Tunnelling and Underground Space Technology 132 (2023) 104881

Contents lists available at ScienceDirect

Tunnelling and Underground Space Technology

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

A novel hybrid model to design fiber-reinforced shotcrete for tunnel linings

Marcello Congro a, b, *, Vitor Moreira de Alencar Monteiro a, Flávio de Andrade Silva a,
Deane Roehl a, b, Amanda L.T. Brandão c
a
Department of Civil and Environmental Engineering, Pontifical Catholic University of Rio de Janeiro, Rua Marquês de São Vicente 225, Gávea – Rio de Janeiro, Brazil
b
Multiphysics Modeling and Simulation Group, Tecgraf Institute/PUC-Rio, Rua Marquês de São Vicente, 225 Gávea – Rio de Janeiro, Brazil
c
Department of Chemical and Materials Engineering (DEQM), Pontifical Catholic University of Rio de Janeiro, Rua Marquês de São Vicente 225, Gávea – Rio de Janeiro,
Brazil

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

Keywords: This article proposes a workflow to design fiber-reinforced shotcrete (FRS) for tunnel linings. The workflow is
Fiber reinforced shotcrete divided into two major stages: artificial neural network (ANN) development for predicting the composite
Artificial neural networks toughness, and tunnel lining design through numerical modeling. First, an artificial neural network was devel­
Tunnel linings
oped to predict FRS toughness from ASTM C1550 and EN 14488–5 panels, with concrete strength and fiber
Steel fibers
Polypropylene fibers
characteristics as input parameters. An extensive literature review was developed to build a database gathering
the relevant FRS properties for this work. The second part of this research consisted of developing an elasto­
plastic finite element model to design a steel fiber-reinforced tunnel lining application. Parameters from a real
case study of a circular tunnel were taken from the literature to calibrate the proposed computational model. The
model results are compared to analytical solutions and indicate good agreement. In this sense, the workflow is an
alternative for designing fiber-reinforced shotcrete for tunnel linings.

1. Introduction models for ground response and the effect of installed support (Carra­
nza-Torres and Fairhurst, 2000; Brown et al., 1983; Brady and Brown,
Shotcrete is one of the most widely used support elements in con­ 1993). The CCM consists of three main components, which are evaluated
structing underground mine excavations and rock tunnels (Louchnikov in the form of graphs: the longitudinal deformation profile (LDP), the
et al., 2014; de Alencar Monteiro and de Andrade Silva, 2021). The support characteristic curve (SCC), and the ground reaction curve
initial shotcrete is applied to the excavation walls to guarantee workers’ (GRC). Although some limitations of the CCM method are present, it
safety in the underground environment. According to Rispin et al. simplifies the 3D rock-support interaction problem through a 2D
(Rispin et al., 2017), the shotcrete support can be summarized in two approach and delivers a displacement profile across the excavation
main functions. Firstly, the shotcrete linings provide full-field support by tunnel (Vlachopoulos and Diederichs, 2014).
holding up large rock pieces. Secondly, they accommodate inward radial According to the improvement of the CCM method and other tunnel
displacement before installing the final support measures (concrete monitoring techniques, numerical simulations also became valuable
liners and rock bolts). tools for designing and constructing tunnels (Nie et al., 2021; Naseri and
In general, underground tunnel design mainly relies on empirical Bahrani, 2021; Oreste, 2003; Neuner et al., 2020; Rafiei Renani et al.,
methods, which strongly depend on the previous work experience of the 2016). In standard numerical tunneling models, based on the finite
engineering team (Wickham et al., 1972; Bieniawski, 1976; Grimstad element method (FEM) within the continuum approach, distinct
and Barton, 1993). On the other hand, analytical techniques have also constitutive models are applied for rock mass and shotcrete as system
been developed to predict deformations of rock masses and the most support. While the rock mass is generally modeled as an elastic, perfectly
suitable support system to contain them (Oreste and Pella, 1997). The plastic material (Alejano et al., 2009; Alonso et al., 2003), shotcrete is
convergence confinement method (CCM) is the most traditional method, assumed to be associated with a more complex damage plasticity model,
which allows the estimation of rock deformations through analytical which incorporates the material mechanical variation through time due

* Corresponding author.
E-mail addresses: [email protected] (M. Congro), [email protected] (V. Moreira de Alencar Monteiro), [email protected] (F. de
Andrade Silva), [email protected] (D. Roehl), [email protected] (A.L.T. Brandão).

https://doi.org/10.1016/j.tust.2022.104881
Received 6 October 2022; Received in revised form 13 November 2022; Accepted 22 November 2022
Available online 2 December 2022
0886-7798/© 2022 Elsevier Ltd. All rights reserved.
M. Congro et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 132 (2023) 104881

Fig. 1. Frequency distribution of the collected fiber and composite parameters of the proposed ANN: (a) fiber length (L), (b) toughness level (T40) for ASTM C1550
standard (Astm c1550., 2012), (c) fiber aspect ratio (AR), (d) nominal fiber volume fraction (Vf) and (e) matrix compressive strength (fc).

to shrinkage, creep, and other factors (Oreste, 2003; Schreter et al., added to the material composition to provide more appropriate strain
2018). Furthermore, it is possible to evaluate the excavation displace­ ability and post-crack reinforcement (Aci 506.1r., 1998; Aci 544.4r.,
ments over time through finite element models. 2018; Trottier and Banthia, 2000). The design of system support in
With the development of shotcrete technology, fibers have been underground excavations, especially in terms of the flexural toughness

Table 1
Upper and lower bounds for each ANN input variable.
Model Fiber Aspect Ratio (AR) Compressive Strength (fc) Nominal Fiber Volumetric Fraction (Vf)
- (MPa) (kg/m3) [%]

ANN 1 45–65 30–60 20–60 [0.25–0.76]

ANN 2 – 30–60 02–10 [0.22–1.10]

Fig. 2. Database generation procedure considering experimental and synthetic data.

2
M. Congro et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 132 (2023) 104881

Fig. 3. General workflow to predict panel ASTM C1550 (Astm c1550., 2012) toughness of steel and polypropylene fiber-reinforced shotcretes.

verification, is given by the execution of two main test standards: ASTM All regulated bending tests on prisms and panels demand expensive
C1550 (Astm c1550., 2012) and EN 14488–5 (EN, 2006). Both panel and sophisticated experimental setups, which are challenging to carry
standards are included on Papworth fiber-reinforced shotcrete design on without collaboration with a well-equipped structural laboratory.
guidelines for system support (Papworth, 2002), oriented by the corre­ Due to the difficulty in performing complex and costly mechanical tests,
lation between the toughness performance levels and the overall rock artificial neural network (ANN) models have been applied as a predic­
mass quality (Q-system). tive tool to estimate the desired mechanical parameters (Thai, 2022). As
Several works (Massone and Nazar, 2018; Sjölander and Ansell, explained by Ding et al. (Ding et al., 2013), ANNs are adaptive non-
2017; de la Fuente et al., 2012) have recently investigated the adoption linear information processing systems that combine several processing
of fiber-reinforced shotcrete models for tunneling design applications. units with various characteristics (e.g., self-adapting, self-organizing,
Generally, a simplified bi-linear model, proposed initially according to and real-time learning).
the RILEM TC-162 standard (Rilem TC 162-TDF, 2003), is carried out. Predictive tools using machine learning techniques have been used
Chiai et al. (Chiai et al., 2009), for instance, verified through their finite to estimate structural parameters, such as buckling (Chuang et al., 1998;
element model that the presence of fibers in the tunnel support system Cascardi et al., 2017), shear (Adhikary and Mutsuyoshi, 2006; Hossain
provided relevant structural advantages in serviceability and ultimate et al., 2017), and axial (Ahmadi et al., 2014; Vu et al., 2021) strengths of
limit states, including crack width control. The stress–strain model reinforced concrete structures (Li et al., 2019; Dung and Anh, 2019).
proposed by RILEM has input parameters for the post-crack residual Moreover, many ANN-based predictive models have been developed to
strength obtained from flexural bending tests on pre-notched prisms estimate the main concrete properties such as compressive strength,
([32]; CEB-FIB: Model Code. International Federation for Structural Young’s Modulus, and tensile strength (Prasad et al., 2009; Tayfur et al.,
Concrete (FIB), 2012). 2014; Kang et al., 2021; Demir, 2008). Kang et al. (Kang et al., 2021), for

3
M. Congro et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 132 (2023) 104881

Fig. 4. Pareto chart of standardized effects for T40 parameter considering (a) Fig. 5. Schematic representation of the ANN architecture: (a) SFRC; (b) PFRC.
SFRS; (b) PFRS.

Table 3
Blind tests selected for ANN 1 (SFRS) and ANN 2 (PFRS) validation.
Table 2
ANN architecture configurations. Blind Test / ANN Fiber Compressive Fiber T40
Parameters Aspect strength (fc) Volumetric
ANN Configuration ANN 1 ANN 2 Ratio Fraction (Vf)
Number of Neurons 100 24 (AR)
Number of Hidden Layers 1 - (-) (MPa) (kg/m3) [%] (J)
Training Algorithm Bayesian Regularization Training Algorithm (TrainBr) Blind Test 1 1 60.0 45.0 35.0 [0.45] 228.4
Transfer Function Hyperbolic Tangent Sigmoid Transfer Function (TanSig) Blind Test 2 1 65.0 40.0 22.5 [0.29] 384.0
Blind Test 3 1 65.0 50.0 50.0 [0.64] 491.1
Blind Test 1 2 – 47.0 9.0 [1.00] 640.5
instance, presented a comparative study on the performance of 11 Blind Test 2 2 – 56.5 9.0 [1.00] 409.0
distinct models to verify the flexural strength of steel fiber-reinforced Blind Test 3 2 – 34.0 4.5 [0.49] 227.7
concrete.
More recently, Congro et al. (Congro et al., 2021) developed an ANN
to predict the residual flexural strength of fiber-reinforced concrete
Table 4
(FRC) in designing commercial floors (TR 34 ([49]). The authors Regression values for each ANN dataset.
calculated the cementitious composite fR,1, fR,3, and fR,4 with the input of
ANN R2 Training R2 Test R2 Global
matrix resistance, fiber properties, fractions, and post-crack strength to
(-) (-) (-)
design slabs-on-ground bending moment distribution. Garcia-Taengua
ANN 1 0.894 0.865 0.885
(Garcia-Taengua, 2020; Garcia-Taengua et al., 2021) conducted a
ANN 2 0.879 0.866 0.876
comprehensive literature review to gather an initial FRC database and
deliver new software (OptiFRC) to estimate flexural post-peak behavior
without needing mechanical tests. proposed by Panet et al. (Panet and Sulem, 2022). Therefore, the nu­
This article proposes a novel hybrid model to design fiber-reinforced merical model can help researchers and engineers simulate tunnel lin­
concrete for tunnel linings considering the NATM (New Austrian ing’s behavior.
Tunneling Method). The methodology involves the development of an
artificial neural network to predict the FRS toughness and applies the 2. Methodology
residual flexural strength parameters estimated from a previous ANN
developed by Congro et al. (Congro et al., 2021) in an elastoplastic finite This research is divided into two main parts: fiber-reinforced shot­
element model. Finally, the numerical model is compared with a real crete toughness prediction through ANN and a tunnel excavation
case tunnel application and validated with an analytical solution simulation through an elastoplastic finite element model. First, two

4
M. Congro et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 132 (2023) 104881

Fig. 6. Error histogram and mean squared error value for (a) ANN 1; (b) ANN 2.

