Block-sparse reconstruction and imaging for Lamb wave structural health monitoring

A frequently investigated paradigm for monitoring the integrity of plate-like structures is a spatially-distributed array of piezoelectric transducers, with each array element capable of both transmitting and receiving ultrasonic guided waves. This configuration is relatively inexpensive and allows interrogation of defects from multiple directions over a relatively large area. Typically, full sets of pairwise transducer signals are acquired by exciting one transducer at a time in a round-robin fashion. Many algorithms that operate on such data use differential signals that are created by subtracting prerecorded baseline signals, leaving only signal differences introduced by scatterers. Analysis methods such as delay-and-sum imaging operate on these signals to detect and locate point-like defects, but such algorithms have limited performance and suffer when potential scatterers have high directionality or unknown phase-shifting behavior. Signal envelopes are commonly used to mitigate the effects of unknown phase shifts, but this further reduces performance. The blocksparse technique presented here uses a different principle to locate damage: each pixel is assumed to have a corresponding multidimensional linear scattering model, allowing any possible amplitude and phase shift for each transducer pair should a scatterer be present. By assuming that the differential signals are linear combinations of a sparse subset of these models, it is possible to split such signals into location-based components. Results are presented here for three experiments using aluminum and composite plates, each with a different type of scatterer. The scatterers in these images have smaller spot sizes than delay-and-sum imaging, and the images themselves have fewer artifacts. Although a propagation model is required, block-sparse imaging performs well even with a small number of transducers or without access to dispersion curves.