Deep-learning-based phased array ultrasonic testing (PAUT) interpretation requires defect-candidate preservation while suppressing false calls from geometry-like responses and noise. We propose the Source-Local Attention Transformer (SLAT), a modular Stage-2 verifier that re-scores candidate bundles exported by replaceable S-, B-, and C-scan Stage-1 scorers rather than operating as a new raw-data detector. SLAT-Std forms up to 40 tokens from a radius-2 scan neighborhood and the top-8 candidates per scan position, using score, coordinate, C-scan continuity, and local 3D context features. Each score source uses its respective scorer/fold threshold selected on the Internal Development Set, and a structural point-pair operating policy fixed before External evaluation converts the learned score surface into final call decisions. The held-out External Test Set contains 49 design/ultrasonic-testing-linked records from specimen sources not used for development: 42 records designated as fatigue cracks and seven designated as lack of fusion, all with a nominal height of 1 mm. Averaged over six Stage-1 scorers and five folds, SLAT-Std improved final fixed-policy F1 from 0.829 to 0.884, recall from 0.856 to 0.943, and precision from 0.805 to 0.833. Mean false-negative count decreased from 51.4 to 20.5, and mean call-level false calls decreased from 1.4 to 0.033. A descriptive frozen-score component replay showed that the point-pair veto accounted for most false-call suppression, while the admission gate had a small incremental effect. Source-clustered uncertainty remained wide for recall and false-negative count; the results therefore provide controlled-corpus operating evidence rather than a population-level reliability or broad flaw-generalization claim.
Although sodium-ion supercapacitors are promising candidates for next-generation electrochemical systems, their performance is often hindered by the relatively large ionic radius of Na⁺, which limits facile ion insertion and thus suppresses capacity enhancement. The high surface area and large hexagonal tunnels make colloidal monoclinic WO2.72 nanocrystals attractive as electrode materials, but their use in Na⁺-based supercapacitors remains unexplored. In this study, we introduce colloidally synthesized monoclinic WO2.72 nanocrystals capable of effectively accommodating Na⁺ ions within their hexagonal sites for Na+-based supercapacitors. By removing insulating organic ligands, the nanocrystals are uniformly deposited onto hydrophilic carbon cloth (CC) through a controllable, additive-free dip-coating process. Furthermore, applying a conformal 1 nm Al2O3 overlayer via atomic layer deposition (ALD) enhances mechanical robustness and prevents nanocrystal delamination, enabling highly reversible charge-storage behavior. A symmetric supercapacitor fabricated with ALD-treated WO2.72/CC and a PVA/Na2SO4 gel electrolyte demonstrates stable performance without degradation under bending, achieving an areal energy density of 10.08 µWh/cm2 at 0.3724 mW/cm2. These results highlight the potential of colloidal nanocrystal-based electrodes for high-performance sodium-ion supercapacitors suitable for practical, wearable electronic applications.
Practical large-field-of-view (FOV), high-resolution imaging of microfeatures is demonstrated through the geometry-optimized design, precision fabrication, and actual application of reattachable optical lenses and deep learning super-resolution with residual and attention mechanisms. The presented approach to the elastomeric reattachable lens addresses key limitations of metal-based lenses and metalenses, including high manufacturing cost and complexity and material constraints. Ray-tracing simulations are performed to optimize lens curvature and geometry, minimizing aberrations and improving effective focal performance. The polydimethylsiloxane (PDMS) lens, with less than 2
We present a high-precision measurement procedure for lens mold asphericity with high dynamic range, incorporating a novel calibration strategy based on radial shearing interferometry (RSI). The RSI system was first modeled using ZEMAX® ray-tracing software with a bulk-optics configuration and experimentally validated. A numerical approach, referred to as iterative reverse optimization, was introduced to adapt the initial lens design model to measured wavefront data. In this study, RSI was employed to rapidly minimize measurement departures, enabling calibration with a spherical target using multiple axial displacements at non-confocal positions combined with reverse optimization. The ZEMAX® model was optimized to simultaneously fit all non-confocal experimental wavefront departures, thereby revealing complete system parameters, including assembly deviations and misalignments. The calibrated system was subsequently applied to measure the asphericity of a lens mold, with the aspheric surface parameters retrieved through the proposed reverse optimization procedure. The extracted asphericity showed excellent agreement with independent measurements performed by the Form Talysurf contact method. This approach demonstrates a flexible and practical route for measuring asphericities of varying degrees in non-null testing conditions.
We demonstrate the feasibility of using polished porous Onggi ceramic tableware as food-presenting plates by modifying the water contact angle of the sintered surface. Mechanical grinding and polishing were found to increase the contact angle to values exceeding 60°. Progressive polishing with finer grit papers led, under specific conditions, to the development of a nanoscale surface layer enriched in potassium (K), aluminum (Al), and sodium (Na). During polishing, the relatively soft, low-melting glassy phase underwent plastic deformation and mixed with wear debris, locally altering its composition and forming a thin, modified post-sintering surface film. When this surface layer was removed using an ultra-fine diamond suspension, the water contact angle reverted to its original lower value. This reversible change supports the hypothesis that the enhanced concentrations of K, Al, and Na in the reconstituted surface glass phase are primarily responsible for the observed increase in hydrophobicity.
