
cal modeling have a common limitation: only the tunnel convergence 2021c; Lisjak et al., 2014b; Lisjak et al., 2016), heterogeneous (Deng
caused by the elastic–plastic deformation of the surrounding rock et al., 2023) surrounding rock and the support or reinforcement (Lisjak
can be obtained. However, in addition to continuous deformation, et al., 2020; Liu et al., 2019) of surrounding rock, most of these studies
the soft surrounding rock mass with high in situ stress also generates focused on the development of new algorithms and research on the
dilatation phenomena caused by fracturing. Therefore, the predic­ deformation and failure mechanism of surrounding rock. There are few
tion results are often different from the actual situations. For studies on the simulation and prediction of the final deformation of
example, research by Zhang et al. (2022) indicated that Hoek’s surrounding rock without support, and there is no empirical equation or
(Hoek and Marinos, 2000) prediction result has large errors with the model addressing the deformation of surrounding rock and the strength-
actual tunnel deformation, and they (Zhang et al., 2022) concluded stress ratio based on FDEM numerical modeling, needless to say a
that the prediction results are generally conservative. demonstration and analysis of the reliability of the empirical model.
• Fitting prediction equations with field monitoring data appears to be
a reliable model. However, unfortunately, it is difficult to obtain the 2. Method
total deformation of unsupported tunnels because almost all tunnel
projects require some support measures before large deformation The rock material is meshed into triangular elements and quadri­
occurs during the construction process. In addition, the tunnel lateral joint elements in the FDEM, and two adjacent triangular elements
deformation monitored by field surveys is often caused by multiple are connected by a quadrilateral element, as shown in Fig. 1(a). The
factors, such as water, stress and structural planes. In fact, when numerical model, as well as the triangular elements, can be built by
fitting the deformation values obtained from different tunnels, the using Gmsh software, while the insertion of quadrilateral elements must
differences in engineering geological conditions, construction use the FDEM itself. Only elastic deformation occurs for triangular ele­
methods (TBM or D&B) and support measures of different tunnels are ments, while the yielding and failure of materials are realized by
often ignored. For example, according to the field monitoring data quadrilateral joint elements. When the quadrilateral joint element fails,
collected by Chen et al. (2019) and Chen et al. (2018), the defor­ the triangular elements on both sides of it change from a bond rela­
mation of different tunnels can vary by 2–10 times at the same tionship to a contact relationship. The triangle element deformation
strength-stress ratio. stress, quadrilateral element bond stress, triangular element contact
• Most prediction models, especially empirical equations taking the stress, node coordinate update and calculation process are as follows:
strength-stress ratio as an independent variable, do not consider the
deformation under extreme conditions. For example, when the
compressive strength of the rock mass is 0, and thus σ0c = 0, i.e., the
σ∗ 2.1. Deformation stress of the triangular element
tunnel is excavated in dry, cohesionless and smooth contact fine
sand, the surrounding rock mass will collapse, and the unsupported Under two-dimensional (2D) plane strain conditions, according to
tunnel space will be buried. The strain of the tunnel under this the general Hooke’s law, the deformation stress of the triangular
extreme condition will be 100%. If this boundary condition is not element is (Deng et al., 2023).
considered when fitting the curve, the prediction error by the ⎧
⎪ E [ ]
⎪
empirical equation will be large for a small strength-stress ratio. For ⎪
⎪
⎪
σxx =
(1 + ν)(1 − 2ν)
(1 − ν)εxx + νεyy + με̇xx
⎪
example, Chen et al. (2019) noted that when the strength-stress ratio ⎪
⎪
⎨ E [ ]
is <0.1, the error of the prediction results proposed by Hoek and σyy = νεxx + (1 − ν)εyy + με̇yy (16)
⎪ (1 + ν)(1 − 2 ν)
Marinos (2000) reached 600–1300%. In addition, most empirical ⎪
⎪
⎪
⎪
⎪
equations, such as those proposed by Hoek and Marinos (2000), ⎪
⎪
⎩ τxy =
E
γ + μγ̇ xy
Meng et al. (2022), and Shao et al. (2021), do not consider another 1 + ν xy
boundary condition, that is, when σ0c ⩾2, the surrounding rock is in an
σ∗
where σxx, σ yy and τxy are normal stress and shear stress, respectively;
elastic state, and the strain at the tunnel surface has a strict analytical
εxx, εyy and γxy are normal strain and shear strain, respectively; ε̇xx , ε̇yy ,
solution, as shown in Eq. (3). Adding these two boundary conditions
and γ̇ xy are the corresponding strain rates; μ is viscous damping

3
P. Deng et al. Computers and Geotechnics 161 (2023) 105622

(a) (b)

Fig. 1. Fundamentals of FDEM. (a) Mesh topology, (b) constitutive model of quadrilateral joint element, and (c) calculation flow.

(Munjiza, 2004), as ⎧ o
√̅̅̅̅̅̅ ⎨ ⋅σ t , o < op
μ = 2h Eρ (17) σn = op (18)
⎩
z⋅σt , op < o < ot
where h is the element size, i.e., the minimum edge length of a triangular ⎧
element, and ρ is the material density. ⎪
⎪ |s|
⎪
⎪ ⋅c, |s| < sp and σn > 0
⎪
⎪ sp
⎪
⎪
2.2. Bond stress of quadrilateral joint element ⎪
⎪
⎨ |s|
τ= sp
⋅(c − σn tanφ), |s| < sp and σn < 0 (19)
⎪
As shown in Fig. 1(b), the constitutive equations of the quadrilateral ⎪
⎪
⎪
⎪
joint element modified by the authors are (Deng et al., 2021a) ⎪
⎪
⎪
⎪
z⋅c, sp < |s| < st and σ n > 0
⎪
⎩ z⋅c − σ tanφ,
n sp < |s| < st and σn < 0

4
P. Deng et al. Computers and Geotechnics 161 (2023) 105622

where σ n and τ are normal and shear stresses (σn > 0 represents tensile
fct = min(Pt ⋅δt , fcn ⋅tanφr ) (26)
stress and σn < 0 is compressive stress), respectively; o and s are opening
and sliding displacements, respectively; σ t is tensile strength; and z is a
where Pt is the tangential contact stiffness, δt is the tangential
postpeak softening function (Munjiza, 2004), as follows:
displacement, and φr is the sliding friction angle.
[ ( ) ][ ]
a′ + b′ − 1 a′ + b′d′ ′
z= 1− ′ ′
exp D ′ ′ ′ ′
a′(1 − D) + b′(1 − D)d 2.4. Update of the node coordinates
a +b (a + b )(1 − a − b )
(20)
The deformation stress of the triangular element, the cohesive stress
where a′, b′ and d′ are fitting parameters equal to 0.63, 1.8, and 6.0, of the quadrilateral joint element and the contact stress between trian­
respectively (Munjiza, 2004). D is a damage coefficient with a value gular elements are equivalent to the corresponding triangular nodes to
between 0 and 1 (Tatone and Grasselli, 2015), that is, obtain node forces. Then, the node coordinates of triangular elements
⎧ are updated by Newton’s second law as
⎪ o − op
⎪ , when o > op and |s| < sp
⎪
⎪
⎪ o t − op mẍ + c′ẋ + Fin + Fex = 0 (27)
⎪
⎪
⎪
⎪
⎪
⎪ |s| − sp
⎪
⎨ st − sp
, when |s| > sp and o < op where m is the node mass; c′ is the hysteretic damping (Deng et al.,
D = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
⎪ 2021b); and Fin and Fex are the internal force and external load,
⎪ ( )2 ( )2
⎪
⎪
⎪ o − op |s| − sp respectively. They are
⎪
⎪ + , when |s| > sp and o > op
⎪
⎪ ot − op st − sp √̅̅̅̅
⎪
⎪
⎪ E
⎩
0, when |s| < sp and o < op
′
c = 2mβ (28)
ρ
(21)
Fin = Fd + Fc + Fco (29)
where op, ot, sp and st are the peak and ultimate displacements in the
normal and tangential directions, respectively, as where β is the hysteretic damping coefficient and Fd, Fc and Fco are the
⎧ σt node forces induced by the deformation of the triangular element,
⎪
⎨ op = h⋅Pf
⎪ bonding of the quadrilateral joint element and contact between trian­
(22) gular elements, respectively.
⎪
⎪ σs
⎩ sp = h⋅
Pf
2.5. Calculation flow
where Pf is the joint penalty and σ s is the shear strength. The modified
calculation equations of the tensile displacement ot and shear displace­ According to the above calculation process, the FDEM calculation
ment st corresponding to the failure moment of a joint element were flow is shown in Fig. 1(c). It should be noted that the bond stress and
given by Tatone (2014) as contact stress between adjacent triangular elements do not exist at the
⎧ same time; that is, only when the quadrilateral joint element is broken
GI
⎪
⎪
⎪ ot = op + 3
⎨ can the contact detection and contact stress calculation program be­
σt tween two adjacent triangular elements be activated (Deng et al., 2021a;
(23)
⎪
⎪
⎪ s
⎩ t = s + 3
GII Miglietta et al., 2017). This modification not only increases the calcu­
p
σs lation efficiency but also improves the accuracy of the simulation results
(Deng et al., 2021a).
where GI and GII are mode I and mode II fracture energies, respectively.
The shear strength σs is 3. Parameters
{
c, σn > 0
σs = (24) 3.1. Basic input parameters
c − σn tanφ, σn < 0

