Feature extractions by proper orthogonal decomposition and bidirectional gated recurrent units for identifying ship collisions to offshore wind turbine

Liu, Lihua, Liu, Huijuan, Liu, Dianzi and Liu, Fushun (2025) Feature extractions by proper orthogonal decomposition and bidirectional gated recurrent units for identifying ship collisions to offshore wind turbine. Ocean Engineering, 332. ISSN 0029-8018

PDF (Accepted Manuscript) - Accepted Version Restricted to Repository staff only until 5 May 2026. Available under License Creative Commons Attribution Non-commercial No Derivatives. Request a copy

Abstract

Timely detection and termination of ship collisions to offshore wind turbine are essential to ensure the structural integrity and sustainability of marine applications. The collision identification through sensors, such as accelerometers mounted on structures, is simpler and more reliable than non-contact collision methods. However, this approach is susceptible to the environmental complexities, leading to potential misjudgements. To mitigate this issue, a collision identification architecture, called ColliNet, utilising a Bidirectional Gated Recurrent Unit (BiGRU) neural network with the integration of Proper Orthogonal Decomposition (POD) and time-frequency fusion (TFF) techniques, is proposed for the impact analysis of ship-offshore wind turbine. This architecture is realized by two phases of feature processing in prediction and classification, respectively. Firstly, the critical features of time evolution coefficients are extracted from the acceleration response predicted by the POD feature extraction block. Afterwards, the fused time-frequency domains of the time evolution coefficient features are fed into the BiGRU feature classification block to further extract the features related to the operation condition. ColliNet, constrained by user-defined loss functions with physical information imposed, reliably explains the complex response patterns of the wind turbines under long-term operating conditions. It also reduces the spatial complexity of the signal due to the removal of structure-related modal information. The effectiveness of the developed architecture is validated by numerical simulations using OpenFAST with the implementation of ship-turbine collision module. The collisions under different operating conditions are successfully identified with an accuracy improvement of over 6.7% on the testing dataset compared to the BiGRU network. Moreover, lab-scale collision tests are conducted on an NREL 5-MW monopile wind turbine (1:70 scale) to further demonstrate the validity of the ColliNet architecture. The proposed architecture is able to provide reliable support for maintenance measures and helps to realize the autonomous monitoring in replacement with visual inspection using high-definition camera systems, thereby mitigating the operation and maintenance cost of wind turbines while improving their reliability and sustainability.

Item Type:	Article
Uncontrolled Keywords:	sdg 14 - life below water ,/dk/atira/pure/sustainabledevelopmentgoals/life_below_water
Faculty \ School:	Faculty of Science > School of Engineering, Mathematics and Physics
UEA Research Groups:	Faculty of Science > Research Groups > Materials, Manufacturing & Process Modelling Faculty of Science > Research Groups > Sustainable Energy
Depositing User:	LivePure Connector
Date Deposited:	09 May 2025 14:30
Last Modified:	09 May 2025 14:30
URI:	https://ueaeprints.uea.ac.uk/id/eprint/99233
DOI:	10.1016/j.oceaneng.2025.121441

Actions (login required)

View Item