Overview and details of the sessions of this conference. Please select a date or location to show only sessions at that day or location. Please select a single session for detailed view (with abstracts and downloads if available).
7.02: Reactor Physics & Analysis-Data and Deterministic Methods
2:00pm - 3:30pm
Session Chair: Jeffrey Charles King, Colorado School of Mines, United States of America Session Chair: Armando Gomez, National Institute for Nuclear Research in Mexico, Mexico
An Active Learning Strategy for the Development of Data-Driven Reactor Models
Cliff Ghiglieri, Michael Bowman, Jeffrey King, Xiaoli Zhang
Colorado School of Mines
There is a recent and growing interest in applying machine learning and artificial intelligence techniques to the monitoring and operation of nuclear reactor systems. One challenge posed by the application of machine learning techniques to nuclear reactor power systems is the need for a large and reliable set of data for the data-driven models to learn from. Thus, in the near term, it is likely that machine learning models for nuclear systems will be built using physics-based simulation data. This poses an additional challenge – many reactor simulations are based on Monte Carlo methods, which can be very computationally expensive. This project developed a machine learning model for the critical control rod positions in a sodium-cooled fast reactor, using an active learning model coupled to a Monte Carlo N-Particle (MCNP) physics-based model. Initial results indicate that including a selection filter in the active learning routines based on a simple control rod worth curve can significantly reduce the required search space and improve the efficiency of the active learning process.
Nuclear criticality safety estimations for MOX-dump powders experiments using ROSFOND/ABBN-RF nuclear data
The paper presents the results of a computational analysis of the OECD/NEA benchmark conducted to estimate the accuracy of the critical safety parameters of multiplying MOX-fueled systems. The computational test is a set of 15 spherical multiplying systems that differ in their compositions and geometries. According to the test conditions, the keff values of the analyzed systems are unknown in advance. As part of the computational analysis of the test involving national codes and nuclear data libraries, along with the keff calculations, it is also necessary to estimate the a priori (due to the accuracy of the nuclear data used) and a posteriori (based on the accumulated experimental information) errors in the calculated keff values. Based on the benchmark, an updated version of the ROSFOND/ABBN-RF nuclear data was tested. The results of estimating the a priori and a posteriori errors in keff using the INDECS system for the proposed test models are presented. The analysis of the calculated data shows that (1) the observed spread in the keff values obtained from the Russian ROSFOND library and foreign evaluated nuclear data libraries (ENDF/B-VII.0, JEFF-3.2, JENDL-4.0) varies from –0.3 up to 0.8%; and (2) the deviation of the calculation results in the keff values obtained from the ROSFOND library and its group version, ABBN-RF, does not exceed 0.1%. The average a priori error in keff for all the tested options of multiplying systems is about 1% and, taking into account the selected set of experimental criticality data for MOX-fueled systems, including experiments at the BFS facilities, the average a posteriori error in keff can be reduced to 0.3%. The performed estimations confirm the high accuracy of the ROSFOND/ABBN-RF nuclear data for calculating the critical safety parameters of multiplying MOX-fueled systems.
Nodal Integral Method for Multi-group Neutron Diffusion Equation in 3D cylindrical Coordinate System
Manish Raj1, Suneet Singh2
1NPCIL; 2IIT Bombay
The nodal methods are quite accurate compared to traditional numerical schemes and have been extensively used for solving neutron diffusion and transport equations. The conventional Nodal methods requires either square or hexagonal lattice/ mesh structures. This kind of mesh structure imposes the requirement of very fine meshing near the boundary to take care the curvature due to cylindrical shape of reactor pressure vessel (in LWRs) or Calandria (in PHWRs). However, if the mesh structure is consistent with cylindrical coordinates, these will not require the fine meshing near the boundaries. For conventional reactor with solid fuel, there is not much loss due the non-inclusion of cylindrical shape of reactor. However, if the fuel is in molten form or reactor size is small as in few of Gen-IV reactors (such as in MSR, PBMR etc.), the methodology developed in cylindrical coordinates would be more suitable.
Recently, a Nodal Integral Method (NIM) was developed for 2D cylindrical polar coordinates for one group neutron diffusion equation. However, its extension to 3D and multi-group equations is neither trivial nor straightforward. In this work, the approach for solving steady state neutron diffusion equation in 3D cylindrical geometry using nodal integral method (NIM) is discussed. The approach adopted using transverse integration process is similar to that used in analytic nodal method to solve the 3D steady state neutron diffusion equation in Cartesian coordinates. However, the problem involved with the transverse integration in θ-direction has been resolved by approximating the average of products by product of averages. The comparison of the results for a source problem obtained by the present approach with benchmark and analytical solutions shows that method maintains its accuracy and order even in 3D multi-group formulation.
Recent developments in the neutronics codes of the AZTLAN platform
Armando Gomez1, Edmundo del Valle2, Julian Duran2, Andres Rodriguez1
1National Institute for Nuclear Research in Mexico; 2National Polytechnic Institute
The Aztlan Platform project is a Mexican national initiative which aims to have several in-house tools integrated and coupled into a common platform for analysis and design of nuclear reactors. The Aztlan platform project started in 2015 and since then several developments have been done in each of the codes. In the present paper, the novel capabilities of the three deterministic neutronic codes AZTRAN, AZKIND and AZNHEX are presented. In all of them, high performing computing capabilities have been implemented and several exercises for V&V have been done. AZTRAN and AZKIND codes deal with square geometry. AZTRAN is a 3D transport code which solves numerically the multi-group time independent Discrete Ordinates neutron transport equation with recently implemented Domain Decomposition capabilities for parallelization. AZKIND is a 3D diffusion nodal module that solves numerically the time dependent neutron diffusion equations and that has been accelerated by means of open source library Paralution that allows to run the code with GPU’s obtaining outstanding speed-up. Also, a simplified TH module have been implemented in AZKIND for feedback considerations by means of multivariable Lagrange interpolations of NEMTAB Cross Section libraries. Finally, AZNHEX is a 3D diffusion module that solves numerically the time dependent neutron diffusion equations in Hexagonal-Z geometry first, by means of a Gordon-Hall transfinite interpolation for transforming the hexagons in four cubes and then by means of nodal methods. AZNHEX has been used in several international Benchmarks related to fast reactors with comparable results as the obtained with well-known codes like DYN3D and PARCS. In this paper, a summary of the codes together with the main results of the exercises performed with each one of the codes is presented. Also, a discussion of the on-fly developments, future work and benefits of the Aztlan platform for the Mexican nuclear field are stated.