(1) Calibration method

2.3. Contact stress between triangular elements
For any new FDEM numerical model, parameter calibration is the
When a quadrilateral joint element fails, the triangular elements on primary task. After more than 20 years of development, especially in the
both sides of the joint element are converted from the bonding rela­ past 10 years, the calibration method of FDEM input parameters is
tionship to the contact relationship. The authors (Deng et al., 2021a) relatively sophisticated (Deng et al., 2022; Tatone and Grasselli, 2015;
previously proposed a new expression for the normal contact stiffness to Vlachopoulos and Vazaios, 2018). In addition, according to previous
realize a smooth transition and avoid abnormal vibrations during the conclusions by the authors (Deng et al., 2022), it was found that only the
stress transition process. The specific normal contact stiffness value re­ mode I and mode II fracture energies, i.e., GI and GII, of the joint element
lies on a joint element penalty and the geometric size of the triangles need to be calibrated, while other parameters can be obtained by lab­
(Deng et al., 2021a). The normal contact force fcn is (Tatone and oratory tests (such as E, ν, ρ, c, φ, σ t and φr), theoretical equations (such
Grasselli, 2015) as viscous damping μ (Munjiza, 2004)), or being cited from previous
∮ empirical conclusions (such as h (Liu and Deng, 2019), Pf (Deng et al.,
fcn = Pn nΓ (φc − φt )dΓ (25) 2022), time step Δt (Guo et al., 2015), normal and tangential contact
Γβt ∩βc stiffness Pn (Deng et al., 2021a) and Pt (Mahabadi, 2012)). According to
a specific laboratory test, some input parameters are obtained, as shown
where Γβt ∩βc is the boundary of the contact overlap; Pn is the normal in Table 1.
contact stiffness; nΓ is the outer normal direction of the boundary; and φc The values of GI and GII can be calibrated by a new method previ­
and φt are potential functions (Munjiza, 2004). The minimum value of ously proposed by the authors (Deng et al., 2022) to obtain accurate
the static and sliding friction forces is taken as the tangential contact values under a static loading state to eliminate the impact of the loading
force, as (Deng et al., 2021a) rate. The Brazilian disc numerical model shown in Fig. 2(a) is used to

5
P. Deng et al. Computers and Geotechnics 161 (2023) 105622

Table 1 respectively. The peak load under static loading can be obtained by
Partial input parameters. fitting the simulation results with different loading rates. When the
Parameter Value Parameter Value simulated load Fs is close to the theoretical load Ft, the GII value is
deemed reasonable. When the rock sample fails along the pre-prepared
Young’s modulus E 5 Internal friction angle φ ( )
◦
26.5
(GPa) path, the theoretical load Ft is (Deng et al., 2022)
Poisson’s ratio ν 0.25 Joint penalty Pf (GPa) (Deng et al., 30E
c⋅Ar
2022) Ft = (30)
Density ρ (kg/m3) 2500 Normal contact stiffness Pn (GPa) ( 0.1448Pf sinαr − cosαr ⋅tanφ
Deng et al., 2021a)
Viscous damping μ (
√̅̅̅̅̅̅
2h Eρ Tangential contact stiffness Pt 5E where Ar is the shear slip surface area and αr is the dip angle of the pre-
Munjiza, 2004) (GPa) (Mahabadi, 2012) prepared failure path. By substituting the data, Ft = 324.66 kN can be
Tensile strength σt 0.5 Sliding friction angle φr (◦ ) 26.5
obtained.
(MPa)
Cohesion c (MPa) 2 Uniaxial compression strength σc 6.47
(MPa) (2) Calibration results

After several trial calculations, when GI = 9 J/m2 and GII = 12 J/m2,

the calibration results are shown in Fig. 3. The Brazilian disc sample
produces a tensile fracture that starts from the rock center, which meets
the requirements of the Brazilian disc test, as shown in Fig. 3(a). In
addition, shear slip fracture of the uniaxial compression sample occurs
along the pre-prepared failure path, as shown in Fig. 3(b), meeting the
requirements of Eq. (30), which shows that the simulation results are
credible. The simulation results under different loading rates are shown
in Fig. 3(c) and (d), which indicates the following:

(1) The tensile strength increases linearly with increasing loading

rate for the Brazilian disc, and the correlation coefficient reaches
R2 = 0.999; according to the fitting curve, when v = 0, σt =
0.4986 MPa, which is very close to the input value of tensile
strength shown in Table 1, indicating that the GI value is
reasonable; in addition, the linear relationship between Brazilian
disc tensile strength and loading rate is consistent with that ob­
tained by Lei (Lei, 2015) through laboratory experiments.
(2) The simulated load increases exponentially with increasing
loading rate for uniaxial compression, and the correlation coef­
ficient is R2 = 0.995; according to the fitting curve, when v = 0, Fs
= 324.15 kN, which is very close to the theoretical load calcu­
lated by Eq. (30), indicating that the GII value is reasonable. It
should be noted that the curve shape in Fig. 3(d) is different from
the previous straight line shape (Deng et al., 2022), which is
determined by the Young’s modulus.
(3) Parameter validation

