Generative Machine Learning Approach for Asymmetric Cellular Architectures with Enhanced Mechanical Properties

Cellular structures have gained remarkable attention in recent years due to their high strength-to-weight ratio, energy absorption, and heat transfer properties. A key feature of cellular structures employed in modern infrastructure and devices are symmetric configurations with repeating unit cells. This periodic design makes it more feasible to fabricate for various applications. However, this kind of repeating unit cell design often undergoes a sudden and catastrophic failure in the structure since all the unit cells will naturally fail simultaneously at a critical loading condition. In this research, we propose an elegant solution to achieve progressive failure by adjusting the diameter of each strut to create asymmetric cellular structures. We develop a simulation-based machine learning framework to achieve materials with extraordinary mechanical properties by properly tuning our design parameters. Additionally, genetic optimization is applied to improve the energy absorption capabilities of structures with a given number of unit cells and relative density. Lastly, our designs are validated using 3D printing and mechanical testing and more progressive failures are observed compared to typical architectures consisting of identical beam elements and gradient structures. This research proposes a novel design model to optimize the mechanical performance of cellular structures without the constraints of symmetry in order to achieve structures with more promising manufacturability and more damage-tolerant failure, greatly broadening the applications of cellular structures.