Fig. 7. Training performance considering each neural network: (a) ANN 1; (b) ANN 2.

ANNs were developed to assess the fiber-reinforced concrete toughness 2019; Bernard et al., 2010; Silva and Monteiro, 2021; Bernard, 2021;
parameter. The first artificial neural network considers 186 data points Bjontegaard and Myren, 2018; Barros, 1998; Aire and Aguilar, 2021;
for steel fiber-reinforced concrete tunnel lining applications collected Sandbakk et al., 2010; Myren and Bjotegaard, 2010; Myren and Bjote­
from different sources in the literature (de Alencar Monteiro and de gaard, 2009; Malmgren, 2007; Rengarajan, 2020). Subsequently; two
Andrade Silva, 2021; Silva and Monteiro, 2021; Silva and Monteiro, artificial neural network models were developed using the Deep
2021; Cengiz and Luranli, 2004; Decker et al., 2012; Bernard, 2015; Learning Optimization Toolbox in MATLAB 2021a ® and trained ac­
Carmona et al., 2020; Bernard et al., 2010; Bernard, 2012; Bernard, cording to the Levenberg-Marquardt optimization method (Marquardt,
2020; Amin et al., 2017; Ryabchikov et al., 2020; Salehian et al., 2014; 1963). Next, blind tests from the literature have been proposed to check
Manfredi and Silva, 2020; Gallo and Mejiá, 2013; Buratti et al., 2019; the ANN accuracy.
Cáceres et al., 2021; Juhasz et al., 2017; Garcia and Aguado, 2011; Furthermore, an elastoplastic finite element model is carried out to
Martin et al., 2015; di Colombo and Prisco, 2009; Ding and Kusterle, alternatively evaluate the displacements and stresses in a tunnel lining
1999; Larsson, 2018; xxxx; Bernard, 2002; Silva et al., 1550; Bernard, application. The ground conditions from the tunnel were derived from
2019; Bernard et al., 2010; Silva and Monteiro, 2021; Bernard, 2021; Panet et al. (Panet and Sulem, 2022) and modified accordingly to show
Bjontegaard and Myren, 2018; Barros, 1998; Aire and Aguilar, 2021; the influence of fiber addition. The ground behavior is simulated
Sandbakk et al., 2010; Myren and Bjotegaard, 2010; Myren and Bjote­ through an elastic, perfectly plastic Mohr-Coulomb model. The results
gaard, 2009; Malmgren, 2007; Rengarajan, 2020). Moreover; the second obtained from the simulation in terms of radial displacement and sup­
ANN gathers 176 data points for polypropylene fiber-reinforced con­ port pressure were validated with the analytical approach. The fiber-
crete tunnel lining applications from distinct authors (de Alencar Mon­ reinforced concrete stress–strain curve defining the constitutive model
teiro and de Andrade Silva, 2021; Silva and Monteiro, 2021; Silva and for the composite is achieved through the application of another ANN
Monteiro, 2021; Cengiz and Luranli, 2004; Decker et al., 2012; Bernard, model proposed by Congro et al. (Congro et al., 2021).
2015; Carmona et al., 2020; Bernard et al., 2010; Bernard, 2012; Ber­
nard, 2020; Amin et al., 2017; Ryabchikov et al., 2020; Salehian et al.,
2014; Manfredi and Silva, 2020; Gallo and Mejiá, 2013; Buratti et al., 2.1. ANN development methodology
2019; Cáceres et al., 2021; Juhasz et al., 2017; Garcia and Aguado, 2011;
Martin et al., 2015; di Colombo and Prisco, 2009; Ding and Kusterle, 2.1.1. ANN database construction
1999; Larsson, 2018; xxxx; Bernard, 2002; Silva et al., 1550; Bernard, According to the literature survey carried out for this research, the
following parameters were analyzed for each ANN:

5
M. Congro et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 132 (2023) 104881

Fig. 8. Result validation for blind tests n◦ 1, 2, and 3 for (a) steel and (b) polypropylene fiber-reinforced shotcrete ANNs.

Table 5
ANN validation results for the blind tests.
Toughness level at 40 mm (ASTM C1550) ANN
1 2

T40 Predicted (J) 228.4 579.2 Blind test 1

Absolute error (J) 42.8 61.3
Relative error 15.80 % 9.60 %
T40 Predicted (J) 384 409 Blind test 2
Absolute error (J) 20 48
Relative error 5.50 % 10.50 %
T40 Predicted (J) 491.1 227.7 Blind test 3
Absolute error (J) 46.5 31.1
Relative error 8.60 % 12.00 %

• Mean compressive strength (fc); values between the standard panel tests, the Bernard correlation (Ber­
• Nominal fiber volume fraction (Vf); nard, 2002) was applied. Bernard (Bernard, 2002) estimated that 100 J
• Fiber length (L); at 25 mm deflection in an EN 14488–5 panel test was equivalent to 400 J
• Fiber aspect ratio (AR), only for hooked-end steel fibers; at 40 mm of ASTM C1550 results. A linear correlation was achieved with
• Results for energy absorption capacity in standard flexural panel a coefficient of determination (R2) of 0.90. Equation (1) brings the
tests: TEN14488-5, 25mm for EN 14488–5 (EN, 2006) and TASTMC1550; mentioned correlation.
40mm for ASTM C1550 (Astm c1550., 2012).
1
TASTMC1550,40 mm = TEN14488− 5,25 mm (1)
2.5
All test results were obtained from EN 14488–5 (EN, 2006) and
ASTM C1550 (Astm c1550., 2012) guidelines. ASTM C1550 (Astm The database was divided into two groups: steel and polypropylene
c1550., 2012) recommends the application of a central point load on fiber-reinforced shotcretes. The energy absorption capacity (T40) given
circular specimens (800 mm in diameter and 75 mm thickness) sup­ by the ASTM C1550 (Astm c1550., 2012) standard is the output for both
ported by three symmetrically disposed pivots. The European guidelines ANN models. For ANN 1 (steel fiber-reinforced shotcrete), the input
EN 14488–5 (EN, 2006) display a centrally loaded square panel test parameters were nominal fiber volume fraction (Vf), matrix strength (fc),
(600 mm × 600 mm and 100 mm thickness) supported on rigid square and fiber aspect ratio (AR). In the case of ANN 2 (polypropylene fiber-
support. The energy absorption capacity (T) of each specimen is ob­ reinforced shotcrete), the input parameters were nominal fiber volume
tained from the area under the force per deflection curve. While the fraction (Vf) and matrix strength (fc). All database points were obtained
circular panel tests reach their limit displacement at 40 mm, the square considering panels with the use of spraying equipment. The complete
panel tests are interrupted at 25 mm. To unify the energy absorption database can be analyzed in the Appendices.

6
M. Congro et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 132 (2023) 104881

Fig. 9. General workflow to estimate the FRS lining post-crack mechanical properties through ANNs and its application to an elastoplastic finite element model.

7
M. Congro et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 132 (2023) 104881

Table 1 brings all input limits for the developed neural networks, where
the nominal fiber volumetric fraction is given in terms of fiber con­
sumption (kg/m3) and volumetric fraction (%). All input data was
linearized between 0 and 1, considering the upper and lower bounds.
The samples were separated into three groups: 70 % for training, 15 %
for validation, and 15 % for testing.
Some authors provided the experimental toughness in terms of its
mean value and standard deviation. The deviation is associated with the
FRC high property scatter, as shown by Cavalaro et al. (Cavalaro and
Aguado, 2015). When variations are assigned, two additional database
points are increased by adding and subtracting the standard deviation
from the mean toughness (T40 = T40,mean ± σ ), as shown in Fig. 2. This
procedure has already been applied in past works from the literature
(Congro et al., 2021; Merma et al., 2020; Rego et al., 2018). When
available, the values of standard deviations are displayed in the tables in
the Appendix section. Fig. 3 brings the complete workflow to predict
panel ASTM C1550 (Astm c1550., 2012) toughness of steel and poly­
Fig. 10. Two-dimensional model with boundary conditions. Detail of concrete propylene fiber-reinforced shotcretes from database construction until
shotcrete and rock. model validation.

2.1.2. Sensitivity analysis

Table 6 Before defining the neural network model and initiating the data
Elastic and mechanical parameters of the training step, it is necessary to check if the experimental input data is
rock. statistically significant to the FRC toughness parameter.
Parameter Value Sensitivity analysis methods are valuable alternatives to establish the
E (GPa) 30.0
relationship between factors and outputs of a given process (McKay
ν (-) 0.2 et al., 1979; Montgomery, 2013; Congro et al., 2019). In this article, the
C (MPa) 1.0 analyses are carried out using Design of Experiments (DOE), particularly
Φ (◦ ) 30 the Response Surface Methodology (RSM).
ψ (◦ ) 0
The sensitivity analyses were carried out in Minitab 18 ® software
and adopted a central circumscribed composite design (CCC) and a full
quadratic polynomial model. The default configuration of statistical
Table 7 analyses considers a 95 % confidence interval for the simulations and a
Elastic and mechanical parameters of the 5 % significance level. Pareto charts are generated to check if the input
SFRS. factors considered for the ANN database influence the composite
Parameter Value toughness parameter. Fig. 4(a) and 4(b) present the Pareto charts with
E (GPa) 35.0 the absolute values of the standardized effects from the most to the least
ν (-) 0.25 significant impact. A reference line (p-value) is also included to indicate
σeq (MPa) 39.06 which effects are statistically significant for the toughness parameter.
ψ (◦ ) 38 Fig. 4(a) and 4(b) indicate that the nominal fiber volume fraction (Vf)
fb0/fc0 1.16
is the most significant parameter that influences the toughness in fiber-
reinforced shotcrete panel tests. When it comes to steel fibers, the in­
crease in aspect ratio (L/D) also contributes to increasing toughness. The
Table 8 effect of fiber geometry and material properties is well described in the
Steel fiber fiber-reinforced shotcrete lining properties from ANN. literature, such as the Mobasher (Mobasher, 2019) and Bentur &
ANN from current research ANN by Congro et al. ( Mindness (Bentur and Mindness, 2007). A higher volume fraction in­
Congro et al., 2021) creases the number of fibers in the concrete member, resulting in a
fc AR Vf T40 TPL/Rock Class fR,1 fR,3 fR,4 significant gain in the post-crack strength. The increase in aspect ratio,
(MPa) – (kg/m3) (J) – (MPa) (MPa) (MPa)
in turn, results in higher bond strength of the fiber inside the concrete
50 65 50 491 IV/E 6.30 5.32 4.82 matrix, which also enhances the strength of the cracked section. Since
TPL: toughness performance level according to Papworth (Papworth, 2002)
toughness is evaluated as the area under the force per deflection curve,
the toughness increase directly raises the values of the post-crack
The distinct inputs are related to the fiber properties. Fig. 1 (a) to (e) strength of the composite.
present the distribution frequency of the input and output ANN pa­ The compressive strength parameter (fc) shows significant statistical
rameters. As shown in Fig. 1 (a), based on the collected parameters, steel relevance for the steel fiber-reinforced shotcrete (Fig. 4(a)), primarily
fiber length is between 30 mm and 35 mm. Therefore, the nominal fiber because of its interaction with the fiber volumetric fraction (Vf) and the
length can be assumed to be the same for ANN 1 database and is not used compressive strength. For the polypropylene fiber-reinforced shotcrete
as an input parameter. On the other hand, polypropylene fiber length in (Fig. 4(b)), the interaction between the compressive strength and fiber
the ANN database is defined in a range between 48 mm and 60 mm. volumetric fraction parameters is very close to the imposed p-value limit,
Polypropylene (PP) fiber manufacturers do not routinely provide the including the compressive strength parameter as input for the network.
fiber aspect ratio due to imperfections in the fiber section. Therefore, the
aspect ratio cannot be precisely determined like in steel fibers. The main 2.1.3. Model definition and training
commercialized PP fibers for FRS applications also report similar This study uses the Bayesian regularization algorithm (Marquardt,
geometrical properties with a length above 45 mm. The ANN only in­ 1963) to estimate the correct targets. All simulations and tests for the
cludes results with nominal fiber length values within those limits. artificial neural network model were carried out using the Deep
Learning Optimization Toolbox in MATLAB ®.