The input parameters shown in Table 1 and the calibrated GI and GII
values, as well as the uniaxial compression numerical model shown in
Fig. 2(b) but without a pre-prepared failure path, are used to verify the
Fig. 2. Numerical models of parameter calibration. (a) Brazilian disc model, (b) reliability of the input parameters. The element size and the time step
uniaxial compression model. are h = 2 mm and Δt = 1.5 × 10-9 s. The loading rate is set as 0.05 m/s,
0.1 m/s, 0.2 m/s, 0.3 m/s, 0.4 m/s, 0.5 m/s, 0.6 m/s, 0.8 m/s and 1.0 m/
calibrate the mode I fracture energy, and the improved uniaxial s, and the simulation results shown in Fig. 4(a)–(c) are obtained.
compression model shown in Fig. 2(b) is employed to calibrate the mode From the validation results, it can be concluded that (1) with
II fracture energy. increasing loading rate, the uniaxial compressive strength of the rock
The diameter of the Brazilian disc model is 50 mm. The cohesion of samples increases as an exponential function, and the correlation coef­
the rock is set as a large value to prevent shear failure at the end of the ficient is 0.9965, as shown in Fig. 4(a); according to the fitting curve,
sample under the action of point load. The element size is h = 2 mm, and when v = 0, the compressive strength is 6.91 MPa, which is close to the
the calculation time step is Δt = 1.5 × 10-9 s. The loading rate is set from test results shown in Table 1. (2) The simulated elastic modulus is
0.1 m/s to 0.5 m/s with a spacing of 0.05 m/s. The tensile strength under defined as the secant modulus at half of the compressive strength, and
static loading can be obtained by fitting the simulation results with the simulated elastic modulus shown in Fig. 4(b) is 5 GPa, which is the
different loading rates. When the fitting result with v = 0 is close to the same as the input value. (3) The rock sample is subject to shear failure at
input value (Table 1), the GI value is deemed reasonable. a dip angle of approximately 60◦ , which is close to the theoretical failure
The height and width of the uniaxial compression model are 100 m angle (φ2 +4π (Lisjak et al., 2014a)) and is similar to the failure mode of the
and 50 mm, respectively. A pre-prepared failure path with a dip angle of laboratory test result shown in Fig. 4(d) (Wu et al., 2013). The above
60◦ is set to ensure that only shear slip failure along the pre-prepared verification results show that the input parameters shown in Table 1 and
path occurs. The element size and calculation time step are the same the calibrated GI and GII values are reliable.
as those of the Brazilian disc rock sample. The loading rates are set to
0.05 m/s, 0.1 m/s, 0.2 m/s, 0.3 m/s, 0.4 m/s, 0.5 m/s and 0.6 m/s,

6
P. Deng et al. Computers and Geotechnics 161 (2023) 105622

Fig. 3. Parameter calibration simulation results. (a) Failure mode of Brazilian disc, (b) failure mode of uniaxial compression, (c) tensile strength–loading rate curve,
and (d) simulated load-loading rate curve.

3.2. Hysteretic damping coefficient process can be divided into two stages, including an in situ stress loading
stage and a core material softening stage.
(1) Tunnel excavation simulation method and numerical model In the in situ stress loading stage, a pure triangle element without a
quadrilateral joint element with a free boundary is used, while in the
Previous research results (Deng et al., 2021b) have shown that under tunnel excavation simulation stage, quadrilateral joint elements are
the premise of a correct numerical algorithm and input parameters, inserted, and the model boundary is fixed to maintain the in situ stress.
hysteretic damping is a decisive parameter affecting tunnel deformation. To achieve a quasistatic excavation process, the total kinetic energy of
Therefore, it is necessary to calibrate an appropriate hysteresis damping the model must be less than the critical kinetic energy before the next
coefficient β shown in Eq. (28) by using tunnel excavation simulation. softening can be continued. The specific softening path, total softening
The circular tunnel excavation numerical model is shown in Fig. 5 step and critical kinetic energy can be selected according to the authors’
with a model size of 100 × 100 m, which is divided into three zones: a previous research results (Deng and Liu, 2020).
tunnel zone or named core material zone, a mesh refinement zone and a The tunnel excavation numerical model shown in Fig. 5 and the input
far-field zone. The existence of the mesh refinement zone indicates that parameters shown in Table 1 are used, and the horizontal and vertical in
only a small element size is used in this zone, while a larger element size situ stresses are set to 14.38 MPa. It must be noted that GI and GII values
can be used in the far-field zone and gradually transition to the model vary linearly with element size. The GI and GII values with different
boundary to reduce the number of elements and thus improve the element sizes are (Deng et al., 2022)
computational efficiency. The diameters of the tunnel and mesh ⎧
9
refinement zone are 3 m and 30 m, respectively, with a minimum ⎪
⎪
⎨ GI = h⋅
0.002
element size of 0.07 m and a calculation time step of Δt = 7.0 × 10-8 s. (31)
⎪
To reflect the weakening process of the radial support effect acted ⎩ GII = h⋅ 12
⎪
upon by the tunnel face (Farrokh et al., 2006), the core material soft­ 0.002
ening method (Deng and Liu, 2020) can be used; that is, the Young’s 9 12
where the equations of 0.002 and 0.002 represent the GI and GII values
modulus E and viscous damping μ of the tunnel core material can be
corresponding to the element size of 2 mm, as calibrated in Section 3.1.
gradually softened until the core material is completely removed to
By using Eq. (31), the constitutive model of joint element can be indi­
simulate the excavation process of an unsupported state. The simulation
rectly transformed from a stress-displacement relationship to a

7
P. Deng et al. Computers and Geotechnics 161 (2023) 105622

Fig. 4. Uniaxial compression simulation results and laboratory test results. (a) The relation between uniaxial compression strength and loading rate, (b) stress–strain
curve with v = 0.05 m/s, (c) failure mode of rock sample with v = 0.05 m/s, and (d) laboratory test result (Wu et al., 2013).

Fig. 5. Tunnel excavation numerical model.

8
P. Deng et al. Computers and Geotechnics 161 (2023) 105622

stress–strain relationship, which is more suitable for different joint excavation numerical model shown in Fig. 5 is used. The simulation
element sizes (Deng et al., 2022). input parameters are shown in Table 1, and Eq. (31). The fracture
propagation process of the surrounding rock is shown in Fig. 7, which
(2) Calibration result of hysteretic damping coefficient indicates the following:

The hysteresis damping coefficient β shown in Eq. (28) is set from (1) At the early stage of unloading, uniform X-shaped conjugate
0.024 to 0.051 with a spacing of 0.003. The calibration result shown in shear cracks are generated around the tunnel (step 700 k, as
Fig. 6 can be obtained, and the results in Fig. 6(a) indicate that with the shown in Fig. 7(a)), which is very similar to the theoretical
increase in the hysteresis damping coefficient β, the displacement complete shear failure model proposed by Aydan et al. (Aydan
around tunnel u gradually decreases and reaches a stable state at β = et al., 1996) and to the shear fracture morphology of the sur­
0.042. This is because a larger hysteresis damping coefficient can rounding rock obtained from the physical test by Gu et al. (2008),
dissipate the kinetic energy of the model faster. Therefore, the hysteresis as shown in Fig. 8(a) and (b). The generalized model obtained by
damping coefficient β can be taken as 0.042. In addition, it can be Luo et al. (2012) through summarizing physical test results and
concluded in Fig. 6(b) that the fracture modes of the surrounding rock FDM numerical simulation results of tunnel excavation and the
with different hysteresis damping coefficients are similar, which are physical test result of transparent rock mass materials by Li and
mainly shear fractures accompanied by a small number of tensile frac­ Lin (2015) also verify the above fracture mode, although they are
tures, indicating that the hysteresis damping coefficient does not affect not in the hydrostatic in situ stress state, as shown in Fig. 8(c) and
the basic fracture mode of the surrounding rock but has a significant (d).
impact on the surrounding rock displacement u around the tunnel. (2) As the core material continues to soften, conjugate shear cracks
continue to propagate toward the deep surrounding rock. During
4. Failure mechanism of soft rock tunnel the X-shaped conjugate shear crack propagation process, the
angle between the crack and the horizontal plane gradually de­
4.1. Fracture propagation process under high in situ stress creases, as shown in Fig. 7(d), which is consistent with the frac­
ture morphology observed by Armand et al. (Armand et al., 2014)
Taking the strength-stress ratio α = 0.20 as an example, i.e., the from field investigations, although the latter was not a circular
hydrostatic in situ stress σ 0 is 32.35 MPa with a hysteretic damping tunnel or in a state of hydrostatic in situ stress.
coefficient of β = 0.042, as calibrated in Section 3.2. The tunnel

Fig. 6. Calibration simulation results for the hysteretic damping coefficient. (a) Simulation results with different hysteresis damping coefficients, (b) failure mode of
surrounding rock with β = 0.024, 0.042 and 0.051.

