Digital Twin of Road Surfaces Rebuilds Asphalt Texture Particle by Particle

Every road surface tells a story in millimeters. The tiny peaks and valleys that make asphalt feel rough to a fingertip are the same features that determine whether a tire grips the pavement in the rain, how efficiently water drains away, and how much noise a highway generates. Yet for decades, engineers have designed these crucial surfaces largely by trial and error, mixing aggregates according to empirical rules and waiting for laboratory samples to reveal how the texture actually turns out. A new study published in Case Studies in Construction Materials changes that equation, presenting a digital framework that can reconstruct a three-dimensional pavement surface texture on a computer before a single stone is ever laid.

The research team, led by Haitao Ge of Chang’an University together with collaborators in France, built their framework around a deceptively simple idea: pavement texture is not an abstract surface property but the direct geometric fingerprint of how individual aggregate particles pack together near the top of an asphalt layer. If you can digitally reproduce the shape, size distribution, and arrangement of those particles, and then account for the smoothing role of the asphalt binder that glues them together, you should be able to predict the surface that a paving machine will produce. The challenge was doing so with enough physical realism to match what laser scanners actually measure on real roads.

To establish a trustworthy benchmark, the researchers first scanned real pavement surfaces using a high-resolution three-dimensional laser scanner, the Freescan UE11, which operates by laser triangulation and achieves a measurement accuracy of 0.02 millimeters while capturing up to 1,020,000 points per second. From the dense point clouds, they extracted representative 90 by 90 millimeter regions and processed them through noise filtering, outlier removal, and plane detrending, fitting and subtracting a least-squares plane so that only texture-related roughness remained. The cleaned data were interpolated onto a regular grid with a fine 0.1 millimeter resolution, producing continuous height fields suitable for rigorous statistical analysis.

Two widely used asphalt mixtures served as the test cases: a dense-graded AC-13 and a stone matrix asphalt SMA-13. The measured surfaces revealed clear and quantifiable differences. SMA-13 showed consistently higher amplitude parameters, with its mean profile depth, a standard macrotexture indicator, exceeding that of AC-13 by roughly 47 percent. Both mixtures displayed negative skewness, indicating valley-dominated surfaces, but SMA-13’s less negative skewness and lower kurtosis revealed a more balanced distribution of peaks and valleys created by large protruding aggregates. In contrast, AC-13 presented a smoother, more continuous surface with sharper, more concentrated asperities, reflecting its denser and more uniformly packed gradation.

The heart of the digital reconstruction lies in how the aggregates themselves were represented. The team laser-scanned approximately 500 real aggregate particles with sieve sizes from 4.75 to 26.5 millimeters, converting each into a triangular surface mesh. From these scans they quantified elongation, flatness, and sphericity, finding that all three descriptors followed normal distributions with high goodness of fit, which allowed them to stochastically generate a particle library of realistic polyhedral shapes. Because full-resolution meshes would be computationally prohibitive, they systematically simplified the geometry, showing that reducing particles to about 60 to 65 faces preserved morphological indices within deviations of roughly 4 percent while dramatically cutting simulation cost.

For the simulation engine, the researchers chose the Contact Dynamics method rather than classical discrete element approaches. Conventional discrete element models rely on stiff elastic contact laws that demand artificially high stiffness parameters and tiny time steps, often causing numerical instability when irregular polyhedral particles collide. Contact Dynamics instead enforces impenetrability through non-smooth mechanics, using the Signorini condition for normal contact and the Coulomb friction law for tangential interaction, solved implicitly with a nonlinear Gauss-Seidel algorithm in the LMGC90 software. This formulation handles point, edge-face, and face-face contacts between angular particles without artificial stiffness, permitting larger time steps and mechanically consistent simulations of large assemblies of irregular grains.

