| Autor: |
Hong, Sunghwan, Lee, Heehwan, Kim, Yoonsuk, Cho, Hyungkeon, Seo, Yudeok, Cho, Sungchan |
| Předmět: |
|
| Zdroj: |
SID Symposium Digest of Technical Papers; Jun2023, Vol. 54 Issue 1, p1284-1286, 3p |
| Abstrakt: |
As customer specifications for the optical properties of OLED display devices are increasing, the flatness of the surface beneath the electrodes must also be considered. The 3D shape of an organic layer be able to simulate using computational fluid dynamics accurately. But, in general CFD (FEM, FVM, etc.) has limitations such as boundary conditions, mesh quality etc. In addition, it has a problem that too many calculations must be performed when simulation for a very large area is required. Accordingly, we propose a new AI model for prediction 3D‐surface profile based on machine learning that can overcome the limitations of physical methods. We can create xyz‐coordinates from top surface of stack structure, and some of parameters used in physical model were selected for input data. Also, a large amount of data set was collected with CFD simulation by fully considering the influence distance of the fluid and the available range of physical parameters. Modeling was performed using a hybrid architecture of CNN and U‐Net. The first achievement of this study is to apply parameters with physical meaning to the training model as a feature channel of an input image. The second is adoption a CNN to the basic U‐ Net structure for correcting the average height of the 3D‐surface. As a result, target image can be obtained quickly and unconditionally, even when predicting many cases or large areas. This study is expected to replace existing CFD analyzes for obtaining 3D profiles. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
Plný text