Effect of Aggregate and Void Characteristics on Shear Resistance of Asphalt Mixtures

Abstract

Shear strength plays a critical role in the rutting resistance of asphalt pavements. In this study, a three-dimensional discrete element method (DEM) model was developed in PFC^3D to systematically investigate the effects of aggregate and void characteristics on the shear resistance of asphalt mixtures under triaxial loading. The simulation results exhibited excellent agreement with laboratory triaxial tests across varying confining pressures and mixture gradations, confirming the reliability of the DEM approach. Both aggregate and void characteristics were found to exert significant influences on shear deformation. Uniform aggregate distributions and higher friction coefficients enhanced interparticle interlocking, thereby increasing cohesion, internal friction angle, and shear resistance. In contrast, elevated air-void contents or nonuniform void distributions significantly degraded shear performance, whereas a well-distributed void size range (1–4 mm) optimized aggregate skeleton interlocking and load-transfer efficiency. These findings provide micromechanical insights into the mechanisms governing shear resistance and offer practical guidance for designing asphalt mixtures with improved rutting resistance.

1. Introduction

Asphalt mixtures are among the most widely used materials in modern pavement engineering, and their long-term performance is crucial to pavement longevity and traffic safety. However, asphalt pavements are particularly vulnerable to rutting under high-temperature and heavy-load conditions—a prevalent form of distress that severely compromises structural integrity and driving safety. Fundamentally, rutting originates from inadequate shear strength within the asphalt mixture, which leads to the progressive accumulation of irreversible plastic deformation [1,2]. Consequently, a thorough understanding of the shear behavior of asphalt mixtures and its governing factors is essential to improving rutting resistance and enhancing the durability and safety of road infrastructure.

The shear strength of asphalt mixtures results from a complex interplay of factors, including intrinsic material properties—such as aggregate gradation, void distribution, and asphalt binder characteristics—as well as the mixture’s mechanical response under complex loading conditions [3,4,5]. Previous studies have shown that the macroscopic shear performance of asphalt mixtures is strongly associated with the interfacial mechanics between aggregates and the asphalt binder [6]. This interfacial shear behavior exhibits a pronounced temperature dependence [7,8,9] and is further influenced by aggregate surface roughness, mineral composition, and the aging state of the asphalt binder [10]. Moreover, the aggregate–binder interaction represents a highly coupled process in which fine aggregate interference and binder lubrication can deteriorate the mixture’s resistance to shear deformation [11]. As a result, the shear strength of asphalt mixtures arises from the combined effects of material properties and complex microscale mechanisms, making accurate characterization of shear behavior essential for improving rutting resistance. Numerous test methods have been developed to evaluate the shear strength of asphalt mixtures, including direct shear, inclined shear, rotational shear, uniaxial penetration, and triaxial shear tests [12,13,14,15,16,17]. Among these, the triaxial shear test is widely recognized as the most reliable method for characterizing shear behavior and key strength parameters—such as cohesion c and internal friction angle φ—because it closely replicates the complex stress conditions encountered in real pavements [18]. However, conventional laboratory-based triaxial tests face significant challenges, including complex specimen preparation, high costs, lengthy testing procedures, and limited capability to isolate individual microscale factors for systematic analysis. These constraints have hindered deeper exploration of the relationship between the macroscopic mechanical behavior of asphalt mixtures and their internal microstructural characteristics.