The virtual specimens were assembled by inserting particles according to the target gradation curves of each mixture, with particles finer than 2.36 millimeters lumped into an implicit mastic phase to save computation. Gravitational deposition created a loose configuration in a 100 by 100 millimeter domain with periodic horizontal boundaries, mimicking an infinitely repeating material volume. Compaction then proceeded under a constant vertical force of 6 kilonewtons, equivalent to about 0.6 megapascals of compressive stress, superimposed with a 40-hertz harmonic vibration representative of real vibratory rollers. A reduced inter-particle friction coefficient of 0.1 accounted for the lubricating effect of hot bitumen, and compaction continued until mechanical stabilization at a specimen thickness of roughly 40 millimeters.

When the researchers extracted surface textures from these compacted digital specimens using a virtual scanning procedure that mimics laser triangulation, the raw results revealed a fundamental problem. The purely aggregate-controlled surfaces overestimated roughness dramatically, with mean profile depth nearly doubled for both mixtures, because the digital surface lacked the binder mastic that in reality fills deep valleys and bridges neighboring stones. To correct this, the team introduced a bridging algorithm applied directly to the height field: a maximum filter that locally elevates narrow valleys, a grayscale morphological closing operation that removes small isolated depressions, and optional Gaussian smoothing. Calibrated once, the same parameters worked for both mixtures, cutting the deviation in mean profile depth from 160.9 percent to 27.4 percent for AC-13 and from 119.1 percent to 18.2 percent for SMA-13, while deviations in the areal roughness parameters Sa and Sq fell by more than 80 percent.

Beyond validation, the framework opened a window onto the particle-scale mechanics of texture formation, and the findings are striking. By classifying particles into size fractions and tracking which ones reach the surface, the researchers found that larger aggregates mainly control surface exposure, while intermediate and small particles dominate the formation of the highest asperities. In AC-13, surface peaks arose from distributed contributions of the 2.36 to 4.75 millimeter and 4.75 to 9.5 millimeter fractions, which together accounted for nearly 98 percent of peak particles, producing a homogeneous texture. SMA-13 showed a decoupled, skeleton-driven pattern in which the 9.5 to 13.2 millimeter fraction was most exposed at the surface, yet the 4.75 to 9.5 millimeter fraction generated 56 percent of the peaks, explaining its broader, more heterogeneous height distribution.

The implications reach well beyond academic curiosity. Because the framework links mixture composition, gradation, and aggregate morphology directly to quantitatively validated surface texture, it transforms pavement design from descriptive measurement into predictive, simulation-driven engineering. Engineers could someday screen candidate mixture designs virtually, optimizing skid resistance and drainage before committing materials to a laboratory or a road. The authors caution that the bridging parameters are numerical rather than intrinsic material properties and may require recalibration for different aggregate lithologies, binders, or interface conditions, and that the computationally demanding polyhedral simulations currently limit the approach to texture reconstruction and microscopic investigation. Even so, the study marks a decisive step toward a future in which the texture of a highway is designed on a computer as deliberately as the shape of an aircraft wing.

Subject of Research: Digital reconstruction of three-dimensional asphalt pavement surface texture using laser-scanned aggregates and the Contact Dynamics method

Article Title: A digital reconstruction framework for three-dimensional asphalt pavement surface texture using the Contact Dynamics method

Article References: Ge, H., Sha, A., Jiang, W., Xing, C., Quezada, J. C., Houerou, V. L., & Chazallon, C. (2026). A digital reconstruction framework for three-dimensional asphalt pavement surface texture using the Contact Dynamics method. Case Studies in Construction Materials, 25, Article e06469. https://doi.org/10.1016/j.cscm.2026.e06469

Image Credits: AI Generated

DOI: 10.1016/j.cscm.2026.e06469

Keywords: asphalt pavement, surface texture, Contact Dynamics, discrete element method, aggregate morphology, 3D laser scanning, skid resistance, macrotexture, virtual compaction, pavement engineering, digital reconstruction, asphalt mastic