Physics-informed dual-network framework for wire-break anomaly imaging in prestressed concrete cylinder pipes

Prestressed Concrete Cylinder Pipes (PCCPs) are widely used in large-diameter and long-distance water conveyance systems, where progressive wire breakage poses a major threat to structural safety. Existing monitoring approaches, including electromagnetic inspection, acoustic emission sensing, and data-driven DAS-based classification, are often limited by discrete measurements, insufficient physical interpretability, and weak robustness under varying field conditions. To address these limitations, we propose a physics-informed vibroacoustic inversion framework for PCCP wire-break-induced anomaly imaging using Distributed Acoustic Sensing (DAS) and a Gabor-enhanced Physics-Informed Neural Network (Gabor-PINN). Unlike empirical filtering or threshold-based DAS signal processing, the proposed framework converts DAS observations into a physics-constrained wavefield and anomaly image. To make the inverse problem computationally tractable, the cylindrical pipe surface is represented using a reduced-order unwrapped model with circumferential periodicity and an axial absorbing treatment. In this model, wire-break-induced damage is described by an equivalent velocity perturbation that represents the acoustic manifestation of localized prestress loss and stiffness disturbance, rather than the complete mechanics of wire fracture. A dual-network inversion architecture is then constructed to jointly reconstruct the scattered wavefield and the equivalent velocity perturbation from derivative-form DAS observations. To overcome the spectral bias of conventional PINNs, localized Gabor basis functions are introduced to improve the representation of impulsive, broadband, and highly oscillatory wavefields. The results show that the proposed Gabor-PINN achieves better convergence and higher reconstruction accuracy than the vanilla PINN and Fourier-feature PINN. Experimental validation on a full-scale PCCP prototype further demonstrates that the proposed framework can extract physically consistent wavefield information from physics-guided preprocessed DAS data and localize wire-break-induced anomalies. The method transforms wire-break monitoring from event classification into physics-informed anomaly imaging, thereby providing a physically interpretable tool for structural health assessment, which is potentially useful for future digital-twin updating after calibration with higher-fidelity structural models.