Rock Mechanics and Rock Engineering (2023) 56:5293–5313

[Link]

ORIGINAL PAPER

Modeling Brazilian Tensile Strength Tests on a Brittle Rock Using

Deterministic, Semi‑deterministic, and Voronoi Bonded Block Models
Carlos E. Contreras Inga1 · Sankhaneel Sinha1,2 · Gabriel Walton1 · Elizabeth Holley3

Received: 23 February 2022 / Accepted: 2 April 2023 / Published online: 25 April 2023
© The Author(s), under exclusive licence to Springer-Verlag GmbH Austria, part of Springer Nature 2023

Abstract
This study aims to numerically investigate how various common simplifications of grain structure representation in bonded
block models affect simulations of rock mechanical behavior. Specimens of Wausau granite were characterized mechani-
cally through Brazilian tensile strength tests for this work. The samples were also characterized petrographically using thin
section microscopy, scanning electron microscopy-based automated mineralogy, and visual inspection. Four types of repre-
sentations of the Wausau granite samples were developed, including 6 detailed manually developed deterministic models, 6
semi-deterministic models, and 120 randomly generated representations (Voronoi models). First, a calibrated set of micro-
properties was determined using the deterministic representations to simulate the Brazilian tensile strength measurements.
Next, the study examined the ability of different Voronoi tessellations to adequately represent the grain structure for the
purposes of accurate tensile strength simulation. This was evaluated by comparing Voronoi model results to the deterministic
grain structure model results and laboratory test results. The findings of the study show that the four types of models used
in this study can all provide realistic representations of the mechanical behavior of rock. The study confirms that standard
Voronoi approximations of grain structures can be reasonably used in lieu of less practical, manually developed representa-
tions of the grain structure. Specifically, Voronoi models can properly replicate the geometric heterogeneity within the grain
structure, even though they simplify some of its geometric attributes.

Highlights

• Four types of models of granite specimens were generated, each type representing the specimen grain structure with a
different degree of realism.
• Brazilian tensile strength simulation results obtained using deterministic, semi-deterministic, and two different Voronoi
structures were compared.
• Validity of randomly generated Voronoi models to adequately approximate the geometric heterogeneity within the grain
structure was investigated.
• Voronoi models provided strengths nearly equivalent to those obtained from the more complex deterministic and semi-
deterministic models.

Keywords Grain structure representation · Heterogeneity · Micro-properties · Bonded block models · Voronoi tessellations

* Carlos E. Contreras Inga 1 Introduction

ccontreras@[Link]; [Link]@[Link]
1
Department of Geology and Geological Engineering, Laboratory tests such as uniaxial compression, triaxial com-
Colorado School of Mines, 1516 Illinois St., Golden, CO, pression, and tensile strength tests are commonly used to
USA characterize the strength of rocks. However, accurate charac-
2
WSP Golder, 426 N 44th St., Suite 375, Phoenix, AZ, USA terization of the fracturing behavior of a rock can sometimes
3
Department of Mining Engineering, Colorado School be difficult to achieve by these means due to the hetero-
of Mines, 1600 Illinois St., Golden, CO, USA geneous nature of the rock’s grain structure. According to

13
Vol.:(0123456789)
5294 C. E. C. Inga et al.

experimental studies (Martin 1993; Eberhardt 1998; Staub Grasselli 2014; Zhang and Wong 2018; Wang and Cai 2019).
et al. 2004), individual specimens of the same rock type Detailed descriptions of these approaches can be found in
can present variations in fracturing behavior and associated the literature (Jing and Hudson 2002; Jing 2003; Jing and
strength because of grain-scale heterogeneities. The hetero- Stephansson 2007; Bobet et al. 2009) and are beyond the
geneities of the rock grain structure cause complex micro- scope of the current study.
mechanical interactions and generate localized tensile stress Discontinuum and hybrid methods can represent the het-
concentrations that result in fracture development within a erogeneity within crystalline rocks more accurately than
rock undergoing compressive loading (Diederichs 2003; Gao continuum methods (Zhang and Wong 2018). The Discrete
et al. 2016; Liu et al. 2018; Wang and Cai 2018). Hetero- Element Method (DEM) and the Finite Discrete Element
geneity can present due to elastic mismatch between con- Method (FDEM) are the most commonly applied approaches
stituent mineral grains, difference in grain shape and size, for rock damage modeling among the discontinuum and
and/or presence of micro-flaws, such as pores, cleavage, and hybrid methods, respectively. The DEM simulates rocks as
strings of grain boundary cavities (Sprunt and Brace 1974; assemblies of discrete particles or blocks, which interact
Tapponnier and Brace 1976; Dey and Wang 1981; Kranz between each other and separate as fractures develop (Cund-
1983; Cunha 1990). Thus, the grain structure heterogeneity all 1971; Jing and Hudson 2002; Jing and Stephansson 2007;
controls the emergent macroscopic mechanical response of Lisjak and Grasselli 2014). The hybrid FDEM, can represent
the intact rock (Lan et al. 2010; Mahabadi 2012; Gao et al. grains of irregular shape by connecting several triangular
2016). Lan et al. (2010) formally classified the sources of elements (Munjiza 2004) starting from a continuum repre-
heterogeneity in intact rocks into three groups: (i) geometric sentation of the rock that progressively develops new frac-
heterogeneity related to variability in size and shape of the tures and separates into discrete bodies (Lisjak and Grasselli
mineral grains, (ii) deformability heterogeneity associated 2014). Among DEM models, the Bonded Particle Model
with the contrasts among mineral grains in terms of density (BPM) and Bonded Block Model (BBM) approaches are
and elastic properties, and (iii) contact heterogeneity linked commonly used for simulating intact rock behavior. BPMs
to the variability of the stiffness, length, orientation, and represent grains as circular or spherical particles bonded
distribution of contacts between mineral grains. Numerous at their contacts (Cundall and Strack 1979; Potyondy and
experimental studies (Olsson 1974; Singh 1988; Fredrich Cundall 2004; Bewick et al. 2014). Since its introduction,
et al. 1990; Shea and Kronenberg 1993; Wong et al. 1996; a number of limitations were identified in BPM (Cho et al.
Eberhardt et al. 1999; Tuǧrul and Zarif 1999; Přikryl 2001; 2007) and modifications were proposed to overcome these
Güneş Yilmaz et al. 2011; Keikha and Keykha 2013; Yesi- limitations (Potyondy and Cundall 2004; Cho et al. 2007;
loglu-Gultekin et al. 2013; Sajid et al. 2016; Cowie and Wal- Potyondy 2012; Scholtés and Donzé 2013; Ding and Zhang
ton 2018) have explored the effects of grain size and mineral 2014). Many of these modifications attempted to overcome
content variability on the mechanical response of rock under the lack of particle–particle interlocking in BPM by fus-
loading, but these studies do not provide a clear consensus ing neighboring spheres/discs. BBMs, on the other hand,
on the influences of specific grain structure attributes. represent grain structures as collections of polygonal or
In recent decades, numerical modeling has been increas- polyhedral blocks (or grains) bonded at their interfaces (or
ingly used to quantitatively investigate damage processes grain–grain contacts) (Garza-Cruz et al. 2014).
of brittle rocks. Notably, numerical modeling allows one The Voronoi tessellation approach is usually applied to
to simulate conditions that are difficult to attain and study generate the grain structure for a BBM, since the mathe-
in the laboratory (Li et al. 2017a, 2019). Numerical mod- matical process of Voronoi tessellation is the most conven-
eling approaches for simulation of brittle rock damage are ient technique to randomly generate polygonal or polyhe-
usually classified into three categories: (i) continuum, (ii) dral shapes. Polygonal blocks with four or more sides (or
discontinuum, and (iii) hybrid continuum–discontinuum complex polyhedral blocks in 3D) generated with Voro-
(Jing 2003; Potyondy and Cundall 2004). The continuum noi tessellations are usually referred to as Voronoi blocks
approach represents the rock as a single continuous body (Ghazvinian et al. 2014; Li et al. 2019; Sinha and Walton
and employs constitutive relations and associated failure 2020). Triangular blocks (or tetrahedral blocks in 3D) pro-
criteria to define rock damage (Jing 2003), but it is unable duced with a modification of the Voronoi logic are called
to explicitly represent fracture development. The common trigons (Gao and Stead 2014). The Grain-Based Model
continuum modeling approaches are Finite Element Method (GBM) approach was introduced in BPMs to improve the
(FEM) and Finite Difference Method (FDM) (Jing 2003). representation of grain structures by adding grain-bound-
In contrast, discontinuum and hybrid continuum–discon- aries defined by smooth joint elements to a BPM, which
tinuum approaches explicitly simulate rock damage devel- are often defined using Voronoi tessellations (Bahrani
opment under various loading conditions without using et al. 2014). Similarly, the hybrid FDEM uses the GBM
pre-defined macroscopic constitutive models (Lisjak and approach to define the grain structure through Voronoi

