Machine learning-aided damage identification of mock-up spent nuclear fuel assemblies in a sealed dry storage canister

Abstract

Spent nuclear fuel (SNF) assemblies (FAs) contain high-level radioactive waste from operation of nuclear power plants (NPPs). Their safe storage in dry casks is critical for nuclear waste management. However, interim storage or transportation of a canister in a cask, the FAs may be damaged due to human errors, extreme events, or long-term degradation. Therefore, non-destructive evaluation (NDE) methods are needed to monitor their structural integrity over time. In this paper, a novel machine learning (ML)-aided method is proposed to identify different levels of FA damage using measurements taken on the exterior surface of a 2/3-scale experimental canister mock-up. The frequency response functions (FRFs) of the fully loaded canister basket (FLCB) system and the canister with interior damage were computed from recorded input excitation and acceleration response measurements. Machine learning classifiers, including Random Forest (RF), Artificial Neural Networks (ANNs), and Gaussian Naive Bayes (GNB), were trained and tested using features extracted from the FRF difference between an intact FLCB and a damaged canister. The results indicated that the RF model produces the best performance with a testing accuracy of 0.916 and Macro-F1 score of 0.920. Notably, the RF model accurately predicted the damage levels regardless of sensor (i.e., accelerometer) locations. The GNB, however, provided a compromised performance due to its inability to handle correlated features and non-ideal Gaussian distribution of the data. The ANN model exhibited an intermediate performance between the RF and GNB models. Besides, ANN and GNB models had a higher rate of false negatives than the RF models. The accuracy of the model needs to be considered in a real-life implementation because false positives are likely to result in added costs in terms of further inspection while false negatives could potentially result in further degradation and failure.