Machine learning for predicting and optimizing energy absorption characteristics of hollow variable cross-section lattice metamaterials

Enhancing the energy absorption performance of mechanical metamaterials while reducing their weight has long been a sought-after topic in the field of impact protection. Lattice structures with hollow struts demonstrate superior energy absorption performance compared to those with solid struts due to the enhancement of the fully plastic bending moment. However, the energy absorption capacity can be further improved by optimizing the geometric configuration to adjust the distribution of the fully plastic bending moment. It is imperative to investigate the design guidelines for the hollow variable cross-section body-centered cubic (HVCB) lattice to enhance its energy absorption. In this work, a new database containing geometric and energy absorption information of 10,000 HVCB lattices was constructed to train the convolutional neural network models which can predict the energy absorption of the HVCB lattices. By combining the convolutional neural network models with the genetic algorithm, an optimization flow can be established to refine the structural configuration of the HVCB lattice, enabling the first systematic exploration of the relationship between complex geometric configurations and nonlinear mechanical properties. The strut shape of the HVCB lattice forms a concave curve when it achieves the highest specific energy absorption (SEA) under limited peak force or mass. However, the energy absorption capacity diminishes significantly when the strut shape of the HVCB lattice resembles a spindle. The energy absorption mechanism of the HVCB lattice with the highest SEA mainly attributes to enhanced compression force and plastic deformation regions. The optimization approach presented in this study efficiently improves the SEA of the HVCB lattice with the highest SEA being up to 2.59 times greater than that of the hollow uniform cross-section BCC lattice with the same mass.