With the rapid advancement of computer technology and numerical simulation techniques, virtual testing has become a powerful tool for investigating the mechanical behavior of asphalt pavements. Unlike traditional laboratory testing, numerical simulations not only overcome experimental limitations but also allow systematic exploration of how multiple factors influence material performance under diverse loading conditions. Currently, analytical methods, the finite element method (FEM), and the discrete element method (DEM) serve as the primary approaches for pavement mechanical analysis. Among these, DEM has proven particularly effective in modeling asphalt mixtures, as it can accurately represent discontinuous media, particulate systems, and heterogeneous material behavior [19,20]. Buttlar et al. were the first to introduce DEM for asphalt mixture simulations, proposing a microstructural modeling approach for asphalt mixture [21]. Since then, researchers have developed two- and three-dimensional DEM-based models that effectively reproduce aggregate spatial distributions and enable high-fidelity virtual simulations [22,23]. DEM has been widely applied to simulate rutting-induced shear deformation and to monitor aggregate particle movement throughout the deformation process [24]. For example, Jiang et al. developed a DEM model for uniaxial penetration tests to examine the effects of experimental conditions and microstructural features on shear strength, ultimately proposing a high-interlock, dense-graded mix with improved high-temperature stability [25]. Similarly, Peng et al. employed DEM to evaluate the effects of aggregate size, temperature, asphalt content, and loading rate on shear strength [26]. Yang et al. employed DEM to simulate triaxial shear tests, reporting strong correlations between numerical predictions and laboratory results [27]. Fan et al. further analyzed the effects of loading rate, confining pressure, porosity, and aggregate properties on the dilatancy behavior of asphalt mixtures under triaxial shear conditions [28], while Zou et al. investigated how aggregate structure, gradation, and stiffness affect shear strength through virtual triaxial shear testing [29]. These studies have greatly advanced the application of DEM in evaluating the shear performance of asphalt mixtures, surpassing the constraints of conventional testing methods. Nevertheless, most existing studies have not fully addressed the effects of heterogeneous air-void distribution and spatial variations in aggregate arrangement on the macroscopic shear behavior of asphalt mixtures. Consequently, the mechanisms linking microstructural heterogeneity to macroscopic shear performance remain insufficiently understood, representing a key knowledge gap and an ongoing challenge for future research.

In this study, a DEM-based virtual triaxial testing framework was developed by using PFC^3D, incorporating realistic aggregate spatial arrangements and internal air-void distributions. Virtual tests under varying confining pressures and gradations were validated against laboratory experiments to ensure model reliability. Compared with previous DEM studies that often focus on single factors, our approach offers three main advancements: (1) integrated consideration of aggregate and void characteristics, enabling systematic evaluation of microstructural heterogeneity; (2) quantitative insight into the optimal void size range (1–4 mm) that enhances aggregate interlocking and load transfer; and (3) validated generalizability across multiple mixture gradations and stress conditions, providing practical guidance for mixture design. These contributions bridge microscale mechanisms and engineering practice, offering a robust numerical tool for understanding and improving the rutting resistance of asphalt pavements.

2. Experimental Program

2.1. Raw Materials

The asphalt used in this study was 70# asphalt and its technical properties are summarized in

Table 1. The principal physical properties of 70# asphalt.

Indices	Softening Point (°C)	15 °C Ductility (cm)	25 °C Penetration (0.1 mm)
Test Value	68	35	53.2
Test Standard [30]	≥45	≥40	60–80

. All filler, fine aggregates, and coarse aggregates were limestone and met the technical requirements of the Chinese Technical Specification for Construction of Highway Asphalt Pavement (JTG F40-2004) [30]. The detailed material properties are not presented here for brevity.

2.2. Laboratory Triaxial Test

The triaxial test was designed to replicate the in-service stress conditions of pavement materials by applying axial stress and confining pressure in multiple directions, thereby enabling the determination of their shear strength parameters. Tests were conducted in displacement control mode using a Universal Testing Machine (UTM, Shanghai Nuolai Technology Co., Ltd., Shanghai, China) under confining pressures of 0, 138, and 276 kPa, with a loading rate of 1.27 mm/min and a test temperature of 60 °C. Cylindrical test specimens (ϕ 100 mm × h 150 mm) were prepared using the gyratory compaction method. Three asphalt mixture gradations were evaluated, as summarized in

Table 2. Mineral aggregate gradation.

Mix Types	Mass Percentage Passing Through Sieves (mm)
	19	16	13.2	9.5	4.75	2.36	1.18	0.6	0.3	0.15	0.075
AC-16	100.0	95.0	86.0	70.0	48.0	34.0	24.5	17.5	12.5	9.5	6.0
SMA-16	100.0	95.0	75.0	55.0	26.0	19.5	18.0	15.0	12.5	11.5	10.0
SUP-16	100.0	95.0	82.5	64.0	41.0	30.0	22.5	18.0	14.0	10.0	7.0

, with optimum asphalt–aggregate ratios of 5.30%, 5.75%, and 4.90% for AC-16, SMA-16, and SUP-16, respectively.

