Delamination Assessment of Glass-Fiber Reinforced Polymer Using Microwave Technique

Composite insulations are routinely deployed as protective measures against high-temperature environments and corrosive influences on metal substrates. However, aging and cyclic processes cause defects, including metal substrate-insulating layer delamination. Despite the growth of artificial intelligence-driven microwave approaches for defect detection, precise quantification of defect severity, particularly in terms of size, presents a low accuracy. These challenges are due to the intrinsic irregularity characterizing glass fiber reinforced polymers (GFRP), coupled with the presence of outliers within microwave measurement data. This study presents a novel microwave approach with a k-medoid clustering algorithm to improve the reliability of defect detection. This method measures and analyzes wave reflection from a waveguide probe operating at 18–26.5 GHz, scanning on a coated metal substrate. A well-calibrated vector network analyzer (VNA) evaluates the reflected wave responses, which are then analyzed using a Gaussian filter and the k-medoids clustering technique. The acquired microwave data are transformed to the time domain to reveal subsurface defects hidden behind the insulation. The k-medoid technique uses these properties to cluster data for accurate defect inspection and sizing. By eliminating dataset outliers, defect size assessment beneath insulating layers is improved. The proposed algorithm is tested using measured data for validation. The proposed algorithm achieves an accuracy of 81.21% in predicting defect size and can locate defects down to 1 mm depth. The proposed algorithm emerges as a robust and promising model for defect detection, providing substantial contributions across diverse industrial realms.