Scholarlink

<p>Conventional impedance-based structural health monitoring (SHM) is often affected by baseline fluctuations induced by operational vibration and measurement noise, which may generate significant impedance deviations even under healthy conditions and reduce diagnostic reliability. To address this challenge, this study proposes a data-driven SHM framework that integrates piezoelectric electromechanical impedance sensing with a hybrid one-dimensional convolutional neural network and self-attention (1DCNN–Self-Attention) model for robust joint condition assessment. The framework directly processes raw impedance signals without handcrafted feature extraction or signal preprocessing. The 1DCNN automatically extracts local frequency-dependent features, while the self-attention mechanism enhances damage-sensitive information and suppresses vibration-induced disturbances across the frequency spectrum. Experimental validation was conducted on a lab-scale wind turbine tower under impact excitation, blade rotation at different wind speeds, and varying levels of added noise, with progressive bolt loosening introduced to simulate joint degradation. Results show that operational excitation alone can cause significant impedance variations in healthy states, reducing the effectiveness of conventional baseline-dependent approaches. Using the same training protocol and databank, the proposed model consistently outperformed three benchmark deep learning models, including two conventional 1DCNN and a hybrid 1DCNN–LSTM architectures. For the joint distant from the impact excitation, the proposed method achieved 100% damage detection accuracy with a testing RMSE of 0.229, whereas the reference models achieved 78.1%, 31.3%, and 96.9% accuracy, respectively. For the joint with impact excitation effects, the proposed framework maintained 81.2% detection accuracy, significantly higher than the competing models. These results demonstrate that integrating self-attention with impedance sensing substantially improves robustness against vibration-induced baseline variations and enables reliable damage diagnosis using low-cost sensing hardware for wind turbine support structures.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

CNN-self-attention framework for enhanced impedance-based damage detection under vibration and noisy environments

  • 期刊论文

  • 发表日期：

    2026年8月4日

查看全文
  • Thanh-Truong Nguyen,
  • Gia Toai Truong,
  • Thanh-Canh Huynh

摘要

Conventional impedance-based structural health monitoring (SHM) is often affected by baseline fluctuations induced by operational vibration and measurement noise, which may generate significant impedance deviations even under healthy conditions and reduce diagnostic reliability. To address this challenge, this study proposes a data-driven SHM framework that integrates piezoelectric electromechanical impedance sensing with a hybrid one-dimensional convolutional neural network and self-attention (1DCNN–Self-Attention) model for robust joint condition assessment. The framework directly processes raw impedance signals without handcrafted feature extraction or signal preprocessing. The 1DCNN automatically extracts local frequency-dependent features, while the self-attention mechanism enhances damage-sensitive information and suppresses vibration-induced disturbances across the frequency spectrum. Experimental validation was conducted on a lab-scale wind turbine tower under impact excitation, blade rotation at different wind speeds, and varying levels of added noise, with progressive bolt loosening introduced to simulate joint degradation. Results show that operational excitation alone can cause significant impedance variations in healthy states, reducing the effectiveness of conventional baseline-dependent approaches. Using the same training protocol and databank, the proposed model consistently outperformed three benchmark deep learning models, including two conventional 1DCNN and a hybrid 1DCNN–LSTM architectures. For the joint distant from the impact excitation, the proposed method achieved 100% damage detection accuracy with a testing RMSE of 0.229, whereas the reference models achieved 78.1%, 31.3%, and 96.9% accuracy, respectively. For the joint with impact excitation effects, the proposed framework maintained 81.2% detection accuracy, significantly higher than the competing models. These results demonstrate that integrating self-attention with impedance sensing substantially improves robustness against vibration-induced baseline variations and enables reliable damage diagnosis using low-cost sensing hardware for wind turbine support structures.