Effect of grain network structure on macroproperties and strain energy of marble under unconfined compression using 2D continuum-based voronoi models Authors: Sasan Ghorbani Published: 2025 ABSTRACT AbstractOne of the most significant elements influencing the mechanical properties and intrinsic energy of rock is its microstructure. In order to study the effects of grain network structure on the macroproperties and strain energy parameters of marble, a 2D continuum-based Voronoi Tessellated model is employed. For this purpose, 36 Voronoi joint models with irregular, medium regular, almost regular polygon shapes and with low, medium, high, and very high densities were generated in different realizations. Then, the stress–strain curves and deformation parameters under unconfined compression were determined for each network. In the next step, for each network, three main strain energy parameters including, elastic strain energy, dissipated strain energy, and total strain energy were calculated. The results showed that in all Voronoi networks, brittle failure of marble with low joint densities occurs at lower strains. Moreover, as the density of the Voronoi joint network increases, the peak strength and elastic modulus of marble decrease. Finally, it was found that the majority of the energy that was absorbed prior to the peak strength was converted into elastic energy and stored within the rock. In rock engineering problems, microstructure can play a significant role in macroscopic behaviors and internal energy of brittle rocks.Keywords: Voronoi models, Strain energy, Brittle failure, Stress–strain curve, MicrostructureSubject terms: Engineering, Materials science Introduction The macroscopic behavior of intact rock is determined by its heterogeneous character at the grain scale1–3. Rock damage and deformation processes are mostly caused by stress heterogeneity, which is introduced by grain shape, grain boundaries, mineral components, grain size, and micro-flaws present in the rock microstructure4,5. In general, it is challenging to carry out studies that can shed light on such intricate processes as heterogeneity evolving at the grain-scale using standard laboratory-based techniques6. Therefore, numerical simulation techniques are employed to analyze the grain scale of the rock materials since they can not only replicate the complicated conditions of rocks but also have the advantages of being inexpensive and yielding findings fast. It is now feasible to include grain-scale heterogeneities, like grains and grain boundaries, in numerical models because of recent developments in computing power and software applications. These models, referred to as Grain-Based Models (GBM), depict a rock medium as a collection of triangular or polygonal blocks that stand in for rock minerals. Numerical “joint” elements or contacts are used to represent the interactions between the grains at their boundaries. By simulating the grain-like structure of rocks, the GBM is intended to represent the pre- and post-peak fracture processes. In order to accomplish this, a numerical model must include a variety of grain-scale heterogeneities. The continuum, discontinuum, and hybrid continuum-discontinuum numerical approaches are commonly employed to investigate the mechanical behavior of rocks at the grain scale7. The mechanical characteristics and failure process of hard brittle rocks can be significantly influenced by the existence of grain heterogeneities. In this regard, the grain-based model has been used frequently by researchers using different numerical programs. For example, Wang et al.8,9 investigated the microcracking behaviors of brittle rock using a polygonal UDEC-GBM. Gao et al.10 used a distinct element grain-based method to simulate the microstructure of brittle rock. Hu et al.11 analyzed the heterogeneities of grain boundary contact to simulate the mechanical behavior of granitic rocks using the PFC program. Also, Wang and Cai5 used 3DEC-GBMs for rock failure process simulation. In the meantime, recently continuum-based Voronoi Tessellated models have garnered significant interest mainly due to lower runtime7,12,13. Amiri and Bahrani7 investigated the brittle damage and failure process of granite using FEM-Voronoi Tessellated Models (VTMs). Their results showed that compared to the homogeneous VTM, the heterogeneous VTM is thought to be a more representative model for hard rocks. Li and Bahrani13 developed the RS2-GBM to simulate the brittle rock failure of intact and granulated marble. Their findings demonstrated that the variation of stresses under compressive loading is controlled by grain-scale heterogeneity. Hamediazad et al.14 developed the 3D Voronoi Tessellated model using the RS3 program to simulate the failure process of hard brittle rocks under unconfined