• Dual-channel self-supervised terahertz signal recovery without clean data. • Label-free 3D defect visualization of voids in composites. • Six-axis terahertz scanning enables full-surface inspection of cylindrical parts. • Coupling-free nondestructive testing solution for industrial quality control of polytetrafluoroethylene components. Glass fiber-reinforced polytetrafluoroethylene (GFR-PTFE) composites are widely used in chemical engineering due to their exceptional chemical resistance and dielectric properties. However, nondestructive testing (NDT) of internal defects in such materials remains challenging. Terahertz (THz) imaging offers a non-contact and non-ionizing alternative with high sensitivity to subsurface structures, but its effectiveness in GFR-PTFE is severely limited by strong material absorption, resulting in low signal-to-noise ratio (SNR) and shallow penetration. To overcome these limitations, we present a Dual-Channel Self-supervised THz Signal Reconstruction (DCSR) framework for high-fidelity signal recovery without clean references or labeled data. By exploiting spatial redundancy in densely sampled 3D scans, DCSR employs two complementary input pathways to simultaneously suppress random noise and fixed pattern noise through self-supervised training, thereby enhancing signal integrity and temporal resolution. Additionally, a customized six-axis THz scanning system has been further developed, which performs full-surface inspection on cylindrical components. Experimental results on real GFR-PTFE samples demonstrate an SNR gain of 9.77 dB (114% improvement) and clear visualization of subsurface defects in reconstructed 3D volumes. This work presents a label-free, contactless and nondestructive solution for high precision inspection of strongly absorbing composites, potentially contributing to the industrial applicability of THz technology in quality control.
Han et al. (Wed,) studied this question.