Unsupervised learning-based damage detection of mooring lines in floating bridges
Floating bridges offer a practical alternative to sea-crossing bridges in regions with deep water and poor seabed conditions. It consists of a superstructure and substructure that includes piers, pontoons, and mooring lines. Damage to individual mooring lines can trigger progressive failures in othe...
Saved in:
| Published in: | Ocean engineering Vol. 343; p. 123604 |
|---|---|
| Main Authors: | , , |
| Format: | Journal Article |
| Language: | English |
| Published: |
Elsevier Ltd
15.01.2026
|
| Subjects: | |
| ISSN: | 0029-8018 |
| Online Access: | Get full text |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Floating bridges offer a practical alternative to sea-crossing bridges in regions with deep water and poor seabed conditions. It consists of a superstructure and substructure that includes piers, pontoons, and mooring lines. Damage to individual mooring lines can trigger progressive failures in other lines, potentially leading to catastrophic accidents. However, direct inspection and monitoring are extremely challenging because mooring lines are located beneath the water surface. To address this problem, this study proposes a deep-learning-based damage-detection methodology that leverages the measurable structural responses of a floating body. Given the difficulty of acquiring damage condition data, the approach is grounded in unsupervised learning trained exclusively on intact condition data. The proposed model adopts an encoder–decoder architecture with long short-term memory (LSTM) networks and integrates multi-head self-attention (MHSA) to extract representative features and reconstruct input time-series data. To verify its effectiveness, the model was compared with a conventional long short-term memory autoencoder (LSTM-AE) that is widely used for unsupervised time-series damage detection. Under untrained irregular wave conditions, the MHSA-LSTM-AE extracted temporal–spatial features more effectively than the standard LSTM-AE. As a result, the proposed model not only distinguishes between intact and damaged conditions, but also correctly identifies the pontoon where mooring line damage occurs. This capability enables rapid localization of failures, which is critical for a timely response. The model achieved high accuracy under single-mooring line failure scenarios; however, its performance was relatively lower in cases of partial top damage compared to complete failure.
•An unsupervised deep learning method is proposed for detecting mooring line failures in floating bridges.•The model employs an encoder–decoder LSTM architecture enhanced with multi-head self-attention (MHSA).•Validation datasets are generated from time-domain hydrodynamic simulations of a reference floating bridge.•The method detects single-line failures and localizes damaged pontoons, though partial damage cases remain challenging. |
|---|---|
| ISSN: | 0029-8018 |
| DOI: | 10.1016/j.oceaneng.2025.123604 |