2.3. Virtual Triaxial Test

2.3.1. DEM Model Construction

A schematic of the model-generation process is shown in Figure 1. The virtual specimen was modeled with dimensions of 100 mm in diameter and 150 mm in height. To balance computational efficiency and accuracy, a particle-size-grouping strategy was employed. Aggregates in the 0.075–2.36 mm range were grouped into a single category, as their small size, high abundance, and uniform role in void filling and strength development justified simplification. Larger particles were divided into five size groups: 2.36–4.75 mm, 4.75–9.5 mm, 9.5–13.2 mm, 13.2–16 mm, and 16–19 mm, with radii uniformly distributed within each size range. After the initial model generation, residual stresses were observed in the numerical specimen and needed to be eliminated [31]. First, all particles were assigned slightly smaller diameters to avoid initial overlaps, followed by a gradual restoration to their design sizes. After this resizing step, the contact network was automatically recalculated by PFC^3D to update all interparticle contacts and forces. Second, the system was subjected to a dynamic relaxation process, during which boundary displacements and particle velocities were continuously monitored until both the unbalanced force ratio fell below 1 × 10⁻⁵ and the mean stress fluctuations were within ±0.5%. This ensured that the model reached a quasi-static equilibrium state prior to loading. Residual stresses were also checked to confirm that no significant preloading effects remained after equilibration. The final stress field was spatially uniform, and the average principal stress difference was less than 1% across the specimen. A detailed description of this equilibration protocol has been added to the revised manuscript to enhance transparency and reproducibility.

Particle numbers and distributions were determined by the target air-void content and gradation parameters, replicating the volume fractions of coarse aggregates, asphalt mastic, and air voids as specified in the laboratory mix design. This approach ensured that the virtual specimens accurately reflected the volumetric characteristics of the physical samples. In addition, spherical particles were adopted to simplify particle geometry and allow a clearer focus on the effects of gradation and void structure. Although spherical particles do not explicitly reproduce the angularity and surface texture of real aggregates, their influence on interparticle interaction—which primarily governs shear strength through mechanical interlocking—was represented by setting the interparticle friction coefficient. This approach, commonly used in discrete element modeling, effectively captures the contribution of aggregate surface characteristic to shear resistance while maintaining computational efficiency and reproducibility [28].

Rigid upper and lower walls were generated in PFC to simulate loading platens for applying axial stress σ₁ to the numerical model, while cylindrical sidewalls were used to replicate the confining pressure σ₃ employed in laboratory triaxial tests. The stress applied to the wall cannot be specified in PFC^3D software 5.0; therefore, the wall stress must be adjusted through the velocity-stress coefficient to ensure that the lateral wall provide a constant target confining pressure. The velocity-stress coefficient is determined using Equations (1)–(3):

The wall velocity is assumed to be given by

$$u=G\left({\sigma }^m-{\sigma }^r\right)=G\Delta \sigma$$

(1)

where u represents the wall velocity; G is the velocity-stress coefficient; the ${\sigma }^m$, ${\sigma }^r$ and $\Delta \sigma$ denote the current stress, target stress, and stress difference, respectively.

Within a computational time step, the incremental force exerted on the wall is expressed as

$$\Delta F=k_n^{\left(w\right)}{N}_{c}u\Delta t$$

(2)

where ΔF represents the incremental force, N_c is the number of contacts between particles and the wall, and $k_n^{\left(w\right)}$ denotes the average contact stiffness.

Consequently, the stress increment on the wall within a computational time step is given by

$$\Delta {\sigma }^{\left(w\right)}={\displaystyle \frac{k_n^{\left(w\right)}{N}_c{u}^{\left(w\right)}\Delta t}{A}}$$

(3)

where A is the wall area and Δt denotes the particle motion time.

To ensure that the cylindrical walls maintain a stable confining pressure, the stress increment within each time step must not exceed the absolute difference between the target and measured stresses, as shown in Equation (4). A relaxation factor α is introduced to regulate this process, as it affects the stability of the confining pressure during the servo adjustment of the lateral walls. In this study, the α value was set to 0.8.

$$\left|\Delta {\sigma }^{\left(w\right)}\right|<\alpha \left|\Delta \sigma \right|$$

(4)

Equation (5) is obtained by substituting Equations (1) and (3) into Equation (4).

$${\displaystyle \frac{k_n^{\left(w\right)}{N}_{c}G\left|\Delta \sigma \right|\Delta t}{A}}<\alpha \left|\Delta \sigma \right|$$

(5)

Taking the critical value from Equation (5) yields Equation (6). The number of contacts between the wall and particles can be obtained by implementing a FISH function; the velocity-stress coefficient G can then be calculated using Equation (6).