8
M. Congro et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 132 (2023) 104881

Fig. 11. Distribution of stresses (in Pa) in distinct steps of the FE model: (a) radial stress during excavation step; (b) radial stress during shotcrete step; (c) tangential
stress during excavation step; (d) tangential stress during shotcrete step.

Fig. 12. Radial displacements evolution (in meters) in the numerical model considering (a) excavation step; (b) shotcrete step.

The training method updates the weight and bias of each neural validation dataset is active during the training process, although these
network layer according to Levenberg-Marquardt optimization. The al­ can lead to a global minimum (MacKay, 1992; Foresee and Hagan,
gorithm reduces the combination between squared errors and weights, 1997). Moreover, the training process with Bayesian regularization ends
adjusting the proper combination to obtain an ANN with an adequate the training with fewer effective parameters, which indicates its
generalization level. Therefore, this function adopts the Jacobian for superiority.
calculations. According to MacKay (Mobasher, 2019), this training al­ Establishing the ANN architecture is another crucial step for model
gorithm does not require a validation dataset since checking validation definition. Several parameters must be defined during this phase: the
aims at verifying if the dataset error increases or not during the training number of neurons, hidden layers, and the transfer function to obtain the
stage. However, higher weights lead to more significant relative errors. correct targets. Given the random variability of artificial intelligence
Therefore, the network might not explore higher weights if the methods, several rounds were performed, considering distinct

9
M. Congro et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 132 (2023) 104881

Stresses around the tunnel polypropylene fiber-reinforced shotcrete’s toughness parameter

30 (Table 3). The joint verification of the regression analysis, the MSE,
absolute difference, and relative error between the experimental values
and the ANN response are assessed to validate the best network.
25

3. Discussion and analysis

20
Stress (MPa)

3.1. ANN performance and regression

15
Regression tests, MSE, and blind tests were the adopted criteria for
choosing the best ANN. Table 4 summarizes the regression values for
10 Radial stress each tested ANN.
The regression values for all five ANNs are above 0.86, which in­
Tangential stress
dicates a good fit between predicted and experimental values. The
5 Radius ratio/Plastic limit analytical solution validation dataset’s regression value is not presented since the Bayesian
Regularization algorithm was used for the ANN training process. Similar
0 values of R2 were also reported in past works applying ANNs to predict
0 5 10 15 20 25
shear strength and other steel fiber-reinforced concrete mechanical pa­
Radius ratio (-)
rameters (Bai, 2022; Rahman et al., 2021; Hossain et al., 2017).
Fig. 13. Simulated radial and tangential stresses. Fig. 6 presents the error histogram considering each of the ANNs
tested in this study. Each histogram gives the mean square error (MSE)
for the performance of each ANN during the training process. The mean
square errors obtained for all ANNs present an approximation order of
10e-3, demonstrating excellent predictive performance. According to
Radial Displacement Evolution Teixeira et al. (Teixeira et al., 2022), if the training and test MSE are
0.01 lower than the training and test variances of T40, the model MSE can be
0.009
accepted. In turn, ANN 1 presents an MSE of the training dataset of
7.83e-3 and an MSE of the test dataset of 7.41e-3. These values are lower
0.008 than the training and test variance parameters (8.70e-3 and 8.55e-3,
0.007
respectively). Next, ANN 2 presents a MSE of the training dataset of
Radial Displacement (m)

6.92e-3 and a MSE of the test dataset of 6.77e-3. These metrics are lower
0.006 than the training and test variance parameters (7.30e-3 and 7.19e-3,
0.005 respectively).
Fig. 7 presents the best validation performance for each ANN. In
0.004
other words, the best validation performance indicates the calculation of
0.003 Displacement evolution the MSE for each interval during the training stage. Both networks show
good performance. ANN 1 is trained considering 946 iterations with a
0.002
gradient equal to 6.02e-4. ANN 2 presents 137 iterations with a gradient
0.001 of 3.33e-4. In a first investigation, these parameters indicate the net­
works show a good fit. However, it is crucial to validate these ANNs with
0
0 5 10 15 20 25 additional studies, using datasets that did not take part in the network
Radius ratio (-) training stage, such as blind tests.

Fig. 14. Radial displacement evolution.

3.2. ANN validation

After assessing the training performance of the neural network, it is

architectures to get the most accurate network. The preliminary crite­ required to analyze its accuracy. Although regression and mean square
rion for network selection is the highest value of test regression (R2). In error are essential parameters to determine if the network has a good
this sense, considering the test regression metric, Table 2 summarizes level of forecasting the output variables, it is common to apply blind
the final architecture configuration for each neural network, where the tests to the artificial system, as presented in Table 3. This research se­
Bayesian regularization (TrainBr) algorithm and the tansigmoid transfer lects three experimental references not part of ANN training to verify the
function (Tansig) were used. Next, Fig. 5 presents a schematic repre­ network accuracy. Also, the experimental blind tests selected for vali­
sentation of the architectures developed for each ANN. dation are within the lower and upper bounds determined for each
It should be pointed out that the ANNs do not represent the flexural neural network parameter.
behavior of concrete in the entire space of independent variables once The verification of the blind tests follows by comparing the absolute
the database is restricted to the boundaries in Table 1. and relative errors between the experimental reference value and the
composite toughness estimated by the ANN. It is essential to highlight
2.1.4. ANN selection and validation that the standardized experimental test for obtaining toughness usually
In addition to the regression analysis, evaluating the neural net­ implies laboratory responses with relative errors up to 20 %, as Bernard
work’s performance and the mean square error value (MSE) is essential. et al. (Silva et al., 2021). These deviations are considered acceptable
Moreover, some experimental tests available in the literature and not since they are within the range of the intrinsic uncertainties of the ex­
included in the ANN training were selected to validate steel and periments. Fig. 8 and Table 5 present the ANN validation results.
Moreover, it is essential to note that the toughness results predicted

10
 M. Congro et al.
Table A1
Model database (part 1).
Authors fc Fiber Vf L AR T25 Standard T40 Standard
type (EN 14488–5) Deviation (ASTM C1550) Deviation
3
(MPa) - (kg/m ) (mm) - (J) (J) (J) (J)

(de Alencar Monteiro and de Andrade Silva, 2021) 50 SF 25 35 65 – – 407 34

(de Alencar Monteiro and de Andrade Silva, 2021) 50 PP 4.2 54 62 – – 189 9
(de Alencar Monteiro and de Andrade Silva, 2021) 50 PP 6 54 62 – – 304 54
(de Alencar Monteiro and de Andrade Silva, 2021) 50 PP 8 54 62 – – 449 46
(de Alencar Monteiro and de Andrade Silva, 2021) 50 PP 4.2 54 62 780 100 312 40
(Silva and Monteiro, 2021) 37 SF 30 35 65 971 69 388 28
(Silva and Monteiro, 2021) 34 PP 6 – – 158 10 63 4
(Cengiz and Luranli, 2004) 50 PP 7 30 33 716 – 286 -
(Cengiz and Luranli, 2004) 50 PP 10 30 33 751 – 300 -
(Cengiz and Luranli, 2004) 50 SF 50 30 50 846 – 338 -
(Cengiz and Luranli, 2004) 50 SF 35 30 50 664 – 266 -
(Cengiz and Luranli, 2004) 30 PP 5 48 – – – 313 42
(Decker et al., 2012) 30 PP 6 48 – – – 355 42
(Decker et al., 2012) 30 PP 7 48 392 51

Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 132 (2023) 104881
– – –
(Decker et al., 2012) 30 PP 5 54 – – – 352 50
(Bernard, 2015) 50 SF 50 35 65 – – 438 42
(Bernard, 2015) 40 SF 50 35 65 – – 650 20
(Bernard, 2015) 25 SF 50 35 65 – – 531 33
(Bernard, 2015) 32 PP 10 48 447 66
11

– – –
(Bernard, 2015) 50 PP 7 50 – – – 354 26
(Bernard, 2015) 40 PP 7 50 – – – 340 31
(Bernard, 2015) 50 PP 7 54 – – – 470 24
(Carmona et al., 2020) 40 PP 4 54 64 805 – 322 -
(Carmona et al., 2020) 41 PP 8 54 64 1047 – 419 -
(Carmona et al., 2020) 42 PP 12 54 64 1254 – 502 -
(Bernard et al., 2010) 32 SF 30 30 45 – – 560 63
(Bernard et al., 2010) 32 PP 8 48 – – – 370 61
(Bernard et al., 2010) 25 PP 9 50 85 – – 521 59
(Bernard, 2012) 44 PP 10 50 85 – – 413 20
(Bernard, 2012) 66 PP 6.7 50 85 – – 337 18
(Bernard, 2020) 61 SF 50 30 55 – – 376 92
(Bernard, 2020) 45 SF 50 30 55 – – 352 11
fc = matrix compressive strength
Vf = fiber volume fraction
AR = aspect ratio
L = fiber length
PP = polypropylene fiber
SF = steel fiber
T25 = toughness level at 25 mm of deflection (EM 14488–5)
T40 = toughness level at 40 mm of deflection (ASTM C1550)
 M. Congro et al.
Table A2
Model database (part 2).
Authors fc Fiber Vf L AR T25 Standard T40 Standard
type (EN 14488–5) Deviation (ASTM C1550) Deviation
(MPa) - (kg/m3) (mm) - (J) (J) (J) (J)

(Bernard, 2020) 48 SF 36 30 55 – – 256 30

(Bernard, 2020) 38.1 SF 60 35 65 – – 480 154
(Bernard, 2020) 35 SF 20 38 45 – – 329 14
(Bernard, 2020) 43.5 PP 6.7 48 – – – 267 8
(Bernard, 2020) 47.4 PP 5.7 50 57 – – 178 35
(Bernard, 2020) 33.3 PP 5 50 – – – 286 25
(Amin et al., 2017) 66 PP 4 58 58 – – 701
(Amin et al., 2017) 63 PP 8 58 58 – – 1075 -
(Ryabchikov et al., 2020) 30 SF 70 50 50 – – 583 5
(Ryabchikov et al., 2020) 30 SF 35 50 75 571 99 228 -
(Ryabchikov et al., 2020) 50 SF 48 60 60 – – 681 167
(Salehian et al., 2014) 61.9 SF 60 35 60 1133 200 453 80
(Salehian et al., 2014) 65.2 SF 45 35 60 922 150 369 60
(Manfredi and Silva, 2020) 73 PP 3 40 74 – – 173 2

Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 132 (2023) 104881
(Manfredi and Silva, 2020) 73 PP 6 40 74 – – 263 30
(Manfredi and Silva, 2020) 73 PP 10 40 74 – – 670 4
(Gallo and Mejiá, 2013) 37 SF 40 30 52 807 – 323 -
(Gallo and Mejiá, 2013) 37 SF 40 30 50 – – 307 -
(Gallo and Mejiá, 2013) 37 SF 30 30 52 542 – 217 -
12

