| Author: | Chen, Siyi |
| Title: | SHM data processing using low-rank techniques : from time-series vibration data to vision-based image data |
| Advisors: | Ni, Yi-qing (CEE) Wang, Youwu (CEE) Zhou, Lu (CEE) |
| Degree: | Ph.D. |
| Year: | 2024 |
| Subject: | Structural health monitoring -- Data processing Structural analysis (Engineering) Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Civil and Environmental Engineering |
| Pages: | xxiv, 281 pages : color illustrations |
| Language: | English |
| Abstract: | Over the long-term monitoring process, the collected structural health monitoring (SHM) data tend to be voluminous, redundant, but relevant. Although these data are in high-dimensional spaces, they may distribute around low-dimensional subspaces. In such scenarios, low-rank models will become useful tools for efficiently and robustly processing high-dimensional SHM data, since they can fully utilize spatial and temporal correlations among data. Currently, the applications of these low-rank techniques in SHM remain in their infancy. It motivates to seek and utilize appropriate low-rank models to solve existing challenges in SHM. In this regard, this thesis aims to explore several low-rank techniques to address the challenges associated with imperfect vibration data and achieve image-based crack detection. Depending on the different SHM data used (time-series vibration data or vision-based image data), this thesis falls into two main parts. In the first part, two structured low-rank based methods are designed to handle imperfect time-series vibration data, aiming to ensure the integrity and accuracy of the monitoring data. Specifically, a novel methodology, termed Hankel-based robust principal component analysis (HRPCA), is first developed for denoising structural dynamic response data contaminated with gross outliers. Making use of the fundamental duality, a low-dimensional (low-rank) subspace could be constructed to represent the originally redundant yet relevant vibration data, which allows for better separation of gross outliers embedded in the monitoring data. Next, this study further harnesses the low-rank nature of the time-series vibration data to tackle the issue of lost data recovery. Based on the low-rank Hankel property, the proposed method can address all three commonly encountered data loss patterns, including random data loss, continuous but not synchronized data loss, as well as continuous and synchronized data loss. The second part of the thesis focuses on image-based crack detection, where low-rank strategies are incorporated to enhance the detection process. Given the problems of high computational cost and scarcity of labeled data, this part first presents a fast and real-time crack detection method that combines transfer learning (TL) and low-rank dictionary learning (LRDL). It leverages on the knowledge from a well-trained deep convolutional neural network (DCNN) model, and “transfers” its learning ability to a crack detection task. Based on the transferred features, a new classifier is generated for crack recognition via LRDL. The LRDL phase benefits from the availability of pre-extracted features and a limited number of parameters, which can significantly reduce the training time and the number of required labeled images. Finally, to tackle the challenges from complex background scenes, hyperspectral imaging (HSI) technique is further introduced to eliminate their influence. Compared to the RGB images that only contain three spectral bands (red, green, blue), HSI images contain hundreds of spectral bands, providing rich spectral information. Due to the high correlations in HSI data, a low-rank representation (LRR)-based method is developed for crack detection. It possesses strong anti-interference ability to identify cracks and other materials under complex scenes. Through a combination of theoretical, numerical, and experimental investigations, this thesis clearly demonstrates the promise of applying the proposed low-rank methods in processing SHM data. |
| Rights: | All rights reserved |
| Access: | open access |
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/200/13034