9
P. Deng et al. Computers and Geotechnics 161 (2023) 105622

Fig. 7. Fracture development process with a strength-stress ratio of 0.20.

(3) Under hydrostatic in situ stress and circular excavation condi­ two main reasons: the macroscopic movement of rock fragments
tions, the final fracture network morphology is circular, as shown induced by contact and squeezing between fragments and the volume
in Fig. 7(e), which is consistent with the theoretical model shown expansion of broken rocks induced by generating a large number of
in Fig. 8(a). The above comparison results indicate that the macroscopic voids.
simulation results in Fig. 7 are reliable. It should be noted that to obtain the final maximum displacement of
the surrounding rock, the simulation process completes only after the
However, most of the previous studies could only observe the final core material has been removed and the model has stabilized. Both
fracture morphology of the surrounding rock, and it was difficult to criteria are used to determine whether the model has reached a stable
characterize the volume expansion mechanism caused by rock failure state: (1) there are no new cracks generated in the model; (2) the kinetic
and the macroscopic movement of rock fragments. In this section, the energy of the model is small enough. As shown in Fig. 10, the failure
FDEM simulation of tunnel excavation not only reproduces the crack degree η of the surrounding rock (the ratio of the number of broken joint
network propagation process but also simulates the volume expansion elements to the total number of joint elements) has reached a stable
effect caused by fracture, as well as the macroscopic movement of rock state, and the model kinetic energy is sufficiently small. In all subse­
fragments. quent simulations, these two indicators are used to determine whether
The final displacement field is shown in Fig. 9(a), indicating that the model has reached a stable state.
under the condition of a hydrostatic in situ stress of 32.35 MPa, a In addition, according to the simulation results shown in Fig. 10, the
maximum displacement of 0.564 m of surrounding rock occurred. As the crack propagation process can be divided into two stages: (1) before the
core material softens, the newly generated rock fragments in the deep removal of the core material, the crack network uniformly propagates
surrounding rock contact and squeeze against the shallow fragments, toward the depth of the surrounding rock, and the model kinetic energy
causing them to move into the tunnel space and thus causing a reduction is stable at approximately 25 kJ; (2) after the removal of the core ma­
in the tunnel cross-sectional area, as shown in Fig. 9(b). terial, the propagation rate of the crack network gradually decreases
In addition, taking zone A shown in Fig. 9(a) as an example, the until no new cracks are generated; furthermore, the model kinetic en­
deformation or movement process of the broken rock fragments around ergy gradually reaches a stable state under the action of critical hys­
the tunnel surface is shown in Fig. 9(c): the surrounding rock around the teresis damping. The rock fragments move slowly toward the tunnel
tunnel surface undergoes a whole process of elastic deformation re­ space under the critical damping effect, avoiding a dynamic response.
covery, crack initiation, shear dilation, and broken expansion. Through the critical damping effect, a progressive deformation process
Compared to intact rock, a large number of macroscopic voids are simulation of soft rock tunnels is achieved.
generated between fragments, as shown in Fig. 9(c), resulting in an in­
crease in rock volume and a decrease in tunnel cross-sectional area. That
is, the reduction in the tunnel cross-sectional area is mainly caused by

10
P. Deng et al. Computers and Geotechnics 161 (2023) 105622

Fig. 8. Previous research results of the fracture morphology of the surrounding rock. (a) Theoretical model by Aydan et al. (1996), (b) physical test result by Gu et al.
(2008), (c) generalized model by Luo et al. (2012), and (d) physical test result by Li and Lin (2015).

4.2. Stress field 5. Deformation modes and prediction model with different in
situ stresses
In polar coordinates, the radial and tangential stresses of the sur­
rounding rock are calculated by using Eq. (32), as 5.1. Deformation modes
{
σr = σxx cos2 θ + σ yy sin2 θ + τxy sin2θ The hydrostatic in situ stresses are respectively set to 3.235 MPa, 3.7
(32)
σθ = σ xx sin2 θ + σ yy cos2 θ − τxy sin2θ
MPa, 4.313 MPa, 5.176 MPa, 6.47 MPa, 8.088 MPa, 10.78 MPa, 14.378
MPa, 21.567 MPa, 43.133 MPa and 64.7 MPa, i.e., the strength-stress
where σr and σθ are radial and tangential stresses, respectively; σxx, σyy
ratios α are 2.0, 1.75, 1.5, 1.25, 1.0, 0.8, 0.6, 0.45, 0.3, 0.2 (Section
and τxy are normal and tangential stresses in Cartesian coordinates; and
4), 0.15 and 0.1. In addition, the hysteresis damping coefficient β is
θ is the angle between polar coordinates and the x-axis.
0.042, as investigated in Section 3.2. The other conditions are the same
As shown in Fig. 11, for radial stress, the surrounding rock around
as in Section 4. The simulation results of surrounding rock failure or
the tunnel undergoes a gradual unloading process. However, within the
tunnel deformation with different in situ stresses are shown in Fig. 13.
excavation damage zone (EDZ), the tangential stress at any point un­
According to the simulation results in Fig. 13, with the increase in in
dergoes a process of increasing and then decreasing. The maximum
situ stress or the decrease in the strength-stress ratio, the surrounding
concentrated tangential stress is always located at the crack tip. The
rock presents different failure or deformation modes, or the decreases in
mechanical model of the tunnel surrounding rock can be simplified as a
tunnel cross-sectional area are due to different main factors:
triaxial compression dynamic model with unloading confining pressure
and increasing axial pressure, as shown in Fig. 11(c) and (d). When the
(1) For α = 2.0 and α = 1.75, the surrounding rock only undergoes
increased tangential stress exceeds the strength under the same
elastic–plastic deformation without fracturing, as shown in
confining pressure, the surrounding rock mass immediately fails. It is
Fig. 13(a) and (b).
under this triaxial compression that X-shaped conjugate shear cracks are
(2) For α = 1.5, α = 1.25 and α = 1.0, the maximum concentrated
generated, which has been verified by a large number of triaxial
tangential stress exceeds the strength of the rock mass, causing
compression test results, as shown in Fig. 12.
the surrounding rock to fail. However, most of the cracks are in a
closed state without a significant shear dilation effect (Fig. 14(a)),

11
P. Deng et al. Computers and Geotechnics 161 (2023) 105622

Fig. 9. Displacement field and movement process. (a) Displacement field, (b) contact and squeezing between rock fragments, and (c) volume expansion process of
the broken rock fragments around the tunnel surface.

Fig. 10. The criterion for the end of the simulation process. (a) Failure degree of the surrounding rock, (b) model kinetic energy.

resulting in relatively small tunnel deformation, as shown in (3) For α = 0.8, α = 0.6 and α = 0.45, the decrease in the tunnel cross-
Fig. 13(c)–(e). The closed fracturing mode is similar to the lab­ sectional area is mainly caused by the sliding of broken rock
oratory test result shown in Fig. 8(d) (Li and Lin, 2015). along the main shear zone and its induced shear dilation effect
(Fig. 14(b)), as shown in Fig. 13(f)–(h). The shear dilation failure

12
P. Deng et al. Computers and Geotechnics 161 (2023) 105622

Fig. 11. The stress fields at different time steps. (a) Radial stress fields at step 700 k and (b) step 4000 k, (c) tangential stress fields at step 700 k and (d) step 4000 k.

mode is similar to the laboratory test result shown in Fig. 8(b) (Gu
et al., 2008).
(4) For α = 0.3, α = 0.2, α = 0.15 and α = 0.1, the rock mass un­
dergoes comminuted fracture under high concentrated tangential
stress, making it difficult to form a main shear zone, as shown in
Fig. 14(c). The broken expansion effect and the macroscopic
movement of rock fragments result in a decrease in the tunnel
cross-sectional area, as shown in Fig. 13(i)–(l). The comminuted
failure and broken expansion effect of the rock mass around the
tunnel surface under high stress are consistent with those
observed in actual tunnel engineering, as shown in Fig. 15 (Fu
et al., 2019).

