Abstract
This paper explores the viability of using learning-based state-of-the-art video motion magnification to extract vibrational signatures for damage detection in structures. Unlike previous research, the proposed model uses learning-based video motion magnification rather than implementing hand-designed filters. This change allows the presented approach to detect more subtle sub-pixel movement and thus allows for greater sensibility to vibration. This novel approach is validated and verified on a laboratory structural benchmark under different damage scenarios. Although the learning-based model was trained on a synthetic and non-related image dataset, the experimental results prove that the system is suitable for identifying natural frequencies and operating deflection shapes, thus enabling damage detection algorithms to identify structural damage reliably. The results demonstrate the feasibility and suitability of this novel monitoring technique and thus open an avenue for further research regarding deep learning and its applications to structural health monitoring.
Original language | English |
---|---|
Article number | 112218 |
Number of pages | 10 |
Journal | Measurement: Journal of the International Measurement Confederation |
Volume | 206 |
DOIs | |
Publication status | Published - Jan 2023 |
Keywords
- Computer vision
- Convolutional Neural Network
- Damage identification
- Deep learning
- Frequency response functions
- Natural frequency
- Operating deflection shape
- Structural assessment
- Structural health monitoring
- Vibration testing