Modal identification of simple structures with high-speed video using motion magnification

Justin G. Chen, Neal Wadhwa, Young‐Jin Cha, Frédo Durand, William T. Freeman, Oral BüyüköztürkJournal of Sound and Vibration 2015 · DOI

TL;DR

This paper introduces a novel method for identifying the vibrational characteristics (modal identification) of structures using high-speed video and a computer vision technique called motion magnification. This non-contact approach offers a promising alternative to traditional methods, enabling detailed analysis of structural behavior with greater ease and potentially lower cost.

What Problem Does This Paper Address?

Modal analysis is crucial for understanding how structures vibrate and for assessing their health. Traditional methods often rely on contact sensors like accelerometers. While accurate, these sensors can be tedious to install, especially on large or complex structures, and their added mass can sometimes influence the vibration being measured. Non-contact methods exist, such as laser vibrometry, but they can be expensive or have limitations. The authors aim to leverage the accessibility and simplicity of video cameras, combined with advanced computer vision, to provide a robust and potentially more cost-effective way to perform modal identification. The challenge lies in extracting precise displacement data from video, which is inherently difficult due to the nature of image data and the small magnitudes of vibrations.

Methods

The core of the paper lies in the application of motion magnification techniques, specifically a phase-based approach, to high-speed video of vibrating structures. Here's a breakdown:

  • Motion Magnification: This technique amplifies subtle motions in video, making them visually apparent. It works by decomposing video frames into spatial amplitude and phase components using steerable pyramid filters. The phase signals are then temporally filtered, amplified, and reconstructed to create a video where small movements are exaggerated. This is particularly useful for visualizing operational deflection shapes at specific resonant frequencies.
  • Displacement Extraction: For quantitative analysis, the authors employ a method related to motion magnification that extracts displacement signals. This technique uses local phase and amplitude information from oriented complex spatial bandpass filters. The local phase, when tracked over time, corresponds to the displacement of the structure. This displacement is computed at edges within the video and is then spatially weighted to improve the signal-to-noise ratio. The displacement is then converted from pixels to millimeters.
  • Operational Deflection Shape (ODS) Derivation:
    • From Extracted Displacements: The extracted displacement time series from various points in the video (acting as "virtual accelerometers") are analyzed using standard modal analysis techniques (e.g., cross-spectrum analysis) to identify resonant frequencies and ODS.
  • Directly from Magnified Video: A simpler, though less accurate, method involves pausing the motion-magnified video at peak displacement and using edge detection (like Canny edge detection) to outline the structure's deformed shape.

Datasets

The paper does not introduce new benchmark datasets. Instead, it utilizes experimental data collected from:

  • A steel cantilever beam: Instrumented with accelerometers, this served as a primary testbed for comparing camera-based measurements against traditional sensors.
  • A 4-inch schedule 40 PVC pipe: Used to demonstrate the methodology on a different geometry, focusing on circumferential-radial modes.

The experiments involved exciting these structures with impact hammers and recording their vibrations with a high-speed Phantom v10 camera.

Key Insights and Ideas

  • Video as a "Virtual Accelerometer": The displacement extraction method effectively turns regions of a video into "virtual accelerometers," providing displacement data at numerous points without physical contact.
  • Motion Magnification for Visualization: The ability to visually amplify subtle vibrations is invaluable for quickly understanding how a structure deforms at its resonant frequencies, especially for complex geometries where intuition might fail.
  • Quantitative Validation: The paper rigorously validates the camera-based displacement extraction against established methods (laser vibrometer and accelerometers), demonstrating its accuracy.
  • Noise Floor Trade-offs: While camera-based methods offer non-contact advantages, their noise floor is higher than that of laser vibrometers or accelerometers. However, the paper suggests ways to improve this, such as increasing resolution or capturing more spatial scales.

Key Results

  • Verification Test: The displacement extracted from the camera signal closely matched the integrated laser vibrometer displacement, with a correlation of 99.6%.
  • Noise Floor: The camera's noise floor was approximately 40 nm for a 9-second measurement, significantly higher than the laser vibrometer (0.2 nm) but still capable of identifying the first four resonant frequencies of the cantilever beam.
  • Cantilever Beam ODS: Operational deflection shapes extracted from camera data compared favorably to accelerometer data, with MAC values above 95% for the first four modes.
  • Edge Detection ODS: Deriving ODS directly from motion-magnified video using edge detection yielded results that were less accurate than displacement extraction but still provided a useful qualitative comparison.
  • Pipe ODS: The method successfully visualized the first three circumferential-radial modes of a PVC pipe.

Limitations and Future Work

  • Noise Floor: The current noise floor of the camera-based method is higher than that of laser vibrometers, limiting its sensitivity for detecting very small vibrations.
  • Automation: While the displacement extraction is largely automated, picking resonant frequencies and cleaning up masks for ODS derivation still require manual intervention.
  • Complex Geometries and Environments: The paper acknowledges challenges in applying the methodology to structures with complex geometries, closely coupled modes, and in outdoor environments (due to camera movement, lighting variations, and atmospheric effects).
  • Future Research Directions: The authors suggest developing algorithms for fully automatic extraction of ODS, improving the accuracy of ODS derived directly from motion-magnified video, and addressing the challenges of complex structures and outdoor measurements. The ultimate goal is to adapt this technology for real-world structural health monitoring and damage detection.

Summary updated

BibTeX
@article{chen2015modal,
  author = {Justin G. Chen and Neal Wadhwa and Young‐Jin Cha and Frédo Durand and William T. Freeman and Oral Büyüköztürk},
  title = {{Modal identification of simple structures with high-speed video using motion magnification}},
  journal = {Journal of Sound and Vibration},
  year = {2015},
  doi = {10.1016/j.jsv.2015.01.024},
}
Figure