Year
2002
Abstract
Artificial neural networks may find wide application at the decision of problems of the control of reduced nuclear arms. The technology of artificial neural nets was applied to address the problems of re-gression and classification by example of processing of joint NMIS experimental results obtained by VNIIEF and ORNL in 2000. In the first case as a result of addressing the regression task absolute values of plutonium assem-blies ? shell mass and thickness were obtained from experimental data. The task of classification consists of an analysis of plutonium assembly conformity to the pattern. Successive NMIS measurements for one and the same assembly are used as the pattern. The measure-ments differ somewhat from each other due to natural spread under experimental conditions, and due to random radiation interaction with the item and recording system. Application of neural nets based on multilevel perceptron and Kohonen nets enables us to analyze conformity of NMIS signatures of pluto-nium assemblies to the identification pattern. The obtained results encourage us to state that the proposed method can be applied in the system of fissile material identification based on NMIS signatures. In case a more complicated item will have chosen and agreed upon identification parameters that allow its identification among other technical items, the artificial neural nets can be also proposed for application.