Efficient simulation of guided wave testing through frequency-domain synthesis: a comparison with time-domain methods

Alvaro Gavilán-Rojas, Aymeric Orhan, Christophe Droz

Guided Wave Testing (GWT) is a cornerstone of nondestructive evaluation and structural health monitoring of critical infrastructure such as pipelines and rails. Due to the dispersive and multi-modal nature of guided waves, their interaction with defects in buried, immersed, fluid-loaded, composite, or welded structures is particularly complex, rendering damage quantification highly uncertain. Physics-based simulations are frequently employed to reduce this uncertainty, yet accurate numerical modeling of wave-defect interactions in such structures continues to represent a major computational bottleneck. Traditional time-step finite element methods, while robust, require extreme mesh densities and infinitesimal time steps due to the ultrasonic nature of guided waves, leading to prohibitive solve times for complex models. This study presents a relevant comparison between these traditional methods included in commercial software and a computationally efficient alternative which performs scattering analyses directly in the frequency domain. We compute the frequency response function (FRF) of a damaged steel pipe using a hybrid semi-analytical finite element (SAFE–FE) approach, with SAFE describing the guided waves and FE modeling the defect region. This approach treats the corrosion defect as a localized scatterer within an infinite homogeneous waveguide, eliminating the need for large spatial domains for the homogeneous part, which often include absorbing boundary layers. The final time-domain signals are obtained via Fourier synthesis, where the FRF is multiplied by the excitation spectrum. Results demonstrate that this methodology is in good agreement with a traditional finite element scheme. More importantly, when parametric studies for defect characterization and machine learning dataset generation are envisaged, the frequency-domain approach yields a reduction in computational overhead by orders of magnitude compared to traditional transient solvers.