Abstract
A critical task for the reliability analysis and risk management of offshore wind turbines (OWT) is to accurately and efficiently predict the fatigue crack propagation over the service life. To realize the goal, a novel long and short-term network (LSTM)-based deep learning model integrated with full probabilistic perspectives is proposed to effectively handle time-varying and multi-source uncertainties associated with long-term fatigue crack propagation in OWTs. Based on the identification of the imbalance in the instance set and the inconsistency of the prediction model within the feasible regions of multiple uncertain parameters, a multi-bin progressive self-supervised learning (MPSL) framework is formulated afterwards. The trustworthiness of this framework is validated by the investigations on the fatigue crack propagation prediction of the National Renewable Energy Laboratory (NREL) 5 MW OWT. Our findings demonstrate significant gains in prediction accuracy and efficiency, juxtaposed with the traditional Paris model-based numerical simulation framework. Ultimately, the proposed trustworthy MPSL framework offers the stakeholders a robust tool for identifying the OWT fatigue crack propagation, advancing early risk perception and management in practice engineering.
Original language | English |
---|---|
Article number | 119739 |
Journal | Ocean Engineering |
Volume | 314 |
ISSN | 0029-8018 |
DOIs | |
Publication status | Published - 15 Dec 2024 |
Bibliographical note
Publisher Copyright:© 2024 Elsevier Ltd
Keywords
- Deep learning
- Fatigue crack propagation prediction
- Imbalanced learning
- Model inconsistency
- Multi-bin analysis
- Multi-source uncertainty