X-RAY DETECTOR EXPOSURE DIAGRAM APPROXIMATION APPLYING AN ARTIFICIAL NEURAL NETWORK

Nowadays the demand for qualified specialists in non-destructive testing is steadily increasing. In order to develop students' skills of knowledge application necessary for individual practical activity, the training simulator should be equipped with software tools for modelling the results of radiation control. The main approaches to modelling of radiation inspection results and their disadvantages are considered. The method of detector fragment optical density estimation based on approximation of exposure nomographs and sensitometric curves is proposed. The application of the KAN artificial neural network with architecture based on the Kolmogorov-Arnold theorem for approximation tasks is described. The results of computational experiments are given and the relative error of optical density for the considered method is estimated. The proposed solution makes it possible to obtain output values with an accuracy close to the accuracy of the detector technical data, significantly reducing the computational complexity of operations performed during simulation on a graphics processor.