Lunch & Learn | Beyond conventional methods: The role of ANN in nuclear waste management

In this talk, Virginie Solans will present the role of Artificial Neural Networks (ANNs) in a nuclear waste management context and explain how ANNs are reshaping conventional methods. ANNs are an example of machine learning, where typically large amounts of data are analysed.

After a brief introduction to ANNs, she will provide examples of where ANNs are proposed to be used in the nuclear context and of relevance to nuclear waste management. Such developments are requested in the safeguards community already today to increase the effectiveness and efficiency of nuclear material verification and can be adopted as an inspiration for the waste community and its plans for the verification of fuel in the coming decades.

One example is detecting and classifying particle defects in spent nuclear fuel assemblies. The Partial Defect Tester (PDET) is an instrument proposed to investigate the integrity of irradiated nuclear fuel assemblies by analysing the gamma and neutron flux inside the fuel assembly. Researchers have proposed to analyse the data from PDET using ANNs, aiming to identify if missing or replaced fuel material in a spent nuclear fuel assembly. Such verifications are particularly important before disposal in a geological repository, after which further verification is not possible. Another application is the anomaly detection using the Next Generation Surveillance System Camera Data. The ANN could assist IAEA safeguards inspectors to identify unexpected activity in nuclear facilities, including in facilities linked to the final disposal of nuclear waste.

Finally, Viriginie will present her own work that specialises in predicting safety parameters for fuel to be encapsulated. She will demonstrate how ANNs can be used to predict the effective multiplication factor (keff), a parameter for criticality-safety, for different canister loading using the radionuclides concentrations. This ANN can take advantage of the heterogeneity of the spent nuclear fuels irradiation history, including how the ANN can even capture changes in the canister keff when spent nuclear fuels are axially rotated. She will also demonstrate how ANNs can be used to predict the decay heat from experimental measurements planned to be performed before encapsulation and how it can be used to verify state-of-the-art calculations.

This presentation aims to show the large range of possibilities that ANN offers to help with the different aspects of nuclear waste management and how it might be used in the future.