Accelerating the Mechanical Characterization of Hierarchical Materials via Machine Learning

Abstract

Hierarchical materials, with unique multiscale mechanics and potential for lightweightyet robust performance, present an immense level of promise in fields ranging from aerospace to biomedical engineering. However, the sheer number of possible combinations of material and topology across length scales can render exhaustive experimental and computational evaluations prohibitively time-consuming. This body of work attempts to overcome this challenge by integrating additive manufacturing of both periodic and stochastic porous structures, experimental validation through mechanical testing and microcomputed tomography, and a machine learning pipeline that translates microstructural descriptors into bulk mechanical properties. First, a series of 3D-printed periodic gyroid lattices spanning a range of porosities and photopolymer materials was examined to quantify properties such as stiffness, strength, and failure modes. Accounting for process-related variability, the modeling framework was expanded to include low- and high- porosity manufacturing artifacts, thus improving prediction accuracy. To further generalize to inherently irregular porous architectures, descriptors based on Minkowski functionals were introduced, enabling a reduction in the complexity of high-resolution scans while retaining critical morphological features. Results demonstrated that simple regressors could be trained using limited experimental data yet still accurately predict mechanical behavior of newly introduced periodic lattice–material combinations, that accounting for manufacturing variability improved prediction robustness, and that Minkowski functionals offered an effective approach for representing and predicting behavior of inherently irregular porous architectures. Overall, the developed tools reduced the computational and experimental resources required for mechanical characterization, paving the way toward optimized design and analysis of hierarchical materials for a wide range of applications.