13
Modeling Brazilian Tensile Strength Tests on a Brittle Rock Using Deterministic,… 5295

tessellations (Abdelaziz et al. 2018; Li et al. 2021a). More average, however, it was observed that Voronoi grain struc-
details about all these methods can be found in the litera- ture models could provide realistic rock strength predictions
ture (Munjiza 2004; Bahrani et al. 2014; Ghazvinian et al. when used in combination with appropriate micro-properties
2014; Lisjak and Grasselli 2014; Abdelaziz et al. 2018; (Contreras Inga et al. 2021).
Zhang and Wong 2018). Although it is well-established that Voronoi block assem-
It is generally recognized that assemblies of Voronoi blies can reasonably approximate grain interlocking in low
polygonal/polyhedral blocks provide more realistic rep- porosity igneous rocks in contrast to assemblies of circular/
resentations of grain structure geometric heterogeneity as spherical or triangular/tetrahedral grains, it has not yet been
compared to assemblies of other block shapes (Ghazvinian determined to what extent the simplifications inherent in the
et al. 2014; Gao et al. 2016; Wang and Cai 2018, 2019; Voronoi approach affect mechanical behaviors compared to
Li et al. 2019). This is because they provide more realistic more detailed and realistic grain structure representations.
grain interlocking within the grain structure. Voronoi tessel- To better understand the impact of grain structure simplifi-
lations have been widely used to simulate grain structures for cations sought in prior studies, multiple Brazilian Tensile
this reason (Ghazvinian et al. 2014; Abdelaziz et al. 2018; Strength (BTS; International Society for Rock Mechanics
Sinha and Walton 2018, 2020; Wang and Cai 2018, 2019). 1978; Li and Wong 2013; Perras and Diederichs 2014; Li
Several studies have numerically examined the influence of et al. 2021b) simulations were performed with BBM grain
grain structure heterogeneity on the mechanical behavior structures of differing degrees of complexity and a calibrated
of rock using Voronoi tessellations. Most of these studies set of micro-properties. This approach differs from previ-
have focused on the effects of average grain size and grain ous studies (e.g., Contreras Inga et al. 2021), which have
size distribution (Kazerani and Zhao 2010; Ghazvinian et al. considered variations in basic grain structure parameters
2014; Nicksiar and Martin 2014; Fabjan et al. 2015; Azocar (e.g., grain size and grain shape) without altering the over-
2016; Gao et al. 2016; Gui et al. 2016; Peng et al. 2017a, all level of complexity of the grain structure representation.
2017b, 2021; Liu et al. 2018; Li et al. 2020, 2021a; Xu et al. The approach is novel in the sense that this is the first-time
2020; Contreras Inga et al. 2021). Some studies have also geometric heterogeneity is studied in the context of BBM
attempted to incorporate true or ‘deterministic’ grain struc- by simplifying block structures that are derived determin-
tures derived from scanning electron microscope or digital istically from real specimens. The ability to systematically
image processing techniques in grain-based models (Tan simplify the same deterministic block structure allowed for
et al. 2016; Li et al. 2017b, 2019; Park et al. 2017), but the modifying grain shape, grain size distribution, and mineral
vast majority have employed stochastically generated block arrangement. The outcomes from this study will assist future
structures (e.g., Kazerani and Zhao 2010; Ghazvinian et al. researchers in selecting a grain structure representation that
2014; Nicksiar and Martin 2014; Fabjan et al. 2015; Sinha is suitable to a particular problem.
et al. 2020). As pointed out by Wang and Cai (2018), BBMs There are three reasons for selecting the BTS test as the
with block structures that are statistically similar to the focus of this study: (1) specimen preparation and testing
actual grain size distribution of the target rock unit are often procedure is much simpler than the direct tensile strength
computationally intensive and impractical for calibration. test, although the stress state in a BTS test is similar to a
The effects of grain shape (Azocar 2016; Mayer and Stead confined direct tensile test (Diederichs 1999). In general,
2017; Zhu et al. 2017; Xu et al. 2020; Contreras Inga et al. BTS yields a higher tensile strength in comparison with
2021; Li et al. 2021a), mineral arrangement (Lan et al. 2010; a direct tension test, and an empirical correction can be
Fabjan et al. 2015; Gui et al. 2016; Liu et al. 2018; Zhou applied to estimate the direct tensile strength (Perras and
et al. 2019; Contreras Inga et al. 2021), micro-properties Diederichs 2014). Since the BTS test was conducted in the
(i.e., properties of grains and grain–grain contacts within the laboratory and thereafter used directly as a calibration target
model) (Sinha and Walton 2020; Contreras Inga et al. 2021), in the BBM, the aforementioned difference in stress state and
mineral composition (Nicksiar and Martin 2014; Li et al. strength was not a point of concern in this study. (2) Given
2021a), and fabric orientation (Ghazvinian et al. 2014) have the high surface area to volume ratio of the specimens, it
also been investigated. Contreras Inga et al. (2021) inves- was possible to map the grain structure in a non-destructive
tigated the influence of various aspects of grain structures manner by simply observing the ends of the specimen, (3)
generated using a Voronoi tessellation approach on rock The BTS test has been successfully simulated by numerical
unconfined compressive strength prediction using BBMs. modeling over the years and is a common calibration target
It was observed that the stochastic effects introduced by the in similar studies (Chen et al. 2004; Kazerani et al. 2012; Li
mineral arrangement represented in Voronoi models and the and Wong 2013; Mahabadi et al. 2014; Dan and Konietzky
randomness of Voronoi grain structure generation have a 2014; Sinha et al. 2020). It is acknowledged here that con-
significant influence on rock strength predictions, whereas tinuum approaches (FEM/FDM) can also be used to model
average grain size and grain shape have a limited effect. On BTS tests, as has been demonstrated by Chen et al. (2004),

13
5296 C. E. C. Inga et al.

Yu et al. (2006), and Villeneuve et al. (2012), among others. the Colorado School of Mines using a TESCAN Integrated
However, continuum models are unable to explicitly simu- Mineral Analyzer (TIMA). The TIMA is based on a scan-
late the separation of grain boundaries undergoing failure ning electron microscope TESCAN–VEGA-3 model LMU
and do not capture the effect of elastic heterogeneity in its VP-SEM and controlled by the TIMA3 software. The system
entirety (Zhang and Wong 2018). Our choice of modeling relies on four energy dispersive X-ray (EDX) spectrometers
approach, i.e., BBM, is based on the aforementioned draw- for spectral data acquisition. Such spectrometers are set at
backs of continuum models and is part of an ongoing effort a beam stepping interval of 15 μm, a beam intensity of 14,
by the authors to understand the capabilities and limitations and an acceleration voltage of 24 keV. The system employs
of the BBM approach (Sinha 2020; West et al. 2020; Con- Monte Carlo simulations to model the interactions between
treras Inga et al. 2021). The conclusions drawn might be the beam and the specimen. Acquired spectral data are com-
equally applicable to other discontinuum and hybrid mod- pared to spectra kept in a look-up table to determine the min-
eling techniques, but further studies are required to confirm eral composition at each acquisition point. Results reported
this by reproducing the results shown here (or similar). by the TIMA3 software are presented as a spreadsheet with
This study utilizes different types of representations of the areal percentage mineral composition in the look-up
the grain structure of the Wausau granite to examine the table. The thin section petrography identified k-feldspar,
effect of the grain structure representation on the capabili- plagioclase, quartz, and biotite as more abundant minerals
ties of BBMs for the simulation of the mechanical behavior in the specimens of Wausau granite. Figure 1 shows photos
of brittle rock. Four types of models (deterministic, semi-
deterministic, Voronoi with heterogeneous grain size, and
Voronoi with uniform grain size) were generated and evalu-
ated in the context of simulated and actual Brazilian tensile
strength tests. The input micro-parameters were calibrated
to match the laboratory test results using the determinis-
tic grain structure model. Each type of model, therefore,
represents the grain structure of the Wausau granite with a
different degree of realism. Overall, this study directly evalu-
ates the ability of randomly generated Voronoi models to
approximate the geometric heterogeneity within the grain
structure of intact rock.

2 Modeling the Wausau Granite Using

Bonded Block Models

This study compares the results of different BBM models

of Wausau granite in Brazilian strength test simulations to
actual tests conducted in the laboratory. Since the study
focuses on the influence of grain structure representation
on BBM response, a single set of micro-properties was used
for all simulations.

2.1 The Wausau Granite

Sims et al. (1993) describe the Wausau granite as a dark

red medium- to coarse-grained alkali–feldspar granite
mostly exposed in the central region of Wisconsin. The
specimens used in this study were obtained from a quarry
in the Marathon County, Wisconsin. Standard thin sec-
tions of the Wausau granite were prepared for petrographic
analysis and Scanning Electron Microscopy (SEM)-based
automated mineralogy. The automated mineralogy analyses
were conducted at the automated mineralogy laboratory of Fig. 1  a Hand specimen, and b Thin-section photomicrograph of the
the Department of Geology and Geological Engineering at Wausau granite under cross-polarized light

13
Modeling Brazilian Tensile Strength Tests on a Brittle Rock Using Deterministic,… 5297

of the Wausau granite in a hand specimen and a thin section. terms of the equivalent diameter of the grains (i.e., the area
LaBerge and Myers (1983) and Sims et al. (1993) reported of each grain was converted to diameter based on a circular
the presence of irregular intergrowth or exsolution of sodic approximation using the equation: Area = 𝜋∕4diameter2 ).
and potassic feldspar in Wausau granite, and these textures Similar to the mineral composition, the grain size distribu-
were also identified in this study. tion varies from specimen to specimen. The grain sizes by
Three disk specimens employed for the Brazilian Tensile mineral were observed to be approximately log-normally
Strength (BTS) tests were characterized in terms of mineral distributed. The average overall apparent diameter varies
composition and grain diameter. The mineralogical com- from 1.0 mm to 1.3 mm among the samples. Table 1 sum-
position of each face of the disk specimens was examined marizes the mineral composition per specimen, whereas
through macroscopic petrographic characterization, and it Table 2 provides grain size mean (𝜇) and standard deviation
was confirmed that the mineral composition varies on each ( 𝜎) parameters per specimen. The log-normal probability
one of the specimen faces, but all of them closely approxi- distribution is mathematically represented by (Papoulis and
mate the average composition obtained from automated min- Pillai 2002)
eralogy analyses. However, feldspar exsolutions within the
(lnx−𝜇)2
� �
Wausau granite samples made it difficult to determine the e
−
2𝜎 2

actual k-feldspar/plagioclase proportion accurately through f (x) = √ (1)

x𝜎 2𝜋
macroscopic examination. Thus, for the purposes of this
study, after defining visually the total feldspar content on Alternatively, grain size can be described using the clas-
each sample, the average k-feldspar/plagioclase proportion sical particle-size distribution curve, and the corresponding
identified using automated mineralogy was used to estimate ­D10, ­D50, and ­D90 values (equivalent diameters correspond-
the sodic and potassic feldspar content of each disk speci- ing to 10%, 50%, and 90% finer, respectively) for each min-
men. The grain size was recorded in all the specimens in eral type and for the whole face can be found in Table 3.
The log-normal distribution parameters are more relevant to
this study as they are direct input to the Voronoi generation
software Neper (discussed in Sect. 2.2).
Table 1  Mineral composition of Wausau granite
The Wausau granite was mechanically characterized
Specimen Modal abundance (%) using Uniaxial Compressive Strength (UCS; Contreras Inga
Biotite Quartz Plagioclase K-feldspar et al. 2021) and Brazilian Tensile Strength (BTS) tests. The
peak strength, crack damage stress (CD), crack initiation
Average (auto- 3 32 41 24 stress (CI), Young’s modulus (E), and Poisson’s ratio (ν)
mated mineral-
ogy) were identified from the UCS tests (Contreras Inga et al.
BTS-1A 3 34 40 23 2021). The BTS was obtained from three tests (new to this
BTS-1B 2 39 37 22 study) conducted following ASTM International (2001)
BTS-2A 3 37 38 22 procedures in the Earth Mechanics Institute laboratory at
BTS-2B 3 30 42 25 Colorado School of Mines, which reached peak strengths
BTS-3A 2 32 42 24 of 11.5 MPa, 11.6 MPa, and 12.4 MPa. The disk-shaped
BTS-3B 3 34 39 24 specimens used in these tests were 51.4 mm in diameter,
with a diameter-to-width ratio of 2:1. More details on these

Table 2  Grain size log-normal Specimen Grain size (mm)

distribution parameters of
Wausau granite Biotite Quartz Plagioclase K-feldspar Overall
μ σ μ σ μ σ μ σ μ σ

BTS-1A 0.7 0.4 2.3 1.3 1.1 0.6 1.2 0.6 1.2 0.7
BTS-1B 0.6 0.3 2.4 1.3 1.2 0.7 1.2 0.7 1.3 0.9
BTS-2A 0.6 0.3 2.2 1.2 1.1 0.6 1.0 0.5 1.1 0.7
BTS-2B 0.7 0.3 2.1 1.3 1.2 0.6 1.2 0.7 1.3 0.8
BTS-3A 0.5 0.3 2.0 1.1 1.0 0.5 1.0 0.5 1.0 0.6
BTS-3B 0.7 0.3 2.2 1.3 1.2 0.7 1.2 0.7 1.2 0.8