(Gallo and Mejiá, 2013) 37 SF 30 30 50 – – 284 -

(Gallo and Mejiá, 2013) 37 SF 20 30 50 507 – 203 -
(Gallo and Mejiá, 2013) 37 SF 20 30 50 – – 227 -
(Gallo and Mejiá, 2013) 37 SF 30 35 70 855 – 342 -
(Gallo and Mejiá, 2013) 37 SF 30 35 70 – – 347 -
(Gallo and Mejiá, 2013) 37 PP 6 47 58 857 – 343 -
(Gallo and Mejiá, 2013) 37 PP 8 47 58 1086 – 434 -
(Gallo and Mejiá, 2013) 37 PP 10 47 58 1268 – 507 -
(Gallo and Mejiá, 2013) 37 PP 6 47 58 – – 327 -
(Gallo and Mejiá, 2013) 37 PP 8 47 58 – – 424 -
(Gallo and Mejiá, 2013) 37 PP 10 47 58 – – 650 -
(Buratti et al., 2019) 46 SF 30 35 65 – – 269 70
(Buratti et al., 2019) 36 SF 30 35 65 – – 297 69
(Buratti et al., 2019) 35 PP 3 55 60 – – 144 13
(Buratti et al., 2019) 38 PP 7 55 60 – – 310 30
(Buratti et al., 2019) 31 PP 3 55 60 – – 122 35
(Buratti et al., 2019) 30 PP 5 55 60 – – 240 55
(Cáceres et al., 2021) 37.2 SF 90 40 25 759 147 304 59
(Cáceres et al., 2021) 38.5 SF 60 40 25 578 37 231 15
(Cáceres et al., 2021) 39.1 SF 30 40 25 398 30 159 12
(Juhasz et al., 2017) 48 PP 2.5 48 – 608 – 243 -
(Juhasz et al., 2017) 48 PP 5 48 – 1029 – 412 –
 M. Congro et al.
Table A3
Model database (part 3).
Authors fc Fiber Vf L AR T25 Standard T40 Standard
type (EN 14488–5) Deviation (ASTM C1550) Deviation
3
(MPa) - (kg/m ) (mm) - (J) (J) (J) (J)

(Juhasz et al., 2017) 48 PP 7.5 48 – 1489 – 596 -

(Garcia and Aguado, 2011) 36 SF 25 35 65 801 86 320 34
(Garcia and Aguado, 2011) 36 SF 25 35 65 – – 360 70
(Garcia and Aguado, 2011) 34 PP 4.5 50 57 647 62 259 25
(Garcia and Aguado, 2011) 34 PP 4.5 50 57 – – 244 26
(Martin et al., 2015) 38 PP 1.9 48 – 160 – 64 -
(Martin et al., 2015) 38 PP 3 48 – 276 – 110 -
(Martin et al., 2015) 38 PP 4.2 48 – 302 – 121 -
(Martin et al., 2015) 38 PP 1.9 48 – 175 – 70 -
(di Colombo and Prisco, 2009) 36 SF 30 30 45 340 102 136 41
(Ding and Kusterle, 1999) 37 SF 60 30 60 1416 – 566 -
(Ding and Kusterle, 1999) 37 SF 40 30 60 1067 – 427 -
(Ding and Kusterle, 1999) 37 SF 20 30 60 828 – 331 -
(Larsson, 2018) 40 SF 30 35 65 484 79

Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 132 (2023) 104881
– –
(Larsson, 2018) 40 SF 25 35 65 – – 433 46
(Larsson, 2018) 40 SF 22.5 35 65 – – 364 13
(Larsson, 2018) 40 SF 20 35 65 – – 415 36
(Larsson, 2018) 40 SF 15 35 65 – – 356 51
(Yang et al., 2017) 52 SF 42 30 60 1027 32 411 13
13

(Bernard, 2002) 35 SF 50 18 30 – – 230 -

(Bernard, 2002) 35 SF 50 18 30 661 – 264 -
(Bernard, 2002) 54 SF 25 18 30 – – 104 -
(Bernard, 2002) 35 SF 25 18 30 356 – 142 -
(Bernard, 2002) 53.3 SF 30 25 30 – – 195 -
(Bernard, 2002) 53.3 SF 30 25 30 549 – 220 -
(Bernard, 2002) 46.9 SF 60 30 45 – – 513 -
(Bernard, 2002) 46.9 SF 60 30 45 1311 – 524 -
(Bernard, 2002) 46.2 SF 60 30 45 – – 347 -
(Bernard, 2002) 46.2 SF 60 30 45 975 – 390 -
(Bernard, 2002) 35 SF 50 30 60 – – 415 -
(Bernard, 2002) 35 SF 50 30 60 1026 – 410 -
(Bernard, 2002) 35 SF 50 30 45 – – 380 -
(Bernard, 2002) 35 SF 50 30 45 1096 – 438 -
(Bernard, 2002) 45.5 SF 50 30 45 – – 416 -
(Bernard, 2002) 45.5 SF 50 30 45 1030 – 412 -
(Bernard, 2002) 42.4 SF 50 30 45 – – 310 -
(Bernard, 2002) 42.4 SF 50 30 45 769 – 308 -
(Bernard, 2002) 53.7 SF 50 30 45 – – 494 -
(Bernard, 2002) 53.7 SF 50 30 45 1271 – 508 -
(Bernard, 2002) 46.1 SF 40 30 45 – – 310 -
(Bernard, 2002) 46.1 SF 40 30 45 1017 – 407 –
 M. Congro et al.
Table A4
Model database (part 4).
Authors fc Fiber Vf L AR T25 Standard T40 Standard
type (EN 14488–5) Deviation (ASTM C1550) Deviation
3
(MPa) - (kg/m ) (mm) - (J) (J) (J) (J)

(Bernard, 2002) 49.4 SF 40 30 45 – – 269 -

(Bernard, 2002) 49.4 SF 40 30 45 724 – 290 -
(Bernard, 2002) 52.3 SF 40 30 45 – – 442 -
(Bernard, 2002) 52.3 SF 40 30 45 1111 – 444 -
(Bernard, 2002) 47.3 SF 37 30 50 – – 297 -
(Bernard, 2002) 47.3 SF 37 30 50 765 – 306 -
(Bernard, 2002) 43.8 SF 37 30 50 – – 274 -
(Bernard, 2002) 43.8 SF 37 30 50 805 – 322 -
(Bernard, 2002) 48 SF 37 30 45 – – 159 -
(Bernard, 2002) 48 SF 37 30 45 394 – 158 -
(Bernard, 2002) 35 SF 25 30 60 – – 254 -
(Bernard, 2002) 35 SF 25 30 60 765 – 306 -
(Bernard, 2002) 56.7 PP 7 30 – – – 204 -
(Bernard, 2002) 56.7 PP 7 30 724 290 -

Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 132 (2023) 104881
– –
(Bernard, 2002) 55 PP 9 30 – – – 319 -
(Bernard, 2002) 55 PP 9 30 – 737 – 295 -
(Bernard, 2002) 54.5 PP 5 30 – – – 137 -
(Bernard, 2002) 54.5 PP 5 30 – 353 – 141 -
(Bernard, 2002) 52.7 PP 10 30 222 -
14

– – –
(Bernard, 2002) 52.7 PP 10 30 – 658 – 263 -
(Bernard, 2002) 51.7 PP 15 30 – – – 371 -
(Bernard, 2002) 51.7 PP 15 30 – 948 – 379 -
(Bernard, 2002) 44.8 SF 60 35 65 – – 701 -
(Bernard, 2002) 44.8 SF 60 35 65 1628 – 651 -
(Bernard, 2002) 48 SF 50 35 60 – – 239 -
(Bernard, 2002) 48 SF 50 35 60 772 – 309 -
(Bernard, 2002) 47.3 SF 50 35 60 – – 377 -
(Bernard, 2002) 47.3 SF 50 35 60 841 – 336 -
(Bernard, 2002) 57.5 SF 50 35 65 – – 761 -
(Bernard, 2002) 57.5 SF 50 35 65 1591 – 636 -
(Bernard, 2002) 52.3 SF 40 35 60 – – 257 -
(Bernard, 2002) 52.3 SF 40 35 60 767 – 307 -
(Bernard, 2002) 52.2 SF 40 35 65 – – 498 -
(Bernard, 2002) 52.2 SF 40 35 65 1471 – 588 -
(Bernard, 2002) 46.4 SF 40 35 60 – – 340 -
(Bernard, 2002) 46.4 SF 40 35 60 756 – 302 -
(Bernard, 2002) 50.8 SF 35 35 45 – – 502 -
(Bernard, 2002) 50.8 SF 35 35 45 1199 – 480 -
(Bernard, 2002) 50.9 SF 30 35 60 – – 306 -
(Bernard, 2002) 50.9 SF 30 35 60 852 – 341 -
(Bernard, 2002) 55.2 SF 30 35 60 – – 241 –
M. Congro et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 132 (2023) 104881

Table A5
Model database (part 5).
Authors fc Fiber Vf L AR T25 Standard T40 Standard
type (EN 14488–5) Deviation (ASTM C1550) Deviation
(MPa) - (kg/m3) (mm) - (J) (J) (J) (J)

(Bernard, 2002) 55.2 SF 30 35 60 547 – 219 -

(Bernard, 2002) 57.8 SF 30 35 65 – – 418 -
(Bernard, 2002) 57.8 SF 30 35 65 1044 – 418 -
(Bernard, 2002) 57.5 SF 30 35 65 – – 418 -
(Bernard, 2002) 57.5 SF 30 35 65 1044 – 418 -
(Bernard, 2002) 42.5 SF 30 35 65 – – 277 -
(Bernard, 2002) 42.5 SF 30 35 65 752 – 301 -
(Bernard, 2002) 54.3 SF 30 35 65 – – 182 -
(Bernard, 2002) 54.3 SF 30 35 65 491 – 196 -
(Bernard, 2002) 35.3 SF 30 35 65 – – 367 -
(Bernard, 2002) 35.3 SF 30 35 65 896 – 358 -
(Bernard, 2002) 57.8 SF 30 35 65 – – 265 -
(Bernard, 2002) 57.8 SF 30 35 65 631 – 252 -
(Bernard, 2002) 53.3 PP 5 50 – – – 225 -
(Bernard, 2002) 53.3 PP 5 50 – 629 – 252 -
(Bernard, 2002) 61.8 PP 7 50 – – – 329 -
(Bernard, 2002) 61.8 PP 7 50 – 944 – 378 -
(Bernard, 2002) 59.8 PP 9 50 – – – 455 -
(Bernard, 2002) 59.8 PP 9 50 – 1171 – 468 -
(Bernard, 2002) 58.5 PP 13.5 50 – – – 506 -
(Bernard, 2002) 58.5 PP 13.5 50 – 1459 – 584 -
(Bernard, 2002) 56.5 PP 9 50 – – – 457 -
(Bernard, 2002) 56.5 PP 9 50 – 1055 – 422 -
(Bernard, 2002) 46.8 PP 4.6 52 – – – 228 -
(Bernard, 2002) 46.8 PP 4.6 52 – 788 – 315 -
(Bernard, 2002) 56.2 PP 6.9 52 – – – 416 -
(Bernard, 2002) 56.2 PP 6.9 52 – 1060 – 424 -
(Bernard, 2002) 46.8 PP 9.2 52 – – – 501 -
(Bernard, 2002) 46.8 PP 9.2 52 – 1223 – 489 -
(Bernard, 2002) 46.8 PP 11.5 52 – – – 550 -
(Bernard, 2002) 46.8 PP 11.5 52 – 1352 – 541 -
(Silva et al., 2021) 50.12 PP 4.2 54 62 – – 258 28
(Silva et al., 2021) 50.12 PP 6 54 62 – – 351 44
(Bernard, 2019) 46.7 PP 6 30 – – – 248 25
(Bernard, 2019) 45.5 PP 8 48 – – – 611 98
(Bernard, 2019) 46.5 PP 2 48 – – – 169 20
(Bernard, 2019) 44.7 PP 3 54 – – – 283 34
(Bernard, 2019) 47.0 PP 9 54 – – – 641 83
(Bernard, 2019) 55 PP 6 57 – – – 514 72
(Bernard, 2019) 67 PP 6 57 – – – 474 57
(Bernard et al., 2010) 32 PP 4 42 – – – 218 22

