Professional Documents
Culture Documents
Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct
a
Department of Mechanical Engineering – Engineering Mechanics, Michigan Technological University, 1400 Townsend Drive, Houghton, MI 49931, USA
b
Los Alamos National Lab – Engineering Institute, PO Box 1663, MS T001, Los Alamos, NM 87545, USA
c
Graduate Aerospace Laboratories, California Institute of Technology, Pasadena, CA 91125, USA
d
Los Alamos National Lab – Engineering Institute, Los Alamos, NM 87545, USA
Keywords: Many operating structures undergo vibration associated with significant rigid-body motion, such as earthquake-
Rigid-body motion excited structures which vibrate while moving (and/or rocking) along with the shaking ground, rotating
Vibration structures such as the blades of helicopter, aircraft, wind turbine, and in UAV’s measurements (images/videos) of
Full-field measurement infrastructure response which is coupled with the motion of the UAV itself. In many such cases, the rigid motion
Operational modal analysis
is very dominant; measurement and identification of the relatively subtle deformation (flexible) motion of the
Photogrammetry
Blind source separation
structure in their operations is very challenging, especially when high-resolution vibration measurement and
high-fidelity dynamics modeling are required to capture and characterize spatially local and temporally transient
dynamics behaviors and features of the structure. Traditional contact-type wired or wireless sensors, such as
accelerometers, strain gauges, and LVDT’s, are discrete point-wise sensors that only allow instrumentations at a
limited number of places, providing very low spatial resolution vibration measurements. Non-contact inter-
ferometry-based optical measurement techniques using laser beams such as laser Doppler vibrometers, electronic
speckle pattern interferometry (ESPI) and holographic interferometry, can provide high spatial resolution vi-
bration measurement without the need to install sensors on the structure. However, they are very vulnerable to
ambient vibration and rigid-body motion. As an alternative non-contact optical measurement method, photo-
grammetry uses white-light imaging of digital video cameras that are relatively low-cost, agile, and provides
simultaneous measurements with very high spatial resolution where every pixel effectively becomes a sensing
unit. Furthermore, it is robust to rigid-body motion. Using image processing algorithms such as optical flow and
image correlation, photogrammetry has been successfully used for vibration measurements and experimental
modal analysis of structures (including rotating-type) where full-field (or as many points as the pixel number of
the image frame on the structure) modal parameters (mode shapes) can be obtained. However, existing such
methods typically rely on installing and tracking the optical markers or speckle paints on the structure’s surface,
which becomes less applicable for large measurement area or in harsh environment (high temperature and
corrosion). In addition, they are computationally extensive.
Aiming to alleviate these challenges, this study presents the development of adapting a recently-proposed
video-based method for output-only, simultaneous identification of both the subtle, full-field deformation modes
and the dominant rigid-body motion from only the video of the vibrating while moving structure. Full-field
dynamic parameters (deformation modes) of structure, as well as its rigid-body motion can be efficiently esti-
mated, respectively. Laboratory experiments were conducted on a bench-scale building structure with “free-free”
boundary condition. Comparisons were made with an analytical finite element model. Results demonstrate that
the method can efficiently and accurately extract the full-field, (total) motion of the structure and further
identify or separate the subtle deformation modes and large rigid-body motion, respectively. Furthermore, vi-
Corresponding author at: Department of Mechanical Engineering – Engineering Mechanics, Michigan Technological University, 1400 Townsend Drive, Houghton,
⁎
MI 49931, USA.
E-mail address: ycyang@mtu.edu (Y. Yang).
1
Formerly Argonne National Laboratory, 9700 S. Cass Ave., Lemont, IL 60439, USA.
https://doi.org/10.1016/j.engstruct.2020.110183
Received 2 July 2018; Received in revised form 4 January 2020; Accepted 6 January 2020
0141-0296/ © 2020 Elsevier Ltd. All rights reserved.
Y. Yang, et al. Engineering Structures 207 (2020) 110183
deos reconstructed by the method (provided in the supplementary materials) show that it enables realistic vi-
sualization of the subtle deformation modes of the structure in presence of large rigid-body motion. Several
practical limitations for implementing the method are also discussed.
2
Y. Yang, et al. Engineering Structures 207 (2020) 110183
3
Y. Yang, et al. Engineering Structures 207 (2020) 110183
Fig. 2. A few temporal frames from the video measurements of the vibrating while moving structure. Linked to Video 1.
