Metamaterials are engineered materials that can show extraordinary properties which are not readily attainable with materials directly available in nature. Metamaterials are particularly promising for applications in dynamics, where the control of elastic energy propagation, dispersion, and attenuation are essential. The ability to control sound and vibrations in metamaterials arises from their ability to support the formation of forbidden frequency bands of propagation (referred to as band gaps) in their dispersion relation, where waves cannot propagate. Such unique properties have prompted engineers to explore their use in different applications like energy harvesting, seismic protection, frequency filtering, and wave guiding. Metamaterials are commonly designed through empirical trial-and-error, intuition, and computationally expensive topology optimizations. More recently, machine learning algorithms have shown significant promise in metamaterials’ design, but a lack of interpretability has limited the scope of these “black box” approaches.
In this project, we aim to design metamaterials through interpretable machine learning algorithms that illuminate structure/property relationships for a variety of unit cell classes and are also superior to the existing black box approaches. Moreover, we are going to create large Findable, Accessible, Interoperable, and Reusable (FAIR) scientific datasets covering a vast space of geometric configurations, which can be contributed to, and used by, others in the research community for further algorithmic refinement.
Active Researcher on the Project:
Rayehe Karimi Mahabadi, Han Zhang