In a comprehensive experimental and numerical investigation of ductile damage evolution, materials researchers have determined that microscale void growth in an aluminum alloy is governed primarily by initial void volume, void aspect ratio, and early localized mechanical stress fields rather than macroscopic stress triaxiality or crystallographic grain orientation. Ductile fracture in metals unfolds through a sequential progression of void nucleation, void growth, and void coalescence, yet the specific microscopic mechanisms controlling individual cavity expansion have remained unresolved. By tracking two hundred and thirty-three distinct internal cavities across multiple tensile strain increments within an Al-2219 notched specimen, the research team demonstrated that classic analytical formulations fail to capture the marked growth heterogeneity observed at the microscale. Instead, explainable machine learning revealed that initial cavity dimensions dominate growth rate predictions, while localized mechanical stress and plastic strain states established at early plastic deformation exert a critical secondary influence.
Ductile fracture mechanics has historically relied upon continuum-level analytical models to predict internal cavity expansion inside a plastically deforming matrix. Foundational frameworks developed by Rice and Tracey, McClintock, Hancock and Mackenzie, and Gurson established that macroscopic stress states strongly dictate volumetric cavity growth. In these formulations, expanded by Huang, void growth rate is modeled as an exponential function of stress triaxiality, defined as the ratio of hydrostatic stress to von Mises equivalent stress. While these representations achieve strong agreement when compared with sample-averaged growth rates from X-ray computed tomography, they cannot account for the severe variability observed among individual voids. In materials such as pure iron, titanium, and aluminum alloys, local void growth rates within single specimens fluctuate by factors of six to ten under nominally uniform remote loading, demonstrating that homogenized macroscopic loading cannot describe the localized microstructural mechanisms driving failure.
To isolate the physical mechanisms governing heterogeneous growth, the researchers examined an Al-2219 alloy obtained as a square rod measuring 25.4 millimeters by 25.4 millimeters. The alloy composition comprised 5.8 to 6.8 percent copper, 0.2 to 0.4 percent manganese, 0.1 to 0.25 percent zirconium, 0.05 to 0.15 percent vanadium, 0.02 to 0.1 percent titanium, and trace iron, silicon, magnesium, and zinc, with aluminum as the balance. To facilitate diffraction contrast tomography, the material underwent specialized thermal and mechanical processing. The rod was annealed at 535 degrees Celsius for two hours to enhance ductility, cold-rolled to a 25 percent reduction in area, annealed again at 535 degrees Celsius for five hours, water quenched, and artificially aged at 190 degrees Celsius for eighteen hours. This procedure produced a fully recrystallized T6 microstructure with equiaxed grains averaging sixty micrometers in equivalent sphere diameter.
A notched tensile specimen was extracted parallel to the rolling direction, featuring an 800-micrometer gauge length, a 900-micrometer gauge width, and a notch radius of 450 micrometers. This notched geometry established an initial nominal stress triaxiality of 0.6, generating an elevated multi-axial stress state to assess whether crystallographic effects intensify under higher triaxiality. Before tensile deformation, the sample’s internal three-dimensional grain structure was characterized non-destructively using laboratory-based diffraction contrast tomography at a 5-micrometer voxel resolution. The grain morphology and orientations were reconstructed using GrainMapper3D software. Simultaneously, a corresponding X-ray absorption dataset was acquired across the same volume at a 3.3-micrometer voxel resolution, enabling semantic segmentation of preexisting voids and particles using Recon3D software. This multimodal imaging sequence provided an exhaustive baseline description of the microstructure prior to mechanical loading.
In-situ tensile loading was performed using a Deben CT5000 stage mounted within a Zeiss Xradia 620 Versa X-ray microscope at a crosshead displacement rate of 0.5 micrometers per second, corresponding to an engineering strain rate of 5.6 times ten to the negative fourth per second. The tensile test was interrupted at eight progressive strain increments to acquire three-dimensional computed tomography volumes of the gauge section at a 1.5-micrometer voxel resolution, segmented using Dragonfly software. Tracking total void counts and mean void volumes across loading steps allowed researchers to delineate active damage regimes. Void nucleation was detected by 4 percent strain, whereas void coalescence initiated after 11 percent strain, characterized by a declining cavity count coupled with an accelerated rise in average void volume. This timeline established an 11 percent strain upper boundary for analyzing isolated void growth.
Individual cavity kinematics were analyzed inside a 500-micrometer diameter spherical region positioned at the specimen’s center, ensuring a uniform macroscopic stress state. An automated tracking algorithm matched voids across 11 percent, 7 percent, 4 percent, and 0 percent strain increments based on center-of-mass coordinates and volume consistency. To avoid spatial resolution artifacts and image noise, the algorithm filtered out cavities smaller than eight voxels, setting a volume cutoff of 27 cubic micrometers. In total, two hundred and thirty-three individual voids were successfully tracked across all four deformation stages. Validation was confirmed by randomly selecting twenty cavities for manual inspection, which demonstrated complete one hundred percent agreement between manual and automated tracking results. Extracted morphological parameters for every cavity included volume, aspect ratio, sphericity, and the ratio of longitudinal radius to transverse radius relative to the tensile direction.