Fig. 3. Demonstrations of arbitrarily-selected 6 pixels: The time histories and their Fourier spectra of the extracted local phases of 6 pixels from the video of the
vibrating and moving structure. Note that the method uses the phase time history of all pixels.
reconstructed as follows: take the 1st mode for example, only the se- 3. Experimental validations
parated 1st deformation mode q1 (t ) as well as the rigid-body motion ¯
are retained from Eq. (6) and fed into the inverse transform of as per Eq. 3.1. Laboratory experimental setup
(5) to Eq. (4) and the multi-scale filtering. Therefore, the reconstructed
video will contain only the 1st deformation mode in presence of the A bench-scale model of a three-story building structure was used to
rigid-body motion for high-spatial (pixel)-resolution visualization in the validate the approach. The structure consists of aluminum columns and
spatial-temporal video domain. Since the deformation motion is very lumped mass plates on each of the three floors, with its base mass slides
subtle and the rigid-body motion is dominant, the deformation mode on the rails to allow free rigid-body motion in the horizontal direction.
q1 (t ) could be magnified by a scalar factor 1 before fed into the re- An impact hammer was used to excite the structure horizontally on the
construction process for its better visualization in presence of the rigid- top floor. A stationary camera (Sony NXCAM with a pixel resolution of
body motion. Note that higher modes usually require larger magnifi- 1920 × 1080) mounted by a Zeiss lens with a fixed focal length of
cation factors since their motion are weaker contained in the structural 24 mm was used to perform video measurements of the moving while
motion. More details are referred to Ref. [39]. vibrating structure at a frame rate of 480 frames per second (Video 1 in
4
Y. Yang, et al. Engineering Structures 207 (2020) 110183
Fig. 4. The estimated first six principal component coordinates and their Fourier spectra, extracted from the video of the vibrating and moving structure by applying
the PCA on the phase time histories (see Fig. 3 for demonstration) of all pixels.
Fig. 5. The estimated modal coordinates and their Fourier spectra, extracted from the video of the vibrating and moving structure by applying the PCA-CP full-field
output-only modal identification method.
5
Y. Yang, et al. Engineering Structures 207 (2020) 110183
Fig. 6. The full-field mode shapes identified from the video measurements of 3-story structure by applying the PCA-CP full-field output-only modal identification
method (Linked to Video 3–5 for visualization of deformation modes in presence of rigid-body motion). (a–c) are associated with the 3 deformation modes in
Fig. 5 (from the top), and (d–f) are with the other 3 rigid-body mode components. The vertical (frame height) and horizontal (frame width) axis represent pixel
locations. The colorbars represent normalized mode shape magnitude scales and are different among plots. Note that the rigid plates are not presented in the plots.
6
Y. Yang, et al. Engineering Structures 207 (2020) 110183
Fig. 7. The computed mode shapes of the analytical finite element mode of the structure. From (a–d): Mode 3, Mode 2, Mode 1, and Mode 0 (rigid-body mode).
downsampled to 384 × 216. It was then processed by the full-field principal components, the deformation modes and rigid-body motions
output-only modal identification method [39], as presented in Sections are clearly separated: it is seen in Fig. 5 that three deformation modes
2.1–2.2. Fig. 3 shows the local phase time histories of a few arbitrarily- are identified associated with the simultaneously-estimated full-field
selected pixels, which are proportional to the structural displacements. deformed mode shapes shown in Fig. 6 (a)-(c), whereas the rest are the
It is seen that the rigid-body motion, whose frequency is very close to rigid-body motion associated with the full-field uniform, non-deformed
zero, dominated the extracted displacements, where the deformation shapes (Fig. 6 ((d–f)).
was very subtle or weakly present in all the pixels. This is consistent
with what can be observed in the original video (Video 1 or Video 2 in
supplementary material). 3.2.2. Validating the video-estimated results
After applying PCA on the high-dimensional motion (displacement) The modal parameters identified by the method from the video
matrix, six principal components are obtained (Fig. 4) to allow possibly measurements were compared with those computed from the lumped-
redundant components. It can be seen that the first principal compo- mass FEM model. Table 1 shows the comparisons of the modal fre-
nent (most active with the largest singular value) is a rigid-body motion quencies, which match closely with each other. The slightly higher
mode, while other principal components contain mixtures of deforma- modal frequencies of the deformation modes computed from the FEM
tion modes and rigid-body motions. After applying BSS on the six than those estimated from the video may be because: (1) the damping is
not included in the FEM; (2) friction occurred when the structure slid
7
Y. Yang, et al. Engineering Structures 207 (2020) 110183
the method can efficiently and accurately extract the full-field, total
motion of the structure from the video measurement and further
identify or separate the subtle deformation modes (subsequently ob-
taining the full-field modal parameters) and large rigid-body motion
(translation-type), respectively. Furthermore, it also enables realistic
visualization of each of the subtle deformation modes in presence of
large rigid-body motion.
