Machine Learning-Based Optimization of Biomimetic Hierarchical Porous Structures Inspired by the Sea Glass Sponge

In recent biological research, the unique architecture found in the sea glass sponge exhibits high yield strength and buckling resistibility during compressive testing. It is composed of a square-grid-like design with a second set of diagonal bracings in the skeletal system resulting in a chequerboard-like pattern. Mimicking this kind of structural arrangement can provide inspiration for innovating lightweight structures with high compressive properties. Such balance properties could help compensate for the weak mechanical properties of hierarchically porous structures without degrading their outstanding adjustability with density and thermal variation. However, the relationships between the various architectural features and mechanical properties of the sea glass sponge are not well understood. In this study, we utilize machine learning algorithms to elucidate the structure-property relationships of structures inspired by the sea glass sponge. Specifically, an active learning-based optimization approach is used to investigate the best spacing between the diagonal bracings and the square grid in the sponge structure. To assess the structure-property relationships, neural networks (NNs) are trained using massive datasets produced from finite element analysis. Genetic optimization is used to identify the crucial spacing that leads to high modulus of the hierarchical porous structure using rapid inference of trained NNs. Next, the final design is fabricated using 3D printing and validated by mechanical testing. This research makes use of advanced algorithms and additive manufacturing techniques to quickly and efficiently identify optimal biological designs for structural applications.