μ and σ are mean and standard deviation, respectively

13
5298 C. E. C. Inga et al.

tests can be found in Appendix A. Table 4 summarizes the uniform grain size. Table 5 shows a comparison of the four
mechanical properties obtained from the laboratory tests. types of models based on an assessment of the features of
The low Coefficient of Variation (CoV) values relative to the true grain structure that they represent. All the BBMs
other rocks and narrow confidence interval for the mean BTS were developed within 2D circular domains with the same
(11.8 ± 0.56) based on the test results signify limited natural diameter as the BTS specimens (i.e., 51.4 mm).
variability in Wausau granite, meaning that three tests rea- The first group of BBMs consists of deterministic rep-
sonably characterize the indirect tensile strength of this rock resentations of each one of the two faces of the three disk-
(Langford and Diederichs 2015). shaped specimens, for a total of six deterministic models.
Deterministic models are the closest approximation of the
2.2 BBM Generation geometry of a grain structure. These models depict the size
and shape of the grains and the actual mineral arrange-
The three disk-shaped specimens used in the BTS tests were ment found within the laboratory specimens. Implement-
used as the basis to generate different representations of the ing these representations requires a manual characterization
grain structure. Accordingly, four different groups of two- of the grain morphology and mineral arrangement within
dimensional BBMs were generated for each one of the two a rock specimen, which is highly time-consuming. To cre-
faces of the disk specimens: deterministic, semi-determinis- ate the models, high-quality photographs of the faces of the
tic, Voronoi with heterogenous grain size, and Voronoi with specimens were digitized by manual creation of a series of

Table 3  Grain size distribution Specimen Biotite Quartz Plagioclase K-feldspar Overall
parameters of Wausau granite
in mm D10 D50 D90 D10 D50 D90 D10 D50 D90 D10 D50 D90 D10 D50 D90

BTS-1A 0.41 0.77 1.35 1.32 2.75 4.09 0.7 1.29 2.06 0.71 1.31 2.05 0.69 1.39 2.88
BTS-1B 0.41 0.67 1.33 1.43 2.74 4.35 0.73 1.35 2.13 0.7 1.49 2.3 0.74 1.56 2.96
BTS-2A 0.37 0.68 1.41 1.25 2.63 4.11 0.67 1.27 1.96 0.65 1.14 1.94 0.66 1.33 2.75
BTS-2B 0.39 0.74 1.41 1.13 2.73 4.07 0.74 1.42 2.19 0.78 1.37 2.2 0.74 1.47 2.75
BTS-3A 0.31 0.61 1.04 1.14 2.15 3.44 0.61 1.09 1.71 0.61 1.13 1.73 0.6 1.18 2.17
BTS-3B 0.4 0.72 1.28 1.29 2.55 3.42 0.69 1.35 2.26 0.66 1.42 2.19 0.67 1.48 2.73

D10, D50, and D90 are equivalent diameters corresponding to 10%, 50%, and 90% finer in a classical parti-
cle-size distribution curve

Table 4  Experimentally Property Number of μ σ CoV (σ/

determined macro-mechanical tests μ) in %
properties of the Wausau granite
Density, ρm (kg/m3) 11 2605 8 0.3
Uniaxial compressive strength, UCS (MPa) 11 226 21 9.3
Crack damage stress, CD (MPa) 11 220 19 8.6
Crack initiation stress, CI (MPa) 4 107 9 8.4
Young’s modulus, E ­ m (GPa) 4 70 2 2.9
Poisson’s ratio, νm 4 0.24 0.02 8.3
Brazilian tensile strength, BTS (MPa) 3 11.8 0.52 4.4

CoV is coefficient of variation

Table 5  Comparison of the Type of BBM Mineral Average Grain size Mineral Grain shape
representation complexity of the composi- grain size heterogene- arrange-
four types of BBMs tion ity ment

Deterministic ✔ ✔ ✔ ✔ ✔
Semi-deterministic ✔ ✔ ✔ ✔ –
Voronoi with heterogeneous grain size ✔ ✔ ✔ – –
Voronoi with uniform grain size ✔ ✔ – – –

*✔ = properly represents this feature; – = not able to represent this feature

13
Modeling Brazilian Tensile Strength Tests on a Brittle Rock Using Deterministic,… 5299

convex polyhedral cells, which intersect along planar faces

and straight edges, and at vertices. The open-source soft-
ware Neper (Quey 2019; Quey et al. 2011) was employed to
generate the Voronoi grain assemblies. Neper uses Voronoi
or Laguerre tessellations to generate polycrystal assemblies
in 3D or 2D from a set of seeds positioned in space (Quey
2019). The main difference between Voronoi and Laguerre
tessellations is that the latter is a generalization of Voronoi
tessellations that allow for geometries that are not possible
to achieve with Voronoi cells. This is possible using dif-
ferent weighted seeds that make boundaries between cells
Fig. 2  Digitized surface of a BTS specimen prepared in CAD soft- non-equidistant between seeds (Quey and Renversade 2018).
ware; the white line segments shown were manually added to digitize Six semi-deterministic models were generated for this
the grain boundary shapes study. Each semi-deterministic model is considered a sim-
plification of a corresponding deterministic model, since not
straight-line segments in computer-aided design (CAD) soft- all features of the actual specimen grain structure are rep-
ware (Autodesk 2019). A length scale was assigned by set- resented in the semi-deterministic version. The semi-deter-
ting the diameter of the photograph to the measured diameter ministic grain structures were generated in Neper using the
of the specimens (i.e., 51.4 mm). Figure 2 shows an example “centroiddiameq” option (Quey 2019). This option employs
of this digitization process. The grain outlines developed in the centroid coordinates and equivalent diameter of each of
the CAD software were later imported into UDEC to build the grains within the assembly, but it does not allow one
the models. Concave and convex grains are represented to define the shape of the grains. The mineral arrangement
in these models. Although these models are referred to as and mineral composition of the actual granite specimens are
deterministic, some simplifications were adopted in their properly represented in this type of model, given that such a
development. Specifically, these models only considered grain structure representation uses the centroids of the grains
grains with an apparent diameter of 0.1 mm or greater and and mineral arrangement in the deterministic models. In
a minimum edge length of 0.05 mm. These simplifications addition, using the grain diameters of the real specimens, the
were established to ensure consistency in the digitization semi-deterministic models are able to accurately depict the
process of the specimen faces and obtain models with com- grain size distribution of the specimen. However, this type of
parable effective resolution. Once the grain structure was model cannot represent the actual shape of the grains, since
built in UDEC, the mineral type was assigned to the grains the generated cells can only be convex polygons.
according to the spatial distribution observed in the actual The third group of BBMs consists of 60 Voronoi models
specimens. Thus, each one of the six deterministic mod- that properly represent the grain size heterogeneity within
els has a unique mineral content proportion and mineral the Wausau specimens; 10 Voronoi models were stochasti-
arrangement (see Table 2). The feldspar exsolutions pre- cally generated per specimen face to properly account for the
sent within the Wausau granite were modeled as grains of potential effects of randomness in the grain structure genera-
either pure k-feldspar or plagioclase. Accordingly, the dis- tion process on the simulated mechanical behavior. The mor-
tribution of k-feldspar and plagioclase grains was assigned phological properties of the grains were defined in Neper
randomly in each model. Considering that both types of using statistical distributions (i.e., mean, standard deviation,
feldspar present similar mechanical properties, variations and distribution type) of grain size and shape. In Neper, the
in the k-feldspar/plagioclase proportion within the exsolu- grain size is defined as the diameter (d) of a sphere of equiv-
tions are presumed to have a negligible effect on the rock alent volume (or circle of equivalent area in 2D). The grain
macro-mechanical behavior. shape is defined using the sphericity (s) of the grain, which
The three other types of models (i.e., semi-deterministic, is the ratio between the area of a sphere of equivalent vol-
Voronoi with heterogeneous grain size, and Voronoi with ume and the area of the grain (Quey and Renversade 2018).
uniform grain size) were generated as assemblies of 2D Sphericity is replaced by circularity in 2D, which is the ratio
Voronoi tessellations, where each tessellation or cell repre- of the perimeter of a circle of equivalent area and the perim-
sents a mineral grain. Mathematically, a Voronoi tessellation eter of the grain (Wadell 1933). Each model in a sub-group
is the partition of an n-D space in an assembly of n-D poly- of 10 represents the same mineral content and grain size
hedral entities defined as zones of influence of a particular heterogeneity of the corresponding deterministic model, but
set of seeds, corresponding to their centers. The polyhedral a different set of randomly generated grains (specifically,
entities fill the space within a domain without overlaps nor 2 block structures were created per face with randomized
gaps (Quey et al. 2011). The resulting Voronoi grains are mineral assignments). The grain size log-normal distribution

13
5300 C. E. C. Inga et al.

parameters identified from the actual specimen faces (see to the blocks within the grain structure using the mineral
Table 2) were used to generate the grain structures in Neper. content proportions estimated in the granite specimens as
A 2D sphericity (i.e., circularity) equal to 0.88 was used in the only constraint. Table 6 summarizes the geometric char-
Neper to develop the models, as this value was qualitatively acteristics employed for the generation of the two different
assessed to produce grain shapes that were approximately types of Voronoi models. Figure 3 shows the actual grain
representative of the actual Wausau granite grains. The min- structure for one Wausau granite specimen face and its rep-
eral types were randomly assigned to the grains within the resentations according to the approaches described above.
models. The mineral content proportions (see Table 1) and
grain size distributions per mineral type (see Table 2) were 2.3 Constitutive Behavior of Intact Rock
used as constraints. Specifically, 100% of the biotite grains and Micro‑Property Assignment
have diameters between 0.0 and 2.0 mm; 16%, 68%, and 16%
of the quartz grains correspond to ranges of 0.0–1.5 mm, The software UDEC (Itasca Consulting Group Inc. 2014)
1.5–3.0 mm, and 3.0–6.0 mm, respectively; and 100% of was used to run the simulations presented in this study.
the k-feldspar and plagioclase feldspar grains have diameters UDEC allows for the simulation of grains within a BBM
between 0.0 and 6.0 mm. as rigid, elastic, or plastic blocks (Ghazvinian et al. 2014;
The last group of BBMs also consists of Voronoi models Wang and Cai 2018; Zhang and Wong 2018). A constitu-
generated in Neper. As opposed to the previous group, the tive relationship can be applied to the gains when they are
models within this group represent the grain size heteroge- modeled as elastic or plastic. The mechanical interactions
neity in the conventional way employed by previous stud- along the common contact of two grains can be simulated
ies (Chen et al. 2004; Chen and Konietzky 2014; Fabjan through a joint constitutive model (Itasca Consulting
et al. 2015; Gui et al. 2016), with a grain size variability Group Inc. 2014). In this study, the mineral grains were
significantly lower than the observed variability in the real modeled as unbreakable elastic blocks with distinct density
specimens. As in the previous case, six sub-groups of 10 (ρ), Young’s modulus (E), and Poisson’s ratio (ν) for each
models were built. The models within each sub-group rep- mineral type. The use of elastic grains only allows fail-
resent the mineral content and average grain size measured ure to occur along the contacts between grains (Ghazvin-
in the Wausau granite specimens and depicted in the cor- ian et al. 2014; Nicksiar and Martin 2014; Wang and Cai
responding deterministic models. In Neper, the grain size 2018). A similar simplification was applied in published
distribution was denoted using the average/overall grain size studies that use BBMs (Kazerani and Zhao 2010; Lan et al.
measured in the granite specimens (see Table 2) and a uni- 2010; Chen and Konietzky 2014; Ghazvinian et al. 2014;
form standard deviation of 0.25 mm for all models. A grain Nicksiar and Martin 2014; Farahmand and Diederichs
2D-sphericity (i.e., circularity) value of 0.88 was applied 2015; Chen et al. 2016). This simplification is appropriate
in all the models. The mineral type was randomly assigned for simulated behavior under unconfined or tensile loading