Several practical limitations for implementing the method are dis-
cussed. Future study is needed to extend the method to account for
rotating-type rigid-body motion of the structure because it involves
different structural dynamics models (formulations). Implementations
of the video-based approach on real-world full-scale cable structures,
e.g., using UAV imagers to measure structural response, also need to
address the issues associated with the video camera measurement under
varying and harsh imaging environment (such as variant illuminations
and perspectives, appearance of obstructions, etc), and therefore testing
Fig. 8. The estimated rigid-body motion of the 3-story structure compared with on large-scale structure is needed in future work to validate the ro-
the laser displacement measurement. Both are normalized by their maxima to bustness of the proposed method.
exclude the calibration factors.
CRediT authorship contribution statement
on the rails while the FEM assumes free-free boundary. In addition, the
frequency resolution of the Fourier Transform (about 0.6 Hz with only Yongchao Yang: Conceptualization, Methodology, Investigation,
less than 2 s in signal length) in estimation of the modal frequencies Writing - original draft, Writing - review & editing. Charles Dorn:
from the modal coordinates extracted from the video also causes errors. Investigation, Writing - review & editing. Charles Farrar:
The vibrating patterns of the estimated full-field, high-spatial-resolu- Conceptualization, Supervision, Resources, Writing - review & editing.
tion mode shapes match closely with those computed from the lumped- David Mascareñas: Conceptualization, Supervision, Resources,
mass FEM of very low spatial resolution (Fig. 7). Writing - review & editing.
3.2.3. Reconstructing the rigid-body motion of the structure Declaration of Competing Interest
The three separated rigid-body motion modes were combined to
reconstruct the overall rigid-body motion using Eq. (8). The re- The authors declare no conflict of interest for the manuscript
constructed rigid-body motion were compared with that directly mea-
sured by the laser displacement sensor. Fig. 8 shows reasonable corre- Acknowledgements
lation between them. This shows that the method is able to estimate
both the subtle deformation modes and large rigid-body motion of the We would like to acknowledge the financial support of the Los
structure from digital video measurements. Alamos National Laboratory Lab Directed Research and Development
(LDRD) program for a Director’s Funded Postdoctoral Fellowship and
3.2.4. Visualization of individual subtle deformation modes in presence of the U.S. Defense Advanced Research Projects Agency (DARPA) “Physics
large rigid-body motion of AI” program for Yongchao Yang. Yongchao Yang would also like to
After the individual deformation modes and rigid-body motion are acknowledge the discussions with Dr. Peter Avitabile of University of
identified (separated), new videos (Video 3–5 in supplementary ma- Massachusetts – Lowell in July 2016 and with Dr. Eric Flynn of Los
terial) were reconstructed to enable visualization of each of the subtle Alamos National Laboratory in April 2017.
deformation modes in presence of large rigid-body motion. Scalar
magnification factors of 10, 20, and 40 are used for Mode 1 (Video 3), Appendix A. Supplementary material
Mode 2 (Video 4), and Mode 3 (Video 5), respectively. Take Video 3
for example, the 1st deformation mode can be clearly visualized while Supplementary data to this article can be found online at https://
the structure was going through large rigid-body motion. doi.org/10.1016/j.engstruct.2020.110183.
This study presents the development of adapting a video-based [1] Ewins D. Modal testing: theory, practice and application, Baldock, Hertfordshire.
method for output-only, simultaneous identification of both the subtle, England: Research Studies Press LTD; 2000.
[2] Friswell M, Mottershead J. Finite Element model updating in structural dynamics.
full-field deformation modes and the dominant rigid-body motion from Springer; 1995.
only the video of the vibrating while moving structure. The method is [3] Doebling S, Farrar C, Prime M. A summary review of vibration-based damage
able to efficiently and blindly estimate the full-field dynamic para- identification methods. Shock Vibrat Digest 1998;30(2):91–105.
[4] Brincker R, Ventura C. Introduction to operational modal analysis. Wiley; 2015.
meters of the deformation modes of structure, as well as its rigid-body [5] Yang Y, Dorn C, Mancini T, Talken Z, Theiler J, Kenyon G, et al. “Reference-free
motion. This approach enhances the previously-developed full-field detection of minute, non-visible, damage using full-field, high-resolution mode
output-only modal analysis method by explicitly allowing retaining shapes output-only identified from digital videos of structures. Struct Health
Monitor Int J 2017. https://doi.org/10.1177/1475921717704385.
redundant components where subtle deformation modes are retained
[6] Wei S, Zhang Z, Li S, Li H. Strain features and condition assessment of orthotropic
and further blindly separated, and the dominant rigid-body motion is steel deck cable-supported bridges subjected to vehicle loads by using dense FBG
reconstructed. This enables estimations of both the subtle, full-field strain sensors. Smart Mater Struct 2017;26(10):104007.
