Session: 02-06-01: Machine Learning in Structural Dynamics and Aeroelasticity

Paper Number: 182078

182078 - Scan Once for Your Observation — Modal Ridge Detection and Material Property Regression Using Lamb Wave F-K Spectra 

Lamb waves are widely used in non-destructive evaluation (NDE) and structural health monitoring (SHM) due to their strong sensitivity to both material properties and structural integrity. In particular, the A0 and S0 modes carry critical dispersion behaviors, which are commonly leveraged to detect damage, characterize materials, and estimate parameters such as plate thickness and Young's modulus. However, analyzing Lamb wave dispersion in practical applications remains highly challenging. One of the most common approaches is transforming the raw signal into a frequency–wavenumber (f-k) spectrum and then extracting the dispersion curves. In noisy environments, especially at low signal-to-noise ratios (SNR), the modal ridges in the f-k spectrum can become severely obscured by measurement noise. Additionally, dispersion curves typically span a wide frequency range, which limits the effectiveness of traditional filtering methods. Filters such as low-pass or band-pass often truncate dispersion curves at their cutoff frequencies, resulting in significant information loss. Moreover, conventional workflows typically treat modal ridge detection and material parameter estimation as separate stages. These methods often rely on human expertise, handcrafted features, or intermediate curve-fitting procedures, which are time-consuming, error-prone, and inefficient—particularly in high-noise or large-scale conditions.

To address these limitations, we propose Scan Once for Your Observation (SOYO), a deep learning-based framework that performs either modal ridge detection or material property regression directly from raw f-k spectra. SOYO consists of two lightweight yet robust models, each tailored to a specific task. The first model, based on a UNet segmentation network, generates heatmaps that localize the A0 and S0 modal ridges within the f-k spectrum, enabling automatic and high-resolution detection of modal ridges, even under low-SNR conditions such as 10 dB. The second model, combining a UNet and a global regression module, directly evaluates quantities associated with material properties, including plate thickness and Young's modulus, by utilizing full-spectrum information. Rather than relying solely on the visible modal ridges, the model learns complex mappings from spectral features to physical parameters.

Both models are trained and evaluated on a large-scale synthetic dataset that includes a wide range of material and excitation conditions. Specifically, we tested simulated wave data obtained from structures with the Young's modulus varying from 40.0 to 71.7 GPa, and with the plate thickness ranges from 0.2 to 2.0 mm, considering isotropic metallic materials typically considered in SHM/NDE inspection. To evaluate performance under noisy conditions, we consider three signal-to-noise ratio (SNR) levels representing high (20 dB), medium (15 dB), and low (10 dB) noise environments. Across all noise levels, the models maintain high prediction accuracy. Under high and medium SNR conditions, none of the test samples exceeds 10% relative error in either modulus or thickness prediction. Even under low SNR conditions, the framework remains reliable, with over 94% of predictions staying within 10% relative error. The segmentation model also demonstrates strong robustness, achieving an average mean squared error (MSE) of 0.00316 in A0 and S0 heatmap prediction across the test set. These results confirm that SOYO is capable of performing reliably in highly noise-contaminated environments.

The proposed SOYO framework provides an efficient, noise-tolerant, and fully automated solution for Lamb wave-based dispersion analysis and material property estimation. Its end-to-end design eliminates the need for manual preprocessing, making it well-suited for real-time deployment in complex SHM applications across industrial settings.

Presenting Author: Boshi Chen University of South Carolina

Presenting Author Biography: Boshi Chen received his B.S. degree in Biomedical Engineering from Hefei University of Technology, Hefei, China, in 2023. He is currently pursuing a Ph.D. in Mechanical Engineering at the University of South Carolina. His research interests include non-destructive evaluation and autonomous systems.