$$G={\displaystyle \frac{\alpha A}{k_n^{\left(w\right)}{N}_c\Delta t}}$$

(6)

During the loading process, the upper and lower walls were set to move at a constant velocity to apply axial loading, with the servo system continuously adjusting the sidewall displacements to ensure the stability of confining pressure σ₃. Loading was terminated when the axial stress reached 90% of its peak value, which corresponds to the occurrence of pronounced shear failure. Finally, the axial peak stresses obtained under different confining pressure levels were fitted using the Mohr–Coulomb criterion to determine the shear strength parameters of the asphalt mixture, specifically cohesion c and internal friction angle φ. To verify the stability and reliability of the DEM simulations, each configuration was repeated three times using different random seeds for particle generation. The variations in cohesion and internal friction angle among repeated runs were within 5% across all confining pressures and gradations, indicating excellent repeatability and numerical stability of the model.

2.3.2. Model Parameter Setting

A linear contact stiffness model was adopted to model the mechanical interactions between coarse aggregate particles, with the normal stiffness k_n and shear stiffness k_s determined by the elastic modulus and Poisson’s ratio of the rock. Based on existing engineering practice and literature [32], the friction angles of limestone typically range from 35° to 40°, the Poisson’s ratio from 0.18 to 0.35, and the elastic modulus from 28 to 41 GPa. Considering both the deformation characteristics and computational efficiency of the model, a friction coefficient of 0.68, a Poisson’s ratio of 0.27, and an elastic modulus of 35 GPa were selected. The stiffness parameters of the limestone aggregates are summarized in

Table 3. Stiffness parameters of limestone aggregate.

Particles (mm)	D1	D2	D3	D4	D5	D6
kn (MPa)	82.6	124.4	249.4	397.3	511.0	612.5
ks (MPa)	32.5	49.0	98.2	156.4	201.2	241.1

. The particle size fractions of 2.36 mm, 2.36–4.75 mm, 4.75–9.5 mm, 9.5–13.2 mm, 13.2–16 mm, and 16–19 mm are denoted as D1, D2, D3, D4, D5, and D6, respectively.

A parallel bond model was employed to characterize the mechanical interactions between coarse aggregate and asphalt mortar units, as well as within the mortar itself [33,34]. This model integrates two key components to capture realistic mechanical behavior. The linear contact component accounts for the elastic response, while the parallel bond component represents asphalt cohesion by transmitting both forces and moments across the contact plane. Model parameters include the following: normal stiffness (k_n), shear stiffness (k_s), and friction coefficient (μ) for the linear component, and average bond stiffness ${\overline{E}}_c$, bond stiffness ratio ($\overline{k}$), bond normal strength (σ_c), bond shear strength (τ_c), and the parallel bond radius multiplier (PBRM) for the bonding component. The compressive strength and elastic modulus of the specimens were used to define the bond normal strength σ_c and bond stiffness ${\overline{E}}_c$, respectively, with both the bond strength ratio and bond stiffness ratio $\overline{k}$ set to 1.0. The PBRM determines the effective bonding area between contacting particles and thus directly influences the strength and stiffness of the bonded contacts. In this study, different PBRM values were assigned to D1–D6 particle groups to reflect the scale-dependent bonding characteristics between aggregates and asphalt mortar. The PBRM values were calibrated to reproduce the peak axial stress and elastic modulus observed in laboratory triaxial tests for the AC-16 mixture under 0 kPa confining pressure, ensuring consistency between simulated and experimental mechanical responses. The complete set of model parameters is summarized in

Table 4. Bonding model parameters.

Parameters	σc (MPa)	τc (MPa)	${\overline{\mathit{E}}}_{\mathit{c}}$ (GPa)	$\overline{\mathit{k}}$	Parallel Bond Radius Multiplier (PBRM) Between Mortar and Different Particles
					D1	D2	D3	D4	D5	D6
Values	0.68	0.68	0.525	1.0	0.50	0.61	0.76	0.81	0.82	0.80

.

3. Results and Discussion

3.1. Evaluation of Virtual Triaxial Test

3.1.1. Numerical Simulation Accuracy

To illustrate the applicability and feasibility of the proposed DEM-based modeling approach, a numerical example of AC-16 virtual specimens is presented. The proportions of each aggregate size fraction in these virtual specimens are summarized in

Table 5. Particle generation quantities of each component in AC-16 virtual specimens.

