An X-ray fluorescence and machine learning based methodology for the direct non-destructive compositional analysis of (ThU)O fuel pellets

• Published in: Journal of analytical atomic spectrometry Vol. 38; no. 9; pp. 1841 - 185
• Main Authors: Kanrar, Buddhadev, Sanyal, Kaushik, Sarkar, Arnab, Pai, Rajesh V
• Format: Journal Article
• Published: 30.08.2023
• ISSN: 0267-9477; 1364-5544
• DOI: 10.1039/d3ja00158j

A highly sensitive analytical method for elemental quantification of U in (Th 1− x U x )O 2 mixed oxide (MOX) fuel pellets is extremely necessary for nuclear fuel quality control. It will be an added advantage if the analytical method is direct and non-destructive. Presently we have demonstrated a direct non-destructive methodology for (Th 1− x U x )O 2 MOX fuel pellets by the X-ray fluorescence technique using U/Th M lines as the analytical line instead of the well resolved U/Th Lα 1 lines. U/Th M lines were selected as they could be excited using low-energy excitation. In the present study, we have used seven (Th 1− x U x )O 2 MOX fuel pellets with varying U/Th concentrations. All the MOX pellets were prepared via the sol-gel micro-sphere pelletization (SGMP) process. All the pellets were presented for μ-XRF measurements. Each pellet was measured at 10 different spots to construct an input data set. Analytical parameters like relative error and precision obtained from the classical FP-based method utilizing U/Th M lines are 22.4% and 4.9%, respectively. To improve the same parameters, we employed a classical chemometric method like partial least square regression (PLSR). It produced the above-mentioned analytical parameters ∼3.0%. Furthermore, an optimized ANN-based modeling methodology generated a relative error and precision for the U determination in the test sample of 3.1% and 4.9%. The comparative study suggests that both the ANN-based methodology and PLSR outperform the classical FP-based methodology for the analytical quantification of U/Th in MOX by employing U/Th M lines as the analytical line. An artificial neural network based methodology for the direct non-destructive compositional analysis of (Th 1− x U x )O 2 fuel pellets using the X-ray fluorescence technique by considering the U/Th M lines as the analytical line.