| Abstract Scope |
Lattice structures are architected materials that enable functional grading through tailored unit cell design. This works proposes a machine learning (ML)-based inverse design framework that maps target properties to multiple admissible design candidates, addressing the inherent “one-to-many” nature of the inverse problem. The framework is built on a novel parameterisation of a curvilinear body-centred cubic (curvy-BCC) unit cell, where non-uniqueness arises from variations in strut orientation and thickness. The design space is further expanded by incorporating multiple materials and unit cell topologies. The framework integrates a material classifier, a topology classifier, mixture-density-network (MDN)-based inverse generators, and a property predictor to generate multiple valid design candidates all with high property satisfaction, enabling further selection based on performance or manufacturability criteria. Moreover, probability measures derived from MDNs offer insights into solution reliability and diversity, as well as the feasibility of the target properties, bridging the gap in existing data-driven inverse design methods. |