Pentamode Structures Optimized by Machine Learning with Adaptive Sampling

Abstract

Pentamode structures, gain increasing interest as insulation or stealth material. The enhancements in computers and clusters make it possible to investigate those structures not only in theory but also with simulations. Their applicability to mechanical wave dampening is the main focus of the present work, which leads to a structure with good damping and enough strength as the goal. Therefore, a parametrized geometry based on the diamond lattice is examined within a design space. A factorial testing plan investigates the boundaries and gives first hints on the structure's behaviour under compressive and oscillatory loading and also reveals the necessity of a multi objective optimization. Feed‐forward neural networks are then trained to predict the material properties action and mass specific stiffness utilizing adaptive sampling in order to save time and computational cost. An optimization procedure to gain the structure with lowest mass, highest stiffness, and best damping capabilities, which means lowest action, is successfully implemented and yields the best compromise solution for an equally balanced optimization. This structure is then investigated by finite element simulations and confirms the optimization as well as the neural network training, thus being the best trade‐off of all optimization targets.