compression. Their results revealed that the peak strength and stress–strain curve of the RS3-VTM are similar to those of the RS2-VTM. Bewick et al.15 investigated the effect of grain heterogeneity on tensile stress generation inside a brittle rock specimen under compression. Their results depicted that tensile stress generation within a brittle rock specimen is controlled by the grain boundary orientation and grain system arrangements. Satheesh et al.2 used the Continuum Grain-Based Model (CGBM) to simulate the brittle rock failure process of intact marble under varying confinement levels. Their results showed that the CGBM’s ability to capture the change in the failure modes with increasing confinement. Rógenes et al.12 presented a numerical model called the Continuum Voronoi Block Model (CVBM) to represent the fracture process in hard rocks. Their proposed model matched some macro-properties such as Poisson’s ratio, Young’s modulus, tensile strength, and peak strength. In the current study, using a finite element method (FEM) based on Voronoi Tessellated structures, various grain networks of marble were generated. These networks were generated in four general categories with low, medium, high, and very high joint density. Also, to investigate the role of grain arrangement in the generated networks, joints with irregular, medium regular, and almost regular polygon shapes were generated. In the next step, the networks were subjected to unconfined loading and the macro-properties of marble, including peak stress, peak strain, and elastic modulus were analyzed. Finally, the strain energy characteristics of intact marble in each Voronoi model were calculated using stress–strain curve. Materials and methods FEM-GBM of marble This study utilizes the calibrated parameters of intact marble, building upon previous work by Li and Bahrani13 in RS2, in order to investigate the effect of grain structures with different densities and joint shapes on macro-properties and strain energy parameters. For this purpose, a series of grain-based models with dimensions of 20 × 50 mm was constructed to represent the studied marble specimen. The bottom boundary of the model was fixed in a vertical direction, while the midpoint was fixed in both vertical and horizontal directions (Fig. 1). To simulate the UCS test, a displacement was applied to the specimen’s top boundary at a rate of 0.002 mm/stage until the numerical specimen failed. To calculate the strain, the applied displacement was divided by the specimen’s length. Boundary conditions of generated Voronoi models under unconfined compression. Calibration assumptions The important step in Voronoi-tessellated models (VTM) is to calibrate the 16 micro-properties and macro-properties. These properties include grain and grain boundary properties. The objective of the GBM calibration of rock was to align its macro-properties, such as the elastic modulus and the unconfined compressive strength, with the laboratory characteristics of intact rock materials. Generally, in order to calibrate FEM-GBM to the laboratory properties of rocks, it is required to adjust the micro-properties, including those of grains and grain boundaries. Energy evolution characteristics of rock under uniaxial compression The total energy density (U) input to the rock mass is converted into elastic strain energy density () and dissipated strain energy density (), assuming that the rock mass element is in a closed system and that there is no energy exchange with the external environment during its deformation and failure process (Eq. (1)). Eq (2) can be used to determine the total energy produced by the axial load applied on the rock sample during uniaxial compression testing16. Numerical integration (trapezoidal rule) is employed as an approximation since the integral is difficult to compute constantly in actual testing. Next, Eq. (2) can be rewritten as Eq. (3)17.. The elastic strain energy is calculated via Eq. (4). It should be mentioned that the initial elastic modulus (Ei) is generally used in place of the unloading elastic modulus for engineering applications, as shown in Eq. (4)17,18. Finally, the dissipated energy is determined through Eq. (5)19,20. where U is the total energy, is the elastic energy, is the dissipated energy, is axial stress, is the axial strain and correspond to the stress values at steps i and i + 1, respectively. Similarly, and denote the strain values of the i and i + 1 steps, respectively, and are employed to calculate the incremental work within discretized intervals of the stress–strain curve. Finally, it is possible to systematically quantify the energy evolution metrics throughout the failure process by applying Eqs. (1)-(5) to the data gathered from uniaxial compression simulations on marble. The relationship between the