Speaker
Description
Improving the prediction of TSL on-the-fly based on deep neural networks using experimentally measured double differential data and dynamic structure factor
Vaibhav Jaiswal$^{1,*}$, Arthur Pignet$^2$, and Luiz Leal$^1$
$^1$Insititut de Radioprotection et de Sûreté Nucléaire (IRSN), Fontenay-aux-Roses, France
$^2$MINES ParisTech, PSL - Research University, Centre for Material Forming (CEMEF), CNRS UMR 7635, CS 10207, Sophia Antipolis Cedex, 06922, France
$^*$Corresponding author: vaibhav.jaiswal@irsn.fr
The slowing down of neutrons generated in a fission reaction, from the MeV range to the meV range, is governed by the scattering process. In the low energy region, when the energy of the neutron is comparable to the energy of the scattering medium, the probability of this interaction is described by Thermal Scattering Law (TSL). TSLs for moderator materials takes into account the dynamical properties of the system and governs the energy and momentum exchange between the neutron and the moderator. The TSLs libraries available to users in the ENDF evaluations are for fixed temperatures, and one makes various approximations and mechanisms to use for non-available temperatures in the evaluations. It is well known that direct interpolation of the TSL to obtain data at intermediate temperature leads to misleading results [1]. A recent approach to overcome this limitation is to interpolate the LEAPR parameters and re-run NJOY to obtain TSL at the required temperatures [2]. This is not a practical solution when multi-physics calculations demand TSL for a very fine grid of temperatures and running NJOY on-the-fly will not be a feasible solution. On the other hand, storing TSLs for a fine grid of temperatures is memory and data intensive to use with Monte Carlo codes.
This paper demonstrates the capabilities of deep neural networks to overcome the above challenges and that TSLs can be reconstructed for any temperatures preserving the dynamics of the system. This approach provides reliable and accurate results within the required memory and size of the data. Several recent works have been carried out in this domain that involves both machine learning and deep learning to carry out nuclear cross-section data processing on-the-fly. In particular, a recent work for TSL, focuses on training the neural network model using the alpha, beta, and temperature as parameters [3]. This paper proposes a new methodology to incorporate not only the alpha, beta and temperatures, but also the experimentally measured double differential data and the dynamic structure factor to improve the deep learning network's prediction capabilities and accuracy.
Presently, the most common thermal scattering cross-section evaluation tool is the LEAPR module of the NJOY code. In the LEAPR framework, the entire physics related to the dynamic structure factor of the material is reduced as a function of the phonon frequency spectrum. This methodology seems to perform well but can be improved involving the experimentally measured dynamic structure factor and the double differential data. Double differential data and the derived dynamic structure factor for light water data measured at the SNS facility at the Oak Ridge national laboratory [4-5] have been used in this work to train the deep learning model. This paper demonstrates an example of TSL for light water, but this method can be applied on a more generalized approach to proceed towards on-the-fly sampling of the TSL in Monte Carlo codes.
[1] W. Haeck and N. Leclaire. “Thermal scattering data and criticality safety”, International Conference on the Physics of Reactors “Nuclear Power: A Sustainable Resource”, Casino-Kursaal Conference Center, Interlaken, Switzerland, September 14-19, 2008.
[2] Vaibhav Jaiswal, “Theoretical and experimental approach towards generation of thermal scattering law for light water”, Ph.D. thesis, University of Lille, 2018.
[3] C. A. Manring, A. I. Hawari, “Development of neural thermal scattering (NeTS) modules for reactor multi-physics simulations”, EPJ Web of Conferences 247, 20004 (2021).
[4] Luiz Leal, Vaibhav Jaiswal, and Alexander I. Kolesnikov, “High-resolution neutron time-of-flight measurements for light water at the Spallation Neutron Source (SNS), Oak Ridge National Laboratory”, EPJ Web of Conferences 239, 14005 (2020).
[5] Vaibhav Jaiswal, Luiz Leal, and Alexander I. Kolesnikov, “Analysis of the time-of-flight neutron scattering cross-section data for light water measured at the SEQUOIA spectrometer, Spallation Neutron Source (SNS)”, EPJ Web of Conferences 239, 14007 (2020).