Stability study of the neural network at particle physics detectors

Neural networks are used as triggers at highenergy physics detectors. These triggers can separate the event that must be saved for later analysis from the other events or noises. Using the raw data of the detector, the signal and the background can be separated offline. After separation, sets of signals and backgrounds can be used to train the neural network. A gas-filled detector (multiwire proportional chamber) was used to study the trigger at different noise levels to find the most stable neural network that tolerates the random hits. The ratio of the recognized and the unrecognized signal and background events is used for the measurement. Its stability is part of the systematical uncertainty.

Fission Products Particular Peak Measurement for UO2-Gd2O3-MOX γ-Scanning Renormalization in 100% MOX ABWR Mock-Up Cores

This paper presents the principles of the peak check measurements by gamma spectrometry. One details the main equations used for the analysis of the raw data as the calculation of the different sources of uncertainties and their propagation on the result. The method is illustrated with actual examples from the French-Japanese BASALA ABWR 100% MOX experimental program. • For each individual fission rate, the systematical uncertainty due to the radioactive decay data is smaller than the statistical uncertainty due to the counting process. • The main part of the final uncertainty on the scaling factor is brought by the systematical uncertainty on the average fission yields. This study enables to propose some recommendations for fission products and nuclear data evaluation used. • Only the analysis of the 140La peak at 1596 keV leads to acceptable uncertainties on the fission rate maps renormalization, with a good consistency with the integral γ-scanning results. • The JEF-2.2 evaluation on the fission yields and their associated uncertainties seems more realistic than the ENDF/B-6 and is recommended for the scaling factor analysis.