Learning Guided Wave Dispersion Curves from Multi-Path Reflections with Compressive Sensing

Abstract

Guided wave dispersion curves characterize materials and the waves that propagate in them. While there have been many efforts to learn dispersion curves from data, multipath reflections cause most methods to either fail or require a significant quantity of data. This paper uses compressive sensing and sparse recovery algorithms to learn dispersion curves from known multi-path characteristics. We assume that reflection signals originate from virtual sources, according to the method of mirrors. We demonstrate our methodology with guided wave simulation data of a 1.125m by 1.00 m convex, nonsquare aluminum plate. We show that our multi-path method can accurately estimate dispersion curves with as few as 20 measurements, whereas more than 100 measurements would be necessary without it.