by the neural network do not consider some typical phenomenon that Júnior, 2013). The study’s novelty is given by adopting a workflow that
occurs during the support system design, such as fiber rebound, batch- combines artificial intelligence techniques with a finite element model
to-batch variation (Kaufmann et al., 2013; Bernard et al., 2010) or for a steel fiber-reinforced shotcrete lining application. The numerical
other unpredictable events during delivery or spraying stages (set workflow aims to provide an alternative way of designing the support
accelerator dosage rate, poor spraying or poor curing, for example). In system through an elastoplastic finite element model. Fig. 9 presents the
this sense, the ANN could estimate the expected mean level of perfor­ complete workflow associated with the investigations carried out in this
mance prior to the commencement of construction and quality control article.
testing. However, this predicted level of performance must be confirmed
by actual physical testing of material samples during construction. In
addition, a safety factor can also be included after the calculations to 4.2. Numerical model
cover these unpredictable events associated with construction.
An elastoplastic finite element model is carried out using the com­
4. ANN application for FRS tunnel excavation linings analysis mercial software ABAQUS®. The tunnel geometry consisted of a gallery
with a circular section of radius R = 3.15 m excavated at 600 m depth.
4.1. Overview The initial stress state is assumed to be isotropic (σ0) and equal to 16.2
MPa. A shotcrete lining of thickness e = 20 cm is installed at a distance d
The evaluation of fiber-reinforced shotcrete lining performance = 5 m. The primary geometrical information regarding the rock tunnel
through full-scale experimental tests is usually time-consuming and was taken from Panet et al. (Panet and Sulem, 2022).
costly. Since there are few studies on this topic, the development of All the materials are modeled using 8-node quadrilateral elements
reliable numerical models emerges as an interesting alternative to pre­ with reduced integration (3006 elements and 3009 nodes). A two-
dict the global behavior of an FRS tunnel section, including the assess­ dimensional plane strain model is adopted with consistent boundary
ment of the maximum tunnel displacements and the evolution of radial/ conditions, as presented in Fig. 10. The model size is defined in such a
tangential stresses around the tunnel. way as to avoid the influence of boundary conditions on final stresses
Therefore, this section proposes developing a finite element model and displacements.
based on a real case of an excavation of a circular tunnel lining (Vissotto The numerical simulation considers the main phases of the tunnel
lining construction: geostatic step, initial confinement of the tunnel,

15
M. Congro et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 132 (2023) 104881

Table A6
Model database (part 6).
Authors fc Fiber Vf L AR T25 Standard T40 Standard
type (EN 14488–5) Deviation (ASTM C1550) Deviation
(MPa) - (kg/m3) (mm) - (J) (J) (J) (J)

(Bernard et al., 2010) 50 PP 7 48 – – – 272 27

(Bernard et al., 2010) 50 PP 6 48 – – – 258 26
(Silva and Monteiro, 2021) 36.6 SF 25 35 65 727 108.34 291 43
(Silva and Monteiro, 2021) 36.6 SF 20 35 65 679 74 272 30
(Silva and Monteiro, 2021) 36.6 SF 15 35 65 380 95.9 152 38
(Bernard, 2021) 52 PP 6 40 – – – 410 -
(Bernard, 2021) 58 PP 8 54 – – – 524 -
(Bernard, 2021) 45 SF 25 35 65 – – 296 -
(Bjontegaard and Myren, 2018) 60 SF 20 35 65 623 – 249 -
(Bjontegaard and Myren, 2018) 60 SF 20 35 65 616 – 246 -
(Bjontegaard and Myren, 2018) 60 SF 20 35 65 634 – 254 -
(Bjontegaard and Myren, 2018) 60 SF 20 35 65 693 – 277 -
(Bjontegaard and Myren, 2018) 60 SF 20 35 65 725 – 290 -
(Bjontegaard and Myren, 2018) 60 SF 20 35 65 765 – 306 -
(Bjontegaard and Myren, 2018) 60 SF 25 35 65 751 – 300 -
(Bjontegaard and Myren, 2018) 60 SF 25 35 65 753 – 301 -
(Bjontegaard and Myren, 2018) 60 SF 25 35 65 734 – 294 -
(Bjontegaard and Myren, 2018) 60 SF 30 35 65 795 – 318 -
(Bjontegaard and Myren, 2018) 60 SF 30 35 65 825 – 330 -
(Bjontegaard and Myren, 2018) 60 SF 30 35 65 851 – 340 -
(Bjontegaard and Myren, 2018) 60 SF 30 35 65 865 – 346 -
(Bjontegaard and Myren, 2018) 60 SF 30 35 65 883 – 353 -
(Bjontegaard and Myren, 2018) 60 SF 30 35 65 888 – 355 -
(Bjontegaard and Myren, 2018) 60 PP 5 54 – 648 – 259 -
(Bjontegaard and Myren, 2018) 60 PP 5 54 – 686 – 274 -
(Bjontegaard and Myren, 2018) 60 PP 5 54 – 837 – 335 -
(Bjontegaard and Myren, 2018) 60 PP 5.6 54 – 890 – 356 -
(Bjontegaard and Myren, 2018) 60 PP 6 54 – 862 – 345 -
(Bjontegaard and Myren, 2018) 60 PP 6 54 – 941 – 376 -
(Bjontegaard and Myren, 2018) 60 PP 7 54 – 1000 – 400 -
(Bjontegaard and Myren, 2018) 60 PP 7 54 – 1018 – 407 -
(Bjontegaard and Myren, 2018) 60 PP 7 54 – 934 – 374 -
(Bjontegaard and Myren, 2018) 60 PP 7 54 – 869 – 348 -
(Bjontegaard and Myren, 2018) 60 PP 7 54 – 879 – 352 -
(Bjontegaard and Myren, 2018) 60 PP 8 54 – 974 – 390 -
(Bjontegaard and Myren, 2018) 60 PP 8 54 – 1151 – 460 -
(Barros, 1998) 45 SF 25 30 60 500 – 200 -
(Barros, 1998) 45 SF 25 30 60 497 – 199 -
(Barros, 1998) 45 SF 25 30 60 421 – 168 -
(Barros, 1998) 45 SF 25 30 60 461 – 184 -
(Barros, 1998) 45 SF 30 30 60 456 – 182 –

excavation operation, and shotcrete (application of the concrete system obtain the constitutive lining model according to RILEM TC 162-TDF
support). The concrete is modeled as an elastoplastic material using the ([106]) and to run the numerical simulations using the concrete
concrete damaged plasticity (CDP) constitutive model. The rock is also damaged plasticity model. Residual flexural strength values are the
considered an elastoplastic material, adopting the Mohr-Coulomb cri­ required input parameters for the constitutive model. More information
terion. Table 6 summarizes the rock’s elastic and plastic material pa­ regarding the constitutive model and determining the residual flexural
rameters, where E refers to the Young modulus, ν is the Poisson’s strength can be assessed at RILEM TC 162-TDF (Rahman et al., 2021)
coefficient, C is the cohesion, Φ is the friction angle of the Mohr- and Congro et al. (Congro et al., 2021), respectively. Equations (2) to (4)
Coulomb model, and ψ is the dilatancy. Table 7 presents the steel bring the formulas to calculate peak and post-peak stresses by RILEM
fiber-reinforced concrete shotcrete’s elastic and plastic material pa­ TC-162-TDF (Rahman et al., 2021). However, other alternatives can be
rameters, where σeq refers to the equivalent strength, ψ is the dilation used to design tunnel linings through empirical methods defined from
angle, and fb0/fc0 is the ratio between the biaxial and uniaxial toughness, such as the Papworth method (Papworth, 2002). More in­
compressive strength of concrete. formation regarding the constitutive model and determining the resid­
The steel fiber-reinforced shotcrete lining was first determined using ual flexural strength can be assessed at RILEM TC 162-TDF (Rahman
the present ANN and applying Papworth (Papworth, 2002) toughness et al., 2021) and Congro et al. (Congro et al., 2021), respectively.
performance level correlation with the ground conditions to classify the Equations (2) to (4) bring the formulas to calculate peak and post-peak
obtained system support. Three compositions were selected in terms of stresses by RILEM TC-162-TDF (Rahman et al., 2021).
concrete compressive strength (fc), steel fiber aspect ratio (AR), and fiber
σ 1 = 0.70 × ffctm,fl (2)
fraction (Vf). The achieved values of toughness performance level and its
classification according to the Q system (Grimstad and Barton, 1993) for
σ 2 = 0.45 × fR,1 × κh (3)
ground conditions are presented in Table 8.
The next step consists of applying the previously developed ANN by σ 3 = 0.37 × fR,4 × κh (4)
Congro et al. (Congro et al., 2021) to obtain the results for residual
flexural strength (fR,1, fR,3 and fR,4). The ANN uses the composite ma­ where σ1, σ2 and σ3 are the peak and post-peak tensile stresses, ffctm,fl is
terial properties as input (fc, AR and Vf) to achieve the FRS post-peak the matrix tensile stress, κh is the size factor, fR,1 and fR,4 are the post-
strength (fR,1, fR,3, and fR,4). The residual strength values are used to peak flexural stresses achieved from EN 14,651 tests.

