A THz Detection Method for GFRP Delamination Based on THz Time-Domain Signal Model and Transfer Learning

Composite materials, with their superior properties, have found extensive applications in aerospace and other fields. Compared to metallic materials, composite materials pose more stringent and specific requirements for non-destructive testing technologies. Data-driven terahertz (THz) non-destructive testing techniques have shown great potential in identifying the composite material damage. However, most previous data-driven methods rely on a sufficient amount of labeled data and assume that the training data (source domain) and test data (target domain) follow similar distributions. Due to variations in the testing environments, materials, and other conditions, THz signals can exhibit significant differences, leading to decreased performance of the models in the target domain. We propose a THz detection technology for composite materials using a THz time-domain signal model and deep domain adaptation, capable of imaging delamination defects without labeled data training. A THz echo model was developed to generate a simulation dataset, addressing dispersion in multi-layered composites. Transfer learning was applied to bridge the gap between THz signals and simulations. Our method shows superior generalization and accuracy in defect recognition compared to traditional deep learning models, as evidenced by experimental results.