Baseline-Free Defect Detection in Curved Surfaces Using a Neural Surrogate Wavefield Model

Abstract

Accurate defect localization in structural health monitoring (SHM) can be performed by using high-resolution wavefield data, which can be challenging to obtain due to hardware limitations and time constraints. In this work, we propose a data-driven surrogate model that reconstructs high-resolution wavefields from sparse measurements, enabling efficient and precise defect localization. Our approach leverages a coordinate-based neural network to interpolate waveforms continuously, allowing for theoretically infinite resolution wavefield predictions. The proposed framework reduces data acquisition demands while preserving the fidelity of wavefield information, making it an efficient tool for SHM. Modeling results using numerical simulation data demonstrate that our approach reconstructs wavefield data and can be used to identify scattering sources associated with defects.