Transforming sensory data into actionable information for timely structural health monitoring (SHM) is essential for the integrity management of pipeline assets. This study develops an automated framework that integrates ultrasonic guided wave testing with deep learning–based data fusion for precise pipeline damage detection. A 20-layer residual network (ResNet) is designed to classify guided wave responses across diverse damage scenarios, including cracks, weld defects, and buried configurations, under varying noise levels. The framework is validated through both finite element simulations and laboratory experiments, demonstrating that the ResNet architecture consistently outperforms support vector machines (SVM) and conventional convolutional neural networks (CNN). The results show near-perfect detection accuracy at high signal-to-noise ratios (≥80 dB) and high robustness under noisy conditions (60 dB). By reducing reliance on manual interpretation and specialized expertise, this study contributes to the automation of pipeline SHM and supports proactive, data-driven identification of structural compromises.
Pan et al. (Fri,) studied this question.