Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor | Department of Civil and Environmental Engineering | en_US |
| dc.contributor.advisor | Ni, Yiqing (CEE) | en_US |
| dc.creator | Tan, Yanke | - |
| dc.identifier.uri | https://theses.lib.polyu.edu.hk/handle/200/14243 | - |
| dc.language | English | en_US |
| dc.publisher | Hong Kong Polytechnic University | en_US |
| dc.rights | All rights reserved | en_US |
| dc.title | Enhancing structural health monitoring through advanced time series-targeted neural networks : from sensor placement to data filtering, inference, and diagnosis | en_US |
| dcterms.abstract | With the accelerated advancement of global urbanization, the coexistence of large-scale infrastructure construction projects and aging existing structures has imposed more severe challenges on structural health monitoring (SHM). The inherent conflict between continuous investment in infrastructure and the demand for rational lifecycle maintenance further highlights the necessity to develop advanced SHM techniques. Current SHM systems still face multiple technical bottlenecks in sensor placement optimization, data denoising, damage identification, and decision-making: sensor placement relies on empirical experience without comprehensive optimization criteria balancing global and local requirements; conventional signal preprocessing methods employing simple frequency-domain analysis struggle with data loss and extreme noise contamination; most damage diagnosis approaches focusing on macro-scale dynamic characteristics like modal parameters exhibit limited sensitivity to early-stage damage. Even machine learning (ML)-based methods suffer from insufficient accuracy, weak generalization capability, and incompatibility with massive monitoring data. This study proposes to reframe SHM data mining as time series analysis tasks, systematically optimizing key SHM technologies through proposing and improving neural networks (NNs) specialized for sequential data, thereby enhancing automation and intelligence in structural safety assessment. The specific research contributions are outlined as follows. | en_US |
| dcterms.abstract | For sensor optimization, a novel optimization criterion integrating local signal representativeness and global modal independence is proposed, with a multi-phase heuristic optimization algorithm addressing combinatorial optimization problem. An automatic candidate selection method based on structural characteristic matrices and graph data theory is developed. In the accelerometer deployment for Rafael Cloud Corridor, the proposed criterion successfully achieved most accurate reconstruction of structural acceleration fields. The improved search algorithm demonstrated faster convergence and higher-quality local optima compared to conventional evolutionary algorithms. By simplifying structures as graphs, the node ranking algorithm enables interpretable automated generation of candidate sets with arbitrary quantities. | en_US |
| dcterms.abstract | In data preprocessing, an integrated solution incorporating dynamic adaptive denoising, missing data reconstruction, and outlier detection is established. An improved Kalman filter with a state transition model dynamically adjusts filtering parameters based on sensor performance, effectively balancing frequency retention and noise suppression while avoiding rigid manual thresholds. The bidirectional echo state network (Bi-ESN) achieves high-precision reconstruction of extreme missing scenarios (including long-term and multi-channel losses) in Canton Tower monitoring data by simultaneously leveraging historical-future information and cross-channel correlations. A frequency-decomposition anomaly detection framework combining generative models and local outlier factor algorithms precisely identifies various outliers in Shanghai Tower monitoring data with distinct types, quantities, distributions, and abnormality levels, while enabling reasonable signal recovery. | en_US |
| dcterms.abstract | Data inferencing tasks focus on three application scenarios: prediction, synchronized mapping, and physical system solving. For time series forecasting, a memory-enhanced network with dynamic feature extractors achieves accurate wake flow prediction above tracks for wind barriers along the Lanzhou-Xinjiang high-speed railway. Addressing synchronized mapping between homologous variables, innovative random reservoir-based positional embedding breaks network input constraints, realizing vibration-based far-field noise substitution measurement in Shenzhen Metro. For differential equation-governed physical systems with environment-dependent correlations, a general generalization performance expanding structure successfully solves chaotic particle damper (PD) excitation-reaction mapping, establishing new paradigms for complex system calibration. | en_US |
| dcterms.abstract | The assessment phase encompasses damage detection and condition evaluation. A supervised multi-dimensional multi-scale convolutional classifier achieves precise multi-position combined damage localization and severity quantification on a three-span model bridge, complemented by a redundant data screening module enhancing noise robustness while reducing dependency on complete damage datasets. An updatable digital twin (DT) framework utilizing long-term operational data enables unsupervised rapid structural assessment, guiding post-disaster rehabilitation. Transfer learning facilitates convenient model adaptation to structural evolution over extended service periods. | en_US |
| dcterms.extent | xxv, 240 pages : color illustrations | en_US |
| dcterms.isPartOf | PolyU Electronic Theses | en_US |
| dcterms.issued | 2025 | en_US |
| dcterms.educationalLevel | Ph.D. | en_US |
| dcterms.educationalLevel | All Doctorate | 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/200/14243