Full-field camera measurements for structural and vibro-acoustic system analysis

Cameras are an aspiring technology for structural and vibro-acoustic system analysis. They offer full-field, transient, and contactless displacement measurements. However, measurement noise is one of the main limitations preventing their widespread application.

This thesis proposes new noise reduction algorithms taking advantage of geometry and physics information. By approximating with local polynomial functions (polynomial filtering) the camera data is denoised and beyond that, geometry information and surface gradients are obtained. In a complementary approach, model-based filtering incorporating a physics-based finite element model is proposed. Thereby, state estimation with the Kalman filter is exploited for superior estimates of the measured quantity. Thanks to the noise reduction, camera measurements are leveraged for applications at multiple stages of a noise and vibration analysis.

3D components with curved geometries, which previously have been rarely investigated due to their geometrical complexity, are targeted. Therefore, robust workflow for the definition of measurement points is developed based on feature matching.

Camera-based modal analysis of academic and industrial structures is validated and provides accurate estimates of eigenfrequencies and mode shapes. Enabled by polynomial filtering, an approach for structural intensity analysis applicable to general shells is developed. Employing a novel finite element model-based approach, camera measurements are exploited to predict the acoustic radiation of curved, potentially unbaffled surfaces. Moreover, it is demonstrated that model-based filtering can extend the capabilities of the camera for the virtual sensing of acceleration.

Validation of the proposed camera measurements compared to accelerometers and microphones yields good agreement. Even though overall not as accurate as conventional transducers even considering advanced noise reduction, the full-field character allows for valuable insights that are inaccessible with single-point sensors. Furthermore, the camera results are less prone to noise than the reference transducers at low frequencies.