Automated Defect Localization Via Low Rank Plus Outlier Modeling of Propagating Wavefields
Stefano Gonella (University of Minnesota), Jarvis Haupt (University of Minnesota, Department of Electrical and Computer Engineering)
Characterization and Imaging of Structural and Material Imperfections
Mon 4:20 - 5:40
Barus-Holley 191
This work proposes an agnostic inference strategy for material diagnostics, based on the concept of spatial saliency. This strategy is conceived within the context of laser-based nondestructive evaluation methods, which attempt to extract the signature of structural anomalies from the analysis of the acoustic wavefields measured on the structure’s surface by means of scanning laser vibrometers. As a consequence of this, the method requires and exploits the availability of a spatially dense acquisition grid. The approach consists of two main steps. The first is a special spatiotemporal windowing strategy designed to partition the structural domain in small regional sub-domains and replicate impinging wave conditions at each discrete location; this has the scope to de facto align the regional wave motion events, thus highlighting the saliency of possible discriminating features associated with the presence of scattering. The second step is the construction of a low rank plus outlier model of the regional wavefields, following a principal component analysis formalism. A region is labeled salient when its behavior does not belong to a common subspace of a certain feature space, i.e. when its signature is markedly different from the typical one observed in the surrounding medium. The technique is capable to identify a priori unknown deviations in the propagating wavefields caused by material inhomogeneities or defects, using virtually no knowledge of the structural and material properties of the medium and only invoking concepts of saliency and data sparsity. This characteristic makes the approach particularly suitable for diagnostics scenarios where the geometric and material models are complex, unknown, or unreliable. We demonstrate our approach in a simulated environment using as benchmark problem the localization of point and line defects in a thin plate by means of flexural waves.