16
M. Congro et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 132 (2023) 104881

Table A7
Model database (part 7).
Authors fc Fiber Vf L AR T25 Standard T40 Standard
type (EN 14488–5) Deviation (ASTM C1550) Deviation
(MPa) - (kg/m3) (mm) - (J) (J) (J) (J)

(Barros, 1998) 45 SF 30 30 60 442 – 177 -

(Barros, 1998) 45 SF 30 30 60 547 – 219 -
(Barros, 1998) 45 SF 30 30 60 467 – 187 -
(Barros, 1998) 45 SF 35 30 60 552 – 221 -
(Barros, 1998) 45 SF 35 30 60 678 – 271 -
(Barros, 1998) 45 SF 35 30 60 543 – 217 -
(Aire and Aguilar, 2021) 40 SF 30 35 65 820 – 328 -
(Aire and Aguilar, 2021) 40 SF 30 35 65 894 – 358 -
(Aire and Aguilar, 2021) 40 SF 30 35 65 814 – 326 -
(Aire and Aguilar, 2021) 40 SF 30 35 65 – – 266 -
(Aire and Aguilar, 2021) 40 SF 30 35 65 – – 287 -
(Aire and Aguilar, 2021) 40 SF 30 35 65 – – 381 -
(Aire and Aguilar, 2021) 40 SF 30 35 65 – – 311 -
(Sandbakk et al., 2010) 60 PP 4.5 50 – – – 186 -
(Sandbakk et al., 2010) 60 PP 4.5 50 – – – 226 -
(Sandbakk et al., 2010) 60 PP 4.5 50 – – – 181 -
(Sandbakk et al., 2010) 60 PP 4.5 50 – – – 239 -
(Sandbakk et al., 2010) 60 PP 9 50 – – – 277 -
(Sandbakk et al., 2010) 60 PP 9 50 – – – 321 -
(Sandbakk et al., 2010) 60 PP 9 50 – – – 347 -
(Sandbakk et al., 2010) 60 PP 9 50 – – – 365 -
(Sandbakk et al., 2010) 60 PP 4.5 50 – – – 257 -
(Sandbakk et al., 2010) 60 PP 4.5 50 – – – 291 -
(Sandbakk et al., 2010) 60 PP 4.5 50 – – – 223 -
(Sandbakk et al., 2010) 60 PP 9 50 – – – 373 -
(Sandbakk et al., 2010) 60 PP 9 50 – – – 486 -
(Sandbakk et al., 2010) 60 PP 9 50 – – – 315 -
(Sandbakk et al., 2010) 60 PP 4.5 50 – – – 245 -
(Sandbakk et al., 2010) 60 PP 4.5 50 – – – 271 -
(Sandbakk et al., 2010) 60 PP 4.5 50 – – – 197 -
(Sandbakk et al., 2010) 60 PP 4.5 50 – – – 272 -
(Sandbakk et al., 2010) 60 PP 9 50 – – – 405 -
(Sandbakk et al., 2010) 60 PP 9 50 – – – 353 -
(Sandbakk et al., 2010) 60 PP 9 50 – – – 365 -
(Sandbakk et al., 2010) 60 PP 9 50 – – – 402 -
(Sandbakk et al., 2010) 60 PP 4.5 50 – – – 209 -
(Sandbakk et al., 2010) 60 PP 4.5 50 – – – 261 -
(Sandbakk et al., 2010) 60 PP 4.5 50 – – – 236 -
(Sandbakk et al., 2010) 60 PP 4.5 50 – – – 291 -
(Sandbakk et al., 2010) 60 PP 9 50 – – – 314 -
(Sandbakk et al., 2010) 60 PP 9 50 – – – 378 –

Fig. 11 (a) to (d) present the map colors for the radial (S11) and model can help engineers from the field to have an approximate per­
tangential (S22) stress evolution during excavation and shotcrete stages. formance level for the mechanical behavior of an FRS tunnel lining
Moreover, Fig. 12 presents the radial displacement evolution in exca­ application. Due to the versatility of the proposed workflow, it is
vation and shotcrete steps. possible to replicate this methodology for specific real FRS lining ap­
The analytical equation for circular tunnel excavation derived by plications, changing the rock/concrete parameters according to each
Panet et al. (Panet and Sulem, 2022), was also applied to validate the case.
proposed finite element model. The equation for the plastic zone radius
(Rp), as shown in Equation (5), from Mohr-Coulomb yield criteria was 5. Conclusions
then compared with the numerical model result.
[ ]K 1− 1 This article presented a workflow for predicting fiber-reinforced
RP
= 2×
(KP − 1)σ 0 + σc P
. (5) shotcrete toughness through ANNs and by applying the flexural pa­
R (1 − λ)(KP − 1)σ0 + σ c rameters of FRS in an elastoplastic finite element model of tunnel
where R is the tunnel excavation radius, σ0 is the initial isotropic excavation considering a steel fiber reinforcement. Steel fiber-reinforced
state of stress, σ C is the uniaxial compressive strength, and λ is the shotcrete lining design considers the toughness and residual flexural
deconfinement ratio. strength parameters. However, executing these tests in the laboratory is
Fig. 13 presents the radial and tangential stresses obtained by the usually costly and time-consuming. Therefore, artificial neural networks
finite element model developed in this article and the plastic and total (ANNs) have proven to be an alternative method for obtaining these
radii ratio (Rp/R) analytical value calculated with Equation (5), which parameters. Two distinct ANNs were trained and evaluated based on a
can also be translated as the plastic limit. The plastic limit obtained by database extracted from the literature considering steel and poly­
the finite element model equals 2.40, while the analytical solution propylene fibers. In this sense, all ANN models presented good accuracy,
radius ratio is 2.30, indicating a relative error of 4.2 %. In this sense, the with regression higher than 86 % and MSE in the order of 10e-3. Blind
model presents a good fit with the analytical solution. Fig. 14 presents tests were carried out to validate each network, presenting absolute
the radial displacement evolution of the numerical model, indicating a error values of less than 60 J and a relative error of less than 16 %. These
maximum radial displacement of 9.46 mm (0.00946 m). metrics are in a good range in terms of Engineering, especially in shot­
In this sense, the development of an elastoplastic finite element crete lining scenarios.

17
M. Congro et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 132 (2023) 104881

Table A8
Model database (part 8).
Authors fc Fiber Vf L AR T25 Standard T40 Standard
type (EN 14488–5) Deviation (ASTM C1550) Deviation
(MPa) - (kg/m3) (mm) - (J) (J) (J) (J)

(Sandbakk et al., 2010) 60 PP 9 50 – – – 332 -

(Sandbakk et al., 2010) 60 PP 9 50 – – – 393 -
(Myren and Bjotegaard, 2010) 56 SF 20 35 65 527 – 211 -
(Myren and Bjotegaard, 2010) 56 SF 20 35 65 453 – 181 -
(Myren and Bjotegaard, 2010) 56 SF 20 35 65 601 – 240 -
(Myren and Bjotegaard, 2010) 60 SF 20 35 65 583 – 233 -
(Myren and Bjotegaard, 2010) 60 SF 20 35 65 681 – 272 -
(Myren and Bjotegaard, 2010) 60 SF 20 35 65 485 – 194 -
(Myren and Bjotegaard, 2010) 60 SF 35 35 65 1645 – 658 -
(Myren and Bjotegaard, 2010) 60 SF 35 35 65 1748 – 699 -
(Myren and Bjotegaard, 2010) 60 SF 35 35 65 1542 – 617 -
(Myren and Bjotegaard, 2010) 51 SF 15 35 65 895 – 358 -
(Myren and Bjotegaard, 2010) 51 SF 15 35 65 1037 – 415 -
(Myren and Bjotegaard, 2010) 51 SF 15 35 65 753 – 301 -
(Myren and Bjotegaard, 2010) 40 SF 30 35 65 561 – 224 -
(Myren and Bjotegaard, 2010) 40 SF 30 35 65 638 – 255 -
(Myren and Bjotegaard, 2010) 40 SF 30 35 65 484 – 194 -
(Myren and Bjotegaard, 2010) 47 SF 45 35 65 1086 – 434 -
(Myren and Bjotegaard, 2010) 47 SF 45 35 65 1278 – 511 -
(Myren and Bjotegaard, 2010) 47 SF 45 35 65 894 – 358 -
(Myren and Bjotegaard, 2010) 52 SF 30 35 65 978 – 391 -
(Myren and Bjotegaard, 2010) 52 SF 30 35 65 1048 – 419 -
(Myren and Bjotegaard, 2010) 52 SF 30 35 65 908 – 363 -
(Myren and Bjotegaard, 2010) 64 PP 5.5 50 – 896 – 358 -
(Myren and Bjotegaard, 2010) 64 PP 5.5 50 – 920 – 368 -
(Myren and Bjotegaard, 2010) 64 PP 5.5 50 – 872 – 349 -
(Myren and Bjotegaard, 2010) 59 PP 8 50 – 1226 – 490 -
(Myren and Bjotegaard, 2010) 59 PP 8 50 – 1459 – 584 -
(Myren and Bjotegaard, 2010) 59 PP 8 50 – 993 – 397 -
(Myren and Bjotegaard, 2010) 64 PP 5.5 50 – 912 – 365 -
(Myren and Bjotegaard, 2010) 64 PP 5.5 50 – 1056 – 422 -
(Myren and Bjotegaard, 2010) 64 PP 5.5 50 – 768 – 307 -
(Myren and Bjotegaard, 2010) 58 PP 3 50 – 563 – 225 -
(Myren and Bjotegaard, 2010) 58 PP 3 50 – 615 – 246 -
(Myren and Bjotegaard, 2010) 58 PP 3 50 – 511 – 204 -
(Myren and Bjotegaard, 2009) 60 SF 20 35 65 1062 – 425 -
(Myren and Bjotegaard, 2009) 60 SF 20 35 65 1261 – 504 -
(Myren and Bjotegaard, 2009) 60 SF 20 35 65 949 – 380 -
(Myren and Bjotegaard, 2009) 60 SF 20 35 65 1358 – 543 -
(Malmgren, 2007) 50 SF 50 35 65 1200 – 480 -
(Malmgren, 2007) 50 SF 50 35 65 1344 – 538 –

Table A9
Model database (part 9).
Authors fc Fiber Vf L AR T25 Standard T40 Standard
type (EN 14488–5) Deviation (ASTM C1550) Deviation
3
(MPa) - (kg/m ) (mm) - (J) (J) (J) (J)

(Malmgren, 2007) 50 SF 50 35 65 1056 – 422 -

(Malmgren, 2007) 50 PP 10 50 – 1100 – 440 -
(Malmgren, 2007) 50 PP 10 50 – 1232 – 493 -
(Malmgren, 2007) 50 PP 10 50 – 968 – 387 -
(Malmgren, 2007) 56 PP 8 50 – 1190 – 476 -
(Malmgren, 2007) 56 PP 8 50 – 1333 – 533 -
(Malmgren, 2007) 56 PP 8 50 – 1047 – 419 -
(Malmgren, 2007) 55 PP 10 50 – 1190 – 476 -
(Malmgren, 2007) 55 PP 10 50 – 1333 – 533 -
(Malmgren, 2007) 55 PP 10 50 – 1047 – 419 -
(Malmgren, 2007) 58 PP 8 55 – 790 – 316 -
(Malmgren, 2007) 58 PP 8 55 – 885 – 354 -
(Malmgren, 2007) 58 PP 8 55 – 695 – 278 -
(Malmgren, 2007) 59 PP 10 55 – 1060 – 424 -
(Malmgren, 2007) 59 PP 10 55 – 1187 – 475 -
(Malmgren, 2007) 59 PP 10 55 – 933 – 373 -
(Rengarajan, 2020) 52 SF 50 35 65 – – 700 -
(Rengarajan, 2020) 52 SF 50 35 65 – – 750 -
(Rengarajan, 2020) 52 SF 50 35 65 – – 710 –

