2-dimensional stress field prediction using deep learning
Computational solid mechanics has been widely conducted using Finite Element Analysis (FEA). However, to do structural analysis on an object, many processes have to be done after setting up the geometry and boundary conditions for final analysis of the structure using Finite Element Method (FEM)....
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
2019
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/10356/76460 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | Computational solid mechanics has been widely conducted using Finite Element
Analysis (FEA). However, to do structural analysis on an object, many processes
have to be done after setting up the geometry and boundary conditions for final
analysis of the structure using Finite Element Method (FEM). With the progress of
deep learning, conditional Generative Adversarial Network (cGAN) has shown
great potential in solving FEA problems quickly and accurately.
In this project, a cGAN architecture was developed and 2D structures with arbitrary geometries and boundary conditions were trained on the network. The network was trained to directly output Von Mises stress distribution plots over the structure. The trained network produced predicted outputs with a mean relative error of 4% with respect to the average ground truth. cGANs have great potential of being generalised and applied to more complex structures with more complex boundary conditions. |
|---|