Related Experiment Video
Updated: Jul 16, 2026
07:46
Data Acquisition Protocol for Determining Embedded Sensitivity Functions
Published on: April 20, 2016
A Novel Residual Dual Attention Multiscale Network for Vibration-Based Damage Recognition in Floating Wind Turbine
Huiming Han1, Yifei Li2, Renqiang Wang1
1School of Nautical Technology, Jiangsu Maritime Institute, Nanjing 210024, China.
Sensors (Basel, Switzerland)
|July 15, 2026
Summary
A new Residual Dual Attention Multiscale Network (RDAMNet) effectively monitors the structural health of floating wind turbines (FWTs). This advanced AI model accurately detects damage by analyzing multi-scale vibration signals, ensuring safer and more efficient deep-sea energy generation.
Area of Science:
- Engineering
- Artificial Intelligence
- Renewable Energy
Background:
- Floating wind turbines (FWTs) are crucial for deep-sea clean energy, but their structural health is vital for safety and output.
- FWT vibration signals present complex non-stationary and multi-scale characteristics, challenging existing damage detection methods.
- Current techniques struggle to adequately extract and integrate damage-sensitive features across various temporal scales.
Purpose of the Study:
- To propose a novel network, the Residual Dual Attention Multiscale Network (RDAMNet), for enhanced structural health monitoring of FWTs.
- To address the limitations of existing methods in extracting and fusing multi-scale damage-sensitive features from FWT vibration signals.
- To develop a robust and efficient solution for intelligent structural health monitoring in complex oceanic environments.
Main Methods:
- Developed RDAMNet featuring a signal-level multi-scale decoupling strategy for extracting damage-sensitive features from complementary signal representations.
- Implemented a multi-branch differentiated architecture to process features at different scales.
- Integrated an ECA-SE dual attention mechanism to enhance damage-related channel responses during feature extraction and fusion.
Main Results:
- RDAMNet achieved a mean damage recognition accuracy of 95.39% and a weighted F1-score of 95.37% on a public dataset, outperforming five other methods.
- Cross-condition generalization experiments showed RDAMNet maintained mean accuracies above 94% across varying wind speeds and directions.
- The network demonstrated a favorable performance-efficiency trade-off with 663,783 parameters and a 5.35 ms single-sample GPU inference time.
Conclusions:
- RDAMNet offers an effective technical approach for intelligent structural health monitoring of FWTs.
- The proposed method excels in extracting and fusing multi-scale damage-sensitive features, crucial for complex oceanic environments.
- RDAMNet demonstrates high accuracy, stability across operating conditions, and computational efficiency, validating its practical applicability.