Therefore, the surrounding rock can be divided into four different

deformation or failure modes, or the tunnel deformation is mainly
caused by four different main driving factors based on different strength-
stress ratios: elastic–plastic deformation (α ≥ 1.75), closed fracturing
(1.0 ≤ α < 1.75), shear dilation (1.0 < α ≤ 0.45) and broken expansion
(α < 0.45). The maximum displacement, strain and failure mode of the
surrounding rock with different strength-stress ratios are shown in
Table 2.
Fig. 12. Triaxial compression laboratory test results. (a) By Zhu et al. (2022),
(b) by Ren et al. (2022). 5.2. Prediction model

Section 5.1 only investigated the soft rock tunnel deformation mode
under specific circumstances, including specific element size (h = 0.07

13
P. Deng et al. Computers and Geotechnics 161 (2023) 105622

Fig. 13. Simulation results of surrounding rock failure or tunnel deformation with different strength-stress ratios.

m), tunnel diameter (D = 3.0 m), and rock parameters (mainly E = 5 of the failure mode of the surrounding rock or the main factors affecting
GPa, ν = 0.25 and internal friction coefficient tanφ = 0.5, which may tunnel deformation is similar to Section 5.1, i.e., from elastic-plastic
influence the fracture mode of the surrounding rock). Therefore, this deformation to closed fracturing and then to shear dilation and finally
section will study the influence of the above parameters on the tunnel to broken expansion. In addition, the conversion threshold of different
deformation mode and further obtain the prediction equation for tunnel deformation modes is also similar to that in Section 5.1. The details of
deformation. the maximum displacement of the surrounding rock with different
strength-stress ratios will be described later. The simulation results in
(1) Element size and tunnel diameter this section indicate that the simulation results are reliable and stable
within the allowable element size range.
The element size h around the tunnel is set to 0.1 m with a tunnel
diameter of 3 m. In addition, the tunnel diameter D is set to 5 m with an (2) Young’s modulus and Poisson’s ratio
element size of 0.1 m. Other conditions are the same as those in Sections
4 and 5.1, including the hysteretic damping coefficient (β = 0.042) and The Young’s modulus and Poisson’s ratio are set to 2.5 GPa and 0.35,
strength-stress ratio interval (α = 0.1, 0.15, 0.2, 0.3, 0.45, 0.6, 0.8, 1.0, respectively. The in situ stress settings are the same as those described in
1.25, 1.5, 1.75 and 2.0). Section 5.1. It should be noted that for a Young’s modulus E = 2.5 GPa,
Partial simulation results with different excavation diameters are the corresponding critical hysteresis damping coefficient needs to be
shown in Figs. 16 and 17. It can be concluded that with the increase in in recalibrated. Using a calibration method similar to that in Section 3.2, a
situ stress or the decrease in the strength-stress ratio, the transformation critical hysteresis damping coefficient of 0.18 is obtained, as shown in

14
P. Deng et al. Computers and Geotechnics 161 (2023) 105622

Fig. 14. Simulation details with different strength-stress ratios.

Fig. 18. In addition, using the same calibration method as Section 3.1,
the mode I and mode II fracture energies with an element size of 2 mm
are obtained to be 18 J/m2 and 24 J/m2, respectively. The uniaxial
compressive strength of the rock sample remains unchanged, i.e., σ c =
6.47 MPa.
According to the partial simulation results shown in Figs. 19 and 20,
it can be concluded that with different Young’s moduli and Poisson’s
ratios, the transformation of the surrounding rock failure mode is
consistent with the research results in Section 5.1.
The detailed strain values of the tunnel under different conditions are
shown in Table 3. In addition, under the same strength-stress ratio, the
average strain with different conditions can also be obtained as

1∑ n
ε= εi (33)
n i=1
Fig. 15. A special example of large deformation caused by broken expansion of
a soft rock tunnel (Fu et al., 2019). where ε is the average strain at the same strength-stress ratio, n = 5, and
εi is a special strain value under different conditions shown in Table 3

Table 2
The maximum displacement and strain with different strength-stress ratios.
Strength-stress ratio, α 0.1 0.15 0.2 0.3 0.45 0.6 0.8 1.0 1.25 1.5 1.75 2.0

The maximum displacement, u (m) 0.993 0.7887 0.564 0.37 0.227 0.121 0.0718 0.037 0.01 0.0065 0.00146 0.00128
The maximum strain, ε (%) 66.2 52.58 37.6 24.67 15.13 8.067 4.787 2.467 0.667 0.433 0.097 0.085
Failure or Deformation mode Broken expansion Shear dilation Closed fracturing Elastic-plastic deformation

Fig. 16. Partial simulation results with an excavation diameter of 3.0 m and element size of 0.1 m.

15
P. Deng et al. Computers and Geotechnics 161 (2023) 105622

Fig. 17. Partial simulation results with an excavation diameter of 5.0 m.

strain value and the average value is within 10%. Therefore, on the one
hand, it can be proven that the aforementioned simulation results are
stable, especially for the large deformation caused by the broken
expansion effect; on the other hand, the average strain shown in Table 3
can be used to fit the tunnel strain prediction equation to eliminate
accidental errors in the simulation results.

(3) Internal friction coefficient

The internal friction coefficient is set to 0.7, and the strength-stress

ratios are set to 0.125, 0.175, 0.25, 0.35, 0.55, 0.7, 0.9, 1.1, 1.3, 1.6
and 1.8 to further refine the strength-stress ratio spacing to improve
fitting accuracy. The uniaxial compression strength is 7.62 MPa, and the
other parameters are the same as those in Section 5.1. The tunnel strains
when tanφ = 0.7 with different strength-stress ratios are shown in
Table 4.
Combining the average strain in Table 3 with the simulation results
in Table 4 and considering the left boundary condition (when α = 0 and
ε = 100%, as described in Section 1), a tunnel strain fitting equation can
Fig. 18. Simulation results with different hysteresis damping coefficients when be obtained, as shown in Fig. 22. It can be concluded that the tunnel
E = 2.5 GPa and strength-stress ratio α = 0.45. strain decreases exponentially with an increase in the strength-stress
ratio, with a correlation coefficient of 0.997, which is consistent with
and Fig. 21. The average strain is also shown in Table 3. the fitting form proposed by Shao et al. (2021). In fact, when the extreme
In Table 3 and Fig. 21, it can be concluded that under the same boundary condition of α = 0 is conceded, the exponential function is
strength-stress ratio, the tunnel strain values simulated with different obviously more appropriate than the power function (Hoek and Mar­
conditions are relatively concentrated; especially under low strength- inos, 2000) because for the power function, when α = 0, the function
stress ratios, such as α ≤ 0.6, the error between the special simulation loses its meaning, and for the power function, the smaller the value of α

Fig. 19. Partial simulation results with E = 2.5 GPa.

16
P. Deng et al. Computers and Geotechnics 161 (2023) 105622

Fig. 20. Partial simulation results with ν = 0.35.

Table 3
The detailed strain values and their average values of tunnel surface surrounding rock under different conditions.
Strength-stress ratio, α 0.1 0.15 0.2 0.3 0.45 0.6 0.8 1.0 1.25 1.5 1.75 2.0

Section 5.1 66.2 52.58 37.6 24.67 15.13 8.067 4.787 2.467 0.667 0.433 0.097 0.085
h = 0.1 m 66.67 52.78 38.23 25.94 16.59 8.85 5.86 2.11 0.58 0.307 0.0975 0.0852
D=5m 68 52.76 37.22 24.29 14.14 8.19 4.98 1.69 0.675 0.42 0.0964 0.085
E = 2.5 GPa 64.89 51.98 40.63 26.09 14.83 9.06 4.18 2.46 0.864 0.516 0.138 0.104
ν = 0.35 68.14 51.72 39.66 25.78 15.35 9.25 4.66 1.99 0.539 0.306 0.1 0.088
Average strain ε 66.78 52.36 38.67 25.35 15.21 8.68 4.89 2.14 0.665 0.396 0.106 0.089

intervals mentioned in Table 3 are also applicable in this section, as

shown in Table 4 and Fig. 23. Taking α = 1.1, α = 0.55 and α = 0.125 as
examples, the surrounding rock reflects three different failure modes:
closed fracturing, shear dilation and broken expansion, as shown in
Fig. 23.

5.3. Comparison with previous results

The FDEM numerical simulation method was used in the afore­

mentioned studies. Although the fracture morphology of tunnel

Fig. 21. The detailed strain values of the tunnel surface surrounding rock under
different conditions.

is, the greater the error, as discussed by Chen et al. (2019). The fitting
equation is
ε(%) = 100 × Bα (34)

where B is the fitting parameter and B = 0.00977 according to Fig. 22.

In addition, the different failure or deformation modes of the sur­
Fig. 22. Tunnel strain fitting curve.
rounding rock corresponding to the different strength-stress ratio

Table 4
Tunnel strain with tanφ = 0.7 at different strength-stress ratios.
Strength-stress ratio, α 0.125 0.175 0.25 0.35 0.55 0.7 0.9 1.1 1.3 1.6 1.8

Tunnel strain, ε (%) 57.20 44.04 28.55 19.61 9.10 6.07 2.16 1.27 0.51 0.125 0.111
Failure or Deformation mode Broken expansion Shear dilation Closed fracturing Elastic-plastic deformation

17
P. Deng et al. Computers and Geotechnics 161 (2023) 105622

Fig. 23. Partial simulation results with tanφ = 0.7.

surrounding rock can be consistent with physical test results or tunnel

field observations, specific deformation values of the surrounding rock
need to be verified, that is, the accuracy of the tunnel deformation
prediction equation shown in Eq. (34) must be verified by comparison
with deformation monitoring results from actual tunnel engineering.
It must be noted that the total displacement of the surrounding rock
is obtained by FDEM numerical modeling, including the pre-generated
deformation in front of the tunnel face, while field monitoring can
generally only observe deformation behind the tunnel face. The longi­
tudinal deformation profile (LDP) of the surrounding rock is shown in
Fig. 24, which indicates that the displacement obtained from the FDEM
numerical simulation is u0, and the displacement monitored on the en­
gineering field is u1 or even u2, which depends on the placement of
monitoring points. According to previous research results (Cui et al.,
2014; Vlachopoulos and Diederichs, 2009), u1 is generally 80% of u0, i.
e., u1≈0.8u0. Therefore, when comparing the monitoring results with
the predicted results, it is necessary to multiply the predicted results
from Eq. (34) by 80%.
According to the comparison results shown in Fig. 25, it can be
concluded that when the strength-stress ratio is relatively high, such as Fig. 25. Comparison between the prediction result in this paper and the pre­
vious prediction results.
α > 0.6, the prediction results are consistent well with the monitoring
results collected by Chen (Chen et al., 2019), Yang (Yang et al., 2022)
and Meng (Meng et al., 2022) et al.; however, when the strength-stress
ratio is relatively small, such as α < 0.6, the prediction results are
generally larger than the monitoring results. Actually, the above com­
parison results are reliable and reasonable because on the one hand,
monitoring points are generally difficult to arrange at the tunnel face,
such as P1 shown in Fig. 24, but lag behind the tunnel face, such as P2
shown in Fig. 24; therefore, the monitoring results tend to be more
conservative than u1. In fact, most of the monitoring data derive from u2
and not u1, as shown in Fig. 24. On the other hand, more importantly,

Fig. 26. Tunnel support measures (Hoek and Marinos, 2000).

actual tunnel engineering must often adopt some support measures, as

shown in Fig. 26 (Hoek and Marinos, 2000), while the FDEM numerical
simulation in this paper does not consider the support effect. Especially
under low strength-stress ratios, support measures are indispensable,
and the support strength is higher; thus, the support effect is better.
Fig. 24. The relationship between the simulated displacement and monitored Therefore, the monitoring results are inevitably smaller than the
displacement.

18
P. Deng et al. Computers and Geotechnics 161 (2023) 105622

prediction results under unsupported conditions. situ stress, that is, the lateral pressure coefficient λ is equal to or
In addition, Fig. 25 once again proves the conservatism of the pre­ close to 1.0;
vious prediction equations. For example, the prediction results by Eqs. (3) This paper simulates the failure mode and deformation prediction
(9) and (13) shown in Fig. 25 are much smaller than the actual moni­ of isotropic and homogeneous rock masses, which may not be
toring results. This paper can predict the maximum displacement of a applicable to anisotropic or heterogeneous rock masses;
soft rock tunnel in an unsupported state, and improved the prediction (4) The large deformation mechanism of soft rock in this paper is
accuracy, which is extremely important for designing appropriate sup­ mainly caused by stress driving. The surrounding rock in this
port measures in advance. paper is assumed to be dry and intact. The expansion and soft­
ening effects of water on rock, as well as the influence of
6. Limitation analysis and discussion macroscopic structural planes were not considered. When
considering the above effects, the failure mode classification and
This paper reveals the deformation or failure modes of soft rock deformation prediction equation obtained in this paper are not
tunnels with different strength-stress ratios by using FDEM numerical suitable. In addition, the time-dependence of rock mass strength
simulations. In addition, according to the displacement values of sur­ was also not considered.
rounding rock with different strength-stress ratios, a prediction equation (5) The scale of the object simulated in this paper is the size of
for tunnel surrounding rock deformation is obtained through data general tunnel excavation. For small-size excavation (such as
fitting, and the prediction model is verified by comparison with the boreholes) or large-size excavation (such as coal mining stopes),
monitoring data from actual tunnel engineering. Although research on the surrounding rock failure mode and deformation prediction
different factors (element size, tunnel diameter, Young’s modulus, equation in this paper are not applicable.
Poisson’s ratio and internal friction coefficient) has confirmed the
robustness of the simulation results, there are still some limitations as 7. Conclusion
follows:
The FDEM numerical simulation method for the progressive large
(1) The maximum Young’s modulus in this paper is 5 GPa, and the deformation process of soft rock tunnels was proposed in this paper, and
maximum uniaxial compressive strength is 7.62 MPa, which in­ this method was used to study the deformation or failure mechanism and
dicates a soft rock range. However, if the rock parameters (mainly modes of surrounding rock with different strength-stress ratios. Ac­
including Young’s modulus and compressive strength) continue cording to the displacement values of the surrounding rock with
to be increased, the failure mode and deformation prediction different strength-stress ratios, a deformation prediction equation was
equation obtained in this paper may not be applicable, as it may obtained through data fitting. The specific conclusions are as follows:
transform into dynamic failure, such as rock burst (Han et al.,
2023); (1) A progressive large deformation process simulation of soft rock
(2) The numerator and denominator of the strength-stress ratio α = σσ0c tunnels can be realized by adopting critical hysteresis damping,
represent the rock uniaxial compressive strength and hydrostatic which avoids a dynamic response;
in situ stress, respectively. σ c is the intact rock strength consid­ (2) The basic failure mechanism of tunnel soft surrounding rock is X-
ering the rock size effect, as shown Fig. 27(a), or the rock mass shaped conjugate shear fracture induced by concentrated
strength considering the structural plane effect and size effect, i. tangential stress; the reduction in tunnel cross-sectional area is
e., σ∗c . The rock mass strength in this paper has taken into account mainly caused by the macroscopic movement of rock fragments
the size effect, and the rock mass structural planes are the random and the volume expansion of fractured rock masses induced by
structural planes that follow statistical laws, rather than a specific generating a large number of macroscopic voids.
large macroscopic structural plane, such as a fault, as shown (3) As the strength-stress ratio decreases, the deformation or failure
Fig. 27(b). The rock mass strength σ∗c can be obtained by labo­ modes of soft tunnel surrounding rock are elastic–plastic defor­
ratory/field tests or using empirical equations proposed by Hoek mation, closed fracturing, shear dilation and broken expansion.
et al (Hoek and Brown, 2019). In addition, σ0 is the hydrostatic in (4) According to the displacement values of the surrounding rock
with different strength-stress ratios, a prediction equation for soft

