Fast uranium enrichment measurement in very low-level wastes by using a neural network algorithm

Abstract

In this work, we developed a neural network model that can analyze uranium enrichment even when the measurement time is less than 10 seconds using a low-resolution detector. Uranium enrichment measurement is essential for the quantitative analysis of uranium activity for the disposal of radioactive waste. The previously studied method of measuring uranium enrichment (infinite thickness method, peak ratio method, and relative efficiency method) is difficult to use in field measurement situations due to many restrictions. The relative efficiency method with the best accuracy among the existing methods is a method using the relative size of the peak, which has good accuracy, but requires a long measurement time, and there is a limitation that a high-resolution detector is essential. In this study, we focused on measuring uranium enrichment by using an artificial neural network with a low-resolution detector. With the proposed model, the enrichment of uranium wastes (ash, soil, concrete) in the Marinelli beaker can be predicted within 5% of the relative error. In particular, when the measurement time was about 10 seconds short, existing methods failed to analyze uranium enrichment, while the model proposed in this study maintained a relative error of less than 5%.