[2B16-21] Decommissioning of Fukushima Daiichi NPP 4

[2B19] Development of Exposure Reduction Technologies by Digitalization of Environment and Radioactive Source Distribution

(10) Understanding and 3D modeling of working environmental data by using deep learning

*Akio Doi¹, Meguru Yamashita¹, Hiroki Takahashi¹, Toru Kato¹, Takashi Imabuchi², Toshihide Hanari², Yuta Tanifuji², Rintaro Ito² (1. Iwate Prefectural University, 2. Japan Atomic Energy Agency)

Keywords:Working environmental data, Deep learning, point cloud, 3D model, CAD

In order to carry out decommissioning work safely and efficiently, we have developed a "radioactive source / dose rate estimation system" that can measure, collect, visualize, and store data such as the state of structures and air dose rate. In this study, in order to easily carry out this update work, point cloud data was automatically recognized by pointnet ++ for the extracted difference information. Pointnet ++ is a neural network that can be learned by directly inputting point cloud data. Furthermore, the recognized point cloud data is automatically converted into a triangular mesh, a plane, a cylinder, and a cube (voxel) as needed, and is used for updating the simulation model of the "source reverse estimation engine".