Adaptive probabilistic estimation of chromium coating thickness for nuclear fuel cladding using non-parametric regression with ECT sensors

To prevent the catastrophic failure of nuclear power plants following unexpected massive earthquakes and subsequent cooling system collapses, various studies have been conducted on nuclear fuel cladding coating. Consequently, measuring the coating thickness has become crucial, leading to the development of eddy current testing (ECT) sensor-based measurement technologies. These technologies typically extract meaningful features from measured signals and employ pre-built functions to correlate these features with the coating thickness, often using linear or polynomial regressions. However, these regressions can be overly simplistic, yielding inaccurate results. Recently, artificial intelligence has been explored to better model the relationship between the features and thickness. However, many studies on sensor development have not gathered sufficient data to construct deep-learning-based models. In this paper, a non-parametric regression-based regression technique is proposed, which outperforms traditional methods, even with limited data. Unlike traditional regression, which requires manually determining the best-fit order, the proposed method adapts automatically to data characteristics, eliminating the need for expert intervention. Furthermore, a probabilistic regression concept is introduced to account for various uncertainties that have not been addressed in previous related research. The effectiveness and adaptability of the proposed method were validated by applying it to real ECT data for thickness estimation.