elastic and dissipated energies of rock materials under uniaxial compression conditions is depicted in Fig. 2. As shown in Fig. 2, the area of the shadow represents the elastic strain energy, and the red area represents the dissipated energy. Energy calculation and analysis model. Results of numerical simulation Generation of various voronoi models The Phase2 program (RS2 in the new version) uses the Poisson distribution for the random seed generation in order to produce a Voronoi structure. Next, lines are created at equal distances between neighboring seed pairs, and these lines together define the Voronoi cell for each seed. Grain representations in an FEM-Voronoi based model are Voronoi blocks made up of several triangular mesh elements. Grain contacts, or block boundaries, are represented as open-ended joint elements in the simulation. In this study, 36 Voronoi models with different joint densities and polygon shapes were generated to investigate the role of grain structure (degree of complexity) on the compressive deformation properties, stress–strain curves, and strain energy of marble. Thus, Voronoi networks were created in three groups, including irregular, medium regular, and almost regular shapes and in four groups with low, moderate, high, and very high joint densities. It should be noted that for each set of Voronoi models, three different realizations were considered in order to reduce probabilistic uncertainties. In the next step, to generate the various Voronoi networks, the calibrated values of grain and grain boundary properties must be applied to the models. For this purpose, in this study, the calibrated grain and grain boundary properties of intact marble presented by Li and Bahrani13 were used (Table 1). In total, as can be seen in Table 2, 36 Voronoi Tessellated models were generated and analyzed. Table 1Grain and grain boundary properties of calibrated RS2-GBM of marble13.Grain properties(peak/residual)/[MPa]/[]/[MPa][GPa]Value45/1537/3714/0.180Grain boundary properties(peak/residual)/[MPa]/[]/[MPa][GPa/m][GPa/m]Value40/0.1503/0.1240,00024,000Note: g and b denote grain and grain boundaries properties, respectively. /: Peak and residual cohesion; /: Peak and residual friction angle; /: Peak and residual tensile strength; : Young’s modulus; /: Peak and residual cohesion; /: Peak and residual friction angle; /: Peak and residual tensile strength; : Normal stiffness; : Shear stiffness. Grain and grain boundary properties of calibrated RS2-GBM of marble13. Grain boundary properties (peak/residual) Note: g and b denote grain and grain boundaries properties, respectively. /: Peak and residual cohesion; /: Peak and residual friction angle; /: Peak and residual tensile strength; : Young’s modulus; /: Peak and residual cohesion; /: Peak and residual friction angle; /: Peak and residual tensile strength; : Normal stiffness; : Shear stiffness. Table 2Different Voronoi joint networks using FEM. Different Voronoi joint networks using FEM. Stress–strain curve under unconfined compression This section presents the findings from the GBMs of intact marble, including the stress–strain curves under unconfined compression. Figs 345 present the stress–strain curves for the various Voronoi models of marble. The GBMs of intact marble in these Figures show a brittle reaction, with a sharp reduction in stress after the peak stress. Stress–strain curves for irregular Voronoi shapes. Stress–strain curves for medium regular Voronoi shapes. Stress–strain curves for almost regular Voronoi shapes. As can be seen in all Voronoi networks, brittle failure of rock with low joint densities occurs at lower strains. In other words, failure occurs in networks with low joint densities at fewer loading stages. Conversely, as the Voronoi joint density increases in the network, the failure of the specimen occurs at higher stages or, in other words, at higher strains. For example, the strain value at failure in the irregular network with low joint density (realization 1) is 0.0016, while in the network with very high joint density, it is 0.00232 (Fig. 3). Based on the simulation and analysis of the results, the highest stage at the moment of failure was obtained for the almost regular network with very high Voronoi joint density (Realization 2). It is also worth noting that different realizations can be generated in each network with different polygon shapes and Voronoi densities. This seems necessary to reduce the uncertainty caused by the arrangement of grains in the numerical simulations. For example, in an irregular network with low joint density, different stress–strain curves are obtained. Therefore, the role of grain arrangement and structure on the macro-properties and stress–strain behavior of rocks can be investigated. According to the analysis of 36 Voronoi