|About this Abstract
||2022 TMS Annual Meeting & Exhibition
||Machine Learning of Symbolic Expressions to Model Dispersion Curves in Metamaterials
||Hongsup Oh, Sharat Paul, Alberdi Ryan, Robbins Joshua, Pai Wang, Jacob Dean Hochhalter
|On-Site Speaker (Planned)
Metamaterials can be tuned to prevent the propagation of a bandwidth of frequencies, referred to as a bandgap. This characteristic presents a unique opportunity to engineer materials and structures for wave control and vibration isolation. Research to date has largely consisted of the design of custom 3D metamaterial structures to tune the band gap. Here, we simulate a parameterization of metamaterial structures to form a relative band gap data training data set. Subsequently, genetic programming for symbolic regression (GPSR) is employed for the development of an interpretable machine learning approach to predict the relative band gap as a function the parameterized metamaterial geometry. This interpretable GPSR model essentially homogenizes the lower length-scale metamaterial behavior as an analytic expression that is natural for inclusion into existing engineering workflows. As a demonstration, an expression parser is implemented to automate the inclusion of GPSR-produced models into the PLATO topology optimization code.
||Machine Learning, Computational Materials Science & Engineering,