Machine Learning-Based Ultrasonic Weld Characterization Using Hierarchical Wave Modeling and Diffusion-Driven Distribution Alignment

TL;DR

This work tackles automated ultrasonic weld inspection under data scarcity and environmental variability by marrying physics-based forward models with modern ML. A hierarchy of forward models (NL for high fidelity and a Lamb-wave–based EM surrogate) generates diverse training data, while diffusion-driven distribution alignment maps real LDV measurements into the simulated data distribution. Inversion is performed by parallel U-Nets enhanced with CBAM, FNO, and MAFE, jointly predicting weld stiffness maps and crack masks; a conditioned diffusion model further enables high-fidelity representations of OOD measurements. Results show EM data enrichment improves NL inversion performance and that guided DDPM can robustly align experimental scans, enabling accurate weld-characterization and crack detection in realistic NDE settings.

Abstract

Automated ultrasonic weld inspection remains a significant challenge in the nondestructive evaluation (NDE) community to factors such as limited training data (due to the complexity of curating experimental specimens or high-fidelity simulations) and environmental volatility of many industrial settings (resulting in the corruption of on-the-fly measurements). Thus, an end-to-end machine learning (ML) workflow for acoustic weld inspection in realistic (i.e., industrial) settings has remained an elusive goal. This work addresses the challenges of data curation and signal corruption by proposing workflow consisting of a reduced-order modeling scheme, diffusion based distribution alignment, and U-Net-based segmentation and inversion. A reduced-order Helmholtz model based on Lamb wave theory is used to generate a comprehensive dataset over varying weld heterogeneity and crack defects. The relatively inexpensive low-order solutions provide a robust training dateset for inversion models which are refined through a transfer learning stage using a limited set of full 3D elastodynamic simulations. To handle out-of-distribution (OOD) real-world measurements with varying and unpredictable noise distributions, i.e., Laser Doppler Vibrometry scans, guided diffusion produces in-distribution representations of OOD experimental LDV scans which are subsequently processed by the inversion models. This integrated framework provides an end-to-end solution for automated weld inspection on real data.

Figures (15)

• Figure 1: Weldline stiffness parameterization depicting (a) the nominal weld stiffness profile for a straight weldline through the domain, (b) the modulated weldline, with beading and Gaussian variation applied, and (c) a meshed 3D geometry based on (b).
• Figure 2: The solutions of the RL dispersion relation in terms of wavenumber, group velocity, and phase velocity versus the frequency-thickness produce, $F\times d$.
• Figure 3: Confirmation of the Lamb-wave assumptions. (a) A full D continuum for computing NL solutions and reduced 2D continuum for generating EM solutions. (b) The resulting out-of-plane displacement fields at $z=h$ with their filtered (c) $(A,0)$ and (d) $(S,0)$ components showing qualitative agreements. (e) The spectra of each wavefield show aggregated energy along pre-computed Lamb wave mode numbers (dashed-lines), confirming the alignment of the high-fidelity problem with Lamb theory.
• Figure 4: Problem Domains for the (a) scattering, (b) cylinder, and (c) coupon problem classes with $w_{\mathsf{PML}}^s$ denoting the PML boundary width in the $s=\{x,y\}$ directions. The variable $\bm{\varphi}$ denotes either $\bm{u}(\bm{x})$ or $\psi^{(m,n)}(\bar{\bm{x}})$ for elastodynamic and EM solutions, respectively.
• Figure 5: The (a) experimental weld sample of 0.25-in. thick steel with labeled crack regions penetrating roughly 30% of the weldine, and (b) surface velocities collected by slow, medium, and fast LDV scans.