models, the regularity of polygon shapes and the Voronoi joint densities are two very important factors in studying the stress–strain behavior of rocks. Determination of macro-properties in each voronoi model In general, important information about the macro-properties of rock materials can be found using the stress–strain curve. A stress–strain curve of a rock sample under uniaxial loading contains four stages, including the compression stage, the linear elastic stage, the yield stage, and the failure stage. In this study, three important deformation parameters, including peak stress at failure (peak strength), peak strain at failure, and elastic modulus were determined from the stress–strain curves of each Voronoi model. The deformation parameters of models are shown in Figs. 678. In these Figures, is the peak strength, Ei is the elastic modulus, and is the peak strain (the axial strain corresponding to the peak strength). The values of for (a) irregular networks, (b) medium regular networks, (c) almost regular networks. The values of for (a) irregular networks, (b) medium regular networks, (c) almost regular networks. The values of for (a) irregular networks, (b) medium regular networks, (c) almost regular networks. According to the results presented in Figs. 678, as the density of the Voronoi joint network increases, the peak strength and elastic modulus of marble decrease. In contrast, the failure strain of marble increases with the increase in the joint network density. Additionally, the results of statistical analyses, including the average, standard deviation, and median of , Ei, and obtained from simulations in different realizations, are presented in Table 3. For example, in the irregular networks with low and very high joint densities, the average peak strength values are 109.5 and 84.6 MPa, respectively. Therefore, the average peak strength value in the low joint density networks is about 23% higher than that in the networks with very high joint density. Also, the average elastic modulus value in the irregular networks with low densities is about 40% higher than that in the network with very high joint density. Table 3Results of statistical analysis of , Ei, and .Regularity of polygon shapeVoronoi joint density[Average, Standard deviation, Median](MPa)Ei (GPa)(%)IrregularLow[109.5, 0.9, 109.3][65.7, 3.4, 67][0.159, 0.005, 0.160]Medium[102.3, 3.9, 104.9][57.6, 6.9, 62][0.173, 0.004, 0.176]High[93.6, 1.9, 92.7][50.4, 4.6, 48.9][0.193, 0.005, 0.192]Very high[84.6, 4.2, 84.7][39.8, 2.6, 38.6][0.233, 0.002, 0.232]Medium regularLow[119.8, 8, 119.9][72.8, 1.5, 73.4][0.163, 0.007, 0.160]Medium[108, 2.2, 108.4][61.5, 1.8, 60.4][0.183, 0.002, 0.184]High[101.4, 7.5, 102.3][50.9, 5.7, 53][0.196, 0.003, 0.196]Very high[91.8, 6.4, 91.3][41.8, 3.1, 43.4][0.233, 0.005, 0.232]Almost regularLow[118.4, 8.1, 114.7][72.8, 1.5, 73.4][0.163, 0.007, 0.160]Medium[112.9, 2.1, 114][60, 1.7, 59.4][0.191, 0.008, 0.192]High[105.9, 4.5, 107][53, 3.9, 52][0.199, 0.007, 0.204]Very high[93.5, 1.4, 93.2][42.6, 3.1, 44.8][0.231, 0.011, 0.232] Results of statistical analysis of , Ei, and . Energy characteristics of marble under uniaxial compression During the uniaxial compression process, the specimen’s energy dissipation is shown in the friction caused by fracture closure, crack initiation, and crack propagation until unstable failure occurs21. After drawing the stress–strain curves and determining the values of the three parameters (, , Ei) for each Voronoi model, the energy characteristics, including elastic energy density, dissipated energy density, and total energy density, were calculated using the Equations presented in section “Energy evolution characteristics of rock under uniaxial compression”. The strain energy transformation processes of the marble specimen by considering various Voronoi models (realization 1) during uniaxial compression are plotted (Figs. 91011). In the initial deformation stage, the input energy is mainly dissipated for the closure of initial cracks. In the elastic stage, the specimen has almost no cracks. When the damage occurs, the dissipated energy of the marble starts to increase significantly. As can be seen in Figs. 91011, in all Voronoi models, the majority of the energy that was absorbed prior to the peak strength was converted into elastic energy and retained in the rock. During the nonlinear deformation process, the rate of increase in elastic energy decreased as the rock approached its peak strength, whereas the rate of increase in dissipated energy increased. It should be noted that in networks with very high joint density, the amount of elastic energy is greater than in networks with low joint density. For example, in the irregular network with low joint density (realization 1), the maximum elastic energy is 77.4 kJ.m−3, while in a network