Transmission Ratio (TR) of surface waves has been widely employed to estimate crack depth in concrete structures, particularly within the normalized depth range of ℎ/λ ≤ 0.3. While TR provides a simple amplitude-based indicator, its sensitivity and reliability can vary depending on the heterogeneous nature of concrete and the influence of scattering and near-field effects around the crack. To explore ways to address these limitations, this study introduces an approach that incorporates multiple surface-wave characteristics obtained from a fully non-contact Air-Coupled Ultrasonic Testing (ACUT) configuration. In addition to the conventional transmission component, reflection behavior and localized scattering features observed near the crack are considered to construct complementary indicators for depth assessment. By integrating these diverse surface-wave responses, the proposed framework aims to provide a more flexible and potentially more informative basis for evaluating cracks in concrete. Preliminary analyses suggest that this multi-feature perspective may help supplement the constraints of TR-based evaluation and could contribute to broadening the applicability of surface-wave–based crack quantification. This study offers an initial step toward developing an expanded and adaptable assessment framework while retaining the advantages of a fully non-contact inspection scheme.

Coatings such as boron nitride (BN) and yttrium oxide (Y₂O₃) are critical in metal casting, as they directly influence casting quality, defect reduction, and mold longevity. Achieving high-precision castings is critical, as even minor imperfections can compromise the mechanical properties and performance of the final product. The quality and thickness of these coatings impact solidification conditions, contamination risks (such as carbon uptake), and dimensional accuracy. Traditionally, coatings are applied manually to casting molds as sprays or paints, with visual inspection used to assess uniformity, which may be insufficient to achieve high-precision castings. In this work, we present the use of laser-based ultrasound, which is a nondestructive, remote, and all-optical technique, to characterize mold coating thickness and uniformity. We excite and detect surface acoustic waves (SAWs) in coated graphite samples and show preliminary results that indicate the time of arrival of the SAWs is sensitive to coating thickness. Our preliminary results demonstrate that laser generation and detection of SAWs offers an effective and powerful approach for the noncontact and nondestructive characterization of advanced coating systems, which is essential for meeting the stringent demands of modern metal casting and maintaining high standards of product reliability and efficiency.

This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344. The document release number is LLNL-ABS-2014038.

The growing use of carbon‑fiber reinforced polymers (CFRPs) in aerospace demands rapid, fully automated non‑destructive testing (NDT). Laser‑ultrasonic testing (LUT) provides a contact‑free, broadband technique, yet the excited waves generate complex time series that are difficult to interpret in both the time and frequency domains, making current inspection workflows time‑intensive. In the literature, various feature representations of the time and frequency data have been used to train deep‑learning models for classifying ultrasonic time series data and automating defect identification. The presented study uniquely determines the most effective data representation for LUT data by comparing multiple modalities, enabling superior defect‑classification accuracy.
A CFRP laminate with embedded artificial delaminations was fabricated by introducing PTFE inlays of varying size between layers at different depths of the laminate. The data set was acquired by recording the transmitted waves with an optical microphone on the opposing side of the exciting laser. Four feature representations were benchmarked as network inputs: time‑amplitude waveforms, Fourier spectra, wavelet‑packet‑transformation (WPT) coefficients, and continuous wavelet transform (CWT) scalograms. A deep residual 1D‑CNN was trained on each of the first three representations, and an equivalent 2D‑CNN on the CWT scalograms. All four models were evaluated via k‑fold cross‑validation and different validation metrics. CWT scalograms performed best across all modalities. Raw amplitude data and WPT coefficients followed second with no significant difference between them, while FFT spectra performed worse yet still yielded very good overall classification results.
Choosing the optimal modality as network input can improve classification accuracy, representing an important step toward automated defect identification in laser‑ultrasonic measurements.