High-resolution defect detection in composite materials using terahertz NDT and deep reconstruction

This study proposes a high-resolution framework for defect detection and structural health monitoring in composite materials using a terahertz (THz)-based non-destructive testing (NDT) approach enhanced by deep image reconstruction. Experimental evaluations were conducted on Kevlar, glass fibre, and carbon fibre composites using THz time-domain spectroscopy. The framework integrates signal processing, machine learning, and statistical modelling to extract features such as absorption, reflection, and defect size. Principal component analysis, Random Forest feature ranking, and K-means clustering were used for defect classification and material characterisation. The proposed framework achieved a detection accuracy of 96.4%, outperforming conventional NDT techniques. Deep image reconstruction improved spatial resolution by 32%, enabling precise identification of micro-cracks, voids, and delamination. Weak-to-moderate correlations were observed between THz spectral features and defect propagation, demonstrating the effectiveness of THz imaging for real-time diagnostics. This research introduces a hybrid THz-based NDT framework combining advanced imaging and machine learning to improve defect quantification, classification, and prediction, supporting predictive maintenance and forensic failure analysis in aerospace, automotive, and civil infrastructure applications.