Designing 2D mechanical metamaterials with printability constraints using interpretable machine learning

Mechanical metamaterials present superior mechanical performance thanks to their designed microstructure. For instance, they can exhibit highly non-standard elastic behavior such as auxeticity or the formation of complete frequency band gaps. Recently, 3D printing has become a practical tool to realize such metamaterials that can be made out of multiple materials or even conveniently realized by printing a single material with voids within it. However, this latter approach introduces additional constraints that need to be considered, e.g., unit-cell connectivity.

We take elastic 2D metamaterials that exhibit band gaps as an example and discuss the application of interpretable machine learning to capture unit-cell features essential for opening band gaps within specified frequency ranges while also incorporating 3D printing considerations. Unlike conventional ‘black box’ machine-learning algorithms, interpretable machine-learning algorithms yield models that can be easily interpreted and checked. Thus, they offer an effective approach towards obtaining interpretable rules spanning the feasible design space (e.g., different unit-cell shapes and topologies) for complex problems, e.g, multi-classification problems or multi-objective problems involving multi-functional metamaterials.