Data-Driven, Sparsity-Based Matched Field Processing for Structural Health Monitoring
2014-05-01T00:00:00Z (GMT) by
This dissertation develops a robust, data-driven localization methodology based on the integration of matched field processing with compressed sensing ℓ<sub>1</sub> recovery techniques and scale transform signal processing. The localization methodology is applied to an ultrasonic guided wave structural health monitoring system for detecting, locating, and imaging damage in civil infrastructures. In these systems, the channels are characterized by complex, multi-modal, and frequency dispersive wave propagation, which severely distort propagating signals. Acquiring the characteristics of these propagation mediums from data represents a difficult inverse problem for which, in general, no readily available solution exists. In this dissertation, we build data-driven models of these complex mediums by integrating experimental guided wave measurements with theoretical wave propagation models and ℓ<sub>1</sub> sparse recovery methods from compressed sensing. The data-driven models are combined with matched field processing, a localization framework extensively studied for underwater acoustics, to localize targets in complex, guided wave environments. The data-driven matched field processing methodology is then refined, through the use of the scale transform, to achieve robustness to environmental variations that distort guided waves. Data-driven matched field processing is experimentally applied to an ultrasound structural health monitoring system to detect and locate damage in aluminum plate structures.