A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
A deep learning-based method for structural modal analysis using computer vision
Abstract Structural modal analysis aims to determine a structure's natural frequency, damping ratio, and mode shape, helping with structural condition assessment and maintenance. In this study, a computer vision-based framework for the identification of structural modal parameters is developed, which consists of two main procedures: First, the one-dimensional (1D) vibration signals of edge pixels on the structure in the video are extracted via edge detection and optical flow theory. Second, a 1D convolutional neural network (CNN) coupled with long short-term memory (LSTM) is generated to extract structural modal parameters from the input 1D signal. The framework's performance has been validated through comparison with baseline values, which were obtained from contact sensors. Additionally, the model's robustness and extrapolability has been analyzed. The good performance of the computer vision-based approach confirms its potential for precise and dependable contact-free modal analysis.
Highlights A deep learning-based method for structural modal analysis using computer vision was proposed. The constructed 1D-CNN-LSTM network was well trained with 1D displacement signals instead of videos. A three-storey acrylic frame specimen experiment was conducted to verify the performance of the proposed method.
A deep learning-based method for structural modal analysis using computer vision
Abstract Structural modal analysis aims to determine a structure's natural frequency, damping ratio, and mode shape, helping with structural condition assessment and maintenance. In this study, a computer vision-based framework for the identification of structural modal parameters is developed, which consists of two main procedures: First, the one-dimensional (1D) vibration signals of edge pixels on the structure in the video are extracted via edge detection and optical flow theory. Second, a 1D convolutional neural network (CNN) coupled with long short-term memory (LSTM) is generated to extract structural modal parameters from the input 1D signal. The framework's performance has been validated through comparison with baseline values, which were obtained from contact sensors. Additionally, the model's robustness and extrapolability has been analyzed. The good performance of the computer vision-based approach confirms its potential for precise and dependable contact-free modal analysis.
Highlights A deep learning-based method for structural modal analysis using computer vision was proposed. The constructed 1D-CNN-LSTM network was well trained with 1D displacement signals instead of videos. A three-storey acrylic frame specimen experiment was conducted to verify the performance of the proposed method.
A deep learning-based method for structural modal analysis using computer vision
Liu, Yingkai (author) / Cao, Ran (author) / Xu, Shaopeng (author) / Deng, Lu (author)
Engineering Structures ; 301
2023-12-02
Article (Journal)
Electronic Resource
English
Convenient Structural Modal Analysis Using Noncontact Vision-Based Displacement Sensor
| British Library Conference Proceedings | 2016
Structural Damage Detection of Steel Corrugated Panels Using Computer Vision and Deep Learning
| Springer Verlag | 2024