This study introduces a practical methodology for deploying a fully lead-free, LiNbO₃(LN)-based sol-gel composite ultrasonic transducer capable of continuous operation at temperatures exceeding 700⁰C, with demonstrated efficacy up to 1000⁰C, without the need for a couplant or backing. Laboratory findings have been translated into actionable steps for industrial application: spray coating the LN-based film onto Inconel, conducting low-temperature poling via positive-corona discharge, and optimizing a 100μm film for 2mm diameter defect echo detection during thermal cycling. The session will elucidate the relationship between performance trends, such as center-frequency shift, bandwidth broadening, and signal-to-noise behavior, and the choices made by attendees regarding materials, substrates, and heat treatment. These insights will be mapped to nondestructive testing (NDT) applications, such as high-temperature pipe monitoring and process monitoring such as light-metal die casting.

In this presentation, we will consider resonant ultrasound spectroscopy (RUS) as a nondestructive technique for evaluating microstructural properties. To this end, we will present RUS results for plasma-sprayed thermal barrier coatings. These coatings exhibit anisotropic elasticity due to their splat-based layered structure, which includes numerous preferentially oriented defects. We determined all the independent elastic constants of plasma-sprayed coatings using RUS coupled with laser Doppler interferometry. This allowed us to overcome the difficulty of identifying resonant modes caused by high internal friction resulting from microstructural defects. The elastic constants and internal friction values obtained from resonance spectrum measurements were closely related to the microstructure of the coatings. For example, longitudinal and shear elastic anisotropies provided information on preferentially oriented inter-splat pores and intra-splat cracks. These results demonstrate that RUS provides a powerful foundation for the nondestructive evaluation of microstructural characteristics.

Conventional ultrasonic testing for crack detection relies heavily on skilled operators to interpret complex signals from weld areas and component geometries. This dependency creates challenges in knowledge transfer and efficiency, especially under anticipated workforce shortages.
To address these issues, we developed automated analysis systems for phased array ultrasonic testing (PAUT) images using machine learning.
A convolutional neural network (CNN) was trained on PAUT images of weld lines with stress corrosion cracking, labeled according to operator classifications. The dataset included inspection images acquired from weld lines of internal structures in boiling water reactor plants, provided by Tokyo Electric Power Company Holdings, Inc., ensuring validation under real-world conditions. The optimized model enables accurate binary classification of crack presence, including challenging cases such as shallow surface-connected flaws and branched tips with weak echoes. Additionally, semantic segmentation was applied to detect crack tips and estimate depth by calculating distances between tip echoes and surface echoes.
The proposed system achieved high accuracy in crack detection and significantly reduced analysis time compared to manual inspection. To enhance trust and usability, we also explored methods for making machine learning decisions interpretable.
Integrating machine learning with PAUT addresses critical challenges in ultrasonic testing, improving reliability and efficiency while supporting sustainable workforce development.