Publication: Delamination under insulation assessment using modified k-medoids based microwave non-destructive testing method.
No Thumbnail Available
Date
2025-03
Authors
Tan, Shin Yee
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Composite materials are becoming more prominent in aerospace, automotive, construction, and oil and gas industries. However, ageing and cyclic processes result in delamination between insulation and metal, cracks, and corrosion on metal. Although microwave non-destructive testing techniques have emerged as viable tools for evaluating the defects under dielectric insulation, they often fail to estimate the defect size and depth accurately. To address this, a hybridisation of ridge waveguide with a machine learning algorithm is proposed. The ridge waveguide, with a large bandwidth and narrow probe aperture, improves the spatial image quality and the sensitivity of defect detection, sizing and depth estimation. The ridge waveguide is employed to perform raster scanning using an XYZ positioner on the surface of low loss glass ceramic (Macor) insulated samples. The magnitude or phase of the microwave reflection coefficients obtained during the scanning is studied. Simulation and measurement data are obtained by leveraging a ridge waveguide with 101 frequency points and an operating frequency between 18 to 40 GHz. The microwave reflection coefficients are analysed using the time domain and principal component analysis (PCA). A modified k-medoids clustering algorithm with weighted Euclidean distance is developed to detect the defects under insulation. Additionally, defect depth prediction is achieved using the Manhattan distance. Key findings include the accurate detection of delamination defects as small as 5 mm × 5 mm, achieving a 97.42 % accuracy rate compared to the actual delamination size. Meanwhile, the proposed method is capable of predicting the depth of delamination down to 0.05 mm with an error rate of 4.15 %. The proposed method presents a significant advancement in comparison to the past methods in detecting defects under insulation in terms of sizing and depth estimation, which is important for predictive maintenance as part of the industry 4.0 scope.