Fig. 3.
Inverse-designed truss metamaterials: generalization beyond the training dataset and functional grading. (A) Inverse design with queries from outside the generated dataset and the corresponding predicted lattice candidates including their elastic surfaces. Two examples are chosen from each of the crystallography-inspired catalogs of periodic truss architectures of ref. 14 and the bovine femoral bone specimens of ref. 16, whose 3D anisotropic stiffness is used as a target for the inverse design. The performance of the inverse design framework is illustrated by projections of the elastic surfaces of the target stiffness and the inverse prediction onto the --, and - planes. Two unique designs (denoted by A and B) are predicted (Pred.) for each target to showcase the variational sampling of the inverse design framework. Microtomography images are adapted with permission from Springer Nature: ref. 16. (B) A functionally graded architecture is generated by interpolating between the two predictions for the queried bone samples, locally closely matching the given anisotropic stiffness (details on the predictions and interpolation scheme are given in SI Appendix, sections 4 and 5). Struts are shown twice as thick for improved visibility.