| Autor: |
Rahmatullah, Helmi Fauzi, Kanaya, Diva Jati, Sigit, Rohmad, Artika, Refa |
| Předmět: |
|
| Zdroj: |
AIP Conference Proceedings; 2024, Vol. 2967 Issue 1, p1-7, 7p |
| Abstrakt: |
The on-board camera manipulator hot cells system in the PRTDBBNLR – ORTN BRIN radiometallurgical installation continues to be developed, both in terms ofsoftware and hardware. In this study, a case study was conducted to classify nuclear fuel images after irradiation from on-board camera manipulator hot cells with operating parameters: 3 cm distance, 3 minutes duration, highest exposure 30 mSv/hour and 60 experiments were conducted. This research was conducted to predict the damage to the on board camera based on image quality using the CNN (ConvolutionNeural Network) model. This CNN model was developed using the open source PythonIDE algorithm. The image dataset is the image surface quality of the production code onthe nuclear fuel plate which is used as reference data. The total dataset is then randomlyexpanded to 500 datasets to meet the minimum amount of data for the CNN model. Before separating the train-validation and random test data, it is necessary to homogenize the image tensor size and normalize the pixel values. The CNN model algorithm that has been compiled is then trained with the results of augmented training datasets, then monitoring the development of the model with dataset validation and endswith testing the dataset test. From the test results of the CNN program built on the python IDE, the model accuracy based on the validation results is 99.75% ± 0.50% and the testdata classification accuracy is 98%. So the classification of image quality on reference data in the form of 'blur' images and 'clear' images based on the CNN algorithm can be used as a reference in predicting damage to the on board camera manipulator due to highradiation in hot laboratory installations (Hot cells). [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
