Título: Identificação da Informação Relevante para a Detecção Neural de Falhas em Varetas de Combustível em Reatores Nucleares
Autores: Dornellas, C. R. R.; Seixas, J. M.; Soares-Filho, W.
Resumo: The detection of failed fuel rods in a nuclear plant with a pressurized-water reactor (PWR) is important for the safe operation of the reactor. Here, ultrasonic pulses are used for automatic detection of failed nuclear fuel rods. The detection is based on processing the received echoes of the emitted ultrasonic pulses by means of a neural network. In order to investigate whether a significant reduction on the dimensionality of the original input data space can be achieved, the relevance of each data sample is evaluated. Using four peaks resulting from the reverberation in the inner wall of the rods, 93% of failed rods are correctly identified, for a false alarm probability of ~4%. When using only the third and fourth peaks, the input dimension is reduced from 135 to only 71 data samples and an efficiency of ~93% in the detection of failed rods is obtained, for a false alarm probability of ~7%.
Código DOI: 10.21528/CBRN2001-046
Artigo em pdf: 5cbrn_046.pdf
Arquivo BibTex: 5cbrn_046.bib