Fig. 27. The size effect of rock strength and rock mass strength. (a) Intact rock sample and (b) rock mass.

19
P. Deng et al. Computers and Geotechnics 161 (2023) 105622

surrounding rock deformation is obtained by data fitting. With Deng, P.H., Liu, Q.S., 2020. Influence of the softening stress path on crack development
around underground excavations: Insights from 2D-FDEM modelling. Comput.
the increase in the strength-stress ratio, the maximum displace­
Geotech. 117, 1–18.
ment of the surrounding rock decreases as an exponential func­ Deng, P.H., Liu, Q.S., Huang, X., Liu, Q., Ma, H., Li, W.W., 2021a. Acquisition of normal
tion, and the correlation coefficient is R2 = 0.997. Compared to contact stiffness and its influence on rock crack propagation for the combined finite-
previous field monitoring results, the fitting equation in this discrete element method (FDEM). Eng. Fract. Mech. 242, 107459.
Deng, P.H., Liu, Q.S., Huang, X., Ma, H., 2021b. A new hysteretic damping model and
paper can predict the maximum possible deformation value of application for the combined finite-discrete element method (FDEM). Eng. Anal.
unsupported tunnels. In addition, compared to previous predic­ Bound. Elem. 132, 370–382.
tion models, the prediction equation in this paper considers the Deng, P.H., Liu, Q.S., Huang, X., Pan, Y.C., Wu, J., 2021c. FDEM numerical modeling of
failure mechanisms of anisotropic rock masses around deep tunnels. Comput.
left boundary condition of strength-stress ratio α = 0, improving Geotech. 104535.
the fitting accuracy, especially for the low strength-stress ratio Deng, P.H., Liu, Q.S., Lu, H.F., 2022. A novel joint element parameter calibration
range. procedure for the combined finite-discrete element method. Eng. Fract. Mech. 276,
108924.
Deng, P.H., Liu, Q.S., Lu, H.F., 2023. FDEM numerical study on the mechanical
However, the prediction equation obtained in this paper has limi­ characteristics and failure behavior of heterogeneous rock based on the Weibull
tations: the prediction model is only applicable to the deformation distribution of mechanical parameters. Comput. Geotech. 154, 105138.
Ding, X.L., Zhang, Y.T., Huang, S.L., Chi, J.J., Zhang, C.J., Liu, D.X., 2023. Mechanism of
prediction of tunnels with a general size in a state of hydrostatic in situ large deformation of surrounding rock masses of tunnels, prediction method of
stress or near hydrostatic in situ stress, and the surrounding rock must be squeezing large deformation and its application. Chin. J. Rock Mech. Eng. 42, 1–24.
assumed to be homogeneous and isotropic; in addition, the effects of the Farrokh, E., Mortazavi, A., Shamsi, G., 2006. Evaluation of ground convergence and
squeezing potential in the TBM driven Ghomroud tunnel project. Tunn. Undergr.
hydraulic expansion effect and macroscopic structural planes were not
Space Technol. Incorpor. Trench. Technol. Res. 21, 504–510.
considered. Fu, Y.K., Sun, Z.Y., Ju, W.J., 2019. Experimental study on static and dynamic mechanical
properties of bolting wire mesh in rock burst roadway. J. China Coal Soc.
2020–2029.
CRediT authorship contribution statement Gu, J.C., Gu, L.Y., Chen, A.M., Xu, J.M., Chen, W., 2008. Model test study on mechanism
of layered fracture within surrounding rock of tunnels in deep stratum. Chin. J. Rock
Mech. Eng. 433–438.
Penghai Deng: Methodology, Software, Writing – original draft. Guo, L.W., Latham, J.P., Xiang, J.S., 2015. Numerical simulation of breakages of concrete
Quansheng Liu: Conceptualization, Supervision. Bin Liu: Visualiza­ armour units using a three-dimensional fracture model in the context of the
tion. Haifeng Lu: Data curation, Investigation. combined finite-discrete element method. Comput. Struct. 117–142.
Han, H.Y., Fukuda, D., Xie, J.B., Fathi Salmi, E., Sellers, E., Liu, H.Y., An, H.M., Chan, A.,
2023. Rock dynamic fracture by destress blasting and application in controlling
Declaration of Competing Interest rockbursts in deep underground. Comput. Geotech. 155, 105228.
Hoek, E., Brown, E.T., 2019. The Hoek–Brown failure criterion and GSI – 2018 edition.
Journal of Rock Mechanics and Geotechnical Engineering 11, 445-463.
The authors declare that they have no known competing financial Hoek, E., Marinos, P., 2000. Predicting tunnel squeezing problems in weak
interests or personal relationships that could have appeared to influence heterogeneous rock masses. Tunn. Tunn. Int. 32, 45–51.
Jimenez, R., Recio, D., 2011. A linear classifier for probabilistic prediction of squeezing
the work reported in this paper. conditions in Himalayan tunnels. Eng. Geol. 121, 101–109.
Lei, J., 2015. Comparative study on size effect of rock tensile and compressive strength
Data availability under different strain rates. Nanjing University, Nanjing.
Li, Y.H., Lin, Z.B., 2015. Innovative experimental method based on development of
transparent rock mass materials for physical tests. Chin. J. Geotech. Eng. 37,
No data was used for the research described in the article. 2030–2039.
The authors do not have permission to share data. Lisjak, A., Figi, D., Grasselli, G., 2014a. Fracture development around deep underground
excavations: insights from FDEM modelling. J. Rock Mech. Geotech. Eng. 6,
493–505.
Acknowledgments Lisjak, A., Grasselli, G., Vietor, T., 2014b. Continuum-discontinuum analysis of failure
mechanisms around unsupported circular excavations in anisotropic clay shales. Int.
J. Rock Mech. Min. Sci. 65, 96–115.
The work in this paper is based on the Y-Code proposed by Munjiza Lisjak, A., Garitte, B., Grasselli, G., Müller, H.R., Vietor, T., 2015. The excavation of a
et al. and the Y-Geo and Y-GUI proposed by Grasselli’s Geomechanics circular tunnel in a bedded argillaceous rock (Opalinus Clay): Short-term rock mass
Group. The authors received support from the Natural Science Foun­ response and FDEM numerical analysis. Tunnel. Undergr. Space Technol. Incorpor.
Trenchless Technol. Res. 227–248.
dation of China (42107171, U21A20153 and 41941018). In addition, Lisjak, A., Tatone, B.S.A., Mahabadi, O.K., Grasselli, G., Nussbaum, C., 2016. Hybrid
this paper was also supported by the State Key Laboratory for Geo­ finite-discrete element simulation of the EDZ formation and mechanical sealing
Mechanics and Deep Underground Engineering, China University of process around a microtunnel in opalinus clay. Rock Mech. Rock Eng. 49,
1849–1873.
Mining & Technology, Beijing (SKLGDUEK2218).
Lisjak, A., Young-Schultz, T., Li, B., He, L., Tatone, B.S.A., Mahabadi, O.K., 2020. A novel
rockbolt formulation for a GPU-accelerated, finite-discrete element method code and
References its application to underground excavations. Int. J. Rock Mech. Min. Sci. 134,
104410.
Liu, Q.S., Deng, P.H., 2019. A numerical investigation of element size and loading/
Armand, G., Leveau, F., Nussbaum, C., de La Vaissiere, R., Noiret, A., Jaeggi, D.,
unloading rate for intact rock in laboratory-scale and field-scale based on the
Landrein, P., Righini, C., 2014. Geometry and properties of the excavation-induced
combined finite-discrete element method. Eng. Fract. Mech. 442–462.
fractures at the Meuse/Haute-Marne URL drifts. Rock Mech. Rock Eng. 47, 21–41.
Liu, Q.S., Xing, H., Kai, S., Liu, X.W., 2013. Jamming mechanism of full face tunnel
Aydan, Ö., Akagi, T., Kawamoto, T., 1993. The squeezing potential of rocks around
boring machine in over thousand-meter depths. J. China Coal Soc. 38, 78–84.
tunnels: theory and prediction. Rock Mech. Rock Eng. 26, 137–163.
Liu, Q.S., Deng, P.H., Bi, C., Li, W.W., Liu, J., 2019. FDEM numerical simulation of the
Aydan, Ö., Akagi, T., Kawamoto, T., 1996. The squeezing potential of rock around
fracture and extraction process of soft surrounding rock mass and its rockbolt-
tunnels: Theory and prediction with examples taken from Japan. Rock Mech. Rock
shotcrete-grouting reinforcement methods in the deep tunnel. Rock Soil Mech. 40,
Eng. 29, 125–143.
4065–4083.
Cai, M.F., He, M.C., Liu, D.Y., 2002. Rock Mechanics and Engineering. China Science
Liu, Z.C., Zhu, Y.Q., Wen-Jiang, L.I., Liu, P.X., 2008. Mechanism and classification
Publishing & Media Ltd., Beijing.
criterion for large deformation of squeezing ground tunnels. Chin. J. Geotech. Eng.
Chen, Z.Q., He, C., Wu, D., Dai, C., Yang, W.B., Xu, G.W., 2018. Study of large
30, 690–697.
deformation classification criterion for layered soft rock tunnels under high
Luo, F., Yang, B.S., Sun, L.H., Yang, W.B., Cui, J.K., Zhang, L., 2012. Experimental
geostress. J. Southwest Jiaotong Univ. 53, 1237–1244.
research on the failure characteristics of surrounding rock under high vertical
Chen, W.Z., Tian, Y., Wang, X.H., Tian, H.M., Cao, H.X., Xie, H.D., 2019. Squeezing
ground stress. J. Min. Saf. Eng. 497–504.
prediction of tunnel in soft rocks based on modified [BQ]. Rock Soil Mech. 40,
Mahabadi, O.K., 2012. Investigating the influence of micro-scale heterogeneity and
3125–3134.
microstructure on the failure and mechanical behaviour of geomaterials. University
Cui, L., Zheng, J., Miao, C., Dong, Y., 2014. Coupling analysis of longitudinal
of Toronto, Toronto, Canada.
deformation profile and ground reaction curve. Chin. J. Geotech. Eng. 36, 707–715.
Meng, L.B., Huang, Y.L., Li, T.B., Chen, B., Zhang, W.J., Chen, H.Q., Li, H.Y., 2022. An
Deng, P.H., Liu, Q.S., Ma, H., He, F., Liu, Q., 2020. Time-dependent crack development
improved classification method of asymmetrical squeezing large deformation of
processes around underground excavations. Tunn. Undergr. Space Technol. 103,
103518.