To evaluate localized mechanical fields inaccessible through direct imaging, the team constructed a crystal plasticity finite element model. The diffraction contrast and absorption tomography datasets were aligned using an affine coordinate transformation to map reconstructed grains and cavities onto a hexahedral finite element mesh comprising 4.1 million elements, capturing 4,369 individual crystallographic grains within the gauge volume. The crystal plasticity constitutive formulation integrated power-law slip kinetics with dislocation density-based hardening across twelve face-centered cubic slip systems, calibrated against empirical stress-strain measurements. The simulation modeled displacement-controlled uniaxial tension, extracting localized state variables including von Mises equivalent stress, equivalent plastic strain, normal stress, shear stress, Lode parameters, and stress triaxiality tensors at true applied strain levels of 0.04, 0.07, 0.11, and 0.15 around each cavity element.
Direct comparisons between measured growth rates and isolated geometric descriptors revealed that void volume exhibited the most noticeable relationship, displaying a clear negative correlation where smaller voids grew faster than larger cavities. This empirical observation contrasts sharply with classic analytical formulations and finite element predictions that assume larger voids grow faster due to magnified stress concentrations. Furthermore, smaller cavities demonstrated greater stochastic variability in their growth kinetics across deformation steps. Other geometric parameters, including cavity aspect ratio, sphericity, and longitudinal-to-transverse radius ratios, showed no strong linear correlation with void growth rate. Investigating neighborhood interactions across interaction distances of 10, 25, 50, and 100 micrometers revealed that void count and neighboring cavity volume had negligible influence, yielding only a weak positive trend between growth rate and the ratio of maximum neighbor volume to void volume.
The researchers subsequently examined crystallographic orientation effects, testing computational hypotheses that grain orientation strongly influences void expansion at elevated triaxiality. Examining grain orientations within 5 micrometers of cavity surfaces on inverse pole figures revealed no correlation with void growth rates, initial void volumes, Schmid factors, Taylor factors, or active slip systems. Similarly, extracting average values of localized stress and strain within 5 micrometers of tracked voids revealed no statistically significant correlation with void growth kinetics. To determine whether localized loading variability accounted for growth differences, the team evaluated the standard deviation of stress and strain fields within 5 micrometers of each cavity, representing microscale stress heterogeneity. Neither average local stress states nor localized stress heterogeneity correlated with void growth rates, indicating that individual linear descriptors cannot explain microscale void expansion.
To discover complex nonlinear interactions among microstructural, geometric, and loading parameters, the team employed a random forest regression model using scikit-learn. The regression framework analyzed six hundred and ninety-nine data points derived from all two hundred and thirty-three tracked voids across 4 percent, 7 percent, and 11 percent strain increments, using an 80/20 train-test split. To prevent redundancy and avoid splitting importance across collinear inputs, initial void volume and aspect ratio were selected as baseline geometric features. Incorporating crystallographic Euler angles did not improve predictive performance, whereas adding local mechanical state variables yielded notable improvements. Hyperparameters were tuned via grid search, restricting maximum tree depth to seven to prevent overfitting on the experimental dataset. Evaluated across five-fold cross-validation, the model demonstrated consistent predictive stability, producing an average coefficient of determination of 0.610 plus or minus 0.05.
The best-performing model achieved a coefficient of determination of 0.686 on the independent test set and 0.760 on the training set, indicating strong generalization. Shapley Additive Explanations identified initial void volume as the primary predictive feature, followed by void aspect ratio, local normal stress at 4 percent strain, local equivalent plastic strain at 4 percent strain, and local stress triaxiality at 4 percent strain. Local mechanical fields from the 4 percent strain increment provided the greatest predictive utility, likely because early plastic deformation exhibits stable stress distributions before unmodeled void evolution causes finite element fields to diverge from physical reality. From a micromechanical perspective, tracked void diameters exceeded 2.8 micrometers, well above the 100-nanometer threshold where diffusion-driven growth operates, indicating that plasticity-driven dislocation absorption governs expansion. Small voids may intersect dense dislocation boundaries, causing rapid, stochastic growth, whereas larger cavities average over microstructural heterogeneity, conforming more closely to homogenized continuum predictions.
Subject of Research: Mechanisms and microstructural factors driving heterogeneous void growth in ductile aluminum alloy 2219
Article Title: Identification of mechanisms driving heterogeneous void growth in ductile aluminum
Article References: Ticknor, C., Lim, H., & Noell, P. (2026). Identification of mechanisms driving heterogeneous void growth in ductile aluminum. Journal of Materials Science: Materials Theory, 10(1), Article 2. https://doi.org/10.1186/s41313-026-00073-9
Image Credits: AI Generated
DOI: 10.1186/s41313-026-00073-9
Keywords: ductile fracture, void growth, Al-2219 alloy, diffraction contrast tomography, in-situ X-ray computed tomography, crystal plasticity finite element, random forest regression, SHAP analysis, stress triaxiality, dislocation absorption, void coalescence, equivalent plastic strain, Rice-Tracey model, microstructural heterogeneity, microvoid kinematics, Deben load frame, Dragonfly segmentation
News Source: Denise Maddox. (November 21, 2026). Initial Void Dimensions Best Predict Heterogeneous Void Growth in Ductile Aluminum. Scienmag.



