Impulsive Source Localization on a Metal Plate Using a Non-Contacting Acoustic Sensor Array

Abstract

Acoustic waves are well-suited for remote sensing and structural health monitoring as they convey source information and can be recorded without contact. Acoustic sensor arrays enable remote recording of structure-borne and airborne sounds for assessing the health of a structure. A key structural health monitoring task is localizing impact excitations, but traditional localization methods like Time Difference of Arrival, Modal Acoustic Emission, Near-Field Beamforming, and Near-Field Holography face challenges due to geometric complexity, dispersive wave propagation, fluid-structure coupling, or difficult implementation. Many techniques also rely on contacting sensors, which can alter the structure and pose maintenance challenges. This dissertation adapts Bartlett Matched Field Processing (MFP), a source localization technique from underwater acoustics, for structural applications. Using remote acoustic array measurements and wave propagation simulations, MFP was applied to a 0.64-cm thick, 91.4 cm diameter aluminum plate. Microphones and hydrophones positioned sufficiently far from the plate captured the source characteristics in the 5-20 kHz frequency bandwidth while excluding evanescent waves. Simulations in COMSOL Multiphysics® modeled fluid-structure interactions from impact excitations. Initially, MFP was implemented using an axisymmetric finite-element wave-propagation model for an infinitely large plate, applicable to all plate impacts with proper time-windowing to exclude plate edge artifacts from experimental measurements. Performance was evaluated in controlled experiments and synthetic noise environments, with accuracy assessed through two-dimensional ambiguity surfaces. Results showed that MFP outperformed near-field beamforming, achieving localization errors as low as 1 cm compared to 10 cm, and maintained accuracy down to a signal-to-noise ratio (SNR) of -7.5 dB. To better represent experimental conditions and reduce time-windowing requirements, the finite element model was refined to include plate edge effects specific to the plate studied. Results indicated that MFP provided accurate localization in a noiseless environment, achieving a 0.5 cm localization error without the need for time-windowing to remove plate edge reflection artifacts. Additionally, MFP maintained over 80% localization accuracy within 4.5 cm of the true source at a SNR of -7.5 dB. A data-driven approach using Neural Networks (NNs) was also explored for source localization. Two architectures were tested: a Feed-Forward Neural Network (FNN) using cross-correlation lags and a Convolutional Neural Network (CNN) leveraging spectrogram images. NNs required only 20% of the sampling density used by MFP, making them attractive for real-time applications. However, they performed poorly in noisy environments, with FNNs achieving 70% accuracy at a SNR of 15 dB and CNNs achieving less than 20% accuracy at a SNR of 30 dB. Finally, an additional complication was added to the MFP source localization problem through the introduction of water beneath the plate to form a three-domain acoustic environment. MFP ambiguity surfaces created using sensors in the water exhibited larger ambiguity zones than the sensors in the air due to the four-times longer wavelengths in water, leading to slightly higher localization errors (1.7 cm vs. 1 cm). A combined microphone-hydrophone configuration mitigated this issue, improving resolution and achieving 0.55 cm localization accuracy. Through these studies, this dissertation advances source localization for structures by adapting MFP for structural health monitoring, using experiments for evaluating robustness across different environments, and benchmarking against data-driven techniques. The findings provide insights into acoustic array configurations, means to improve impact localization accuracy, and practical implementation considerations for applying MFP in structural health monitoring and non-destructive testing.