Unsupervised structural damage detection and localization using deep learning and machine learning
| dc.contributor.author | Wang, Zilong | |
| dc.contributor.examiningcommittee | Svecova, Dagmar (Civil Engineering) Wang, Yang (Computer Science) Gupta, Rishi (University of Victoria) | en_US |
| dc.contributor.supervisor | Cha, Young-Jin (Civil Engineering) | en_US |
| dc.date.accessioned | 2021-05-27T13:14:50Z | |
| dc.date.available | 2021-05-27T13:14:50Z | |
| dc.date.copyright | 2021-05-03 | |
| dc.date.issued | 2021 | en_US |
| dc.date.submitted | 2021-05-03T20:51:17Z | en_US |
| dc.degree.discipline | Civil Engineering | en_US |
| dc.degree.level | Doctor of Philosophy (Ph.D.) | en_US |
| dc.description.abstract | Many data-driven approaches have been developed in recent decades to address problems with damage detection for civil infrastructure. According to training modes of the statistical models or neural networks adopted in the studies, these data-driven damage detection methods can be roughly categorized into supervised modes and unsupervised modes. Supervised damage detection approaches require the recorded data (i.e., ground truth data) from the undamaged and various damaged structural scenarios to train statistical models or neural networks. Then, the trained models or networks can be utilized to detect damage using future data measured from unknown structural scenarios. However, acquiring numerous training datasets from various damage scenarios for the monitored structures is time-consuming and costly, and it is hard to obtain many damage scenarios for the infrastructures in service. To address these challenges encountered in practice, structural damage detection in unsupervised learning mode has become increasingly interesting to researchers. The proposed unsupervised damage detection methods in my study require only the data measured from undamaged structural scenarios or baseline structures in their training processes. This thesis aims to propose novel unsupervised damage detection methods to address the problems facing structural damage detection and localization. Specifically, a novel unsupervised damage detection approach using a deep learning technique is proposed for detecting damage in a simulated multi-story frame and a laboratory-scale steel bridge model in Chapter 3. Additionally, a comparative study with an advanced unsupervised damage detection approach using deep restricted Boltzmann machines is carried out to evaluate their effectiveness of detecting light damage in the steel bridge. In Chapter 4, an unsupervised novelty detection method based on an original technique of fast clustering is developed to roughly locate the damage positions in a small-scale building frame. To verify the effectiveness of the developed method for structural damage localization, several existing machine learning and deep learning methods are developed and converted to the uniform unsupervised novelty detection mode in Chapter 5 for extensive comparative studies. | en_US |
| dc.description.note | October 2021 | en_US |
| dc.identifier.uri | http://hdl.handle.net/1993/35666 | |
| dc.language.iso | eng | en_US |
| dc.rights | open access | en_US |
| dc.subject | Unsupervised damage detection, Structural damage localization, Damage sensitive feature, Deep learning, Machine learning | en_US |
| dc.title | Unsupervised structural damage detection and localization using deep learning and machine learning | en_US |
| dc.type | doctoral thesis | en_US |