Enhanced defect detection with autoencoder based analysis for Golay coded thermal wave imaging for inspection of carbon fiber reinforced polymers.

Abstract

Active thermography is increasingly used in non-destructive testing (NDT) due to its ability to inspect materials remotely and reveal subsurface flaws without damaging the structure. Among the various thermographic techniques, pulse compression-based thermal wave imaging has shown promise for its improved sensitivity, depth resolution, and accuracy in identifying hidden defects. This study explores the use of Golay-Coded Thermal Wave Imaging (GCTWI) for detecting internal defects in a carbon fiber reinforced polymer specimen. The sample includes three sections with different thicknesses, each containing engineered slit-shaped flaws. To improve the clarity of defect visualization and accurately assess thickness variations, several post-processing techniques are applied. The GCTWI results are compared using three approaches: traditional pulse compression, principal component thermography, and a deep learning method known as Autoencoder-based Thermography (AET). Key enhancements to the autoencoder's loss function were introduced to better capture defect features in the thermal data. Experimental outcomes show that GCTWI combined with autoencoder-based processing significantly improves defect visibility, especially by increasing the signal-to-noise ratio. Among the tested factors, the non-correlation of Golay codes played a critical role in enhancing defect detection. These results support the integration of coded excitation with AET based processing for advanced NDT applications.