Table 6  Geometric Type of Voronoi model Bonded Grain diameter Grain spheric- Number of grains
characteristics of the 3D block model (mm) ity, s
Bonded Block Models
μ σ μ σ

Voronoi with grain heterogeneity 1-A 1.2 0.7 0.88 0.03 1361/1364
1-B 1.3 0.8 0.88 0.03 1169/1159
2-A 1.1 0.7 0.88 0.03 1475/1470
2-B 1.3 0.7 0.88 0.03 1285/1289
3-A 1.0 0.6 0.88 0.03 1952/1963
3-B 1.2 0.7 0.88 0.03 1368/1379
Voronoi with uniform grain size 1-A 1.2 0.25 0.88 0.03 1366/1347
1-B 1.3 0.25 0.88 0.03 1159/1165
2-A 1.1 0.25 0.88 0.03 1479/1468
2-B 1.3 0.25 0.88 0.03 1267/1276
3-A 1.0 0.25 0.88 0.03 1939/1936
3-B 1.2 0.25 0.88 0.03 1364/1358

The number of grains corresponds to the two grain structures built per face. For each of these grain struc-
tures, 5 realizations were generated by randomizing the mineral assignment (a total of 5 × 2 = 10 realiza-
tions per face)

13
Modeling Brazilian Tensile Strength Tests on a Brittle Rock Using Deterministic,… 5301

Fig. 3  a Disk-shaped specimen BTS-2, face B, of Wausau granite, and four corresponding representations: b deterministic, c semi-deterministic,
d Voronoi with heterogeneous grain size, and e Voronoi with uniform grain size

conditions (Sinha and Walton 2020), as in this study. The 2014; Ghazvinian et al. 2014; Nicksiar and Martin 2014;
grains were discretized into a mesh of deformable trian- Farahmand and Diederichs 2015; Chen et al. 2016). To
gular finite-difference zones. Given the sensitivity of the calibrate the model, the micro-properties of Farahmand
simulation results to mesh size (Cai and Zhao 2000; Chen and Diederichs (2015) were used as a starting point, fol-
et al. 2000), a maximum triangular zone edge length of lowed by systematically modifying the parameters until
0.8 mm was applied in all models to minimize the effect the simulated BTS for the deterministic BBM matched
of mesh size. The resulting average grain edge length to those measured in the laboratory. Tables 7 and 8 show the
zone edge length ratios are greater than or equal to 1.75, calibrated micro-properties used in this study.
which agrees with the recommended ratio to achieve stable
numerical results (Fabjan et al. 2015). A Coulomb slip- Table 7  Grain micro-properties used by Farahmand and Diederichs
joint constitutive model with residual strength parameters (2015) and applied in the present study
was applied to the grain–grain contacts considering differ-
Mineral type Young’s Poisson’s ratio ν Density ρ (g/cc)
ent attributes for each type of contact. Each contact was modulus E
assigned a normal stiffness (­ kn), shear stiffness (­ ks), peak (GPa)
friction angle (φ), peak cohesion (C), peak tensile strength
K-feldspar 96.8 0.28 2.56
(σt), and residual friction angle (φr). The residual cohesion
Plagioclase 88.1 0.26 2.63
­(Cr), and residual tensile strength (σtr) were assumed to be
Quartz 94.5 0.08 2.65
zero based on previous studies (Kazerani and Zhao 2010;
Biotite 33.8 0.36 3.05
Lan et al. 2010; Chen and Konietzky 2014; Gao and Stead

13
5302 C. E. C. Inga et al.

Table 8  Contact micro-properties calibrated in the present study of dynamic oscillations that could arise, while failure occurs
Contact type kn (GPa/m) ks/kn C (MPa) φ, φr (°) σt (MPa) within the model (Kazerani et al. 2012; Gao et al. 2016).
Axial displacements were obtained by averaging the dis-
KF/KF 2.3E + 5 0.65 110.0 62.0, 5.0 22.8 placements between three pairs of grid points located on the
KF/PL 2.1E + 5 0.65 108.0 61.0, 5.0 20.8 surface of the loading platens using a FISH script (Itasca
KF/QZ 2.7E + 5 0.65 76.0 53.0, 5.0 18.3 Consulting Group Inc. 2014). The tensile stress was calcu-
KF/BT 2.3E + 5 0.65 60.0 48.0, 5.0 7.4 lated from the total reaction forces measured on the surfaces
PL/PL 2.5E + 5 0.65 112.0 63.0, 5.0 24.0 of the platens using the standard BTS (σt) equation (Interna-
PL/QZ 2.3E + 5 0.65 80.0 49.0, 5.0 18.3 tional Society for Rock Mechanics 1978):
PL/BT 2.3E + 5 0.65 54.0 45.0, 5.0 14.6
QZ/QZ 2.8E + 5 0.65 130.0 65.0, 5.0 22.8 2P
𝜎t = (2)
QZ/BT 2.3E + 5 0.65 57.0 52.0, 5.0 15.2 𝜋Dt
BT/BT 1.3E + 5 0.65 88.0 55.0, 5.0 16.4 where P is the load at failure, D is the diameter of the speci-
men, and t is the thickness of the specimen. In addition, the
development of tensile and shear fractures was tracked using
2.4 Numerical Test Setup a FISH script.

For the BTS test simulations, compressive loading was

applied to the disk-shaped specimens through two rigid 3 BTS Simulation Results and Discussion
platens (Fig. 4). A constant vertical velocity was applied on
both platens (top and bottom) to produce an effective loading 3.1 Simulation Using Deterministic
velocity, v (i.e., − v/2 and v/2 applied to the top and bottom and Semi‑deterministic Grain Structure
platens, respectively). The loading velocity and the damping Representations
mode strongly influence the simulation results (Sinha et al.
2022). Thus, to reach realistic simulation results, the loading Figure 5 and Table 9 present the results of the determin-
velocity must be sufficiently slow enough given the damp- istic and semi-deterministic BTS test simulations for the
ing conditions that are enforced to ensure that the model six 2D grain structures modeled in this study (i.e., grain
remains in a quasi-static equilibrium condition (Kazerani structures developed for each of the two faces of three
and Zhao 2010; Mayer and Stead 2017). Through a sen- disk specimens). The results indicate that compared to a
sitivity analysis, a constant rate of 0.05 m/s was identified target BTS value (i.e., BTS obtained in a laboratory test),
as a loading velocity below which changes in velocity had both types of models approximate the actual test values
a negligible (i.e., < 4%) influence on the modeled strength. with varying degrees of accuracy. The strengths from
Since UDEC automatically calculates a timestep of approxi-
mately ­10–8 s, the 0.05 m/s loading rate can be interpreted
as approximately 5*10–6 mm/step. The damping mode for
the simulations was set to “local” with a default damping
coefficient of 0.8 (Itasca Consulting Group Inc., 2014). This
form of velocity-proportional damping minimizes the effect

Fig. 5  Comparison of laboratory results (red dots) against corre-

Fig. 4  Loading conditions and displacement tracking points for BTS sponding deterministic (blue dots) and semi-deterministic (green
test simulations dots) BTS per grain structure (i.e., grain structures described on each
of the two faces of three disk specimens) (color figure online)

13
Modeling Brazilian Tensile Strength Tests on a Brittle Rock Using Deterministic,… 5303

Table 9  Summary of simulated BTS with deterministic and semi- semi-deterministic models provided strengths outside the
deterministic models observed BTS variability range.
Grain structure Laboratory Deterministic Semi-deterministic Figure 6 shows the stress–displacement curves and cor-
responding cumulative number of fractures (tensile and
(MPa) (MPa) Dif. (%) (MPa) Dif. a (%)
a
shear) obtained from the simulations with deterministic and
BTS-1A 12.4 11.7 6 11.1 11 semi-deterministic grain structures. The stress–displacement
BTS-1B 12.4 0 12.9 −4 curves obtained from the BTS simulations using determinis-
BTS-2A 11.5 11.2 3 11.5 0 tic and semi-deterministic models (see Fig. 6) show similar
BTS-2B 13.5 − 18 12.4 −8 pre-peak behavior in all cases. This similarity is more notice-
BTS-3A 11.6 10 13 10.6 9 able when comparing the curves obtained with a determinis-
BTS-3B 12.5 −8 10.4 10 tic model and its corresponding semi-deterministic model;
Overall 11.8 11.8 11.5 such pairs of curves closely match along the pre-peak sec-
a tion. Given that deterministic and semi-deterministic models
Dif. = difference expressed as a percentage of the laboratory target
value have in common a detailed representation of the mineral
arrangement and mineral size distribution, we can infer that
both features have a strong effect on the pre-peak behavior
deterministic models vary from 18% below to 13% above of the rock. The simulations properly capture the exponential
the target, whereas the strengths from semi-deterministic increase of the number of cracks, with a notably higher rate
models fluctuate from 8% below to 11% above the tar- for the tensile cracks than the shear cracks as the simulation
get. In addition, the results indicate that both determin- reaches the peak tensile strength.
istic and semi-deterministic models were able to achieve
approximate predictions of Wausau granite’s BTS (i.e., 3.2 Simulation Using Voronoi Grain Structure
within the range of variability estimated from laboratory Representations
tests), with the overall strength of the deterministic models
matching the laboratory measurements. Although deter- Voronoi models are not intended to be exact depictions of
ministic models closely replicate the grain structure geom- the mineral arrangement and grain geometry found in actual
etry, 2 out of 6 strengths fall outside the BTS variability specimens of a rock. However, Voronoi models are expected
range registered in laboratory tests. Similarly, 2 out of 6 to provide proper approximations of the grain structure

Fig. 6  Stress—displacement
curves and cumulative number
of tensile and shear cracks
simulated using deterministic
and semi-deterministic models
of a specimen BTS-1, b speci-
men BTS-2, and c specimen
BTS-3

13
5304 C. E. C. Inga et al.

