Machine learning–based estimation method for the mechanical response of composite cellular structures

Cellular materials, including porous materials, are widely utilized in engineered and natural systems because their mechanical performance undergoes changes such as compressive deformation and energy absorption under impact loading. The mechanical response of these materials is notably influenced by their inherent cellular structure, specifically the geometric arrangement pattern. Nonuniform arrangements can result in significant variations in mechanical performance, posing challenges for material selection and the geometrical design of cellular structures. In this study, we established a machine learning (ML)–based approach to design the geometric arrangement (architecture) of cellular materials, aiming to achieve improved mechanical performance under uniaxial compression. In particular, we investigated the peak force at the plateau region and the work of energy absorption until structural densification occurs. Various patterns of internal geometry were modeled using the finite element method, and uniaxial deformation behavior was simulated to generate the training data for the ML approach. A neural network was employed as the ML method, correlating the cellular geometric patterns with the mechanical performance, including force–displacement curves and the relationship between peak force and work during energy absorption. The results indicate that the proposed method can accurately predict the mechanical response of any given geometric pattern within the defined scope. Thus, this approach is valuable for discovering cellular structures that can achieve desired mechanical responses.