with very high density, it is 98.2 kJ.m−3. In other words, the elastic energy in irregular networks with very high joint density is about 22% higher than in networks with low joint density. This ratio is equal to 29% and 14% for medium and almost regular networks with low and very high joint densities, respectively. It should be noted that all energy characteristics related to the peak strength point of the Voronoi models have been calculated and presented in Table 4. As can be seen in this Table, in addition to the three energy characteristics, two ratios and have also been calculated and presented. Based on the results, it is clear that in most simulated Voronoi models, the amount of is less than 10%. Physically, when an external force is applied to a rock, part of the external force is converted into dissipated energy, which leads to a gradual deterioration of the strength, while another part of the external force is converted into releasable strain energy that is stored in the rock. Accordingly, the reason for the lower values of Ud/U (< 10%) in the results related to the initiation and propagation of cracks in a rock sample. In other words, when the energy stored in the rock reaches a critical value, the elastic energy stored in the sample is released. Therefore, cracks begin to grow inside the sample. As the energy sorted in the rock increases, the cracks propagate more rapidly, ultimately leading to the failure of the rock. The physical manifestations of this phenomenon include plastic deformation, crack propagation, and crack friction, which can be expressed as energy dissipation. Energy evolution curve of irregular Voronoi models with different joint densities. Energy evolution curve of medium regular Voronoi models with different joint densities. Energy evolution curve of almost regular Voronoi models with different joint densities. Table 4Energy parameters corresponding to the peak strength of each Voronoi joint network.Grain network shapeJoint densityRealization(MPa)(kJ.m−3)(kJ.m−3)(kJ.m−3)(%)(%)IrregularLow1110.777.479.21.897.72.32108.56471790.29.83109.391.894.32.597.32.7Medium196.797.499.52.197.92.12105.395.598.73.296.83.23104.965.165.90.898.81.2High192.786.587.20.799.20.8291.889.791.92.297.62.4396.4100.7105.64.995.44.6Very high184.798.21023.896.33.7289.7108.3110.21.998.31.7379.4111.91197.1946Medium regularLow1119.986.387.71.498.41.62109.987.994.76.892.87.23129.7111.6118.6798.31.7Medium1110.6110.3110.40.199.90.12105.294.798.94.295.84.23108.499.7102.83.1973High1110.2110.2113.43.297.22.82102.382.586.13.695.84.2391.888.198.210.189.710.3Very high199.9121.9128.56.694.95.1284.2116.6118.21.698.61.4391.3106112694.65.4Almost regularLow1114.795.799.13.496.63.42110.8103.6105.72.19823129.7104106.12.1982Medium1114102.9103.20.399.70.32109.9110.8117.16.394.65.43114.7110.1110.60.599.50.5High1110.7104.7107.22.597.72.32107109.1111.22.198.11.9399.9101.9103.81.998.21.8Very high195.4110.6119.58.992.67.4291.9112.1114.22.198.21.8393.2100.7103.62.997.22.8 Energy parameters corresponding to the peak strength of each Voronoi joint network. The patterns of yielded joint elements shown in Fig. 12 are used to identify the mode of failure at different realizations and networks. As shown in this Figure, all three types of tensile, shear, and combined tensile-shear yielded elements are observed in the failure path of the models. In other words, the elastic energy stored in the models is released according to the patterns shown and causes the specimen to fail. Also, in the unconfined compression test, macro-fractures are mainly formed by the accumulation of inter-grain tension cracks that are sub-parallel to the loading direction (Fig. 12, a-c). Moreover, conjugate shear bands have been observed in almost regular network with very high joint density (Fig. 12-d). Failure modes of intact marble at (a) realization 3 from irregular-high density, (b) realization 2 from medium regular-medium density, (c) realization 1 from medium regular-high density, (d) realization 2 from almost regular-very high density. Discussions Voronoi structures divide space into cells based on proximity to a set of points. In rock modeling, these cells can represent individual grains or crystals. Increasing the number of Voronoi cells (smaller cell size) leads to a more heterogeneous microstructure with a higher density of grain boundaries. These grain boundaries are potential sites for microcrack initiation and propagation under stress. Hence, as the complexity of the Voronoi model increases, the density of grain boundaries also increases. This leads to a higher probability of crack initiation and propagation, ultimately reducing the overall strength and elastic modulus of the rock material. This is because a more complex Voronoi structure often results in a higher density of grain boundaries or