Parameters	Coarse Aggregates					Fine Aggregates	Asphalt Mortar	Voids
	D6	D5	D4	D3	D2	D1
Mass fraction (%)	4.75	8.55	15.19	20.89	13.30	32.29	5.03	0.00
Volume fraction (%)	4.27	7.68	13.65	18.69	12.00	30.29	9.32	4.10
Particle quantities	12	39	147	812	4195	36,196	11,135	4899

Note: volume fractions calculated assuming densities: limestone = 2.70 g/cm³, asphalt = 1.03 g/cm³, air = 0.

. Based on the methodology described above, both virtual triaxial tests and corresponding laboratory triaxial shear tests were conducted to evaluate the shear strength parameters of the asphalt mixture specimens. The experimental and numerical results are presented in

Table 6. The triaxial test results of AC-16 specimens.

σ3 (kPa)	Measured Values			Simulated Values			Relative Error
	σ1 (kPa)	φ (°)	c (kPa)	σ1 (kPa)	φ (°)	c (kPa)	δσ1	δc	δφ
0	601	26.38	194.98	622	27.63	189.58	3.49%	2.77%	4.55%
138	1058			1012			4.35%
276	1303			1375			5.53%

. As shown, the axial peak stress consistently increases with rising confining pressure, closely matching the laboratory observations. The relative errors between the simulated and measured values of axial peak stress σ₁, cohesion c, and internal friction angle φ are all minimal, confirming that the model can accurately capture the mechanical response of asphalt mixture specimens under varying confining pressures.

A comparative analysis of the axial stress–strain responses under different confining pressures was conducted, as illustrated in Figure 2. The virtual triaxial simulations showed excellent agreement with the laboratory results across the entire deformation process. Both the experimental and numerical curves exhibited a linear increase in axial stress during the elastic stage, reaching peak stress at the yield point, with relative errors of only 3.49%, 4.35%, and 5.53% at confining pressures of 0, 138, and 276 kPa, respectively. Thereafter, the stress–strain curves entered the strain-softening phase, where axial stress gradually decreased with increasing strain. This close correspondence confirms that the numerical model accurately reproduces the full mechanical behavior of asphalt mixtures—from initial elasticity to shear failure.

Figure 3 presents the shear dilation behavior of the AC-16 mixture obtained from both laboratory and numerical triaxial tests. At the initial loading stage, slight discrepancies are observed between the two curves; however, their overall evolution follows a consistent pattern. The volumetric strain first decreases, indicating specimen contraction under low axial strain, and subsequently increases once the critical strain threshold is exceeded—reflecting the onset of dilation. The transition points occur at approximately 1.5% and 1.9% axial strain for the virtual and laboratory tests, respectively, which are in close agreement and confirm that the DEM-based model successfully captures the contraction–dilation mechanism observed experimentally.

The minor deviations in the early phase can be attributed to inherent simplifications in the numerical model, particularly the idealized representation of particle geometry and contact laws, which cannot fully reproduce the complex microstructural and environmental heterogeneity present in physical specimens [34]. Nevertheless, the strong correspondence between simulated and measured responses across the entire loading process demonstrates that the model can reliably reproduce the elastic stage, yield behavior, and post-peak softening. These findings validate the model’s robustness and applicability for accurately characterizing the shear mechanical behavior of asphalt mixtures under confining stress conditions.

3.1.2. Numerical Simulation Applicability

The gradation of asphalt mixtures exerts a significant influence on their internal skeleton structure, asphalt content, and consequently, their shear performance. To evaluate the applicability of the proposed virtual triaxial testing method across different gradations, two representative mixtures, SMA-16 and SUP-16, were analyzed through both numerical simulations and laboratory tests. The mineral aggregate gradations are listed in Table 2, the numerical model parameters are summarized in

Table 7. Model parameters of SMA-16 and SUP-16 specimens.

Mix Type	Parameters	Coarse Aggregates					Fine Aggregates	Asphalt Mortar	Voids
		D6	D5	D4	D3	D2	D1
SMA-16	Mass fraction (%)	4.73	18.91	18.91	27.42	6.15	18.44	5.44	0.00
	Volume fraction (%)	4.19	16.76	16.76	24.20	5.47	17.06	11.96	3.60
	Particle quantities	12	85	180	1051	1913	20,389	14,287	4301
	PBRM value	0.80	0.84	0.81	0.76	0.60	0.48
SUP-16	Mass fraction (%)	4.77	11.92	17.64	21.93	10.49	28.60	4.67	0.00
	Volume fraction (%)	4.19	10.47	15.50	19.19	9.26	26.25	11.14	4.00
	Particle quantities	12	53	166	833	3237	31,365	13,307	4779
	PBRM value	0.80	0.84	0.81	0.76	0.61	0.50

Note: volume fractions calculated assuming densities: limestone = 2.70 g/cm³, asphalt = 1.03 g/cm³, air = 0.

