Offshore structures, such as monopile Offshore Wind Turbines (OWTs), are subjected to various dynamic loads including waves, wind, and operational vibrations, which can lead to different types of damage. A key consideration in Structural Health Monitoring (SHM) for offshore structures is how soil-structure interaction influences vibration-based damage detection systems. Extracting features manually from vibration signals is often complex, time-consuming, highlighting the need for automatic methods that can learn relevant features straight from raw data. This paper presents a novel vibration-based method for automatic feature learning and damage detection in offshore structures, taking soil interaction into account. A combined deep Convolutional Neural Network and Long Short-Term Memory (CNN-LSTM) network is developed to extract the relevant features from vibration signals reconstructed using the Variational Mode Decomposition (VMD) technique. Integrating the LSTM network with the CNN enhances the detection accuracy and stability while reducing the oscillation. Notably, the proposed method applies VMD-reconstructed vibration signals directly to the deep CNN-LSTM network without requiring separate feature extraction or selection. The VMD technique removes irrelevant components of the vibration signals that do not pertain to the structure’s nature, thereby refining the signals for a more accurate representation of the structure’s condition. The suggested method is verified utilizing experimental data from a lab-scale monopile offshore model that incorporates soil interaction. Vibration data were collected using various accelerometer sensors across different states, including one healthy state and eight damaged states. The results demonstrate that the proposed method effectively learns features from reconstructed vibration data and outperforms comparative methods, making it a promising approach for SHM system development in offshore structures.
扫码关注我们
求助内容:
应助结果提醒方式:
