Physical-guided graph deep learning for composite pipelines structural health monitoring

Abstract

Composite pipelines, widely applied in the oil and gas industry due to their high strength-to-weight ratio and corrosion resistance, are nonetheless vulnerable to damage under harsh operating conditions, making structural health monitoring (SHM) crucial for enhancing safety and preventing potential failures. However, the complex service environment and anisotropic properties of composite pipelines pose challenges for traditional SHM methods, which often struggle to extract underlying features and identify structural defects. To address these challenges, a new graph deep learning method, termed physical-guided graph deep learning (PGGDL), which leverages physical rules to enhance defect detection accuracy in composite pipeline health monitoring, is proposed in this study. The PGGDL method constructs graph data by integrating signals based on physical rules, including the sensor arrangement and the guided wave propagation mechanism, enabling efficient data fusion. Meanwhile, the PGGDL constructs a temporal model and a graph attention network to capture both node-level and global spatial features. Furthermore, an adaptive learning rate strategy is designed to dynamically adjust the learning rate, improving training efficiency, achieving faster convergence, and enhancing model stability. Four experimental cases are conducted to demonstrate that the PGGDL outperforms traditional deep learning models in accurately detecting and localizing pipeline defects with limited data.