Damage identification of spatial structures based on machine learning techniques
- Publication Type:
- Thesis
- Issue Date:
- 2024
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
Spatial structures are widely used in large buildings and bridges. These structures are usually composed of thousands of members and joints. The damage of members and joints may result in changes in the loading path, reducing the load-bearing capacity and even progressive collapse of the entire structure. This study deals with the critical challenge of structural health monitoring (SHM) for large spatial bridge structures. With machine learning techniques, this work introduces a comprehensive approach including supervised, semi-supervised, and unsupervised methods to refine and enhance structural damage detection capabilities.
Firstly, a novel generic element for nonuniform beams with semi-rigid joints is developed, addressing a main error that caused by assuming semi-rigid joints as rigid and nonuniform members as uniform elements. The proposed approach is proven to improve the accuracy of numerical modelling and damage detection for spatial structures. Moreover, an updated Mean Square Error (UMSE) loss function is proposed, specifically designed for supervised deep-learning-based structural damage detection problems. It is proved to achieve significant high convergence speed over the traditional MSE. Additionally, the NMSEC- PCA (normalized modal strain energy change index processed by PCA) index is proposed as damage feature, resulting significant low computational costs while achieving faster convergence and high detection accuracy. Furthermore, to enable the structural health monitoring in practice, a semi-supervised learning frame is proposed for structural damage detection, together with indices used for damage quantification. The efficacy and applicability of the proposed method are validated using the experimental bridge model, showing their practicality for real-world scenarios. Finally, Continuous Wavelet Transform (CWT) based damage quantification of bridges using a convolutional Variational Autoencoder (VAE) is implemented, demonstrating the VAE's effectiveness in accurately reconstructing images and damage quantification, bridging the gap between theoretical models and real-world structural health monitoring applications.
Overall, the thesis presents comprehensive research using supervised, semi supervised, and unsupervised methods for structural damage detection, all illustrated using numerical and experimental bridge models. The study provides machine learning based practical solutions for structural damage detection that are essential for maintaining the safety and integrity of large-scale spatial structures.
Please use this identifier to cite or link to this item: