Defects under insulation evaluation using convolutional neural network-based microwave technique

The detection of subsurface defects has increasingly benefited from the integration of machine learning techniques, particularly in data-driven inspection methods. While convolutional neural networks (CNNs) have shown promising capabilities, their performance in identifying fine-scale defects remains suboptimal. This study proposes a microwave nondestructive testing framework integrating Q-band open-ended rectangular waveguide sensing with short-time Fourier transform (STFT) based time-frequency feature extraction and CNN classification to improve the detection of small-scale delamination beneath ceramic insulation. The methodology involves capturing reflected signals from ceramic insulation using an open-ended rectangular waveguide operating between 33 and 50 GHz. These reflections undergo preprocessing via a hybrid signal processing analysis, wherein the STFT extracts localized frequency-dependent features. Outlier suppression and data normalization are performed using the Z-score method to enhance data quality. The refined features are then input into a CNN classifier trained to distinguish between defective and non-defective regions. The findings reveal that this integrated approach achieves a classification accuracy of 97.84%, demonstrating a notable enhancement in detecting subtle delamination compared to conventional inspection techniques.