Multi-scale structural design considering cellular connectivity using machine learning approaches

Abstract

Accompanied by the fast evolution of additive manufacturing, multi-scale porous structures are gaining ever-increasing popularity in high-performance structure design. Given the connectivity requirement imposed on neighbouring cellular microstructures for their successful printing, we propose in the current work a criterion for connectivity evaluation based on Dijkstra’s shortest path algorithm, and a unit cell generation model is established with the aid of the Generative Adversarial Network (GAN) approach. A family of 9 lattice units satisfying a prescribed connectivity condition is subsequently optimized under various load conditions. Lastly, a multi-scale structural optimization design approach is developed under the neural network framework, and the best combination of the a priori optimized lattice units is found. The effectiveness of the proposed protocol is verified on a series of numerical examples considering structural stiffness/toughness.