Debonding imaging in fibre reinforced concrete columns by deep learning assisted-guided wave technique

This study proposes a novel deep learning-assisted framework for detecting interfacial debonding defects in fibre-reinforced polymer (FRP)-wrapped concrete columns using ultrasonic guided waves. Traditional non-destructive testing methods face significant challenges in curved and anisotropic structures due to complex wave dispersion and multimodal propagation characteristics. To overcome these limitations, we developed an advanced hybrid deep neural network (DNN) architecture that synergistically combines time-domain ultrasonic signals with an enhanced elliptical imaging algorithm (ELIA) to achieve superior defect localization accuracy. A finite element (FE) model was established to simulate guided wave propagation in FRP-strengthened concrete columns, generating synthetic training samples that capture diverse debonding scenarios. The proposed DNN employs a dual-encoder structure to extract both temporal and spatial features, followed by a decoder with skip connections for precise damage reconstruction. Experimental validation on FRP-retrofitted concrete specimens with artificially induced debonding demonstrated the model’s robust performance, achieving accurate defect localization and shape prediction despite variations in real-world conditions. Comparative analysis revealed significant improvements over conventional ELIA, particularly in suppressing imaging artifacts and enhancing edge definition. This research contributes an efficient, cost-effective solution for structural health monitoring (SHM) by leveraging simulated data to minimize experimental requirements while maintaining high detection reliability. The framework shows promising potential for practical implementation in civil infrastructure monitoring systems.