Abstract Non-Destructive Testing (NDT) has become an essential technology for modern industrial maintenance and the inspection of various materials. Amongst different NDT technologies, Infrared Thermography (IRT) offers a robust framework to reveal the subsurface anomalies inside the sample. However, conventional Infrared Imaging (IR) methods are often time consuming, manual and incapable of automatic distinguishing defects from non-defective regions. To address the issue, an automatic defect detection framework is proposed, which integrates an unsupervised autoencoder learning approach with image fusion. Autoencoder is trained on non-defective regions to model the heat distribution patterns. During testing, defective frames show large reconstruction errors and obtained frames are further used for segmentation. The defective frames are segmented using adaptive thresholding to highlight regions with high reconstruction error and generate binary masks that localize potential defect areas. Subsequently, Signal to Noise Ratio (SNR) is computed across each defective region to estimate the thermal contrast. This SNR directs the fusion across all the defective frames of the given defect at given spatial location. The proposed approach estimated the reconstructed and accurate frame with automatic identification of defective frames, achieving high detection sensitivity across varying defect depths. The integration of infrared imaging with neural network based autoencoder substantially enhances the efficiency, reliability, and depth resolving capability of thermal-based defect detection system.

Abstract This paper proposes a novel T-shaped periodic relativistic extended interaction oscillator (TSP-REIO) based on the T-shaped periodic slow-wave structure. The structure is analyzed using the equivalent circuit method, and the electron motion behavior is theoretically examined to derive the optimized configuration for efficient beam-wave interaction. Methodologically, the coupling section/cavity for the TSP-REIO equal inductance/capacitance; Kirchhoff’s law gives inter-cavity field distribution. Substituting this field into the 1D kinematic electron motion equation yields electron final velocity (varying initial phases) and energy conversion efficiency across periods. Four-period structure’s calculated efficiency: 43.78%. Parameters of the T-shaped four-period structure are optimized by Support Vector Regression (SVR) during modeling. The validity of the optimization is confirmed by the predicted efficiency (40.27%). Simulation results show that the T-shaped four-period structure achieves an efficiency of 40.44% when driven by a 450 kV and 400 A annular electron beam. The theoretical analysis method proposed herein effectively predicts the energy conversion efficiency of TSP-REIO, enabling rapid optimization of the T-shaped periodic resonant slow-wave structure. Its correctness is further validated by three-dimensional simulation results, providing an effective and viable scheme for the compact and efficient development of high-power microwave devices.

Abstract he semiconductor–metal transition (SMT) is a key property of VO₂. During the SMT in nanocrystalline VO₂ films, coexisting semiconductor and metallic phases necessitate accurate determination of their fractions to understand evolving optical and electrical properties. We develop a numerical approach using the effective medium approximation to extract these phase fractions by fitting optical spectra, enabling precise tracking throughout the SMT. The depolarization factor is an important parameter describing how the particle shape affects the internal electric field, which bridges the the material’s dielectric properties and particle geometry. The results showed that depolarization factor tends to decrease in the mid-SMT region, revealing a corresponding change in particle shape. It is suggested that VO₂ nanoparticles undergo the SMT independently, consistent with a coalescence-dominated transition mechanism that affects the depolarization factor.