Vision and Learning based Sensing for Structural Health Monitoring
| dc.contributor.advisor | Nagarajaiah, Satish | |
| dc.creator | Jana, Debasish | |
| dc.date.accessioned | 2022-09-28T16:19:53Z | |
| dc.date.available | 2023-05-01T05:01:15Z | |
| dc.date.created | 2022-05 | |
| dc.date.issued | 2022-02-28 | |
| dc.date.submitted | May 2022 | |
| dc.date.updated | 2022-09-28T16:19:53Z | |
| dc.description.abstract | Civil infrastructures like bridges play an essential role in maintaining the socio-economic growth of a nation. These important structures are designed to have a long service life, but the structural health may degrade due to natural hazards, aging, and man-made operation. Therefore, Structural Health Monitoring (SHM) is necessary to circumvent any system malfunctioning that might cause severe economic loss. Vibration measuring sensors are deployed for data acquisition to the structure of interest, a prerequisite to the vibration-based SHM. This dissertation focuses explicitly on increasing the efficiency of state-of-the-art sensing techniques – for both non-contact (vision-based) and contact-based sensors. All the proposed methods are derived and modified from the computer vision and machine learning algorithms; and applied to various structural components of the bridge. Algorithms developed in the first two chapters are applied to the stay-cable tension estimation of a cable-stayed bridge; frameworks proposed in the following three chapters are applied to the beams (girders) and plates (deck) of the bridge; deep learning technique presented in the last chapter is applied to deck pavement crack detection. First, a non-contact video-based cable tension estimation technique is proposed where a video is recorded using a moving handheld camera at a significant distance from the structure. Here, a camera movement cancellation technique is proposed to use vision-based motion estimation algorithms to estimate the cable tension. Second, contact-based wireless sensors are generally mounted on the stay-cables for acquiring vibration data. But these wireless sensors often undergo data-packet loss during data transmission – which may interrupt the online/real-time health monitoring process. Therefore, a sliding windowed compressive sensing-based data reconstruction technique is proposed to estimate the real-time stay-cable tension. Third, full-field sensing is mandatory to discover the possible damage location of a structure. A compressive sensing-based framework is proposed for estimating the full-field response time history from the time histories of a few sensors. Both `fixed sensors' and `mobile sensor' scenarios are explored. In both these scenarios, two subcases are considered – (a) If the system dynamics is unknown, the basis functions for compressive sensing is obtained by Dictionary Learning from a full-field experiment; (b) if a partial knowledge of the model dynamics is available, the basis functions are derived from the closed-form solution of the governing generalized partial differential equation. For mobile sensing-based full-field response estimation – instead of manually driving the vehicles – a multi-agent formation control strategy is proposed to automate the driving process. Finally, a deep learning (DL) framework is proposed for crack segmentation which can process the image data from an additional camera mounted on the mobile robots. This DL framework is based on the encoder-decoder convolution neural network (UNet) with transfer learning, outperforming the state-of-the-art segmentation framework in terms of performance and computational efficiency. Even though all the proposed techniques are applied to different bridges components, this is also implementable to any other structure. Real-time application of these techniques will lead to better maintenance, recovery time, and resilience estimation. | |
| dc.embargo.terms | 2023-05-01 | |
| dc.format.mimetype | application/pdf | |
| dc.identifier.citation | Jana, Debasish. "Vision and Learning based Sensing for Structural Health Monitoring." (2022) Diss., Rice University. <a href="https://hdl.handle.net/1911/113407">https://hdl.handle.net/1911/113407</a>. | |
| dc.identifier.uri | https://hdl.handle.net/1911/113407 | |
| dc.language.iso | eng | |
| dc.rights | Copyright is held by the author, unless otherwise indicated. Permission to reuse, publish, or reproduce the work beyond the bounds of fair use or other exemptions to copyright law must be obtained from the copyright holder. | |
| dc.subject | Vision-based | |
| dc.subject | Wireless | |
| dc.subject | Cable Tension | |
| dc.subject | Source separation | |
| dc.subject | Full-field | |
| dc.subject | Dictionary Learning | |
| dc.subject | Compressive Sensing | |
| dc.subject | Moving robots | |
| dc.subject | Formation Control | |
| dc.subject | Crack Segmentation | |
| dc.subject | Machine Learning | |
| dc.subject | Sensing | |
| dc.subject | Deep Learning | |
| dc.title | Vision and Learning based Sensing for Structural Health Monitoring | |
| dc.type | Thesis | |
| dc.type.material | Text | |
| thesis.degree.department | Civil and Environmental Engineering | |
| thesis.degree.discipline | Engineering | |
| thesis.degree.grantor | Rice University | |
| thesis.degree.level | Doctoral | |
| thesis.degree.name | Doctor of Philosophy |
Files
Original bundle
1 - 1 of 1