| 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 |
Use of Convolutional Neural Networks for Microscale Simulations in Progressive Damage Analysis in Composites |
| Author(s) |
Mohammadamin Alibakhshi, Parisa Khodabakhshi |
| On-Site Speaker (Planned) |
Mohammadamin Alibakhshi |
| Abstract Scope |
Due to the inherent heterogeneity of composite materials, their part-scale properties will strongly depend on constituent properties at microscale. Multiscale simulations help link constituent-level features to the performance of the composite structure through a two-way coupling between the micro- and macroscale simulations and applying homogenization and localization approaches to move from microscale to macroscale and vice versa. The finite element method, a high-fidelity model that can be used at microscale to obtain the homogenized properties of a repeating unit cell (RUC) to be reported to macroscale, has computational costs that are prohibitively high for multiscale simulations. On the other hand, micromechanics models are a class of semi-analytical models that predict the homogenized properties and local fields of composite materials for a given RUC at a relatively lower computational cost. However, the existence of damage, thermal effects, or plasticity necessitates the iterative evaluation of the homogenized properties which would result in high computational costs even with the use of micromechanics models. This study is focused on the use of the convolutional neural network (CNN) to effectively and accurately map the constituent properties and a given damage pattern in the RUC to its effective properties. The CNN will be trained on data obtained from the Generalized Method of Cells (GMC) for different RUC geometries, and material properties. It is expected that the CNN will significantly reduce the computational burden associated with evaluating the homogenized properties in progressive damage analysis. |
| Proceedings Inclusion? |
Definite: Post-meeting proceedings |