Session: 04-06-01: AI for Heterogeneous Materials Design, Discovery, and Manufacturing I

Paper Number: 174066

Unraveling the Mechanics of Damage Nucleation With Computer Vision 

Structural metals used in aerospace, defense, and military applications are frequently subjected to high strain-rate loading conditions, such as those produced by ballistic impacts, explosions, and space debris collisions. Under these extreme conditions, spallation becomes a primary mode of failure. Spallation involves a sequence of microstructural events, beginning with void nucleation, followed by void growth, coalescence, and ultimately complete material separation. Accurately predicting the locations of void nucleation is critical for understanding and mitigating failure, but this remains a major challenge due to the highly heterogeneous and anisotropic nature of polycrystalline microstructures. Factors such as grain orientation, grain boundary character, and local stress states all interact in complex ways that are not easily captured by conventional physics-based or empirical models.

This work introduces a machine learning framework for predicting void nucleation sites in polycrystalline metals subjected to dynamic tensile loading. Synthetic datasets are generated using crystal plasticity simulations of shock-induced spall in statistically representative microstructures. These datasets include detailed spatial information such as grain orientation maps and grain boundary networks. A U-Net convolutional neural network is trained to take these microstructural fields as input and produce binary masks that highlight regions with high void nucleation probability. The model achieves strong performance across a range of microstructural configurations, with high pixel-level accuracy and reliable generalization to previously unseen microstructures.

Beyond accurate prediction, the proposed method provides interpretability by examining the intermediate feature maps within the U-Net. This analysis reveals which grains, orientations, and boundaries are most influential in determining nucleation sites. Notably, the model identifies physically meaningful patterns, such as increased nucleation propensity at triple junctions and in grains with orientations that promote high local stress concentrations. These insights offer a deeper understanding of the mechanisms underlying dynamic failure and can inform both experimental investigations and the development of more robust design criteria for dynamic loading environments.

An additional benefit of the method is its compatibility with physics-based models. The predicted binary masks can be used as input fields in phase field simulations, enabling a hybrid multiscale modeling approach that combines data-driven void nucleation with physically grounded damage evolution. This integration supports faster and more predictive simulations of spall behavior. To illustrate, we present results for a phase field damage model in which the predicted spall nucleation behavior is integrated into a massively parallel multiphysics solver. This demonstrates the ability and versatility of the method.  Overall, this framework establishes a new pathway for understanding and modeling early-stage dynamic failure, offering both predictive power and physical insight for the design of advanced structural materials.

Presenting Author: Brandon Runnels Iowa State University

Presenting Author Biography: Brandon Runnels, Ph.D. is an Associate Professor of Aerospace Engineering at Iowa State University. His research lies at the intersection of computational mechanics, materials science, and high-performance computing, with particular focus on mesoscale modeling of microstructure, grain boundaries, and energetic materials such as solid rocket propellants. Dr. Runnels earned his Ph.D. and M.S. in Mechanical Engineering from Caltech, and a B.S. with Highest Honors from New Mexico Tech. Prior to joining Iowa State, he was a faculty member at the University of Colorado Colorado Springs and held research positions at Los Alamos National Laboratory. He is the recipient of a National Science Foundation CAREER Award and has led numerous externally funded projects from agencies including ONR, AFRL, NSF, LANL, and LBNL. His work has appeared in top journals such as Acta Materialia, Journal of the Mechanics and Physics of Solids, and Computer Methods in Applied Mechanics and Engineering. Dr. Runnels is also an active member of the scientific community, serving as referee for over 30 journals and funding agencies.

Authors:

Abhijith Anantharanga Iowa State University
Brandon Runnels Iowa State University