| 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 |
Using Neural Network to Improve Failure Predictions |
| Author(s) |
Ashutosh Maurya, Dharanidharan Arumugam, Ravi Kiran Yellavajjala, Subramaniam D. Rajan |
| On-Site Speaker (Planned) |
Ravi Kiran Yellavajjala |
| Abstract Scope |
A new failure criterion has been developed to improve modeling of orthotropic structural composites subjected to impact loadings. Instead of using an analytical expression as failure predictor, a discrete set of points called point cloud failure surface is constructed in the stress/strain space. Multi-scale modeling scheme based on a combination of virtual and laboratory testing is used. Representative volume elements (RVE) are used to model the constituent parts of the composite that are then subjected to multi-axial state of stress and the first instance of failure in the RVE is detected and tagged as a point in the point cloud data. One of the challenges in using the generated point cloud data in a finite element simulation is the ability to correctly predict when the onset of failure takes place in a finite element without being too conservative. A Neural Network (NN)-based predictive approach is developed and used. A unidirectional composite, the T800-F3900, commonly used for aerospace applications, is used to compare the performance of these methods. The predictive scheme is implemented in a commercial finite element program, LS-DYNA, and the performance is evaluated by simulating a laboratory test. Results indicate that the NN implementation is robust, efficient, and reasonably accurate. |
| Proceedings Inclusion? |
Definite: Post-meeting proceedings |