Data-driven approaches for damage-type classification in vibrating square plates

Abstract

Acoustic radiation from a mechanical structure due to broadband forcing is inherently dependent on the structure’s material, geometry, boundary conditions, and mechanical status. Measurements of this acoustic or vibrational response can be used to detect mechanical changes (i.e., damage) when compared to known baseline measurements of the structure. Often, however, knowledge of the type of damage is useful for subsequent considerations. Even for relatively simple structures, like flat plates, the lack of simple solutions to problems involving localized damage makes analytical treatments challenging. This is further complicated since the location and severity of damage are typically unknown a priori. We present a data-driven approach for classification of various forms of damage—including cuts, localized mass changes, delamination, and boundary changes—in a vibrating clamped square plate. Frequency response curves are generated using FEA of a 30-by-30-by-0.3-cm aluminum plate with various types and severity of damage. Features (including changes in peak-locations, -amplitudes, and -widths compared to baseline) are extracted and used to train classifiers, with performance quantified using auxiliary test data. Comparisons are made between classifier types, including nearest neighbor methods, discriminant analysis, and support vector machines. [Work supported by NAVSEA through the NEEC and the US DoD through an NDSEG Fellowship].Acoustic radiation from a mechanical structure due to broadband forcing is inherently dependent on the structure’s material, geometry, boundary conditions, and mechanical status. Measurements of this acoustic or vibrational response can be used to detect mechanical changes (i.e., damage) when compared to known baseline measurements of the structure. Often, however, knowledge of the type of damage is useful for subsequent considerations. Even for relatively simple structures, like flat plates, the lack of simple solutions to problems involving localized damage makes analytical treatments challenging. This is further complicated since the location and severity of damage are typically unknown a priori. We present a data-driven approach for classification of various forms of damage—including cuts, localized mass changes, delamination, and boundary changes—in a vibrating clamped square plate. Frequency response curves are generated using FEA of a 30-by-30-by-0.3-cm aluminum plate with various types and severit...