Measurement of uranium quantities by fluorescence-X using neuronal techniques

Vincent Vigneron (1) , Jean-Marc Martinez (1) , Anne-Catherine Simon (2) , Robert Junca (2)

1 SERMA - Service d’Études des Réacteurs et de Mathématiques Appliquées
2 LOAD - Laboratoire Outils d'Analyse des Données

Layered neural networds are a class of models based on neural computation in biological systems. Connexionists models are made of a large number of simple computing structures, highly interconnectec. The weights assigned to the connections enable the encoding of the knowledge required for a task. They can be trained to learn any input-output relation after selecting a suitable architecture. This method appears useful in those cases in which a simple operation and a fast response are needed, together with a reasonnable accuracy. They are applied here to the automatic analysis of X-ray fluorescence spectra, obtained with plastic bottles irradiated by collimated beam of photons, emitted by a sealed source of iridium. The method could allow the surveillance of the measurement of uranium quantities in nuclear fuel cycles in processing plants.

Keywords: artiticial intelligence; pattern recognition; neural network; data analysis; signal processing; Supervised learning; regression; Poisson distribution; online learning; machine learning; spectrum analysis

Domains: Signal and Image Processing

Vincent Vigneron, Jean-Marc Martinez, Anne-Catherine Simon, Robert Junca. Measurement of uranium quantities by fluorescence-X using neuronal techniques. ERSADA 17th symposium on safeguards and nuclear material management, May 1995, Aachen, Germany. ⟨cea-02524489⟩