, and the comparison between experimental and simulated results is presented in

Table 8. The triaxial test results of SMA-16 and SUP-16 specimens.

Mix Type	σ3 (kPa)	Measured Values			Simulated Values			Relative Error
		σ1 (kPa)	φ (°)	c (kPa)	σ1 (kPa)	φ (°)	c (kPa)	δσ1	δc	δφ
SMA-16	0	708	33.92	178.25	731	33.66	182.25	3.25%	0.77%	2.24%
	138	1089			1078			1.01%
	276	1670			1676			0.36%
SUP-16	0	676	28.46	195.52	693	30.13	191.05	2.51%	5.86%	2.29%
	138	1011			1028			1.68%
	276	1451			1518			4.62%

.

To further validate the DEM model’s applicability across gradations, results from Table 6 (AC-16) and Table 8 (SMA-16, SUP-16) reveal pronounced differences in shear strength among the mixtures. The SMA-16 mixture, with its gap-graded design, forms a stable, dense skeleton structure characterized by well-balanced coarse and fine aggregates. This gradation enhances both aggregate interlocking and asphalt mortar bonding, resulting in the highest shear strength and internal friction angle, particularly under increasing confining pressures [35]. In contrast, SUP-16 and AC-16 mixtures employ continuous gradations, resulting in dense-suspension structures with comparatively lower shear strength. Nevertheless, SUP-16 exhibits a more uniform gradation and better coordination between coarse and fine aggregates than AC-16, leading to stronger frictional and cohesive forces, a more stable skeleton framework, and ultimately higher shear strength [36].

These findings confirm that the virtual triaxial testing method effectively distinguishes the effects of gradation on the shear strength of asphalt mixtures. Moreover, the close agreement between simulated and measured results—with relative errors remaining within acceptable limits—demonstrates that the proposed modeling approach and parameter selection are appropriate for different gradation types. Overall, the DEM-based micromechanical contact model accurately captures the shear failure behavior of asphalt mixtures under triaxial loading, as evidenced by its strong consistency with laboratory experiments [26,27].

3.2. Influence of Aggragates Characteristics on Shear Tests

3.2.1. Radial Aggregate Distribution

During the molding of asphalt mixtures, segregation can occur due to variations in processing parameters and material composition, which can lead to nonuniform aggregate distributions in both the longitudinal and radial directions. To examine the influence of aggregate distribution on the shear performance of asphalt mixtures, the regional partitioning scheme illustrated in Figure 4 was adopted in this study.

Based on the partitioning scheme shown in Figure 4a, three outer-to-inner aggregate volume ratios (6:4, 5:5, and 4:6) were designed, and DEM simulations were performed using the AC-16 gradation. The DEM simulation results are summarized in Figure 5. As shown, a uniform aggregate distribution (5:5 ratio) achieves the highest cohesion and internal friction angle. Quantitatively, the 5:5 configuration exhibited approximately 21.3% higher cohesion and 6.2% greater internal friction angle than the 6:4 case, and 11.0% and 4.6% higher, respectively, compared with the 4:6 case. This finding indicates that uniform radial aggregate distribution improves particle contact quality and mechanical interactions, thereby enhancing shear resistance. Conversely, when the inner region dominates (6:4 ratio), the increased density within the inner region is counteracted by a sparse outer-region distribution, weakening overall structural integrity and reducing both cohesion and internal friction angle [37]. Similarly, when the outer region predominates (4:6 ratio), the insufficient skeleton strength in the inner region compromises structural support, disrupts the continuity of stress transmission paths, and further diminishes shear performance.

