Learning multi-dimensional sensor relationships for robust marine pipeline leakage non-destructive detection via adaptive graph networks
Lu, Y.; Li, Y.; Liu, H.; Zhang, Y.; Wang, X.; Chen, M.; Zhao, C.; Abdel Wahab, M.A. (2026). Learning multi-dimensional sensor relationships for robust marine pipeline leakage non-destructive detection via adaptive graph networks. Eng. Struct. 350: 121983. https://dx.doi.org/10.1016/j.engstruct.2025.121983
In: Engineering structures. IPC Science and Technology Press: Guildford. ISSN 0141-0296; e-ISSN 1873-7323, meer
|  |
| Author keywords |
Marine pipeline leakage detection;Sensor networks;Acoustic Emission;Adaptive graph networks;platform |
| Auteurs | | Top |
- Lu, Y.
- Li, Y.
- Liu, H.
- Zhang, Y.
|
- Wang, X.
- Chen, M.
- Zhao, C.
- Abdel Wahab, M.A., meer
|
|
| Abstract |
Marine pipeline leakages pose severe environmental and economic threats, demanding accurate and robust detection systems. Sensor network-based detection faces three key challenges: highly dynamic network structures, weak leak signals with complex sensor correlations, and strong marine environmental noise. Existing methods have struggled significantly with these challenges. To address these limitations, this paper proposes a novel Multi-dimensional Adaptive Graph Integration Framework (MAGI). Within MAGI, an innovative Multidimensional Sensor Edge Interaction (MSEI) layer is introduced. MSEI tackles time-varying network modeling by dynamically integrating local interactions and global context, thereby generating rich multi-dimensional edge features that overcome the constraints of traditional single-weight edges. Furthermore, the framework incorporates a Feature Aggregation Network (FAN). FAN employs a parallel graph convolutional architecture and adaptive gating mechanisms, achieving noise robustness and efficient heterogeneous data integration. Experimental results validate MAGI's superior performance. It achieves up to 97 % accuracy across diverse pressures, maintains over 95 % stability in various pipeline topologies, and exceeds 90 % accuracy under extreme noise conditions. MAGI also demonstrates extremely low processing latency and excellent generalization on highly imbalanced data. With only 0.599 M parameters, 76 % fewer than state-of-the-art Graph Neural Networks (GNNs), MAGI is highly efficient. This work establishes a new technical paradigm for real-time marine pipeline leak monitoring and early warning. |
|