18
M. Congro et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 132 (2023) 104881

The toughness parameter predicted by each ANN can be used in References

tunnel linings design, adopting a design safety factor to consider phe­
nomena such as fiber rebound, batch-to-batch variation on other un­ [1] Materials and Structures 36, 2003, 560–567.
[2] Materials and Structures 36, 2003, 560–567.
predictable events that occurred during the construction phase. It is [3] European Committee for Standardization, 2005.
important to note that the ANN can be used to estimate the FRS’s mean [4] Techincal report 34, 2016.
performance level. However, experiments and material quality control Aci 506.1r., 1998. Guide to fiber-reinforced shotcrete. American Concrete Institute.
Aci 544.4r., 2018. Guide to design with fiber-reinforced concrete. ACI Committee.
testing must be additionally carried out during the design stage. Alter­ Adhikary, B. Mutsuyoshi, H. Prediction of shear of steel fiber RC beams using neural
native methodologies can also be carried out for tunnel lining design. networks. Construction and Building Materials (2006), 20:801-811, 10.1016/j.
One example involves adopting an elastoplastic finite element model to conbuildmat.2005.01.047.
Ahmadi, M. Naderpour, H. Kheyroddin, A. Utilization of artificial neural networks to
evaluate the displacements and stresses around the tunnel lining. prediction of the capacity of CCFT short columns subject to short term axial load.
The concrete damaged plasticity constitutive model is selected to Archives of Civil and Mechanical Engineering (2014), 14:510-517, 10.1016/j.
model the concrete constitutive behavior. In this sense, an ANN model acme.2014.01.006.
Aire, C., Aguilar, L., 2021. Fiber reinforced shotcrete control tests in the Mexico City
proposed by Congro et al. (Congro et al., 2021) was used to assess the
metro line 12 tunnel. Revista Alconpat 11, 73–87.
residual flexural strength parameters to define the post-peak stress–­ Alejano, L. Rodriguez-Dono, A. Alonso, E. Fdez.-Manín, G. Ground reaction curves for
strain curve in tension and compression according to the equations tunnels excavated in different quality rock masses showing several types of post-
proposed by RILEM TC 162-TDF. Therefore, numerical simulation was failure behavior. Tunnelling and Underground Space Technology (2009), 24:689-
705.
carried out, from which radial displacements and the stress distribution Alonso, E. Alejano, L. Varas, F. Fdez.-Manín, G. Carranza-Torres, C. Ground reaction
around the tunnel were assessed. The results from the numerical model curves for rock masses exhibiting Hoek-Brown strain-softening behavior.
were then compared to the analytical solution of circular tunnel exca­ International Journal for Numerical and Analytical Methods in Geomechanics
(2003), 23:1153-1185.
vations. A relative error of 4.2 % between the analytical and numerical Amin, A., Foster, S., Gilbert, R., Kaufmann, W., 2017. Material characterization of macro
plastic and total radii ratio (Rp/R) was reported, validating the numer­ synthetic fibre reinforced concrete. Cement and Concrete Composites 84, 124–133.
ical model. Astm c1550., 2012. Standard Method for Flexural Toughness of Fiber Reinforced
Concrete. American Society for Testing and Material.
Given the difficulty in carrying out experiments in the laboratory or Bai, L., 2022. Evaluation of interface shear transfer strength of steel fiber-reinforced
field to assess information concerning the fiber-reinforced tunnel lining, concrete based on artificial neural network and regression method. Structural
this framework emerges as an interesting alternative. With its applica­ Concrete 23 (2), 1049–1064.
Barros, J., 1998. Experimental behavior of mesh shotcrete and steel fiber reinforced
tion, it is possible to obtain the parameters of toughness and residual shotcrete panels. Materials Science Journal.
flexural strength in a more straightforward, faster, and economical way. Bentur, A., Mindness, S., 2007. Fibre reinforced cementitious composites, 2nd edition.
Hence, the results provided here are promising from an Engineering Taylor & Francis, London.
Bernard, E., 2002. Correlations in the behavior of fibre reinforced shotcrete beam and
point of view.
panel specimens. Materials and Structures 35, 156–164.
Bernard, E., 2019. Effect of Friction on Performance of Fiber-Reinforced Concrete in the
CRediT authorship contribution statement ASTM C1550 Panel Test. Advances in Civil Engineering Materials 8, 258–297.
Bernard, E., 2020. Estimating residual flexural strength of fiber-reinforced concrete using
the ASTM C1550 panel test. Advances in Civil Engineering Materials 9, 494–1492.
Marcello Congro: Conceptualization, Methodology, Software, Bernard, S., 2021. Influence of fiber geometry and type on creep rupture of cracked fiber-
Validation, Formal analysis, Writing – original draft, Writing – review & reinforced shotcrete mixtures. ACI Materials Journal 118, 3–12.
Bernard, E. Xu, G. Carino, N. Precision of the ASTM C1550 panel test and field variation
editing, Visualization. Vitor Moreira de Alencar Monteiro: Concep­ in measured FRS performance. Shotcrete: Elements of a System. 1st edition. CRC
tualization, Methodology, Resources, Formal analysis, Writing – original Press; 2010.
draft, Writing – review & editing. Flávio de Andrade Silva: Concep­ Bernard, E.S., Xu, G.G., Carino, N.J., 2010. Influence of the number of replicates in a
batch on apparent variability in FRC and FRS performance assessed using ASTM
tualization, Methodology, Supervision, Writing – review & editing, C1550 panels. In: Shotcrete: Elements of a System, 1st edition. CRC Press, p. 10 p..
Project administration. Deane Roehl: Conceptualization, Methodology, https://doi.org/10.1201/b10545-8.
Writing – review & editing, Supervision, Project administration. Bernard, E., Xu, G., Carino, N., 2010. Influence of the number of replicates in a batch on
apparent variability in FRS performance assessed using ASTM C1550 panels.
Amanda L.T. Brandão: Conceptualization, Methodology, Writing – Shotcrete: Elements of a System, 1st edition. CRC Press.
review & editing, Supervision. Bernard, E. The influence of curing on the mechanical performance of fibre reinforced
shotcrete. In: Proceedings of the 10th International Conference on Shotcrete for
Underground Supporte (2012).
Bernard, E., 2015. Age-dependent changes in post-crack performance of fibre reinforced
Declaration of Competing Interest shotcrete linings. Tunnelling and Underground Space Technology 49, 241–248.
Bieniawski, Z. Engineering rock mass classification in rock engineering. In: proceedings
for rock engineering symposium (1976).
The authors declare that they have no known competing financial
Bjontegaard, O. Myren, S. Fibre reinforced spreyed concrete panel test - test procedures
interests or personal relationships that could have appeared to influence and influencing factors. In: Spritzbeton Tagung (2018).
the work reported in this paper. Brady, B., Brown, E., 1993. Rock mechanics for underground mining, 2nd editions.
Chapman and Hall.
Brown, E.T., Bray, J.W., Ladanyi, B., Hoek, E., 1983. Ground response curves for rock
Data availability tunnels. Journal of Geotechnical Engineering 109 (1), 15–39.
Buratti, N., Incerti, A., Tilocca, A., Mazzotti, C., Paparella, M., Draconte, M., 2019.
Data will be made available on request. Energy absorption tests on fibre-reinforced-shotcrete round and square panels.
Tunnels and Underground Cities: Engineering and Innovation Meet Archaeology,
Architecture and Art, 1st edition. CRC Press.
Acknowledgements Cáceres, A., Cavalaro, S., Figueredo, A., 2021. Evaluation of steel fiber reinforced
sprayed concrete by energy absorption tests. Journal of Materials in Civil
Engineering 33.
This study was financed in part by the Coordenação de Aperfeiçoa­ Carmona, S., Molins, C., Garcia, S., 2020. Application of Barcelona test for controlling
mento de Pessoal de Nível Superior – Brasil (CAPES) – Finance Code 001, energy absorption capacity of FRS in underground mining works. Construction and
Building Materials 246.
Conselho Nacional de Pesquisa (CNPq) – Process 309384/2019-2, and Carranza-Torres, C., Fairhurst, C., 2000. Application of convergence-confinement
Fundação de Amparo à Pesquisa do Estado do Rio de Janeiro (FAPERJ) – method of tunnel design to rock masses that satisfy the Hoek-Brown failure criteria.
Process E-26/202.928/2019. Tunnelling and Underground Space Technology 15, 187–213.
Cascardi, A., Micelli, F., Aiello, M., 2017. An artificial neural networks model for the
prediction of the compressive strength of FRP-confined concrete circular columns.
Appendices Engineering Structures 140, 199–208.
Cavalaro, S., Aguado, A., 2015. Intrinsic scatter of FRC: an alternative philosophy to
estimate characteristic values. Materials and Structures 48, 3537–3555.
CEB-FIB: Model Code. International Federation for Structural Concrete (FIB) (2012).

