Multi-objective design of multi-material truss lattices utilizing graph neural networks

The rapid advancements in additive manufacturing (AM) across different scales and material classes have enabled the creation of architected materials with highly tailored properties. Beyond geometric flexibility, multi-material AM further expands design possibilities by combining materials with distinct characteristics. While machine learning has recently shown great potential for the fast inverse design of lattice structures, its application has largely been limited to single-material systems. In this work, we propose a novel approach that incorporates material properties as edge features within the graph representation of multi-material truss lattices, utilizing graph neural networks (GNNs) to develop a fast and efficient inverse design framework. We validate this framework by designing lattices with tunable thermal expansion and stiffness properties, showcasing its ability to explore a broad and flexible design space. We showcase the framework's inverse design capabilities for both single and multi-objective optimization tasks and assess its limitations. Additionally, we demonstrate the superior capacity of GNNs in capturing structure-property relationships for multi-material systems. We anticipate that the continued advancement of GNN-assisted inverse design will play a key role in unlocking the full potential of multi-material truss lattices.