heterogeneity. Given the stochastic nature of the Voronoi to reproduce the typical failure pattern (i.e., dominated by
tessellation approach, multiple BBMs with equivalent grain tension cracks) obtained in laboratory BTS tests.
structure heterogeneity can be generated. This portion of the In the three cases corresponding to the three different
study employs two types of Voronoi BBMs with different laboratory specimens (see Fig. 7a–c), the stochastically
degrees of realism: (1) Voronoi models with heterogeneous generated grain structures represented in Voronoi HGS and
grain size (hereafter referred to as Voronoi HGS), which Voronoi UGS models yield a variety of BTS values. This is
represents the grain size variability for each mineral type; expected, as each of the Voronoi grain structures is unique,
and (2) Voronoi models with uniform grain size (hereaf- and the grain structure is known to have a significant effect
ter referred to as Voronoi UGS), which is the conventional on the macroscopic response of BBMs (Ghazvinian et al.
Voronoi grain structure representation that only represents 2014; Fabjan et al. 2015; Azocar 2016; Insana et al. 2016).
the rock’s overall average grain size with limited variability. Voronoi HGS models provide average BTS 4–5% lower
For the three specimens of Wausau granite used in this study, than the average experimental BTS (see Table 10) and an
the respective experimental BTS values were compared overall variability greater than the experimental variability
against the deterministic and semi-deterministic results (i.e., (see Table 11). The average BTS with Voronoi HGS models
mean predictions, obtained from two simulations in each falls within the Wausau granite’s BTS variability range, as
case), as well as the Voronoi HGS and Voronoi UGS results observed in laboratory tests. Similarly, Voronoi UGS mod-
(i.e., mean prediction and corresponding variability, based els result in average BTS 2–4% higher than the actual BTS
on 40 simulations in each case, twenty per face). Figure 7 (see Table 10) and an overall variability greater than the
shows such a comparison of BTS predictions made with experimental variability (see Table 11). The average BTS
the different models against laboratory test results. Figure 8 with Voronoi UGS models falls within the experimentally
shows the final cracking state obtained with the four types of observed variability range. In addition, the average strength
models presented in this study. All types of models are able based on Voronoi UGS models is ~ 1 MPa higher than the

Fig. 7  Comparison of labora-

tory results (LAB) against
corresponding deterministic
(DET), semi-deterministic
(SEM), Voronoi HGS (VO-1),
and Voronoi UGS (VO-2) BTS.
a comparison for specimen 1,
b comparison for specimen 2, c
comparison for specimen 3, and
d overall comparison

13
Modeling Brazilian Tensile Strength Tests on a Brittle Rock Using Deterministic,… 5305

Fig. 8  a Fracture pattern observed in specimen BTS-2, face B, compared to the patterns obtained with the four corresponding types of models: b
deterministic, c semi-deterministic, d Voronoi with heterogeneous grain size, and e Voronoi with uniform grain size

Table 10  Summary of the average BTS generated using each type of model
Grain structure Laboratory Deterministic Semi-deterministic Voronoi HGS Voronoi UGS
μ (MPa) μ (MPa) Dif. a (%) μ (MPa) Dif. a (%) μ (MPa) Dif. a (%) μ (MPa) Dif. a (%)

Specimen 1 11.8 n/a n/a n/a n/a 11.4 −4 12.1 2.1

Specimen 2 n/a n/a n/a n/a 11.3 − 4.7 12.3 3.8
Specimen 3 n/a n/a n/a n/a 11.3 −5 12.3 3.6
Overall simulations 11.9 0.4 11.5 −3 11.3 − 4.6 12.2 3.2
a
Dif. = difference expressed as a percentage of the laboratory mean value

Table 11  Summary of the overall standard deviations generated by each type of model. CI is confidence interval (Sheskin 2003)
Grain structure Laboratory Deterministic Semi-deterministic Voronoi HGS Voronoi UGS
σ (MPa) 95% CI σ (MPa) 95% CI σ (MPa) 95% CI σ (MPa) 95% CI σ (MPa) 95% CI

Overall simulations 0.5 [0.3, 3.3] 1.2 [0.8, 3.0] 1.0 [0.6, 2.5] 1.0 [0.8, 1.2] 0.8 [0.7, 1.0]

13
5306 C. E. C. Inga et al.

average strength from the Voronoi HGS models. This behav- not statistically significant (refer to the 95% confidence inter-
ior is consistent with the findings of Peng et al. (2017a, b) vals in Table 11). As mentioned previously, the similari-
and Xu et al. (2020), who examined the effect of grain size ties between the distributions suggest that the four types of
heterogeneity and identified that as grain size heterogeneity models considered all reasonably represent the heterogeneity
increases, the rock strength decreases. of the grain structure of Wausau granite for the purposes of
Figure 7a–c also shows that the average strengths result- tensile strength simulation.
ing from deterministic and semi-deterministic models fall Finally, to ensure that the effect of grain structure/geo-
within the variability range obtained from Voronoi HGS metric heterogeneity is not substantially dependent on the
and Voronoi UGS models. This indicates that Voronoi rep- choice of micro-properties or material heterogeneity, two
resentations are able to capture the grain structure hetero- tasks were undertaken: (1) a sensitivity analysis on the con-
geneity depicted in detailed models (i.e., deterministic and tact tensile strength, and (2) running models with homog-
semi-deterministic models). In addition, the similarities of enized (i.e., single parameter set) grain (Table 7) and contact
the distributions of BTS for specimens 1, 2, and 3 (using properties (Table 8). The approach and associated results can
Voronoi HGS and Voronoi UGS models) indicate that vari- be found in Appendix B. Although some interesting trends
ations in average mineral content (see Table 2) and average were observed, the results are generally consistent with those
grain size (see Table 4) of the rock have a limited effect on presented here.
the model behaviors.
Figure 7d summarizes the overall distributions for the 3.3 Discussion on Model Results
four types of models developed in this study (i.e., all models
for all specimens and faces). Such distributions present simi- Prior studies (e.g., Nicksiar and Martin 2014; Fabjan et al.
lar average strengths (11.3–12.2 MPa) and limited variability 2015; Mayer and Stead 2017; Xu et al. 2020; Contreras Inga
in strength standard deviations (0.8–1.2 MPa). Given that the et al. 2021) evaluated the ability of Voronoi models to real-
same set of micro-properties was applied in the simulations istically represent grain structure heterogeneity by varying
with the four types of models, the similarities between the individual parameters, such as grain size and grain shape.
strength distributions suggest that even the simplest Voronoi However, the degree to which the simplifications intrinsic
representations appropriately represent the geometrical het- in the Voronoi approach affect the accuracy of mechanical
erogeneity within the Wausau granite’s grain structure that is behavior predictions has not been evaluated in the literature.
relevant to tensile strength as effectively as the more detailed In contrast to previous work, this study presented models of
representations (i.e., deterministic, and semi-deterministic the same specimens with different degrees of grain structure
representations). Nevertheless, as identified previously (see realism. These models allowed us to examine to what extent
Fig. 7a–c), the average BTS obtained with Voronoi HGS results obtained with Voronoi representations differ from
tend to be lower than the average obtained with Voronoi results obtained using more detailed and realistic depictions
UGS. As the Voronoi HGS models are closer to the detailed of grain structure.
models in terms of realism than the Voronoi UGS models Many of the model results presented in this study show
are, it was originally expected that Voronoi HGS models noticeable differences compared to the laboratory results.
would provide strength estimates that closely match those In general terms, once the constraints that control a given
obtained with detailed models. However, Fig. 7d shows that numerical simulation are properly defined, the ability to real-
the average BTS obtained with Voronoi UGS models is istically simulate the mechanical behavior of rocks depends
closer to the average BTS provided by the detailed models. on the representation of the grain structure and the micro-
This result could indicate that Voronoi UGS models provide properties applied in a BBM. With that said, we hypothesize
a more realistic approximation of the degree of interlock- several potential explanations for the observed differences
ing within the grain structure. The distributions provided between the laboratory and model results.
by deterministic and semi-deterministic models may not Using 2D models to represent the grain-structure can
provide conclusive evidence to confirm this presumption, have a significant effect on the validity of the results, as
since a small number of cases (six) were considered. Fig- the 2D representations are simplifications of 3D grain
ure 7d also shows how the experimental variability range structures. For Wausau granite, the 2D geometrical infor-
of the Wausau granite compares to the distributions for the mation obtained by digitizing the faces of the BTS speci-
four types of models. The distributions obtained with the mens cannot be directly converted to a unique 3D structure,
four types of models have higher and lower average strength because the grain size is smaller than the thickness of the
(− 5% to 3% higher) and higher standard deviation than the specimens. Therefore, even in the case of deterministic and
experimental BTS distribution. In addition, the UGS, HGS, semi-deterministic models, the 2D grain-structure represen-
and semi-deterministic models exhibit lower apparent vari- tations developed using individual faces are imperfect depic-
ability than the deterministic models, but the difference is tions of the complete 3D grain-structure of the laboratory