19
M. Congro et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 132 (2023) 104881

Cengiz, O., Luranli, L., 2004. Comparative of steel mesh, steel fibre and high- Martin, L., Clark, C., Seymour, J., Stepan, M., 2015. Shotcrete design and installation
performance polypropylene fibre reinforced shotcrete in panel test. Ccement and compliance testing: early strength, load capacity, toughness, adhesion strength and
Concrete Research 34, 1357–1364. applied quality. Report of investigation 9697.
Chiai, B. Fantilli, A. Vallini, P. Combining fiber-reinforced concrete with traditional Massone, L., Nazar, F., 2018. Analytical and experimental evaluation of the use of fibers
reinforcement in tunnel linings. Engineering Structures (2009), 31-1600-1606. as partial reinforcement in shotcrete for tunnels in Chile. Tunnelling and
Chuang, P., Goh, A., Wu, X., 1998. Modeling the capacity of pin-ended slender reinforced Underground Space Technology 77, 13–25.
concrete columns using neural networks. Journal of Structural Engineering 124, McKay, M.D., Beckman, R.J., Conover, W.J., 1979. A Comparison of Three Methods for
830–838. Selecting Values of Input Variables in the Analysis of Output from a Computer Code.
Congro, M. Pereira, F L G. Souza, L M S, Roehl, D. Parameter sensitivity analysis of steel Technometrics 21 (2), 239–245.
fiber reinforced concrete. Proceedings of XL CILAMCE – Ibero-Latin American Merma, A., Santos, B., Rego, A., Hacha, R., Torem, M., 2020. Treatment of oily
Congress on Computational Methods in Engineering (2019). Natal/RN, Brazil. wastewater from mining industry using electrocoagulation: Fundamentals and
Congro, M., Monteiro, V., Brandão, A., Santos, B., Roehl, D., Silva, F., 2021. Prediction of process optimization. Journal of Materials Research and Technology 9,
the residual flexural strength of fiber reinforced concrete using artificial neural 15164–15176.
networks. Construction and Building Materials 303. Mobasher, B., 2019. Mechanics of fiber and textile reinforced cement composites, 1st
de Alencar Monteiro, V.M., de Andrade Silva, F., 2021. on the design of the fiber edition. CRC Press.
reinforced shotcrete applied as primary rock support in the Cuiabá underground Montgomery, D. Design and Analysis of Experiments. 8th edition, John Wiley & Sons
mining excavations: a case study. Case Studies. Construction Materials 15, e00784. (2013). New York: USA.
de la Fuente, A., Pujadas, P., Blanco, A., Aguado, A., 2012. Experiences in Barcelona with Myren, S., Bjotegaard, O., 2009. Energy absorption capacity for fibre reinforced sprayed
the use of fibres in segmental linings. Tunnelling and Underground Space concrete: effect of friction in round and square panel tests with continuous support
Technology 27, 60–71. (Series 4). Technology report 2534.
Decker, J., Madsen, P., Gall, V., 2012. Use od synthetic, fiber reinforced, initial shotcrete Myren, S., Bjotegaard, O., 2010. Round and square panel tests - a comparative study.
lining at Devil’s Slide tunnel. Journal of the Transportation Research Board 2313, Shotcrete: Elements of a System, 1st edition. CRC Press.
147–154. Naseri, S., Bahrani, N., 2021. Design of initial shotcrete lining for a mine shaft using two-
Demir, F. Prediction of elastic modulus of normal and high strength concrete by artificial dimensional finite element models considering excavation advance rate.
neural networks. Construction and Building Materials (2008), 22:1428-1435, Geotechnical Geological Engineering Journal 39 (7), 4709–4732.
10.1016/j.conbuildmat.2007.04.004. Neuner, M., Schreter, M., Gamnitzer, P., Hofstetter, G., 2020. On discrepancies between
di Colombo, M., Prisco, M., 2009. Sprayed tunnel linings: A comparison between several time-dependent nonlinear 3D and 2D finite element simulations of deep tunnel
reinforcement solutions. Materials and Structures 42, 1295–1311. advance: A numerical study on the Brenner Base Tunnel. Computers and Geotechnics
Ding, S., Hui, L., Su, C., Yu, J., Jin, F., 2013. Evolutionary artificial neural networks: a Journal 119, 103355.
review. Artificial Intelligence Review 39, 251–260. Nie, Z.H., Jin, X.G., Luo, W., 2021. Numerical simulation analysis of tunnel construction
Ding, Y., Kusterle, W., 1999. Comparative study of steel fibre-reinforced concrete and mechanical response considering shotcrete hardening process. J. Phys.: Conf. Ser.
steel mesh-reinforced concrete at early ages in panel tests. Cement and Concrete 2045 (1), 012026.
Research 29, 1827–1934. Oreste, P.P., 2003. A procedure for determining the reaction curve of shotcrete lining
Dung, C., Anh, L., 2019. Autonomous concrete crack detection using deep fully considering transient conditions. Rock Mechanics and Rock Engineering Journal 36
convolutional neural network. Automation in Construction 99, 52–58. (3), 209–236.
EN 14488-5. Testing Sprayed Concrete – Part 5: determination of energy absorption Oreste, P., 2003. Analysis of structural interaction in tunnels using the convergence-
capacity of fibre reinforced slab specimens European Committee for Standardization confinement approach. Tunnelling and Underground Space Technology 18,
(2006). 347–363.
Foresee, F. Hagan, M. Gauss-Newton approximation to Bayesian learning. Proceedings of Oreste, P.P., Pella, D., 1997. Modelling progressive hardening of shotcrete in
the International Joint Conference on Neural Networks (1997). convergence-confinement approach to tunnel design. Tunnelling and Underground
Gallo, A., Mejiá, C., 2013. Fiber reinforced concrete performance parameters using ASTM Space Technology 12 (3), 425–431.
C1550 and EN 14488–5 for rock support in tunnels, Mexico. In: Proceedings of 47th Panet, M. Sulem, J. (2022). Longitudinal Displacement Profile. In: Convergence-
U.S. Rock Mechanics/Geomechanics Symposium. Confinement Method for Tunnel Design. Springer Tracts in Civil Engineering (2022).
Garcia, T., Aguado, A., 2011. Evaluación de la tenacidad en el hormigón proyectado Springer, Cham. doi.org/10.1007/978-3-030-93193-3_5.
reforzado con fibras poliméricas de alto módulo. Boletin de la Sociedad Espanola de Papworth, F. Design guidelines for the use of fibre reinforced shotcrete in ground
Ceramica y Vidrio 43, 552–555. support. In: proceedings of the 30th conference on our world in concrete and
Garcia-Taengua, E., Bakhshi, M., Ferrara, L., 2021. Meta-analysis of steel fiber-reinforced structures (2002).
concrete mixtures leads to practical mix design methodology. Materials 14. Prasad, B. Eskandari, H. Reddy, B. Prediction of compressive strength of SCC and HPC
E. Garcia-Taengua, Using decades of data to rethink proportioning and optimization of with high volume fly ash using ANN. Construction and Building Materials (2009),
FRC mixes: the OptiFRC project. In: RILEM-fib X International Symposium on Fibre 23:117-128, 10.1016/j.conbuildmat.2008.01.014.
Reinforced Cocnrete, 2020. Rafiei Renani, H., Martin, C.D., Hudson, R., 2016. Back analysis of rock mass
Grimstad, E. Barton, N. Updating the Q-system for NMT. In: proceedings of the displacements around a deep shaft using two- and three-dimensional continuum
International Symposium on Sprayed Concrete (1993). modeling. Rock Mechanics and Rock Engineering 49 (4), 1313–1327.
Grimstad, E. Barton, N. Updating the Q-system for NMT. In: proceedings of modern use of Rahman, J., Sakil, A.H., Khan, N.I., Islam, K., Mangalathu, S., 2021. Data-driven shear
wet mix sprayed concrete for underground support symposium (1993). strength prediction of steel fiber reinforced concrete beams using machine learning
Hossain, K. Gladson, L. Anwar, M. Modeling shear strength of medium- to ultra-high- approach. Engineering Structures 233.
strength steel fiber reinforced concrete beams using artificial neural networks. Rego, A., Valim, I., Vieira, A., Vilani, C., Santos, B., 2018. Optimization of sugarcane
Neural Computing and Applications (2017), 28:1119-1130, doi.org/10.1007/ bagasse pretreatment using alcaline hydrogen peroxidethrough ANN and ANFIS
s00521-016-2417-2. modelling. Bioresource Technology 267, 634–641.
Hossain, K.M.A., Gladson, L.R., Anwar, M.S., 2017. Modeling shear strength of medium- Rengarajan, M. Laboratory testing of shotcrete with fibres of steel, basalt or synthetic
to ultra-high-strength steel fiber-reinforced concrete beams using artificial neural materials. MSc. Dissertation (2020): KTH Royal Institute of Technology, Sweden.
network. Neural Computing and Applications 28, 1119–1130. Rispin, M. Kleven, O. Dimmock, R. Myrdal, R. Shotcrete: early strength and re-entry
Juhasz, P., Nagy, L., Schaul, P., 2017. Correlawtion of the results of the standard beam revisited – practices and technology. In: proceedings of the 1st International
and EFNARC panel test. In: Proceedings of the World Tunnel Congress. Conference on underground Mining Technology (2017).
Kang, M. Yoo, D. Gupta, R. Machine learning-based prediction for compressive and Ryabchikov, A., Kiviste, M., Udras, M., Lindpere, M., Vassijev, A., Korb, N., 2020. The
flexural strengths of steel fiber-reinforced concrete. Construction and Building experimental investigation of the mechanical properties of steel fibre-reinforced
Materials (2021), 266, 10.1016/j.conbuildmat.2020.121117. concrete according to different testing standards. Agronomy Research 18, 969–979.
Kaufmann, J., Frech, K., Schuetz, P., et al., 2013. Rebound and orientation of fibers in Salehian, H., Barros, J., Taheri, M., 2014. Evaluation of the influence of post-cracking
wet sprayed concrete applications. Construction and Building Materials 49, 15–22. response of steel fibre reinforced concrete (SFRC) on load carrying capacity of SFRC
https://doi.org/10.1016/j.conbuildmat.2013.07.051. panels. Construction and Building Materials 73, 289–304.
Larsson, P. Optimization of the steel-fiber dosage in shotcrete used in the Kankberg mine. Sandbakk, S., Kanstad, T., Bjontegaard, O., Vandewalle, L., Parmentier, B., 2010.
MSc dissertation (2018): Luleå University of Technology, Sweden. International round robin testing of circular FRC slabs. COIN Project Report 23.
Li, S., Zhao, X., Zhou, G., 2019. Automatic pixel-level multiple damage detection of Schreter, M., Neuner, M., Unteregger, D., Hofstetter, G., 2018. On the importance of
concrete structure using fully convolutional network. Computer-Aided Civil and advanced constitutive models in finite element simulations of deep tunnel advance.
Infrastructure Engineering 34, 616–634. Tunnelling and Underground Space Technology 80, 103–113.
Louchnikov, V. Sandy, M. Watson, O. Orunesu, M. Eremenko, V. An overview of surface Silva, F. Monteiro, V. Estudo da tenacidade de concreto projetado reforçado com fibras
rock support for deformable ground conditions. In: proceedings of the 12th AUSIMM de aço de acordo com a norma EN14488-5. Final Report (2021). PUC-Rio: Pontifícia
Underground Operators’ Conference (2014). Universidade Católica do Rio de Janeiro, Brazil.
MacKay, J., 1992. Bayesian interpolation. Neural computation 4, 415–447. Silva, F. Monteiro, V. Avaliação das propriedades mecânicas do concreto projetado com
Malmgren, L., 2007. Strength, ductility and stiffness of fibre-reinforced shotcrete. fibras da Mina Caraíba. Final Report (2021). PUC-Rio: Pontifícia Universidade
Magazine of Concrete Research 59, 287–296. Católica do Rio de Janeiro, Brazil.
Manfredi, R., Silva, F., 2020. Test methods for the characterization of polypropylene Silva, F. Monteiro, V. Estudo da tenacidade de concreto projetado reforçado com 15, 20 e
fiber reinforced concrete: a comparative analysis. KSCE Journal of Civil Engineering 25 kg/m3 fibras de aço de acordo com a norma EN14488-5. Final Report (2021),
24, 856–866. PUC-Rio: Pontifícia Universidade Católica do Rio de Janeiro, Brazil.
Marquardt, D., 1963. An Algorithm for Least-Squares Estimation of Nonlinear
Parameters. SIAM Journal on Applied Mathematics 11, 431–441.

20
M. Congro et al. Tunnelling and Underground Space Technology incorporating Trenchless Technology Research 132 (2023) 104881

Silva, F. Manfredi, R. Lima, V. Estudo da tenacidade de concreto reforçados com fibras de Vissotto Júnior, L. Análise tridimensional de mina subterrânea com ênfase na interação
acordo com a norma ASTM 1550. Final Report (2021), PUC-Rio: Pontifícia entre maciço e preenchimento (Caso de estudo: Mina Cuiabá). MSc dissertation
Universidade Católica do Rio de Janeiro, Brazil. (2013): Universidade de Brasília, Brazil.
Sjölander, A., Ansell, A., 2017. Numerical simulations of restrained shrinkage cracking in Vlachopoulos, N., Diederichs, M.S., 2014. Appropriate uses and practical limitations of
glass fibre reinforced shotcrete slabs. Advances in Civil Engineering. 2D numerical analysis of tunnels and tunnel support response. Geotechnical and
Tayfur, G. Erdem, T. Kirca, O. Strength prediction of high-strength concrete by fuzzy Geological Engineering 32 (2), 469–488.
logic and artificial neural networks. Journal of Materials in Civil Engineering (2014), Vu, Q. Truong, V. Thai, H. Machine learning-based prediction of CFST columns using
26, 10.1061/(ASCE)MT.1943-5533.0000985. gradient tree boosting algorithm. Composite Structures (2021), 259, 10.1016/j.
Teixeira, M.C., Brandão, A.L.T., Parente, A.P., Pereira, M.V., 2022. Artificial intelligence compstruct.2020.113505.
modeling of ultrasonic fatigue test to predict the temperature increase. International Wickham, G. Tiedemann, H. Skinner, E. Support determination based on geologic
Journal of Fatigue 163, 106999. predictions. In: proceedings of the North American rapid excavation tunneling
Thai, H., 2022. Machine learning for structural engineering: A state-of-the-art review. conference (1972).
Structures 38. Yang, J., Kim, J., Yoo, D., 2017. Performance of shotcrete containing amorphous fibers
Trottier, J., Banthia, N., 2000. Toughness characterization of fibre reinforced concrete by for tunnel applications. Tunnelling and Underground Space Technology 64, 85–94.
JSCE approach. Cement and Concrete research 4, 593–597.

21