3.2.2. Longitudinal Aggregate Distribution

To evaluate the impact of longitudinal aggregate distribution on the shear performance of asphalt mixture numerical models, the longitudinal regional partitioning approach illustrated in Figure 4b was adopted. Aggregate volume ratios between regions A and B were set to 6:4, 5:5, and 4:6, and simulations were performed using the AC-16 gradation. The results are presented in Figure 6. As shown, the model with an even aggregate distribution (5:5) exhibited the highest shear strength parameters across all configurations. Specifically, the 5:5 model achieved about 7.0% higher cohesion and 4.6% greater internal friction angle than the 6:4 case, and 14.2% and 7.4% higher, respectively, than the 4:6 configuration. This improvement can be attributed to the uniform distribution enhancing the skeleton’s load-bearing capacity, reducing local stress concentrations, and promoting a stable shear failure mode [12]. When aggregates were enriched in the upper region (6:4), the shear strength was slightly lower than that of the uniform distribution model. Upper-region enrichment weakened the lower region, where stress concentrated in void-rich zones during shear loading, triggering localized failures. Nevertheless, the strong skeleton support in the upper region maintained moderate shear resistance. In contrast, the model with lower-region enrichment (4:6) showed a further reduction in shear strength. Although the lower region provided higher load-bearing capacity, insufficient coarse aggregates in the upper region facilitated shear slippage, ultimately leading to a significant decline in overall shear performance [38].

These radial and longitudinal distribution findings (Section 3.2.1 and Section 3.2.2) imply that insufficient compaction leading to inner-region aggregate depletion (6:4 ratio) may induce premature rutting even if overall air-void content meets specifications. Field compaction protocols should prioritize homogenization over mere density targets.

3.2.3. Aggregate Surface Texture

The surface texture of aggregates is a key factor influencing their frictional properties. To assess its impact on the shear performance of asphalt mixture, simulations were performed on the AC-16 model with varying friction coefficient parameters, as shown in Figure 7. The results demonstrate that an increase in the friction coefficient systematically enhances the shear strength parameters [25]. In particular, for every 0.1 increase in the friction coefficient, cohesion, and internal friction angle rise by 7.2% and 5.6%, respectively. This trend arises from the greater sliding resistance between particles at higher friction levels, which enhances interparticle interlocking and facilitates the development of a more stable skeleton structure [1].

3.3. Influence of Void Characteristics on Shear Tests

3.3.1. Longitudinal Void Distribution

Masad et al. reported that voids in asphalt mixture specimens exhibit a “∑-shaped” longitudinal distribution pattern [39]. To investigate the effect of void distribution on shear performance, the longitudinal partitioning scheme shown in Figure 8 was employed. Based on the measured 4.1% air-void content of AC-16 specimens, four numerical models were developed with void ratios in regions A: B: C set to 5.3%: 1.7%: 5.3%, 4.7%: 2.9%: 4.7%, 4.1%: 4.1%: 4.1%, and 2.5%: 4.1%: 5.7%, which were designated as A1, A2, A3, and A4, respectively.

The simulation test results are presented in Figure 9, which shows that longitudinal void nonuniformity exerts a pronounced impact on the shear performance of asphalt mixtures. The specimen with a uniform void distribution (A3) exhibited the highest cohesion and internal friction angle, indicating that such uniformity facilitates more effective stress transmission and maximizes shear resistance. In contrast, all nonuniform models (A1, A2, A4) demonstrated reduced shear performance, with the A4 model (characterized by a higher void content in the top region) showing the lowest cohesion and internal friction angle [40]. This reduction can be attributed to the following mechanism: uniform void distribution enhances interparticle contact and stabilizes stress transmission paths, whereas nonuniform distributions create localized high-void zones that either induce stress concentrations or disrupt stress transfer, ultimately diminishing the mixture’s overall shear capacity [41].

3.3.2. Air-Void Content

Numerical models with air-void contents of 2%, 4%, and 6% were developed to examine the influence of air-void content on the shear performance of asphalt mixtures. As shown in Figure 10, the results reveal a pronounced decline in both cohesion and internal friction angle as air-void content increases. Specifically, under a confining pressure of 138 kPa, each 2% increase in air-void content resulted in an average reduction of 11.9% in shear strength, clearly demonstrating that elevated air-void levels substantially degrade the mixture’s shear resistance. This deterioration can be attributed to two primary mechanisms: (i) higher air-void content reduces the effective contact area and bonding between particles, thereby weakening the overall structural integrity, and (ii) the diminished interlocking effect of the aggregate skeleton lowers internal friction, which further compromises shear strength [18].