13
Modeling Brazilian Tensile Strength Tests on a Brittle Rock Using Deterministic,… 5307

specimens. Accordingly, we acknowledge that even if a (e.g., the actual properties of the same minerals varied
2D grain-structure representation is highly accurate (e.g., slightly between the two rocks) or numerical reasons. It is,
deterministic representation), the 3D mechanical interac- therefore, uncertain to what degree these properties can be
tions between grains are not completely captured in a 2D universally applied to the same mineral grains and grain-to-
representation. In addition, there are multiple possible 2D grain contact types in other rocks.
representations for the same 3D grain-structure (i.e., gener-
ated from diametrical sections, in the case of a disk-shaped
specimen). Each one of these possible 2D simplifications has 4 Conclusions
a unique mineral arrangement and grain geometric heteroge-
neity that could lead to different predictions of mechanical This study examined the capabilities of four different types
behavior. This is evident from a comparison of the model of grain structure representations with different degrees of
strengths for the two faces of the deterministic specimens realism for the simulation of brittle rock tensile strength.
(Table 9), where the degree of variation can be considered The grain structure representations were implemented in
a rough proxy for potential magnitude of the effect of 2D Bonded Block Models (BBMs) using a calibrated set of
simplification. In any case, the four types of 2D models used micro-properties. Detailed 2D representations of the grain
in this study produced similar average strengths with limited structure (i.e., deterministic and semi-deterministic models)
variability. This suggests that the heterogeneity of the grain provided realistic estimates of the BTS of Wausau granite.
structure geometry is adequately represented in each of the These models closely approximated the mineral content,
2D Voronoi models with unique grain structure configura- mineral arrangement, grain size heterogeneity, and grain
tions. For practical reasons, conventional Voronoi models shape heterogeneity (the shape only in the deterministic
are recommended over the more detailed alternatives (i.e., models) observed in tested specimens.
deterministic and semi-deterministic models). In contrast to Voronoi models were confirmed to provide strength esti-
more detailed models, multiple randomly generated Voronoi mates following distributions nearly equivalent to those
models can be easily and rapidly generated to evaluate the obtained using the more detailed models. This indicates
range in modeling results depending on stochastic aspects that Voronoi models are as effective as models using more
of the grain structure representation. sophisticated grain structure representations in simulation
In addition, other simplifications applied in the models of rock tensile fracturing behavior. In addition, this suggests
may have a greater effect than was originally expected. For that Voronoi models properly approximate the geometric
example, modeling the grains as unbreakable blocks may heterogeneity of the grain structure, even without including
limit the representation of fracture propagation within the many details of the grain structure geometry. We believe
grain structure, as the development of transgranular fractures this is a highly significant outcome, since it effectively vali-
is not represented in the simulations. Although intergranular dates the simplified modeling approaches that have been
fracturing is expected to be the dominant process, in reality, used by the scientific community and practitioners to date.
some non-zero number of transgranular fractures do occur Accordingly, a Voronoi grain structure representation that
before the peak tensile strength is reached. Modeling only properly captures the mineral content proportions and the
intergranular fracturing potentially leads to overestimations overall average grain size can be considered an appropriate
of the peak tensile strength due to the enhanced interlocking approximation of reality, at least in the context of tensile
among grains. damage processes.
Realistic predictions may not be obtained if the heteroge-
neity of grain and contact properties (i.e., micro-properties)
is not adequately represented in the model, even when the Appendix A: Brazilian Tensile Tests
geometric heterogeneity within the grain structure is prop-
erly depicted. The representation of property heterogene- Three BTS tests on Wausau Granite were conducted fol-
ity is achieved in this study by assigning different sets of lowing the ASTM International D3967-95a (2001) pro-
micro-properties to each mineral type. However, property cedures in the Earth Mechanics Institute laboratory at the
heterogeneity within a single mineral type is not considered Colorado School of Mines. The specimens were loaded at
in the models. This simplification of property heterogene- 110 N/sec until failure occurred via formation of diame-
ity could potentially lead to an unrealistic model response. tral/axial fractures. Figure 9 shows the three specimens
Besides the above, as some of the micro-properties used in before and after completion of the tests. The specimen
the simulations are based on calibration to Lac du Bonnet geometrical details are as follows: (1) BTS-1—diam-
granite properties (Farahmand and Diederichs 2015), they eter = 51.4 mm, thickness = 25.1 mm; (2) BTS-2—diam-
are not a perfect representation of the corresponding micro- eter = 51.4 mm, thickness = 25.2 mm; (2) BTS-3—diam-
parameters for Wausau granite, either for physical reasons eter = 51.4 mm, thickness = 24.7 mm. The load recorded

13
5308 C. E. C. Inga et al.

Appendix B

The results in Sect. 3.2 correspond to BBMs that have four

different grain types and 10 corresponding mineral–min-
eral associations, with properties listed in Tables 7 and 8,
respectively. To ensure that the observed behaviors are not
a function of the specific set of micro-properties or material
heterogeneity (i.e., mismatch in grain elastic and contact
properties) and more generally depict the role of geomet-
ric heterogeneity, additional models were run with different
micro-properties.

(1) Effect of contact tensile strength

Fig. 9  Images of the specimens before and after completion of the
tests Contact tensile strength directly controls the emergent
macroscopic tensile strength of BBMs (Kazerani and Zhao
2010; Fabjan et al. 2015). To understand its influence, two
by the hydraulic press was converted to indirect tensile sets of models were run, where the contact tensile strengths
strength using the following equation (International Soci- were varied by ± 25% (with respect to those listed in
ety for Rock Mechanics 1978): Table 8). Each set consisted of 72 models: 6 deterministic (3
specimens, 2 faces each), 6 semi-deterministic (3 specimens,
2P
𝜎t = 2 faces each), 30 Voronoi HGS (3 specimens, 2 faces each, 5
𝜋Dt
realizations per face), and 30 Voronoi UGS (3 specimens, 2
where P is the load at failure, D is the diameter of the speci- faces each, 5 realizations per face). Model results and overall
men, and t is the thickness of the specimen. statistics are presented in Fig. 10 and Table 12, respectively.
For ease of comparison, model strengths corresponding to

Fig. 10  Comparison of labora-

tory results (LAB) against cor-
responding deterministic (DET),
semi-deterministic (SEM),
Voronoi HGS (VO-1), and
Voronoi UGS (VO-2) BTS. a
Comparison with 25% decrease
in contact tensile strength, b
Comparison with 25% increase
in contact tensile strength, and
c Comparison with no change
in contact tensile strength, i.e.,
calibrated properties

13
Modeling Brazilian Tensile Strength Tests on a Brittle Rock Using Deterministic,… 5309

Table 12  Summary of the Case Deterministic Semi-deterministic Voronoi HGS Voronoi UGS
average BTS generated using
each type of model μ (MPa) σ (MPa) μ (MPa) σ (MPa) μ (MPa) σ (MPa) μ (MPa) σ (MPa)

Calibrated 11.9 1.2 11.5 1.0 11.4 0.9 12.4 0.8

− 25% 9.7 0.7 9.0 0.8 8.8 0.7 9.7 0.6
+ 25% 14.2 1.6 13.5 1.4 13.8 1.2 14.7 1.1

the calibrated set of micro-properties are added to Fig. 10 homogenized by weight averaging the properties listed in
and Table 12. Table 8 with the associated total contact length in every
As expected, the BTS increased and decreased for all four model. Since the contact lengths for the 10 mineral–mineral
model types with an increase or decrease in contact ten- associations varied across the models, the weighted average
sile strength, respectively. The change in simulated BTS is calculation had to be performed individually for each model.
less than 25% in the majority of the models because of the Simply stated, in these models, a single set of grain and con-
complex non-linear relationship between contact strength tact properties was assigned to all the grains and contacts.
and local fracture development. With respect to the cali- Similar to the previous analysis considering variations in
brated case, the average mean strengths changed by simi- tensile strength, 72 models were run for each case—6 deter-
lar amounts (2–2.7 MPa; Table 12), such that the strengths ministic (3 specimens, 2 faces each), 6 semi-deterministic (3
are similar across the four model types in both the + 25% specimens, 2 faces each), 30 Voronoi HGS (3 specimens, 2
and − 25% cases. This supports the general observation in faces each, 5 realizations per face), and 30 Voronoi UGS (3
Sect. 3.2 that there is a limited effect of geometric heteroge- specimens, 2 faces each, 5 realizations per face). Results are
neity on modeled BTS. presented in Fig. 11 and Table 13. The following observa-
The variability in BTS follows the same trend as the tions are made:
change introduced in the contact tensile strength, i.e.,
standard deviation (STD) of − 25% case < STD cali- • BTS increases with material property homogenization,
brated < STD + 25% case. This is explained by the greater consistent with the previous analysis. Homogenization
absolute heterogeneity in the contact tensile strength in reduces the local tensile stresses developing within a
the + 25% case in comparison with the − 25% case. For BBM (Dey and Wang 1981; Kranz 1983; Sinha and Wal-
example, if the tensile strengths of two contact types are ton 2020) and thereby delays fracture development and
2 MPa and 3 MPa, then the difference between them changes failure.
to 0.75 MPa (lower heterogeneity) and 1.25 MPa (higher • With respect to the calibrated case, the average BTS for
heterogeneity) with − 25% and + 25% change, respectively. grain property homogenized models increased by 1.1–
It follows that the − 25% model is more homogeneous in 1.9 MPa and the variability is similar across all model
an absolute sense than the + 25% model, and hence demon- types (Fig. 11a). It is not clear why the VO-HGS model
strates lower variability in BTS. This also means that con- exhibited ~ 1 MPa lower average strength than the other
tact strength heterogeneity has some influence on BTS, but three types. In any case, the simple VO-UGS model dem-
the influence is similar across the different grain structures onstrated similar strength distribution as the determinis-
considered. tic and semi-deterministic models, consistent with the
observations in Sect. 3.2.
(B) Effect of material heterogeneity • For the complete homogenized models (Fig. 11b), the
average strengths are similar across the four model types.
For this analysis, two different types of parameter homog- It is noted here that 5 realizations were run per face for
enization were considered (Sinha and Walton 2020). both VO-HGS and VO-UGS models, but no/minimal
Grain property homogenization: 6 sets of grain elastic deviation was observed (as a result, there is no scatter;
properties were computed for the 6 faces (BTS-1A, BTS-1B, see Fig. 11b). This is because the grain structure in the 5
BTS-2A, BTS-2B, BTS-3A, BTS-3B) by weight averaging realizations for a face was the same (refer to Sect. 2.2 and
the properties listed in Table 7 with the mineral proportions Table 6) and homogenizing both the contact and grain
in Table 1. All grains in the model were assigned 1 set (out properties essentially led to very similar realizations
of 6 sets) of elastic property depending on which face it (same grain structure, same grain properties, and very
represented. No homogenization of the contact property was similar contact properties).
performed. • Overall, the average strengths are similar across the four
Complete homogenization: In addition to the grain grain structures considered, meaning that the simplest
property homogenization, contact properties were also UGS model can be used instead of the more complex

13
5310 C. E. C. Inga et al.

Fig. 11  Comparison of labora-

tory results (LAB) against cor-
responding deterministic (DET),
semi-deterministic (SEM),
Voronoi HGS (VO-1), and
Voronoi UGS (VO-2) BTS. a
Comparison with grain property
homogenization, b Comparison
with grain and contact property
homogenization

Table 13  Summary of the average BTS generated using each type of model
Case Deterministic Semi-deterministic Voronoi HGS Voronoi UGS
μ (MPa) σ (MPa) μ (MPa) σ (MPa) μ (MPa) σ (MPa) μ (MPa) σ (MPa)

Calibrated 11.9 1.2 11.5 1.0 11.4 0.9 12.4 0.8

Grain homogenization 13.5 1.1 13.4 0.9 12.5 1.0 13.8 0.7
Complete homogenization 14.4 1.2 14 0.5 13.8 0.9 14.4 0.5

deterministic or semi-deterministic model for practi- Data Availability The data that support the findings of this study are
cal purposes, whether homogeneous or heterogeneous available from the corresponding author upon reasonable request.
micro-properties are used. The similar performance
Declarations
obtained using the different grain structure assumptions
suggests that the conclusions of this study regarding Conflict of Interest The authors declare that they have no known com-
the validity of the Voronoi approach could perhaps be peting financial interests or personal relationships that could have ap-
extended to the simulation of monomineralic rocks. peared to influence the work reported in this paper.

