Artificial Neural Network-Based Tomography Reconstruction of Plasma Radiation Distribution at GOLEM Tokamak

Abstract

The paper presents an artificial neural network-based model for tomography reconstruction of visible plasma radiation distribution at the GOLEM tokamak. The model was trained using a dataset from emissivity phantoms and associated synthetic measurements from a poloidal cross-section of the GOLEM tokamak. The model validation was performed on the prediction of various unseen phantom samples with shapes similar to those in the training dataset. The backfit of line-integrated measurements indicates the considerable potential of the proposed model for reconstructing the position, size, shape and intensity of the radiation function of one cross section. Additionally, the neural network-based model offers a significantly shorter prediction time compared to traditional tomography methods, providing a substantial advantage.