Ultrasonics. 2026 Jun 9;168:108192. doi: 10.1016/j.ultras.2026.108192. Online ahead of print.
ABSTRACT
The strengthening of steel structures with carbon fiber reinforced polymer (CFRP) composites has gained wide acceptance in civil engineering, yet interfacial bonding defects such as inclusion, delamination, and porosity can severely degrade structural performance. Conventional ultrasonic testing often suffers from noise contamination and limited feature extraction, restricting the reliability of defect identification. To address this issue, this study integrates phased array ultrasonic testing (PAUT) with encoder-assisted acquisition and advanced signal processing. Wavelet packet transform (WPT) was applied to denoise and decompose A-scan signals, from which eight statistical energy-based features were extracted to construct discriminative feature vectors. These vectors were then used to train and compare three backpropagation artificial neural network (BP-ANN) variants. The BP-ANN trained with the conjugate gradient algorithm achieved the highest overall accuracy of 95.83% among the three investigated optimization strategies. The results indicate that WPT-based feature engineering improves the discriminative capability of ultrasonic features under controlled laboratory conditions, while encoder-assisted acquisition improves the consistency of PAUT signal collection. The proposed framework provides a feasible approach for intelligent classification of interfacial defects in CFRP-steel hybrid structures and may serve as a reference for future studies involving real manufacturing defects and practical engineering scenarios.
PMID:42430857 | DOI:10.1016/j.ultras.2026.108192