Acknowledgements The authors would like to extend their gratitude References

for the financial support. Special thanks to Dr. Katharina Pfaff of the
automated mineralogy laboratory, Jae Erickson of the thin-section labo- Abdelaziz A, Zhao Q, Grasselli G (2018) Grain based modelling of
ratory, and Bruce Yoshioka, Brent Duncan, Omid Frough, and Muthu rocks using the combined finite-discrete element method. Com-
Vinayak of the Earth Mechanics Institute at the Colorado School of put Geotech 103:73–81. [Link] oi.o​ rg/1​ 0.1​ 016/j.c​ ompge​ o.2​ 018.​
Mines for their help during the laboratory tests and analyses. 07.​003
ASTM International (2001) ASTM D 3967–95a (reapproved 2001):
Author Contributions CECI: conceptualization, methodology, soft- standard test method for splitting tensile strength of intact rock
ware, validation, formal analysis, investigation, data curation, writ- core specimens. In Annual Book of ASTM Standards: Vol.
ing—original draft, writing—review and editing, visualization. SS: 04.08. West Conshohocken, PA: ASTM International
methodology, software, validation, formal analysis, investigation, Autodesk Inc., (2019) AutoCAD 2020 Product Help. [Link]
data curation, writing—original draft, writing—review and editing, edge.​a utod​e sk.​c om/​s uppo​r t/​a utoc​a d/​d ownl​o ads/​c aas/​d ownl​
visualization. GW: conceptualization, methodology, resources, writ- oads/​conte​nt/​downl​oad-​insta​ll-​autoc​ad-​2020-​produ​ct-​help.​html.
ing—review and editing, supervision, project administration, funding Accessed 17 July 2022
acquisition. EH: conceptualization, methodology, resources, writing— Azocar K (2016) Investigating the mesh dependency and upscaling
review and editing. of 3D grain-based models for the simulation of brittle fracture
processes in low-porosity crystalline rock. Queen’s University,
Funding This work was supported by the National Institute for Kingston
Occupational Safety and Health (NIOSH) under Grant Number
200-2016-90154.

13
Modeling Brazilian Tensile Strength Tests on a Brittle Rock Using Deterministic,… 5311

Bahrani N, Kaiser PK, Valley B (2014) Distinct element method simu- Eberhardt E (1998) Brittle rock fracture and progressive damage in
lation of an analogue for a highly interlocked, non-persistently uniaxial compression, PhD Thesis. ed, Ph.D. Thesis, University
jointed rockmass. Int J Rock Mech Min Sci 71:117–130. [Link] of Saskatchewan, Saskatoon, Saskatchewan, Canada. University
doi.​org/​10.​1016/j.​ijrmms.​2014.​07.​005 of Saskatchewan, Saskatoon, Saskatchewan. [Link]
Bewick RP, Kaiser PK, Bawden WF, Bahrani N (2014) DEM simu- 16953/​deusb​ed.​74839
lation of direct shear: 1. rupture under constant normal stress Eberhardt E, Stimpson B, Stead D (1999) Effects of grain size on the
boundary conditions. Rock Mech Rock Eng 47:1647–1671. initiation and propagation thresholds of stress-induced brittle
[Link] fractures. Rock Mech Rock Eng 32:81–99. [Link]
Bobet A, Fakhimi A, Johnson S, Morris J, Tonon F, Yeung MR (2009) 1007/​s0060​30050​026
Numerical models in discontinuous media: review of advances Fabjan T, Ivars DM, Vukadin V (2015) Numerical simulation of
for rock mechanics applications. J Geotech Geoenviron Eng intact rock behaviour via the continuum and Voronoi tessella-
135(11):1547–1561 tion models: a sensitivity analysis. Acta Geotechnica Slovenica
Cai JG, Zhao J (2000) Effects of multiple parallel fractures on apparent 12:5–23
attenuation of stress waves in rock masses. Int J Rock Mech Min Farahmand K, Diederichs MS (2015) A calibrated synthetic rock mass
Sci 37:661–682. [Link] oi.o​ rg/1​ 0.1​ 016/S
​ 1365-1​ 609(00)0​ 0013-7 (SRM) model for simulating crack growth in granitic rock con-
Chen W, Konietzky H (2014) Simulation of heterogeneity, creep, sidering grain scale heterogeneity of polycrystalline rock, in:
damage and lifetime for loaded brittle rocks. Tectonophysics 49th US Rock Mechanics/Geomechanics Symposium. American
633:164–175. [Link] Rock Mechanics Association, San Francisco, California
Chen SG, Cai JG, Zhao J, Zhou YX (2000) Discrete element modelling Fredrich JT, Evans B, Wong T-F (1990) Effect of grain size on brit-
of an underground explosion in a jointed rock mass. Geotech tle and semibrittle strength: implications for micromechanical
Geol Eng 18:59–78. [Link] modelling of failure in compression. J Geophys Res. [Link]
Chen S, Yue ZQ, Tham LG (2004) Digital image-based numerical org/​10.​1029/​jb095​ib07p​10907
modeling method for prediction of inhomogeneous rock failure. Gao FQ, Stead D (2014) The application of a modified Voronoi logic
Int J Rock Mech Min Sci 41:939–957. [Link] to brittle fracture modelling at the laboratory and field scale. Int
ijrmms.​2004.​03.​002 J Rock Mech Min Sci 68:1–14. [Link]
Chen W, Konietzky H, Tan X, Frühwirt T (2016) Pre-failure damage 2014.​02.​003
analysis for brittle rocks under triaxial compression. Comput Gao F, Stead D, Elmo D (2016) Numerical simulation of microstruc-
Geotech 74:45–55. [Link] ture of brittle rock using a grain-breakable distinct element grain-
018 based model. Comput Geotech 78:203–217. [Link]
Cho NA, Martin CD, Sego DC (2007) A clumped particle model for 1016/j.​compg​eo.​2016.​05.​019
rock. Int J Rock Mech Min Sci 44(7):997–1010. [Link] Garza-Cruz T, Pierce M, Kaiser P (2014) Use of 3DEC to study
10.​1016/j.​ijrmms.​2007.​02.​002 spalling and deformation associated with tunnelling at depth.
Contreras Inga CE, Walton G, Holley E (2021) Statistical assessment of Proceedings of the Seventh International Conference on Deep
the effects of grain structure representation and micro-properties and High Stress Mining 421–434. [Link] rg/​10.​36487/​acg_​
on the behavior of bonded block models for brittle rock dam- rep/​1410_​28_​garza-​cruz
age prediction. Sustainability 13:7889. [Link] Ghazvinian E, Diederichs MS, Quey R (2014) 3D random Voronoi
su131​47889 grain-based models for simulation of brittle rock damage and
Cowie S, Walton G (2018) The effect of mineralogical parameters on fabric-guided micro-fracturing. J Rock Mech Geotech Eng
the mechanical properties of granitic rocks. Eng Geol 240:204– 6:506–521. [Link]
225. [Link] Gui YL, Zhao ZY, Ji J, Wang XM, Zhou KP, Ma SQ (2016) The grain
Cundall PA (1971) A computer model for simulating progressive effect of intact rock modelling using discrete element method
large-scale movements in blocky rock systems. Proceedings of with voronoi grains. Geotech Lett 6:136–143. [Link]
Int. Society for Rock Mechanics Symposium: Rock Fracture, 1680/​jgele.​16.​00005
129–136 Güneş Yilmaz N, Mete Goktan R, Kibici Y (2011) Relations between
Cundall PA, Strack ODL (1979) A discrete numerical model for gran- some quantitative petrographic characteristics and mechanical
ular assemblies. Geotechnique 29(1):47–65. [Link] strength properties of granitic building stones. Int J Rock Mech
1680/​geot.​1979.​29.1.​47 Min Sci 48:506–513. [Link]
Cunha AP (1990) Scale effects in rock mass. Proceeding of Interna- 003
tional Workshop on Scale Effects in Rock Masses, Balkema Insana A, Marla M, Elmo D (2016) Multiscale numerical modelling
Dan DQ, Konietzky H (2014) Numerical simulations and interpreta- related to hydrofracking for deep geothermal energy exploration.
tions of Brazilian tensile tests on transversely isotropic rocks. Int Proc Eng 158:314–319
J Rock Mech Min Sci 71:53–63 International Society for Rock Mechanics (1978) Suggested methods
Dey TN, Wang CY (1981) Some mechanisms of microcrack growth for determining tensile strength of rock materials. Int J Rock
and interaction in compressive rock failure. Int J Rock Mech Mech Min Sci 15:99–103. [Link] oi.o​ rg/1​ 0.1​ 016/0​ 148-9​ 062(78)​
Mining Sci Geomech Abstr 18(3):199–209 91494-8
Diederichs MS (1999) Instability of hard rock masses: The role of ten- Itasca Consulting Group Inc. (2014) UDEC version 6.0: Universal
sile damage and relaxation. Ph.D. Thesis, University of Waterloo, disctinct element code, user’s guide, 4th edn. Itasca Consulting
Waterloo Group Inc., Minneapolis
Diederichs MS (2003) Rock fracture and collapse under low confine- Jing L (2003) A review of techniques, advances and outstanding issues
ment conditions. Rock Mech Rock Eng 36:339–381. [Link] in numerical modelling for rock mechanics and rock engineering.
org/​10.​1007/​s00603-​003-​0015-y Int J Rock Mech Min Sci 40:283–353. [Link]
Ding X, Zhang L (2014) A new contact model to improve the simulated S1365-​1609(03)​00013-3
ratio of unconfined compressive strength to tensile strength in Jing L, Hudson JA (2002) Numerical methods in rock mechanics. Int J
bonded particle models. Int J Rock Mech Min Sci 69:111–119. Rock Mech Min Sci 39:409–427. [Link] oi.o​ rg/1​ 0.1​ 016/S ​ 1365-​
[Link] 1609(02)​00065-5

13
5312 C. E. C. Inga et al.

