Lithium niobate (LiNbO${}_3$) is known for its high Curie temperature, making it an attractive candidate for high-temperature piezoelectric applications (>200 °C); however, the literature suffers from a paucity of reliable material properties data at high temperatures. This paper therefore provides a complete set of elastic and piezoelectric coefficients, as well as complex dielectric constants and the electrical conductivity, for congruent monocrystalline LiNbO${}_3$ from 25 °C to 900 °C at atmospheric pressure. An inverse approach using the electrochemical impedance spectroscopy (EIS) resonance method was used to determine the materials’ coefficients and constants. Single crystal Y-cut and Z-cut samples were used to estimate the twelve coefficients defining the electromechanical coupling of LiNbO${}_3$. We employed an analytical model inversion to calculate the coefficients based on a linear superposition of nine different bulk acoustic waves (three longitudinal waves and six shear waves), in addition to considering the thermal expansion of the crystal. The results are reported and compared with those of other studies for which the literature has available values. The dominant piezoelectric stress constant was found to be $e_{15}$, which remained virtually constant between 25 °C and 600 °C; thereafter, it decreased by approximately 10% between 600 °C and 900 °C. The elastic stiffness coefficients $c_{11}^E$, $c_{12}^E$, and $c_{33}^E$ all decreased as the temperature increased. The two dielectric constants ${ϵ}_{11}^S$ and ${ϵ}_{33}^S$ increased exponentially as a function of temperature.

Lithium niobate is a promising candidate for use in high-temperature piezoelectric devices due to its high Curie temperature ($\approx$ 1483 K) and strong piezoelectric properties. However, the piezoelectric behavior has, in practice, been found to degrade at various temperatures as low as 573 K, with no satisfactory explanation available in the literature. We, therefore, studied the electrical conductivity of congruent lithium niobate single crystals in the temperature range of 293–1273 K with an 500 mV excitation at frequencies between 20 Hz and 20 MHz. An analytical model that generalizes the universal dielectric relaxation law with the Arrhenius equation was found to describe the experimental temperature and frequency dependence and helped discriminate between conduction mechanisms. Electronic conduction was found to dominate at low temperatures, leading to low overall electrical conductivity. However, at high temperatures, the overall electrical conductivity increases significantly due to ionic conduction, primarily with lithium ions (Li⁺) as charge carriers. This increase in electrical conductivity can, therefore, cause an internal short in the lithium niobate crystal, thereby reducing observable piezoelectricity. Interestingly, the temperature above which ionic conductivity dominates depends greatly on the excitation frequency: at a sufficiently high frequency, lithium niobate does not exhibit appreciable ionic conductivity at high temperature, helping explain the conflicting observations reported in the literature. These findings enable an appropriate implementation of lithium niobate to realize previously elusive high-temperature piezoelectric applications.

The inspection of structures operating at high temperatures is a major challenge in a variety of industries, including the energy and petrochemical industries. Operators are typically performing nondestructive evaluations using ultrasound to monitor component thicknesses during scheduled shutdowns, thereby ensuring safe operation of their plants. However, despite being costly, this calendar-based approach may lead to undetected corrosion, which can potentially result in catastrophic failures. There is therefore a need for ultrasonic transducers designed to withstand permanent exposure to high temperatures, so as to continuously monitor the remnant thicknesses of structures in real time. This paper discusses the design of a heat-resistant ultrasonic transducer based on a piezoelectric element. The piezoelectric material, the electrodes, the backing layer, the wires and the casing are presented in detail from the acoustic and thermal expansion point of view. Four transducers optimized for 3 MHz were manufactured and tested to destruction in different conditions: (1) 72-h temperature steps from room temperature to 750 ${}^{\circ }$C, (2) thermal cycles from room temperature to 500 ${}^{\circ }$C and (3) 60 days of continuous operation at >550 ${}^{\circ }$C. The paper discusses the results, as well as the effect of temperature over time on the properties of the transducer.

