Speaker
Description
The data analysis of neutron cross section measurements is a critical step for generating high quality and reliable nuclear databases [1-4] to be used in the fields of nuclear technologies, radioprotection, nuclear medicine or astrophysics, among others. Artificial intelligence techniques, and in particular deep learning, have proven to be a very useful tool for data recognition and analysis. We have investigated the suitability of deep learning and neural network techniques as a complementary tool for the analysis of neutron capture cross section measurements.
Some of the neutron capture cross section measurements performed at the n_TOF facility at CERN [5] use the so-called Total Absorption Calorimeter (TAC) [6]. This device is made of 40 BaF$_2$ crystals capable of detecting in coincidence the γ-rays emitted after the capture reactions. In a standard analysis, conditions (cuts) on the total energy deposition in the TAC and on the event multiplicity (number of crystals with signals in coincidence) serve to discriminate capture from background events.
We have trained a neural network with the list of energies deposited in the different crystals with Monte Carlo simulated capture and background events (due to scattered neutrons and competing fission reactions) and evaluated its capability of selecting the γ-ray cascades in a virtual capture experiment. We have compared different strategies and combinations of the neural network topologies and standard analysis methods in order to obtain at the same time large signal-to-background ratios and large capture detection efficiencies. The use of a neural network with 5 hidden fully connected layers and some conditions on the total energy and multiplicity of the crystals has achieved up to 5 times better signal-to-background ratio and ~15% more remaining capture events in the resulting dataset than with the standard analysis method.
In this conference, we will present these results and describe the deep learning methods used.
References
[1] OTUKA, N., et al., Towards a more complete and accurate experimental nuclear reaction data library (EXFOR): international collaboration between nuclear reaction data centres (NRDC), Nucl. Data Sheets 120, 272 (2014).
[2] BROWN, D.A., The 8th major release of the nuclear reaction data library with CIELO-project cross sections, new standards and thermal scattering data, Nucl. Data Sheets, 148 (2028) 1-142.
[3] CABELLOS, O., et al., Benchmarking and validation activities within JEFF project, EPJ Web Conf., 146 (2017), Article 06004.
[4] SHIBATA, K., et al., JENDL-4.0: A new library for nuclear science and engineering, J. Nucl. Sci. Technol., 48 1 (2011) 1.
[5] GUERRERO, C., et al. Performance of the neutron time-of-flight facility n_TOF at CERN. Eur. Phys. J. A 49, 27 (2013).
[6] GUERRERO, C., et al., The n_TOF total absorption calorimeter for neutron capture measurements at CERN, Nucl. Instrum. Methods A, 608 3 (2009) 424-433.
[7] GARSONS, G.D., Interpreting neural networks connection weights, AI Expert 6 (1991) 47-51.