Neural operator-enabled forward and inverse modeling of laser-induced surface acoustic waves and applications in nondestructive evaluation

Abstract

Laser-induced surface acoustic wave (SAW)-driven nondestructive evaluation offers high-resolution, non-contact characterization of subsurface microstructures. However, its practical application is often limited by the high computational costs associated with traditional numerical simulation methods. Recently, machine learning has emerged as an attractive alternative to accelerate these simulations. In this paper, we develop a neural operator-enabled framework for both forward and inverse modeling of laser-induced SAW propagation. A general dataset with randomly generated subsurface structures is used to evaluate and quantify the model's performance in both wave propagation and subsurface inversion problems. Three potential applications are then investigated: subsurface crack localization, multilayer structure characterization and polycrystalline grain imaging. The results demonstrate that the neural operator-enabled model achieves satisfactory accuracy even in the presence of noise and source waveform variations, underscoring its potential as an efficient and accurate surrogate model for practical nondestructive evaluation using laser-induced SAWs.