Vibration Measurement of Wind Turbine Tower Based on Computer Vision

The use of contact sensors in structural health monitoring presents several challenges, including inconvenience and low spatial resolution, which negatively impact the accuracy of structural health assessments. Recently, visual-based vibration measurement method have gained significant attention in the field of structural health monitoring. However, these methods face limitations such as the need for target installation, poor algorithm robustness, and susceptibility to failure in complex environments. In this paper, we propose a deep learning method to extract the dynamic response signals of a wind turbine tower. Specifically, we introduce a Faster R-CNN-based feature point detection and tracking model, combined with a Kalman filter for signal processing. The effectiveness of our method is validated through vibration measurement experiments conducted on a wind turbine tower model. The results show that the errors in both time and frequency domains are around $1 \%$ compared to the standard sensor. And our method shows good accuracy and stability under different lighting conditions, with accuracy kept within $\mathbf{2 \%}$.