20
P. Deng et al. Computers and Geotechnics 161 (2023) 105622

layered soft rock tunnels under high geo-stresses. Chin. J. Rock Mech. Eng. 41, Vlachopoulos, N., Vazaios, I., 2018. The numerical simulation of hard rocks for
147–156. tunnelling purposes at great depths: a comparison between the hybrid FDEM method
Miglietta, P.C., Bentz, E.C., Grasselli, G., 2017. Finite/discrete element modelling of and continuous techniques. Adv. Civil Eng. 1–18.
reversed cyclic tests on unreinforced masonry structures. Eng. Struct. 159–169. Wu, B.T., Zhu, H.H., Xu, Q.W., Ming, J., 2013. Experimental study of similar material for
Munjiza, A., 2004. The Combined Finite-Discrete Element Method. John Wiley & Sons weak surrounding rock mass of class IV. Rock Soil Mech. 34, 109–116.
Ltd, London, UK. Yang, W.B., Wang, Z.X., Tian, H.S., Wu, F.Y., Yang, Z.C., 2022. Large deformation
Ren, H., Zhu, Y.J., Wang, P., Yu, W.J., Li, P., 2022. Experimental study on the law of rock prediction method of layered soft rock tunnel based on PSO-SVM algorithm. Hazard
damage evolution under the effect of roadway excavation. Chin. J. Undergr. Space Control Tunnel. Undergr. Eng. 4, 29–37.
Eng. 18, 1615–1622. Yao, G.S., Li, J.P., Gu, S.C., 2009. Analytic solution to deformation of soft rock tunnel
Shao, J.J., Liu, Z.C., Sun, M.L., 2021. Research on deformation calculation method of considering dilatancy and plastic softening of rock mass. Rock Soil Mech. 30,
squeezing rock tunnel based on deformation classification. Traffic Eng. Technol. 463–467.
National Defence 19, 13–17. Zhang, Z.D., 2003. Discussion and study on large deformation of tunnel in squeezing
Singh, M., Singh, B., Choudhari, J., 2007. Critical strain and squeezing of rock mass in ground. Modern Tunnel. Technol. 40, 5–13.
tunnels. Tunnel. Undergr. Space Technol. Incorp. Trench. Technol. Res. 22, Zhang, P.Y., Zhang, X.P., Zhang, H., Cui, J.H., Liu, Q.S., Fan, W.S., Li, Y.S., 2022.
343–350. Applicability evaluation of predicton method for large deformation of soft
Tatone, B., 2014. Investigating the Evolution of Rock Discontinuity Asperity Degradation surrounding rock in lianh. J. Eng. Geol. 1689–1702.
and Void Space Morphology Under Direct Shear. University of Toronto. Zhang, C.Q., Zhou, H., Zhu, Y., Yang, F.J., 2016. An evaluation method of squeezing
Tatone, B.S.A., Grasselli, G., 2015. A calibration procedure for two-dimensional degree based on potential squeezing ratio. Chin. J. Rock Mech. Eng. 939–947.
laboratory-scale hybrid finite–discrete element simulations. Int. J. Rock Mech. Min. Zhu, Y.J., Ren, H., Wang, P., Li, P., Wang, X.Z., Wei, M.X., 2022. Grouting test and
Sci. 56–72. reinforcement mechanism analysis of rock with single penetrated fracture surface.
Vlachopoulos, N., Diederichs, M.S., 2009. Improved longitudinal displacement profiles Rock Soil Mech. 43, 3221–3230.
for convergence confinement analysis of deep tunnels. Rock Mech. Rock Eng. 42,
131–146.

21