Research on 3D Radiation Fields Reconstruction based on FNN

Accurate radiation distribution could provide solid guidance for the protection of both occupational workers and the public. However, in some cases , such as nuclear accident, it is still hardly to obtain high accurate radiation distribution over the entire concerned region via numerical calculation due to complex geometry and/or radiation sources. In these cases, the radiation field reconstruction technique based on the limited measured and/or calculated radiation results has attracted more and more attention. In this work, an forward neural network (FNN) framework consisting of two fully connected layers connected by several residual blocks was proposed to study the reconstruction for 3D radiation with typical shielding layout including bulk shielding and streaming. With the help of the idea of piece-wise functions, a good prediction accuracy was achieved by dividing the entire zone into small training zone based on the magnitude of dose. The fraction of relative deviation between predicted and Monte Carlo results within 20% were 80.7% and 90% for bulking shielding case and the streaming case, respectively.