New AI‐Assisted Approach for Expanding the Solution Space: Application to Lattice Structure Design

The design of architected materials with targeted mechanical properties remains a major challenge owing to the complex interplay among geometry, material composition, and effective performance. Among these materials, lattice structures offer significant potential due to their high design flexibility. However, exploration of their design space is difficult and is often restricted to finite catalogs of unit‐cell architectures.The design space can be expanded beyond a finite set of lattice structures by introducing intermediate geometries between existing architectures. Inverse design frameworks that generate such geometries typically rely on large databases, making them difficult to deploy when only a limited number of lattice structures are available. To address this limitation, this study proposes an interpretable reparameterization framework that transforms categorical geometric and material variables into structurally continuous descriptors. This approach enables continuous and controlled expansion of the design space, allowing the generation of intermediate lattice configurations. Coupled with XGBoost‐based surrogate models within a multi‐objective optimization framework, the proposed method identifies lattice architectures with improved trade‐offs among targeted mechanical properties. It also provides insights into the underlying structural mechanisms governing these improvements.