Introducing optimum parameters of separation cascades for 123Te using GWO based on ANN. (1st December 2021)

• Record Type: Journal Article
• Authors: Imani, Morteza; Aghaie, M.; Adelikhah, Mohammademad
• Is Part Of: Annals of nuclear energy. Volume 163(2021)
• Publication Date: 2021-12-01
• Subjects: Square cascade -- Separation -- 123Te -- Gray wolf algorithm -- Artificial Neural network; Nuclear energy -- Periodicals; Nuclear engineering -- Periodicals; 621.4805
• DOI: 10.1016/j.anucene.2021.108545
• Languages: English
• ISSNs: 0306-4549

Abstract

Highlights: Two different ANNs are trained to predict the multicomponent separation calculations. Four step cascades are numerically evaluated for production of 123 Te for medical uses. Combining the GWO and ANN enhanced the fitness function evaluations. Two test cases with different objective functions are studied and parameters are reported. Using GWO-ANN, optimal parameters for production of 123 Te (99.9%) from 15 kg of natural Te is introduced. Abstract: In recent decades, separation of stable isotopes due to their substantial role in human health has been widely increased. The present research deals with square cascades optimization in order to separate the 123 Te by the Gray Wolf Optimization algorithm (GWO). The separation of 123 Te has significant application in medical science, and production of radioisotopes. In this study, attempts have been made to find the desired concentration of product (99.9%) for a given amount of natural Tellurium feed within four connected cascades. In this analysis, instead of solving nonlinear equations of concentration distribution in cascades, two different artificial neural networks (ANN) are trained to predict the objective functions. Two test cases for 123 Te separation with different objective functions have been considered. The aim is to gain the maximum product from a specified amount of feed in different configurations. In the first case, the neural network has 20 inputs and considers four connected cascades. To train the network, 5000 randomly generated data from the results is used. In the second case, the network has 22 inputs and 10, 000 random data is used. In both cases, the Levenberg-Marquardt algorithm with 40 hidden layers is selected to train the networks. Prediction of the objective functions using a neural network leads to a 98% reduction in execution time and significantly improves the speed of the optimization process. Using this method, the optimal cascades for separation of 123 Te with 99.9% concentration from 15 kg of natural Tellurium during a year are introduced.