Physics-informed hypergraph neural network for packaging simulation

Overview

The Physics-Informed Hypergraph Neural Network (HG-PINN) is a cutting-edge solution for packaging design and simulation. By embedding mechanical equilibrium and integrating metrics like structural strength, cost, and usability, HG-PINN provides a fast and accurate digital twin for packaging. This solution significantly reduces the need for costly prototyping and computationally intensive finite-element simulations by achieving near-FEM accuracy without meshing or solver bottlenecks. It allows for rapid evaluation of new packaging geometries or materials and supports AI-driven virtual prototyping and design iteration.

Technical specifications

• Physics-consistent AI surrogate: Generalizes across designs and materials by enforcing material laws and energy consistency.
• Node-element representation: Each packaging component is a node–element pair, learning structural interactions and material responses.
• Integration with topology optimization: Works with methods like SIMP, with learnable stiffness and density features.
• Validation and adaptability: Validated on 2D structural benchmarks and generalizes across mesh sizes and load cases. Adaptable to partner-provided CAD/FEA data and packaging material datasets.
• Differentiable solver: Enables gradient-based optimization within seconds, facilitating rapid design iteration.

Technology readiness level

Currently at TRL 6, the HG-PINN has been validated on structural benchmarks and is in the process of being integrated into industry-scale simulations. The development phases include adaptation, training, validation, design optimization, and integration into a digital twin for comprehensive packaging design and optimization.