About this Abstract |
| Meeting |
2024 ASC Technical Conference, US-Japan Joint Symposium, D30 Meeting
|
| Symposium
|
2024 ASC Technical Conference, US-Japan Joint Symposium, D30 Meeting
|
| Presentation Title |
AI-Empowered LS-Dyna ICME Simulation Technique for Multiscale Predictive Modeling of Composites |
| Author(s) |
Haoyan Wei, Wei Hu, CT Wu, Fabio Pavia |
| On-Site Speaker (Planned) |
Haoyan Wei |
| Abstract Scope |
Predictive ICME (Integrated Computational Material Engineering) modeling tools are transformational for advanced composite design and analysis, for which the influences of manufacturing processes and material microstructures on mechanical properties must be properly considered. One essential component of ICME is computational homogenization, which has proven effective in capturing the material multiscale effects. However, FEM-based homogenization models are computationally expensive, especially for industrial applications. Recently, we developed an AI-based accelerated multiscale composite analysis technique named DMN (Deep Material Network) in the engineering simulation software LS-Dyna. Through offline training, DMN learns the morphologies hidden in representative volume elements of composites and gains the ability to accurately and quickly predict homogenized composite responses based on the material microstructures. We have further implemented this AI technique under an ICME framework for nonlinear analysis of fiber-reinforced composites, where injection molding data are integrated with the structural analysis to consider the effects of heterogeneous fiber orientation and volume fraction distributions. In this presentation, we will discuss the main features of DMN and present case studies to demonstrate the promising performance of this AI-empowered and physics-based simulation technique. |
| Proceedings Inclusion? |
Definite: Post-meeting proceedings |