2D topology optimization simulator implemented by Unet

2D topology optimization simulator implemented by deep learning algorithm

2D topology optimization simulator

User friendly topology simulator using deep learning network(U-Net)

Topology optimization's goal is not a difficult concept : the way to reach the best performance under limited amount of materials.

Although topology optimization is a powerful solution saving materials or money, the way to arrive the destination of

optimization is not easy. Especially, it takes too long time because it contains the process of stress analysis.

So, we came up with time-saving system using U-Net (U-shape deep neural network). The system is divided to three

components : code, GUI, trained network.

--------------------------------------------------

The procedure of whole process is as follows:

1. Run Final_Activate_code.m.
2. GUI imports the geometry of rectangular shape. (For demo version (v1.0.1), click four points along the counter-clockwise direction)
3. After clicking four points that user wants, figure is saved as "1.png" and you can click again for the next step.
4. 1st Unet-based model(ResUnet) predicts the stress distribution of the imported geometry. Load is applied to third point with downward direction.
5. 2nd Unet-based model(ResUnet) predicts optimized shape based on the output from 1st model.
6. 3rd Unet model is employed for upsampling with higher resolution. This is referred by MATLAB example code (Image-to-Image Regression in Deep Network Designer, https://kr.mathworks.com/help/deeplearning/ug/image-to-image-regression-in-deep-network-designer.html)

------------------------------------------------------

The explanation of three components :

1. Code - Final activate 1, and 2 : code controlling entire process
2. GUI - Image file generator(uts_gui_exported.m) : matlab program making the initial structure images using the four points position users want
3. Trained network -

1) ResUnet 1 : using the initial structure images, this network predicts stress distribution of that structure

2) ResUnet 2 : using ResUnet 1 output, this network predicts topology optimized shapes and their stress distribution

3) Unet 1 : this network increases the resolution of ResUnet 2 output