In the Nuclear facilities, especially Fukushima daiichi nuclear power plant, radiation exposure reduction measures have to be carried out appropriately so as to be able to work in the place. Therefore, we need to grasp the radioactive contaminations level in the area. In order to specify the place and the density of the radioactive contamination, we had to estimate the radioactive contamination density of various locations by material sampling measurement, surface smear measurement, or surface dose rate measurement with collimated radiation detectors conventionally. However, these methods require a lot of time and work. To solve this problem, we are developing the estimation method of the radioactive contamination distribution with machine learning from the spatial dose rate that can be acquired easily.
The estimation of the radioactive contamination from the spatial dose has two issues mainly. One is the difficulty of the improving estimation accuracy because of radiation scattering and attenuation with the structure in the building. The other is that it takes much time to make the accurate model with simulation and so on. With machine learning, we will be able to estimate the contamination distribution quickly, and it will lead to exposure reduction of workers. In this study, we constructed the building model of the Operating floor of Fukushima daiichi unit3(1F-3), and set the radioactive contamination on the floor divided to 10×13 mesh. We trained the relationship of the spatial dose distribution with the radioactive contamination densities, locations, and the material structures in the area.
As the result, in the case of setting the various contamination densities to the each mesh, the estimated contamination densities were consistent with the setting contamination densities. Therefore, the feasibility of this method was confirmed.