Connected multiscale generation of complete cloak architectures

Develop a full multiscale image of the Rayleigh-wave carpet cloak in which neighbouring microstructural cells connect and vary smoothly across the domain, using a generative model such as diffusion models to address the stated open problem.

Background

The paper designs a two-dimensional Rayleigh-wave carpet cloak by coupling a coordinate-conditioned neural field to a differentiable finite-element solver. The optimised cloak is subsequently projected onto manufacturable two-phase solid/void microstructures, with each macro-cell generated independently through inverse design, conditional diffusion, or nearest-neighbour selection.

Although the independently generated cells reproduce the target homogenised properties reasonably well, the authors note that the current generative model produces only one unit cell at a time. A complete manufacturable cloak requires a full multiscale architecture whose neighbouring cells connect and whose microstructural features vary smoothly across the domain. The paper identifies constructing such a globally coordinated representation as an open problem and suggests diffusion models as a possible approach.