3.3.3. Air-Void Size

Numerical models with void sizes of 1.86 mm, 2.36 mm, and a continuous range of 1–4 mm were developed to investigate the influence of void size on the shear performance of asphalt mixtures. The 1–4 mm void range was selected based on CT-scan observations, which indicate that most interconnected air voids in dense-graded asphalt mixtures fall within this interval. As shown in Figure 11, the model with a 1–4 mm void range exhibited the highest axial peak stress, cohesion, and internal friction angle, representing an approximate 2.5–47.8% increase in shear performance parameters relative to the models with 1.86 mm and 2.36 mm voids—it thus demonstrated optimal shear performance. These results indicate that a single void size can lead to localized stress concentrations and uneven asphalt bonding, whereas a well-distributed void size range optimizes the interlocking structure of the aggregate skeleton, enhances interparticle mechanical interactions, and thereby effectively improves the shear resistance of asphalt mixtures [28]. Mechanistically, the 1–4 mm void size range facilitates interlocking by providing an effective transition zone between the fine and coarse aggregate skeletons. Voids smaller than 1 mm tend to be filled with asphalt mastic, limiting aggregate movement and leading to localized stiffness heterogeneity, whereas voids larger than 4 mm reduce particle confinement and weaken force chains. In contrast, a continuous 1–4 mm range allows partial interpenetration of neighboring aggregate surfaces and accommodates fine particles that stabilize contact bridges, forming a more continuous and efficient load-transfer network. This intermediate void structure enhances particle interlock density, improves stress diffusion, and mitigates localized stress concentrations, thereby explaining the observed improvement in shear strength and internal friction angle. These micromechanical interactions demonstrate that void size heterogeneity within this range promotes an optimized aggregate skeleton with superior shear resistance. While our models suggest optimal performance with a 1–4 mm void size range, this may be partially confounded by the associated gradation characteristics. Future work should isolate void size variability by fixing gradation and artificially introducing controlled void size distributions.

4. Conclusions

This study developed a DEM-based virtual triaxial shear testing method for asphalt mixtures and investigated the effects of aggregate and void characteristics on their shear behavior. The main conclusions are as follows:

(1)

The DEM model accurately reproduced the shear response of asphalt mixtures under different confining pressures and gradations, showing strong agreement with laboratory results. In practical applications, this model can partially substitute physical triaxial tests to reduce specimen preparation time and enable rapid parametric analysis, thereby supporting efficient pre-evaluation of mixture performance during mix design and compaction standard development.

(2)

Aggregate characteristics exert a dominant influence on shear resistance. A uniform radial or longitudinal aggregate distribution (outer-to-inner or upper-to-lower ratio of 5:5) increased cohesion by 21.3–27.6% and internal friction angle by 4.6–7.4% compared with nonuniform configurations (6:4 or 4:6 ratios), indicating that field compaction standards should emphasize aggregate homogeneity rather than density alone. Moreover, each 0.1 increase in aggregate friction coefficient enhanced cohesion by 7.2% and internal friction angle by 5.6%, suggesting that the use of high-friction, angular aggregates should be prioritized in mix design to improve skeleton stability instead of relying solely on binder modification.

(3)

Void characteristics also play a critical role in shear performance. A uniform longitudinal void distribution optimized stress transmission and particle contact, whereas nonuniform distributions—especially those with higher voids in the upper region—caused stress concentrations and reduced shear strength. Increasing air-void content led to a distinct loss of shear resistance, with each 2% increase in air-void content resulting in an average 11.9% reduction under 138 kPa confinement. Furthermore, a continuous void size range of 1–4 mm improved axial peak stress, cohesion, and internal friction angle by 2.5–47.8% compared with single-sized voids, owing to enhanced particle interlocking and stress diffusion. These results suggest that both void content and size distribution should be included as quantitative indicators in laboratory mix design and field compaction quality control.

Overall, this study elucidates the micromechanical mechanisms through which aggregate and void characteristics govern the shear resistance of asphalt mixtures and provides a scientific basis for refining mix design and compaction standards. The present DEM model simplifies certain aspects—such as temperature effects, binder viscoelasticity, aggregate angularity, and parameter calibration range—through the use of calibrated contact parameters and spherical particles. While this simplification allows efficient computation, it may limit the full representation of viscoelastic and angular effects. Future work will incorporate temperature-sensitive contact laws, viscoelastic binder elements, non-spherical particles, and systematic sensitivity analyses to further enhance model realism and predictive accuracy.