About this Abstract |
Meeting |
12th International Conference on Magnesium Alloys and their Applications (Mg2021)
|
Symposium
|
12th International Conference on Magnesium Alloys and their Applications (Mg 2021)
|
Presentation Title |
A Neural Network Approach for Approximating Simulation Predicted State Variables in a Preform Optimization Process |
Author(s) |
Tharindu Abesin Kodippili, Stephan Lambert, Arash Arami |
On-Site Speaker (Planned) |
Tharindu Abesin Kodippili |
Abstract Scope |
An artificial neural network (ANN) assisted optimization framework is developed to improve the preform design process of a cast-forged magnesium AZ80 I-beam component. An entirely simulation-based optimization process would be computationally expensive and resource-intensive, limiting the searchable design space. A set of ANNs are trained – using finite element method (FEM) simulation data – on a subset of preform designs that are intelligently sampled from an optimized design space to predict state variable responses throughout spatially varying regions of the forging. The prediction accuracy of the ANNs will be discussed and compared with FEM simulations. The influence of the preform design on the material flow behavior will also be addressed. |
Proceedings Inclusion? |
Planned: At-meeting proceedings |