Physics-Guided Graph Convolutional Network Framework for Inverse Design of Bioinspired Helicoidal Laminates

Bioinspired helicoidal laminated composites have attracted considerable attention for their enhanced damage tolerance and shear load-carrying capability. In this work, an inverse design framework is developed to control transverse shear stress distribution in helicoidal laminated composite plates using graph convolutional networks (GCNs). The analytical response of the plates is obtained through a quasi-3D shear deformation theory, considering simply supported boundary conditions and sinusoidal transverse loading. The primary objective is to predict the optimal orientation of the helicoidal plies that achieves a target transverse shear stress for a given plate geometry and layer configuration. By representing the laminate as a graph structure, where nodes denote plies and edges capture interply relationships, the GCN model learns complex mappings between structural parameters and stress outcomes. The trained model demonstrates strong predictive capability and efficiency, enabling rapid identification of helicoidal configurations without iterative trial-and-error simulations. The results highlight the potential of combining advanced shear deformation theories with graph-based machine learning for the inverse design of bioinspired composites, offering new pathways for tailoring stress responses in lightweight structural applications.