Ultrasonic imaging is widely used in defect detection, localization, and characterization, and has proven effective for identifying defects larger than the wavelength of the incident wave. However, the characterization of subwavelength-scale defects remains a major challenge in ultrasonic non-destructive evaluation. Such defects can be characterized by analysing the scattering information encoded in the transmitter–receiver ultrasonic array data. However, local scattering signals from three-dimensional (3D) features of interest are often contaminated by interference from neighbouring scatterers, leading to reduced accuracy. Although 2D arrays are theoretically capable of resolving 3D local scattering fields, their application is hindered by the vast amount of measurement data and the number of elements required to produce an aperture of sufficient size to allow adequate focussing.
In this study, a volumetric ultrasound imaging method using 1D array is proposed. Ultrasound data are acquired by rotating a 1D array and a 3D image is reconstructed from 2D images corresponding to different angular array positions. During the defect characterisation stage, local regions of interest are identified in 3D imaging volume, and 3D scattering information is extracted using the inverse imaging approach. The method is tested on experimental data obtained from an aluminium and steel samples that contain several flat-bottom holes, and the results are compared with numerical simulations. It is shown that the proposed method enables characterization of sub-wavelength scatterers in 3D space while retaining the high image quality advantages to 1D arrays.

Access-limited welds in nuclear piping often fail to satisfy required volumetric examination coverage, and far-side inspection of austenitic stainless steel welds is especially challenging because weld anisotropy distorts beam paths and time-of-flight (ToF). This paper presents a conference-length summary of an ongoing far-side phased array ultrasonic testing (PAUT) program, with emphasis on a physics-informed anisotropy-estimation framework derived from the simulation stage of the project. Previous virtual mock-up studies improved far-side simulation realism, but the manually tuned weld models remained specimen-specific. To address the fundamental source of error, vertical-incidence full matrix capture (FMC) data were used to estimate weld grain-orientation maps via physics-informed deep learning (PIDL). A differentiable Softmin-Fermat ToF (SF-ToF) loss was introduced to enforce travel-time consistency during training. Simulated FMC data generated from MINA-based anisotropic weld models were used to compare ResUNet, CRNN, and ConvLSTM under out-of-distribution (OOD) conditions. Gradient-ratio analysis identified a balanced regime in which the physics term became active without overwhelming data fitting. The best ConvLSTM setting reduced OOD RMSE from 13.14° to 9.94°, P99 from 52.93° to 33.47°, and the physics violation rate from 2.32% to 1.69%. When the predicted anisotropy was incorporated into ray-bending total focusing method (TFM) imaging, substantial focusing improvements were obtained, including lateral full width at half maximum (FWHM) reduction from 3.6 mm to 0.8 mm for SDH #1 and axial FWHM reduction from 10.4 mm to 1.6 mm for SDH #2. The results indicate that reliable far-side PAUT in anisotropic welds requires not only better imaging algorithms but also physically consistent ToF modeling.

This research explores the strengths and limitations of using the Born approximation in ultrasonic phased array setups for characterizing microstructures. Traditionally, the Born approximation has been applied under strict conditions—mainly assuming weak scattering scenarios and focusing on average material properties such as grain size. However, previous studies have rarely examined the validity of imaging results outside these strict conditions or focused on local microstructural details.
In this study, we investigate the performance of the Born approximation specifically for imaging purposes using ultrasonic phased arrays. We use an analytical model based on the Born approximation, which considers only single scattering, to calculate scattering signals and produce simulated images. To evaluate the accuracy and applicability of this approach, we compare these results directly with Finite Element (FE) simulations, which naturally account for both single and multiple scattering phenomena.
By generating realistic polycrystalline microstructures, our analysis specifically considers how grain size and material anisotropy affect the single scattering rate, which directly impacts the agreement between FE and analytical results. Through this analysis, we have established clear parametric conditions defining when the Born approximation is valid for imaging. Additionally, we explore more complex scenarios, including microstructures with multiple scales and large macrozones embedded in finer grains, to test the robustness of the analytical model.
After validating the analytical approach in 2D scenarios, we demonstrate that this method can naturally extend to 3D problems. The analytical model has shown significant speed advantages, making it highly efficient for simulating extensive microstructural configurations in three dimensions. This study provides foundational evidence that validates the analytical model’s effectiveness in both 2D and potential 3D applications, thereby streamlining and accelerating material characterization methods.