Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor | Department of Civil and Environmental Engineering | en_US |
| dc.contributor.advisor | Xia, Yong (CEE) | en_US |
| dc.creator | Du, Yao | - |
| dc.identifier.uri | https://theses.lib.polyu.edu.hk/handle/200/12194 | - |
| dc.language | English | en_US |
| dc.publisher | Hong Kong Polytechnic University | en_US |
| dc.rights | All rights reserved | en_US |
| dc.title | Deep learning empowered data anomaly detection for structural health monitoring systems | en_US |
| dcterms.abstract | Structural health monitoring (SHM) systems have been widely implemented for managing and maintaining civil infrastructure. A large volume of data are generated every day from the SHM systems. It is inevitable that the collected SHM data contain multiple patterns of anomalies caused by sensor faults, system malfunction, communication interference, and harsh operational and environmental conditions. These anomalous data may provide false information for model verification, damage detection, condition assessment and other decision makings. However, modelling the data anomalies is challenging because of their inherent heterogenous patterns, limited anomaly samples and severe class imbalance. Considering the necessity of data anomaly detection, and the difficult and heavy workloads to be accomplished, this study aims to develop new accurate and efficient methods for automatic data anomaly detection. | en_US |
| dcterms.abstract | A transfer learning (TL)-based approach is first investigated for data anomaly detection. Raw sensor data are transformed into time-frequency dual domain images. A pretrained deep convolutional neural network, namely ResNet18, is utilised to learn representative features from visualised data. The adoption of TL strategy avoids training with massive labelled data while maintaining satisfactory learning performance. A new loss function, namely focal loss, is introduced to address class imbalance during the training process. The effectiveness of the proposed method is demonstrated by detecting the abnormal acceleration data of a long-span cable-stayed bridge in China. The proposed approach detects and classifies data anomalies with high accuracy. | en_US |
| dcterms.abstract | Next, a semi-supervised learning (SSL)-based method is developed in order to further mitigate the reliance on labelled data. The proposed method is based on MixMatch, an efficient SSL framework, which could maintain an acceptable learning performance with only a small set of labelled data and massive unlabelled data. Entropy minimisation, consistency regularisation and traditional regularisation techniques are combined into a unified loss function during the model updating process to mine the underlying useful information in the unlabelled data, from which the optimal decision boundaries among different data patterns will be learned. Compared with traditional supervised learning methods, the time and cost on data labelling are significantly reduced. In addition, customised data augmentation (DA) techniques for time series are further developed to enrich the input space of data classes with minor samples. The integration of the DA with SSL further improves the performance of anomaly detection when both labelled and unlabelled data are insufficient. | en_US |
| dcterms.abstract | The last contribution of this thesis is to design and apply a new wireless SHM system for the next-generation wireless SHM paradigm. Three cutting-edge technologies are closely incorporated into the system, including 5G wireless communication, artificial intelligence (AI) embedded edge computing device, and advanced AI algorithms. The new SHM system is designed and implemented in the Hong Kong-Zhuhai-Macao Bridge for technological demonstration. This wireless SHM system is expected to be a new SHM paradigm for future civil infrastructure. | en_US |
| dcterms.extent | xviii, 98 pages : color illustrations | en_US |
| dcterms.isPartOf | PolyU Electronic Theses | en_US |
| dcterms.issued | 2022 | en_US |
| dcterms.educationalLevel | M.Phil. | en_US |
| dcterms.educationalLevel | All Master | en_US |
| dcterms.LCSH | Structural health monitoring | en_US |
| dcterms.LCSH | Structural health monitoring -- Data processing | en_US |
| dcterms.LCSH | Anomaly detection (Computer security) | en_US |
| dcterms.LCSH | Hong Kong Polytechnic University -- Dissertations | en_US |
| dcterms.accessRights | open access | en_US |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/12194