Machine learning-optimized stochastic Voronoi lattices for enhanced mechanical performance
Michael Thompson, Hamidreza Yazdani Sarvestani, Ahmad Sohrabi Kashani, Elham Kiyani, Enzo Filippi, Derek Aranguren van Egmond, Meysam Rahmat, Behnam Ashrafi, Mikko Karttunen
Abstract
Lattice structures, traditionally composed of periodic networks of interconnected struts, offer an excellent balance of high strength and low density. However, their periodicity limits adaptability to complex and unpredictable loading conditions. Stochastic Voronoi lattices, characterized by their irregular, non-periodic geometry, provide a promising alternative with enhanced energy absorption and mechanical robustness. In this study, we present a machine learning (ML)-driven framework integrating finite element analysis (FEA), a multilayer perceptron (MLP) neural network, and three-dimensional (3D) printing to optimize Voronoi lattice structures for targeted mechanical properties. To systematically control structural disorder, we introduce Relaxation Iteration (RI), an ordering parameter inspired by Lloyd’s algorithm. Based on RI, we show that there is a range of RI ≈ 1500 − 2000 which gives enhanced mechanical performance for Voronoi lattices. The ML-FEA optimized Voronoi lattices demonstrate double the stiffness and four times the toughness compared to conventional periodic lattices. These findings underscore the potential of ML-driven design strategies in developing tailored architected materials for applications requiring high energy absorption and structural integrity, including aerospace, automotive crash protection, and biomedical implants.