Skip to Main Content

Bridge Modal Identification Via Video Processing & Quantification of Uncertainties

PI: Zhu Mao
Institution: University of Massachusetts Lowell
PI Email: Zhu_Mao@uml.edu
PI Phone: 978-934-5937

Background / Need:

Bridges form a critical category of the U.S. transportation infrastructure, yet the current structural condition is only evaluated at “C+” according to the 2017 ASCE Infrastructure Report Card. In addition to the fact that 9.1% of the bridges in U.S. are structurally deficient, the bridges in New England are especially experiencing the burden of busy traffic and harsh wintery weather. There is a variety of factors that may affect the bridge dynamics and deteriorate the structures, such as creeping, corrosion, cyclic thermal loadings and accidental damages, and identification modal properties provides a global evaluation capability with rich physical meaning. However, this complicated scenario brings up the demanding in conducting the heterogeneous data acquisition and in-situ modal analysis, as well as quantifying the enormous amount of uncertainties that may come across. The problem we are trying to solve is to adopt portable video cameras and by processing the acquired videos, bridge dynamic systems, especially full-field mode shapes will be extracted to enhance the status awareness. The challenges exist while dealing with the rapidly changing environments and traffics, so that the statistical modeling is needed when interpreting the extracted information.

Graphic illustration for Bridge Modal Identification Via Video Processing & Quantification of Uncertainties. Fig. 1 demonstrate the experimental modal analysis flow via stereo-cameras and the algorithm of phase-based motion extraction and amplification. The non-contact data are employed for the nature of its full-field and easy implementation characteristics.

Fig. 1. Overview of the computer vision approach for modal identification

Fig. 1 demonstrates the experimental modal analysis flow via stereo-cameras and the algorithm of phase-based motion extraction and amplification.

The non-contact data are employed for the nature of its full-field and easy implementation characteristics.

Statistical modeling will be deployed once the modal information is extract, and the uncertainty will be studied via data-driven modeling and Bayesian inference, as demonstrated in Fig. 2.

Objectives:

The objectives of this project are to:

Graphic illustration for Bridge Modal Identification Via Video Processing & Quantification of Uncertainties. Statistical modeling will be deployed once the modal information is extract, and the uncertainty will be studied via data-driven modeling and Bayesian inference, as demonstrated in Fig. 2.

Fig. 2. Statistical modeling and decision-making for damages identification

  • Investigate the non-contact image/video processing techniques for global bridge dynamics evaluation, and modal-based properties will be extracted from the pixel data.
  • Computer vision algorithms will be applied to process the acquired data, and uncertainties, both aleatory and epistemic, will be quantified. With statistical modeling and classification, structural deficiency will be detected and evaluated with statistical significance.
  • Investigate the feasibility of the video-processing techniques in dealing with influence of environmental and traffic variabilities on bridges.
Bridge Model with car passing by

Bridge Model with car passing by

Updates:

December 30, 2019: This project has as objective investigate the capability of applying non-contact optical sensing and motion magnification to identify structural dynamics and damages in bridges. Lab-scale structures are adopted for system identification using video data under a variety of excitations.
Bridge Model with a damaged element

Bridge Model with a damaged element

In real bridges, there are many different inputs simultaneously, such as change of mass distribution by the movement of vehicles and external excitations caused by wind, scour, temperature loadings, therefore, tests are designed aiming to obtain a coherent frequency spectrum applying a random non-stationary excitation during a period of time.
Single-Sided Amplitude Spectrum of Phase for Bridge model with a damaged element

Single-Sided Amplitude Spectrum of Phase for Bridge model with a damaged element

By using the phase-based motion extraction, we are able to identify the vibrational frequencies of bridges, as well as the operational deflection shapes under each mode.
Single-Sided Amplitude Spectrum of Phase for Bridge model with car passing by

Single-Sided Amplitude Spectrum of Phase for Bridge model with car passing by

For extreme damages, such as missing components, the damages are identified and localized via computer vision techniques directly.
We have also customized kernel-based optical targets and compared the performance with LVDT sensors for validation.​
Frame subtraction of the bridge model
Frame subtraction of the bridge model
April 10, 2020: In addition to the resonant frequencies and vibrational deflections of each mode, using motion magnification is also possible to obtain a full-field operational deflection shape of the structure. 
As the frequencies can change due to the load conditions with the same amplitude as a damage. The extracted shapes for healthy and damaged structures are compared. Focusing on how to identify damages using the operational deflection shapes collected from optical sensing, a machine learning approach is established.
Using computer vision to subtract the frames of the collected data of the bridge from a static view, the motion of the structure is extracted which can be adopted as input to the artificial neural nets. As the frames contain different modes of deflections due to broad band excitation, the machine learning algorithm can identify damages in a versatile fashion.

  • Assistant Professor Zhu Mao, Ph.D., Mechanical Engineering