Jing L, Stephansson O (2007) Fundamentals of discrete element Martin CD (1993) The strength of massive Lac du Bonnet granite
methods for rock engineering—theory and applications. Devel- around underground openings. University of Manitoba, Winnipeg
opments in geotechnical engineering. Elsevier, Amsterdam, pp Mayer JM, Stead D (2017) Exploration into the causes of uncertainty
365–398 in UDEC grain boundary models. Comput Geotech 82:110–123.
Kazerani T, Zhao J (2010) Micromechanical parameters in bonded [Link]
particle method for modeling of brittle material failure. Int J Munjiza A (2004) The combined finite-discrete element method. John
Numer Anal Meth Geomech 34:1877–1895. [Link] Wiley & Sons, Chichester
1002/​nag.​884 Nicksiar M, Martin CD (2014) Factors affecting crack initiation in low
Kazerani T, Yang ZY, Zhao J (2012) A discrete element model for porosity crystalline rocks. Rock Mech Rock Eng 47:1165–1181.
predicting shear strength and degradation of rock joint by using [Link]
compressive and tensile test data. Rock Mech Rock Eng 45:695– Olsson WA (1974) Grain size dependence of yield stress in marble. J
709. [Link] Geophys Res 79:4859–4862. [Link] oi.o​ rg/1​ 0.1​ 029/j​ b079i​ 032p​
Keikha T, Keykha HA (2013) Correlation between mineralogical char- 04859
acteristics and engineering properties of granitic rocks. Electron Papoulis A, Pillai SU (2002) Probability, random variables, and sto-
J Geotech Eng 18:4055–4065 chastic processes. Tata McGraw-Hill Education, New York
Kranz RL (1983) Microcracks in rocks: a review. Tectonophysics Park JW, Park C, Song JW, Park ES, Song JJ (2017) Polygonal grain-
100(1–3):449–480 based distinct element modeling for mechanical behavior of brit-
LaBerge GL, Myers PE (1983) Precambrian geology of Marathon tle rock. Int J Numer Anal Meth Geomech 41(6):880–898
county, Wisconsin. Wis Geol Nat Hist Surv Inf Circ 45:88 Peng J, Wong LNY, Teh CI (2017a) Effects of grain size-to-particle
Lan H, Martin CD, Hu B (2010) Effect of heterogeneity of brittle rock size ratio on micro-cracking behavior using a bonded-particle
on micromechanical extensile behavior during compression load- grain-based model. Int J Rock Mech Min Sci 100:207–217.
ing. J Geophys Res. [Link] [Link]
Langford JC, Diederichs MS (2015) Quantifying uncertainty in Hoek- Peng J, Wong LNY, Teh CI (2017b) Influence of grain size hetero-
Brown intact strength envelopes. Int J Rock Mech Min Sci geneity on strength and microcracking behavior of crystalline
74:91–102. [Link] rocks. J Geophys Res 122:1054–1073. [Link]
Li D, Wong LN (2013) The Brazilian disc test for rock mechanics 2016J​B0134​69
applications: review and new insights. Rock Mech Rock Eng Peng J, Yuen Wong LN, Teh CI (2021) Influence of grain size on
46(2):269–287 strength of polymineralic crystalline rock: new insights from
Li XF, Li HB, Zhao J (2017a) 3D polycrystalline discrete element DEM grain-based modeling. J Rock Mech Geotech Eng. [Link]
method (3PDEM) for simulation of crack initiation and propaga- doi.​org/​10.​1016/j.​jrmge.​2021.​01.​011
tion in granular rock. Comput Geotech 90:96–112. [Link] Perras MA, Diederichs MS (2014) A review of the tensile strength of
org/​10.​1016/j.​compg​eo.​2017.​05.​023 rock: concepts and testing. Geotech Geol Eng 32(2):525–546.
Li J, Konietzky H, Frühwirt T (2017b) Voronoi-based DEM simulation [Link]
approach for sandstone considering grain structure and pore size. Potyondy DO, Cundall PA (2004) A bonded-particle model for rock.
Rock Mech Rock Eng 50(10):2749–2761 Int J Rock Mech Min Sci 41:1329–1364. [Link]
Li XF, Li HB, Zhao J (2019) The role of transgranular capability in 1016/j.​ijrmms.​2004.​09.​011
grain-based modelling of crystalline rocks. Comput Geotech Potyondy DO (2012) A flat-jointed bonded-particle material for hard
110:161–183. [Link] rock. Proceedings of the 46th US Rock Mechanics/Geomechan-
Li H, Ma H, Shi X, Zhou J, Zhang H, Daemen JJK (2020) A 3D grain- ics Symposium, Chicago, Illinois, Paper No. 501
based model for simulating the micromechanical behavior of salt Přikryl R (2001) Some microstructural aspects of strength variation in
rock. Rock Mech Rock Eng 53:2819–2837. [Link] rocks. Int J Rock Mech Min Sci 38:671–682. [Link]
1007/​s00603-​020-​02085-4 1016/​S1365-​1609(01)​00031-4
Li XF, Li HB, Zhao J (2021a) Transgranular fracturing of crystalline Quey R (2019) Neper reference manual: the documentation for Neper
rocks and its influence on rock strengths: insights from a grain- 3.5.1. Romain Quey
scale continuum–discontinuum approach. Computer Methods Quey R, Renversade L (2018) Optimal polyhedral description of 3D
Appl Mech Eng. [Link] polycrystals: method and application to statistical and synchro-
Li D, Li B, Han Z, Zhu Q, Liu M (2021b) Evaluation of bi-modular tron X-ray diffraction data. Comput Methods Appl Mech Eng
behavior of rocks subjected to uniaxial compression and Bra- 330:308–333. [Link]
zilian tensile testing. Rock Mech Rock Eng 54(8):3961–3975. Quey R, Dawson PR, Barbe F (2011) Large-scale 3D random poly-
[Link] crystals for the finite element method: generation, meshing and
Lisjak A, Grasselli G (2014) A review of discrete modeling techniques remeshing. Comput Methods Appl Mech Eng 200:1729–1745.
for fracturing processes in discontinuous rock masses. J Rock [Link]
Mech Geotech Eng 6:301–314. [Link] Sajid M, Coggan J, Arif M, Andersen J, Rollinson G (2016) Petro-
2013.​12.​007 graphic features as an effective indicator for the variation in
Liu G, Cai M, Huang M (2018) Mechanical properties of brittle rock strength of granites. Eng Geol 202:44–54. [Link]
governed by micro-geometric heterogeneity. Comput Geotech 1016/j.​enggeo.​2016.​01.​001
104:358–372. [Link] Scholtès L, Donzé FV (2013) A DEM model for soft and hard rocks:
Mahabadi OK (2012) Investigating the influence of micro-scale hetero- role of grain interlocking on strength. J Mech Phys Solids
geneity and microstructure on the failure and mechanical behav- 61(2):352–369. [Link]
iour of geomaterials. University of Toronto, Toronto. [Link] Shea WT, Kronenberg AK (1993) Strength and anisotropy of foliated
org/​10.​13140/2.​1.​2218.​9762 rocks with varied mica contents. J Struct Geol 15:1097–1121.
Mahabadi OK, Tatone BS, Grasselli G (2014) Influence of microscale [Link]
heterogeneity and microstructure on the tensile behavior of crys- Sheskin DJ (2003) Handbook of parametric and nonparametric statisti-
talline rocks. J Geophys Res 119(7):5324–5341. [Link] cal procedures. Chapman and Hall/CRC, Boca Raton
10.​1002/​2014J​B0110​64 Sims PK, Schulz KJ, Dewitt E, Brasaemle B (1993) Petrography and
geochemistry of early Proterozoic granitoid rocks in Wisconsin

13
Modeling Brazilian Tensile Strength Tests on a Brittle Rock Using Deterministic,… 5313

magmatic terranes of Penokean Orogen, Northern Wisconsin. Wang X, Cai M (2019) A comprehensive parametric study of grain-
US Geological Survey, Bulletin 1904-J, p 31. [Link] based models for rock failure process simulation. Int J Rock
3133/​b1904J Mech Min Sci 115:60–76. [Link]
Singh SK (1988) Relationship among fatigue strength, mean grain 2019.​01.​008
size and compressive strength of a rock. Rock Mech Rock Eng West I, Walton G, Sinha S, (2020) Simulating the Behavior of Com-
21:271–276. [Link] pressively Loaded Blanco Mera Granite Using Bonded Block
Sinha S (2020) Advancing continuum and discontinuum models of Models. Proceedings of the 54th US Rock Mechanics/Geome-
brittle rock damage and rock-support interaction. Ph.D. Thesis, chanics Symposium, Paper No. 1209
Colorado School of Mines, Golden, Colorado, USA Wong RHC, Chau KT, Wang P (1996) Microcracking and grain size
Sinha S, Walton G, (2018) Application of micromechanical modeling effect in Yuen Long marbles. Int J Rock Mech Mining Sci
to prediction of in-situ rock behavior. In: 52nd US Rock Mechan- Geomech 33:479–485. [Link]
ics/Geomechanics Symposium. American Rock Mechanics Asso- 00007-1
ciation, Seattle, Washington Xu T, Fu TF, Heap MJ, Meredith PG, Mitchell TM, Baud P (2020)
Sinha S, Walton G (2020) A study on Bonded Block Model (BBM) Mesoscopic damage and fracturing of heterogeneous brittle
complexity for simulation of laboratory-scale stress-strain behav- rocks based on three-dimensional polycrystalline discrete ele-
ior in granitic rocks. Computers Geotech 118:103363. [Link] oi.​ ment method. Rock Mech Rock Eng. [Link]
org/​10.​1016/j.​compg​eo.​2019.​103363 s00603-​020-​02223-y
Sinha S, Shirole D, Walton G (2020) Investigation of the micromechan- Yesiloglu-Gultekin N, Sezer EA, Gokceoglu C, Bayhan H (2013) An
ical damage process in a granitic rock using an inelastic bonded application of adaptive neuro fuzzy inference system for estimat-
block model (BBM). J Geophys Res 125(3):e2019JB018844. ing the uniaxial compressive strength of certain granitic rocks
[Link] from their mineral contents. Expert Syst Appl 40:921–928.
Sinha S, Abousleiman R, Walton G (2022) Effect of damping mode [Link]
in laboratory and field-scale Universal Distinct Element Code Yu Y, Yin J, Zhong Z (2006) Shape effects in the Brazilian tensile
(UDEC) models. Rock Mech Rock Eng 55(5):2899–2915 strength test and a 3D FEM correction. Int J Rock Mech Min
Sprunt ES, Brace WF (1974) Direct observation of microcavities in Sci 43(4):623–627
crystalline rocks. Int J Rock Mech Mining Sci Geomech Abstr Zhang Y, Wong LNY (2018) A review of numerical techniques
11(4):139–150 approaching microstructures of crystalline rocks. Comput Geo-
Staub I, Andersson JC, Magnor B (2004) Äspö pillar stability experi- sci 115:167–187. [Link]
ment: geology and mechanical properties of the rock in TASQ. Zhou J, Lan H, Zhang L, Yang D, Song J, Wang S (2019) Novel grain-
Svensk Kar. ed., Stockholm based model for simulation of brittle failure of Alxa porphy-
Tan X, Konietzky H, Chen W (2016) Numerical simulation of het- ritic granite. Eng Geol 251:100–114. [Link]
erogeneous rock using discrete element model based on digital enggeo.​2019.​02.​005
image processing. Rock Mech Rock Eng 49(12):4957–4964 Zhu D-F, Tu S-H, Ma H, Zhang X (2017) A 3D Voronoi and subdi-
Tapponnier P, Brace WF (1976) Development of stress-induced micro- vision model for calibration of rock properties. Modell Simul
cracks in Westerly granite. Int J Rock Mech Mining Sci Geomech Mater Sci Eng 25:25
Abstr 13(4):103–112
Tuǧrul A, Zarif IH (1999) Correlation of mineralogical and textural Publisher's Note Springer Nature remains neutral with regard to
characteristics with engineering properties of selected granitic jurisdictional claims in published maps and institutional affiliations.
rocks from Turkey. Eng Geol 51:303–317. [Link]
1016/​S0013-​7952(98)​00071-4 Springer Nature or its licensor (e.g. a society or other partner) holds
Villeneuve MC, Diederichs MS, Kaiser PK (2012) Effects of grain exclusive rights to this article under a publishing agreement with the
scale heterogeneity on rock strength and the chipping process. author(s) or other rightsholder(s); author self-archiving of the accepted
Int J Geomech 12(6):632–647 manuscript version of this article is solely governed by the terms of
Wadell H (1933) Sphericity and roundness of rock particles. J Geol such publishing agreement and applicable law.
41:310–331. [Link]
Wang X, Cai M (2018) Modeling of brittle rock failure considering
inter- and intra-grain contact failures. Comput Geotech 101:224–
244. [Link]

13