Phased Array Ultrasonic Testing (PAUT) is one of the most widely usedtechniques in nondestructive testing (NDT), for identifying and characteriz-ing defects within materials, particularly in welds. However, interpretationof PAUT data significantly depends on the inspector’s expertise. Moreover,geometric features associated with the weld often introduce artefacts, whichrender the analysis more complex and increase interpretation time. While ar-tificial intelligence (AI) has demonstrated outstanding capabilities in objectdetection across various fields, its application to ultrasonic testing has alsoadvanced significantly, despite the limited data availability due to confiden-tiality constraints. This work investigates the integration of AI into PAUTanalysis by comparing two state-of-the-art deep learning architectures forautomatic flaw detection and characterization: 1) a two-step detector basedon the Faster R-CNN, and 2) the latest version of YOLO. To explore theimpact of contextual information on model performance, three types of datapreprocessing were analyzed: raw PAUT data used as a baseline, data en-hanced with overlay annotations to indicate the weld geometry, and datapreprocessed by subtracting echoes associated with the geometry, aimingto suppress structural artefacts and highlight defects. Experimental resultsshow that the enhanced Faster R-CNN architecture, combined with mediangeometry subtraction, achieved the highest performance with a mean Av-erage Precision (mAP) of 58.05%, outperforming YOLOv5 and YOLOv8across all dataset configurations. These findings provide practical guidancefor deploying deep learning in industrial PAUT inspection workflows, demon-strating that preprocessing strategy selection must account for defect-type-specific acoustic signatures. The study identifies geometry-induced artefactsuppression as beneficial for most defect classes (10 percentage point mAPimprovement). Results are contextualized within realistic deployment con-straints including limited training specimens (13 defects), cross-manufacturergeneralization challenges, and computational efficiency trade-offs between ar-chitectures.

Phased Array Ultrasonic Testing (PAUT) is one of the most widely usedtechniques in nondestructive testing (NDT), for identifying and characteriz-ing defects within materials, particularly in welds. However, interpretationof PAUT data significantly depends on the inspector’s expertise. Moreover,geometric features associated with the weld often introduce artefacts, whichrender the analysis more complex and increase interpretation time. While ar-tificial intelligence (AI) has demonstrated outstanding capabilities in objectdetection across various fields, its application to ultrasonic testing has alsoadvanced significantly, despite the limited data availability due to confiden-tiality constraints. This work investigates the integration of AI into PAUTanalysis by comparing two state-of-the-art deep learning architectures forautomatic flaw detection and characterization: 1) a two-step detector basedon the Faster R-CNN, and 2) the latest version of YOLO. To explore theimpact of contextual information on model performance, three types of datapreprocessing were analyzed: raw PAUT data used as a baseline, data en-hanced with overlay annotations to indicate the weld geometry, and datapreprocessed by subtracting echoes associated with the geometry, aimingto suppress structural artefacts and highlight defects. Experimental resultsshow that the enhanced Faster R-CNN architecture, combined with mediangeometry subtraction, achieved the highest performance with a mean Av-erage Precision (mAP) of 58.05%, outperforming YOLOv5 and YOLOv8across all dataset configurations. These findings provide practical guidancefor deploying deep learning in industrial PAUT inspection workflows, demon-strating that preprocessing strategy selection must account for defect-type-specific acoustic signatures. The study identifies geometry-induced artefactsuppression as beneficial for most defect classes (10 percentage point mAPimprovement). Results are contextualized within realistic deployment con-straints including limited training specimens (13 defects), cross-manufacturergeneralization challenges, and computational efficiency trade-offs between ar-chitectures.

L'attaque par hydrogène à haute température (HTHA) est un mécanisme de dégradation critique affectant les aciers au carbone exposés à des environnements chargés en hydrogènes sous pression (T > 220 °C). La principale difficulté de son inspection réside dans la similarité de ses signatures ultrasonores avec celles de la porosité ou des inclusions, des types de défauts aux conséquences très différentes sur l'intégrité structurale. Cet article présente une approche par matrice de diffusion extraite d'une capture matricielle complète (FMC) permettant d'identifier une signature physique distinctive du HTHA. Des simulations par éléments finis (POGO FEA) démontrent qu'un lobe dominant caractérise le HTHA tandis que plusieurs lobes multiples caractérisent la porosité. Ces signatures sont confirmées expérimentalement sur blocs certifiés de HTHA et de porosité à 5 MHz et 10 MHz, avec des CNR de défauts allant de 10 à 25 dB.