Model-Based and Data-Based Approaches for Monitoring Elastic Structures Using Acoustics and Ultrasounds

TIME: 3:30 PM

LOCATION: EIS 104

SPEAKER: Sandrine T. Rakotonarivo, Marine Physical Laboratory, Scripps Institute of Oceanography, University of California, San Diego

ABSTRACT:

This talk will present different strategies for monitoring of elastic structures using acoustics and ultrasound in the context of non-destructive testing, structural health monitoring and underwater acoustics. The development of localization and characterization methods is based on solving a signal-to-noise ratio (SNR) problem, which can be maximized by using low-noise measurement devices with as many sensors as possible or by measuring in a simple environment, i.e. isotropic, homogeneous, infinite and low attenuation. However, in several practical applications, the SNR is compromised because the set of sensors may be sparse and the structure or environment under test is generally finite and attenuating and, depending on the case, may be anisotropic (e.g., multi-pass welding), reverberant (e.g., waveguide, shell), and diffuse (e.g., acoustic noise field, concrete specimen). In this context, this talk presents active and passive methods for detecting and localizing a local change of parameters on an elastic structure based on the perturbation approach. In the presence of sufficient a priori knowledge of the physics, parametric approaches based on the adjoint formalism and matched filtering are investigated. In the opposite case of low a priori information, a data-based approach is adopted to feed the forward model with measurements. Finally, we discuss the use of physics-informed neural networks to combine measurement data with partial knowledge of the physics to construct a surrogate model that can be used for either localization or characterization purposes.

HOST: Margherita Capriotti

VIDEO: