Radiation transmission test (RT) is a non-destructive test technique that detects defects using changes in the intensity of transmitted radiation when radiation is irradiated to a test body. The existing method in which experts perform direct inspection with the naked eye has a high inspection cost and an inspection error problem due to human error. In this study, we would like to propose a faster and more accurate inspection process using deep learning techniques.
A virtual defect image was created and used for learning as a series of methods to solve data imbalance, and based on the predicted results of the learned model, images that require final visual inspection by experts can be recommended to reduce inspection costs and improve defect judgment accuracy.
Junhyeok Choi|Changhyun Lee|Hyungun Jo
Sudong Lee
Junhyeok Choi : Data preprocessing, Modeling
Changhyun Lee : Virtual flaw
Hyungun Jo : Data preprocessing
- Anaconda
- Python 3.8
- Tensorflow
- Pandas
- Opencv-python
- Numpy
- Docker