Enhancing Oxide Film Thickness Measurement Accuracy via Temperature Drift Compensation With GAN-Augmented Data

Accurate measurement of oxide film thickness on nuclear fuel cladding is critical for ensuring reactor safety, as excessive oxidation can degrade thermal conductivity and increase the risk of structural failure. Eddy current testing (ECT) is the predominant technology for measuring oxide film thickness due to its high sensitivity and suitability for in-situ inspections. However, the presence of temperature fluctuations during measurement introduces drift in the ECT signal, compromising its accuracy. This article proposes a deep learning-based framework for temperature drift compensation. A generative adversarial network is used to augment limited experimental datasets by synthesizing physically consistent temperature drift profiles. These augmented datasets are then used to train a random forest (RF) model that performs nonlinear compensation of the ECT signal. A correction approach is developed to mitigate the effects of temperature measurement delay, enhancing compensation performance. Experimental validation demonstrates that the proposed method reduces the maximum measurement error from 28.38μm to 0.98μm and the average error from 26.76μm to 0.62μm. Repeated trials confirm the robustness and reliability of the method. The study provides an innovative technique for accurate measurement of oxide film thickness on nuclear fuel cladding, which can also be extended to other small-sample sensing tasks.