interfaces, which can act as potential failure points. In other words, this is primarily due to the increased heterogeneity and the presence of more potential crack initiation sites as the Voronoi cells become smaller and more numerous. For this purpose, The results of this study are consistent with the results obtained from the studies of Xi et al.22. In their studies, the uniaxial compressive strength ​​of the models decreased with increasing triangulation joint network density from 100 to 5000. Based on the above explanation, each Voronoi Tessellated model has internal energy. The absorbed energy of rock samples is mostly stored as elastic strain energy, and the three types of strain energy progressively increase with the loading process before the yield point. Although the specimen still absorbs some external energy, the rate of total energy absorption is significantly decreased. At this time, the dissipative strain energy is increasing quickly, meaning that the rock sample’s material strength is gradually declining, the elastic strain energy can be released quickly, and the entire specimen becomes unstable as a result of the macro-crack penetration and slip dislocation of the macro crack surface. It should be noted that there are limitations in 2D modeling, and therefore, the limitations can be reduced by using 3D programs. In 2D modeling, single mineral and contact properties are considered and equivalent 2D plane strain simulations are performed instead of full 3D modeling. Considering these limitations, the results of this study can be used to guide the calibration process in 3D models such as block-based model (BBM)23, 3DEC-GBM5,24, bonded particle models (BPM)25, X-ray microtomography26, etc. Conclusions In the current research, a 2D Voronoi Tessellated finite element model is used to simulate the microstructure of marble. 36 unconfined compression simulations are carried out on the Voronoi joint networks with irregular, medium regular, almost regular polygon shapes and with low, medium, high, and very high densities. The main objective of this study is to investigate the effect of different grain network structures on the macro-properties and strain energy parameters. Based on the numerical simulation results, the following conclusions can be drawn: In all Voronoi networks, brittle failure of marble with low joint densities occurs at lower strains. Conversely, as the Voronoi joint density increases in the network, the failure of the specimen occurs at higher strains. In all Voronoi networks, brittle failure of marble with low joint densities occurs at lower strains. Conversely, as the Voronoi joint density increases in the network, the failure of the specimen occurs at higher strains. As the density of the Voronoi joint network increases, the peak strength and elastic modulus of marble decrease. In contrast, the failure strain of marble increases with the increase in the joint network density. As the density of the Voronoi joint network increases, the peak strength and elastic modulus of marble decrease. In contrast, the failure strain of marble increases with the increase in the joint network density. All Voronoi models show that the majority of the energy that was absorbed prior to the rock yielding was converted into elastic energy and retained inside the rock. During the nonlinear deformation process, the rate of increase in elastic energy decreased as the rock approached its peak strength, whereas the rate of increase in dissipated energy increased. All Voronoi models show that the majority of the energy that was absorbed prior to the rock yielding was converted into elastic energy and retained inside the rock. During the nonlinear deformation process, the rate of increase in elastic energy decreased as the rock approached its peak strength, whereas the rate of increase in dissipated energy increased. Finally, in this study, a new procedure for investigating the macro-properties and energy parameters of marble was presented based on FEM-GBM. The presented approach in this study can also be used on other types of rocks. Due to the high costs of laboratory experiments, the presented procedure in this study can best demonstrate, at the lowest cost, the role of microstructure and texture in determining the internal energy of rocks. Moreover, it is proposed to simulate the micro-failure evolution of rock masses using the GBM approach. Author contributions Sasan Ghorbani: Data curation, Writing – original draft, Conceptualization, Methodology, Investigation. Funding This research received no funding. Data availability All data generated and analyzed during this study are included in this article. Declaration Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement All data generated and analyzed during this study are included in this article.