[7] Li S, Wei S, Bao Y, Li H. Condition assessment of cables by pattern recognition of
deformation modes (modal parameters) and the dominant rigid-body
vehicle-induced cable tension ratio. Eng Struct 2018;155:1–15.
motion, respectively, which is very challenging for existing vibration [8] Stanbridge A, Ewins D. Modal testing using a scanning laser Doppler vibrometer.
measurement and analysis techniques. Mech Syst Sig Process 1999;13(2):255–70.
Laboratory experiments were conducted on a bench-scale building [9] Castellini P. Laser Doppler Vibrometry: Development of advanced solutions an-
swering to technology's needs. Mech Syst Sig Process 2006;20(6):1265–85.
structure with “free-free” boundary condition. Results demonstrate that
8
Y. Yang, et al. Engineering Structures 207 (2020) 110183
[10] Di Maio D, Ewins D. Continuous Scan, a method for performing modal testing using [33] Lundstrom T, Niezrecki C, Avitabile P. Appropriate rigid body correction of ste-
meaningful measurement parameters; Part I. Mech Syst Sig Process reophotogrammetry measurements made on rotating systems. Exp Tech
2011;25(8):3027–42. 2015;39(6):25–34.
[11] Pedrini G, Wolfgang O, Gusev M. High-speed digital holographic interferometry for [34] Baqersad J, Poozesh P, Niezrecki C, Avitabile P. A noncontacting approach for full-
vibration measurement. Appl Opt 2006;45(15):3456–62. field strain monitoring of rotating structures. J Vib Acoust 2016;138(3):031008.
[12] Steton K. calculating modal assurance criterion from electronic holography data. [35] Fleet D, Allan D. Computation of component image velocity from local phase in-
Sound Vibrat Mag 2008;42(3):6–8. formation. Int J Comput Vision 1990;5(1):77–104.
[13] Schmidt T, Tyson J, Galanulis K. Full-field dynamic displacement and strain mea- [36] Wadhwa N, Rubinstein M, Durand F, Freeman W. Phase-based video motion pro-
surement using advanced 3d image correlation photogrammetry: Part 1. Exp Tech cessing. ACM Trans Graph 2013;32(4).
2003;27(3):47–50. [37] Gautama T, Van Hulle M. A phase-based approach to the estimation of the optical
[14] Baqersad J, Poozesh P, Niezrecki C, Avitabile P. Photogrammetry and optical flow field using spatial filtering. Neural Netw IEEE Trans 2002;13(5):1127–36.
methods in structural dynamics – a review. Mech Syst Sig Process 2017;86:98–115. [38] Chen J, Wadhwa N, Cha Y, Durand F, Freeman W, Buyukozturk O. Modal identi-
[15] Horn B, Schunck B. Determining optical flow. Artif Intell 1981;17:185–204. fication of simple structures with high-speed video using motion magnification. J
[16] Sutton M, Orteu J, Schreier H. Image correlation for shape, motion and deformation Sound Vibrat 2015;345:58–71.
measurements: basic concepts, theory and applications. Springer Science & Business [39] Yang Y, Dorn C, Mancini T, Talken Z, Kenyon G, Farrar C, et al. Blind identification
Media; 2009. of full-field vibration modes from video measurements with phase-based video
[17] Olaszek P. Investigation of the dynamic characteristic of bridge structures using a motion magnification. Mech Syst Sig Process 2017;85:567–90.
computer vision method. Measurement 1999;25(3):227–36. [40] Yang Y, Farrar C, Mascarenas D. Full-field structural dynamics by video motion
[18] Wahbeh A, Caffrey J, Masri S. A vision-based approach for the direct measurement manipulation. 58th AIAA/ASCE/AHS/ASC structures, structural dynamics, and
of displacements in vibrating systems. Smart Mater Struct 2003;12(5):785. materials conference. 2017.
