Machine Learning-Based Biomimetic Optimization Models for Engineered Materials

The slime mold Physarum polycephalum develops networks that can solve high combinatorial problems and optimize material properties. This emergent behavior is heavily based on the complex microstructural tuning of its cytoplasm. In this study, we part from a phase-field model of the slime mold and propose a machine-learning strategy to optimize the microstructural response, hence capturing the optimization of engineered materials with embedded networks. By representing transport networks as composite mesh structures—analogous to biological organisms with distinct regions mimicking functionalities like nuclei and cellular matrices—we bridge the gap between biological phenomena and computational modeling of advanced materials. Using continuum equations, we simulate protoplasmic flux through these meshes, generating data that captures the essence of transport dynamics. A machine learning model is trained to predict the resulting pressure and flow fields given a composite mesh structure. The trained model rapidly approximates the physics and enables iterative optimization of the mesh topology to maximize material efficiency. This framework synergistically combines composite mesh modeling, biophysics simulations, and machine learning to discover new bio-inspired strategies for optimizing engineered material designs with embedded networks.