Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : September 08, 2024 - September 11, 2024
This research aims to develop the method for classifying the break-up patterns of PLTFs using a deep learning. First, PLTFs judged to be good at Stage_1 (from after-blink to PLTF break-up) were classified into [Normal_1] class. On the other hand, for bad PLTFs in Stage_1, their images were classified into three classes [Dimple_1], [Line_1], and [TAL_1], where dimples, lines, and thinner aqueous layers (TALs) were produced in Stage_2 (after PLTF break-up), respectively. Then the 4-calss (Normal_1, Dimple_1, Line_1, TAL_1) classification of images in Stage_1 was performed using the transfer learning of VGG19 that is the model developed by the Visual Geometry Group (VGG) at Oxford University. For Stage_2, the images which were judged as the bad PLTF in Stage_1 were classified in the 3-classes of [Dimple_2], [Line_2] and [TAL_2] using the transfer learning of VGG19.