[19] Lee J, Shinozuka M. A vision-based system for remote sensing of bridge displace- [41] Dorn C, Dasari S, Yang Y, Kenyon G, Welch P, Mascarenas D. Efficient full-field
ment. NDT E Int 2006;39(5):425–31. operational modal analysis using neuromorphic event-based imaging. Shock & vi-
[20] Chang C, Ji Y. Flexible videogrammetric technique for three-dimensional structural bration, aircraft/aerospace, energy harvesting, acoustics & optics. 2017.
vibration measurement. J Eng Mech 2007;133(6):656–64. [42] Roeder A, Zhang H, Sanchez L, Yang Y, Farrar C, Mascarenas D. Identification of
[21] Yoon H, Elanwar H, Choi H, Golparvar-Fard M, Spencer B. Target‐free approach for full-field dynamic loads on structures using computer vision and unsupervised
vision‐based structural system identification using consumer‐grade cameras. Struct machine learning. Shock & vibration, aircraft/aerospace, energy harvesting,
Control Health Monitor 2016. https://doi.org/10.1002/stc.1850. acoustics & optics. 2017.
[22] Feng D, Feng M. Vision-based multipoint displacement measurement for structural [43] Sanchez L, Zhang H, Roeder A, Bowlan J, Crochet J, Yang Y, et al. Establishment of
health monitoring. Struct Control Health Monitor 2016;23(5):876–90. full-field, full-order dynamic model of cable vibration by video motion manipula-
[23] Khuc T, Catbas F. Completely contactless structural health monitoring of real‐life tions. Special Top Struct Dyn 2017.
structures using cameras and computer vision. Struct Control Health Monitor 2016. [44] Javh J, Slavič J, Boltežar M. The subpixel resolution of optical-flow-based modal
https://doi.org/10.1002/stc.1852. analysis. Mech Syst Sig Process 2017;88:89–99.
[24] Zimmermann M, Gülan U, Harmanci Y, Chatzi E, Holzner M. Structural health [45] Sarrafi A, Mao Z, Niezrecki C, Poozesh P. Vibration-based damage detection in wind
monitoring through video recording. EWSHM. 2016. turbine blades using Phase-based Motion Estimation and motion magnification. J
[25] Siebert T, Wood R, Splitthof K. High speed image correlation for vibration analysis. Sound Vib 2018;421:300–18.
J Phys Conf Ser 2009;181(1):012064. [46] Zhao J, Bao Y, Guan Z, Zuo W, Li J, Li H. Video-based multiscale identification
[26] Helfrick MN, Niezrecki C, Avitabile P, Schmidt T. 3D digital image correlation approach for tower vibration of a cable-stayed bridge model under earthquake
methods for full-field vibration measurement. Mech Syst Sig Process ground motions. Struct Control Health Monitor 2019;26(3):e2314.
2011;25(3):917–27. [47] Bao Y, Tang Z, Li H, Zhang Y. Computer vision and deep learning–based data
[27] Wang W, Mottershead J, Ihle J, Siebert T, Schubach H. Finite element model up- anomaly detection method for structural health monitoring. Struct Control Health
dating from full-field vibration measurement using digital image correlation. J Monitor 2018. DOI: 1475921718757405.
Sound Vib 2011;330(8):1599–620. [48] Yang Y, Dorn C, Mancini T, Talken Z, Nagarajaiah SKG, Farrar C, et al. Blind
[28] Stasicki B, Boden F. Application of high-speed videography for in-flight deformation identification of full-field vibration modes of output-only structures from uniformly
measurements of aircraft propellers. 28th International Congress on High-Speed sampled, possibly temporally aliased (sub-Nyquist), full-field video measurements.
Imaging and Photonics, International Society for Optics and Photonics. 2008. J Sound Vib 2017;390:232–56.
[29] Helfrick M, Pingle P, Niezrecki C, Avitabile P. Optical non-contacting vibration [49] Portilla J, Simoncelli E. A parametric texture model based on joint statistics of
measurement of rotating turbine blades. Proceedings of the IMAC-XXVII. 2009. complex wavelet coefficients. Int J Comput Vision 2000;40(1):49–70.
[30] Ozbek M, Rixen D, Eme O, Sanow G. Feasibility of monitoring large wind turbines [50] Feeny B, Kappagantu R. On the physical interpretation of proper orthogonal modes
using photogrammetry. Energy 2010;35(12):4802–11. in vibration. J Sound Vib 1998;211:607–16.
[31] Abrego A, Olson L, Romander E, Barrows D, Burner A. Blade displacement mea- [51] Yang Y, Nagarajaiah S. Blind modal identification of output-only structures in time-
surement technique applied to a full-scale rotor test. NASA 2012. domain based on complexity pursuit. Earthquake Eng Struct Dyn
[32] Sicard J, Sirohi J. Measurement of the deformation of an extremely flexible rotor 2013;42(13):1885–905.
blade using digital image correlation. Meas Sci Technol 2013;24(6):065203.