Investigation of a Methodology for the Detection of Diversions in Spent Nuclear Fuel
Doctoral thesis, 2024
One of the main tasks in nuclear safeguards is the inspection of Spent Nuclear Fuel (SNF) to detect possible diversions of their special nuclear material content, e.g., U-235 and Pu-239. These inspections verify the declared SNF via passive measurements of characteristic signatures such as the emissions of neutrons and gamma rays. The current PhD research investigates different aspects for the development of a novel non-intrusive methodology that can enhance safeguards inspections of SNF assemblies, and it includes two main parts. In the first part, simulations are performed to evaluate the feasibility of measuring the neutron flux and its gradient inside the empty guide tubes of a SNF assembly with a miniaturized detector made of an array of optical fiber-based neutron scintillators. In addition, experiments are carried out to characterize these types of neutron scintillators. The results of this preparatory work show that neutron flux gradient measurements in SNF assemblies may be a viable option and provide insights for the construction of a prototype of a detector for the purpose. In the second part of the research, the application of machine learning models based on Artificial Neural Networks (ANNs) is studied to process measured SNF signatures and reconstruct the arrangement of the fuel pins in an assembly. The objective of this part is two-fold. On one hand, ANN models are explored for the task of determining possible diversion patterns from SNF signatures collected inside the accessible guide tubes. On the other hand, the advantage of providing the neutron flux gradient as input feature to the algorithm is evaluated. The training and testing of the ANN models are performed with synthetic datasets generated from Monte-Carlo simulations of a typical PWR SNF assembly, considering the intact configuration and different degrees and patterns of diversion. The results show that the models effectively predict diversions and characterize most of them to a good extent. In addition, the use of the neutron flux gradient, which is not analyzed during standard inspections, is proven to be advantageous.
flux gradient detector
artificial neural networks
partial defects
neutron scintillator
spent nuclear fuel
nuclear safeguards
machine learning
PJ lecture room, Physics Origo building, Chalmers University of Technology
Opponent: Prof. Anna Erickson, Georgia Institute of Technology, USA
Author
Moad al-Dbissi
Chalmers, Physics, Subatomic, High Energy and Plasma Physics
Identification of diversions in spent PWR fuel assemblies by PDET signatures using Artificial Neural Networks (ANNs)
Annals of Nuclear Energy,; Vol. 193(2023)
Journal article
Conceptual design and initial evaluation of a neutron flux gradient detector
Nuclear Instruments and Methods in Physics Research, Section A: Accelerators, Spectrometers, Detectors and Associated Equipment,; Vol. 1026(2022)
Journal article
Aldbissi, M., Pázsit, I., Rossa, R., Borella, A., Vinai, P. On the use of neutron flux gradient with ANNs for the detection of diverted spent nuclear fuel.
Nuclear safeguards are a critical component of ensuring the responsible and peaceful use of nuclear technology around the world. Safeguards involve systematic examinations of nuclear facilities and activities to verify compliance with international agreements and commitments.
The majority of nuclear materials placed under safeguards are in the form of Spent Nuclear Fuel (SNF) assemblies that are discharged from nuclear power reactors. SNF contains residual amounts of fissile materials such as Uranium-235 and Plutonium-239 which makes it attractive for proliferation purposes, e.g., the construction of nuclear weapons. Therefore, safeguards inspections are performed regularly to ensure that these materials are not being used for illicit activities.
The PhD research presented in this thesis investigates different aspects of a novel non-destructive methodology that can support more accurate safeguards inspections of SNF assemblies. The first aspect is related to the evaluation of the concept of a new detector and the preparatory experimental work for the construction of a prototype. The detector can measure the neutrons emitted inside SNF assemblies and provide the spatial distribution of the neutron flux and its gradient. The gradient, which has not been considered in safeguards inspections before, contains more and independent information compared to the simple neutron flux.
The second aspect of the methodology concerns the development of machine learning models based on Artificial Neural Networks (ANNs) for processing measurements from SNF assemblies and predicting whether sensitive nuclear material has been removed from the assemblies or not. The models show good performance in classifying SNF assemblies based on the percentage of replaced fuel pins, and in reconstructing diversion patterns. The use of the neutron flux gradient as input future to the ANN models was proved advantageous in detecting replaced fuel pins.
The described methodology has the potential of reducing the amount of expert judgement required for the interpretation of the measurements and providing more detailed estimations, and therefore it can facilitate the safeguards inspection process.
The majority of nuclear materials placed under safeguards are in the form of Spent Nuclear Fuel (SNF) assemblies that are discharged from nuclear power reactors. SNF contains residual amounts of fissile materials such as Uranium-235 and Plutonium-239 which makes it attractive for proliferation purposes, e.g., the construction of nuclear weapons. Therefore, safeguards inspections are performed regularly to ensure that these materials are not being used for illicit activities.
The PhD research presented in this thesis investigates different aspects of a novel non-destructive methodology that can support more accurate safeguards inspections of SNF assemblies. The first aspect is related to the evaluation of the concept of a new detector and the preparatory experimental work for the construction of a prototype. The detector can measure the neutrons emitted inside SNF assemblies and provide the spatial distribution of the neutron flux and its gradient. The gradient, which has not been considered in safeguards inspections before, contains more and independent information compared to the simple neutron flux.
The second aspect of the methodology concerns the development of machine learning models based on Artificial Neural Networks (ANNs) for processing measurements from SNF assemblies and predicting whether sensitive nuclear material has been removed from the assemblies or not. The models show good performance in classifying SNF assemblies based on the percentage of replaced fuel pins, and in reconstructing diversion patterns. The use of the neutron flux gradient as input future to the ANN models was proved advantageous in detecting replaced fuel pins.
The described methodology has the potential of reducing the amount of expert judgement required for the interpretation of the measurements and providing more detailed estimations, and therefore it can facilitate the safeguards inspection process.
A new approach to partial defect testing of spent nuclear fuel for safeguards applications
The Swedish Radiation Safety authority (SSM) (SSM2021-709), 2021-07-01 -- 2022-05-31.
Subject Categories
Subatomic Physics
Other Engineering and Technologies not elsewhere specified
Roots
Basic sciences
Infrastructure
C3SE (Chalmers Centre for Computational Science and Engineering)
Driving Forces
Innovation and entrepreneurship
ISBN
978-91-7905-987-3
Doktorsavhandlingar vid Chalmers tekniska högskola. Ny serie: 5453
CTH-NT - Chalmers University of Technology, Nuclear Engineering: 352
Publisher
Chalmers
PJ lecture room, Physics Origo building, Chalmers University of Technology
Opponent: Prof. Anna Erickson, Georgia Institute of Technology, USA