國立虎尾科技大學 |

Structural Dynamics Identification via Computer Vision Approaches.

紀錄類型:	書目-語言資料,印刷品 : Monograph/item
正題名/作者:	Structural Dynamics Identification via Computer Vision Approaches./
作者:	Sarrafi, Aral.
出版者:	Ann Arbor : ProQuest Dissertations & Theses, : 2019,
面頁冊數:	168 p.
附註:	Source: Dissertations Abstracts International, Volume: 81-04, Section: B.
Contained By:	Dissertations Abstracts International81-04B.
標題:	Mechanical engineering. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=22621044
ISBN:	9781687938534

Structural Dynamics Identification via Computer Vision Approaches.
Sarrafi, Aral.

Structural Dynamics Identification via Computer Vision Approaches. - Ann Arbor : ProQuest Dissertations & Theses, 2019 - 168 p.

Source: Dissertations Abstracts International, Volume: 81-04, Section: B.

Thesis (Ph.D.)--University of Massachusetts Lowell, 2019.

This item must not be sold to any third party vendors.

Digital cameras have been widely used to record structural vibrations due to external dynamic loadings. The structural dynamics information of the vibrating structure is embedded in the subtle pixel variations of the recorded video, and the challenge is to recover the imperceptible structural motion from the video data. Estimating and processing the structural dynamics from videos, i.e. a sequence of images, have the advantages over other traditional sensing modalities, such as being non-contact, having no mass loading effect, and reduced testing time given the full-field observability. This information may be used for a variety of applications such as system identification (SI), model validation, model updating, structural health monitoring (SHM), and facilitating the structural testing, and certification procedures.In this dissertation, several novel computer vision and signal processing approaches for optical-sensing-based structural dynamics identification on a diverse range of structures are developed and studied. The methodologies are validated through both simulated signals as well as videos recorded from real structures, including lab-scale benchmark structures and industrial large-scale structures. Phase-based motion estimation and motion magnification (video magnification) are extensively studied for structural dynamics identification and SHM, on four different structures with complex geometry and material properties. In order to map the motion magnified videos to quantified operating deflection shape (ODS) vectors automatically, a novel hybrid computer vision approach is developed in the dissertation work. This approach reduces the human supervision for estimating the quantified ODS vectors significantly.In the real practice, camera data are more prone to noise; therefore, the uncertainty quantification of estimated motion is crucial for having a measured confidence in any subsequent analysis. A method which is a combination of analytical and numerical approaches has been developed to predict the effect of noise and contamination on the estimated motion. Moreover, the feasibility of an end-to-end data-driven method for structural health monitoring is investigated, to establish the optimal filters only based on learning from the observed data. This approach replaces the feature design, and feature selection stages with tuning the learnable filters during the training, and shows promising performance.

ISBN: 9781687938534Subjects--Topical Terms:

557493
Mechanical engineering.
Subjects--Index Terms:

Computer Vision

Structural Dynamics Identification via Computer Vision Approaches.
LDR:03648nam a2200373 4500 001 951856
005 20200821052210.5
008 200914s2019 ||||||||||||||||| ||eng d
020 $a 9781687938534
035 $a (MiAaPQ)AAI22621044
035 $a AAI22621044
040 $a MiAaPQ $c MiAaPQ
100 1 $a Sarrafi, Aral. $3 1241346
245 1 0 $a Structural Dynamics Identification via Computer Vision Approaches.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2019
300 $a 168 p.
500 $a Source: Dissertations Abstracts International, Volume: 81-04, Section: B.
500 $a Advisor: Mao, Zhu.
502 $a Thesis (Ph.D.)--University of Massachusetts Lowell, 2019.
506 $a This item must not be sold to any third party vendors.
520 $a Digital cameras have been widely used to record structural vibrations due to external dynamic loadings. The structural dynamics information of the vibrating structure is embedded in the subtle pixel variations of the recorded video, and the challenge is to recover the imperceptible structural motion from the video data. Estimating and processing the structural dynamics from videos, i.e. a sequence of images, have the advantages over other traditional sensing modalities, such as being non-contact, having no mass loading effect, and reduced testing time given the full-field observability. This information may be used for a variety of applications such as system identification (SI), model validation, model updating, structural health monitoring (SHM), and facilitating the structural testing, and certification procedures.In this dissertation, several novel computer vision and signal processing approaches for optical-sensing-based structural dynamics identification on a diverse range of structures are developed and studied. The methodologies are validated through both simulated signals as well as videos recorded from real structures, including lab-scale benchmark structures and industrial large-scale structures. Phase-based motion estimation and motion magnification (video magnification) are extensively studied for structural dynamics identification and SHM, on four different structures with complex geometry and material properties. In order to map the motion magnified videos to quantified operating deflection shape (ODS) vectors automatically, a novel hybrid computer vision approach is developed in the dissertation work. This approach reduces the human supervision for estimating the quantified ODS vectors significantly.In the real practice, camera data are more prone to noise; therefore, the uncertainty quantification of estimated motion is crucial for having a measured confidence in any subsequent analysis. A method which is a combination of analytical and numerical approaches has been developed to predict the effect of noise and contamination on the estimated motion. Moreover, the feasibility of an end-to-end data-driven method for structural health monitoring is investigated, to establish the optimal filters only based on learning from the observed data. This approach replaces the feature design, and feature selection stages with tuning the learnable filters during the training, and shows promising performance.
590 $a School code: 0111.
650 4 $a Mechanical engineering. $3 557493
650 4 $a Computer science. $3 573171
653 $a Computer Vision
653 $a Deep Learning
653 $a Motion Magnification
653 $a Structural Dynamics
653 $a Structural Health Monitoring
653 $a Uncertainty Quantification
690 $a 0548
690 $a 0984
710 2 $a University of Massachusetts Lowell. $b Mechanical Engineering. $3 1241347
773 0 $t Dissertations Abstracts International $g 81-04B.
790 $a 0111
791 $a Ph.D.
792 $a 2019
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=22621044