Radar-based human activity recognition (HAR) offers a non-contact technique with privacy protection and lighting robustness for many advanced applications. Complex deep neural networks demonstrate significant performance advantages when classifying the radar micro-Doppler signals that have unique correspondences with human behavior. However, in embedded applications, the demand for lightweight and low latency poses challenges to the radar-based HAR network construction. In this paper, an efficient network based on a lightweight hybrid Vision Transformer (LH-ViT) is proposed to address the HAR accuracy and network lightweight simultaneously. This network combines the efficient convolution operations with the strength of the self-attention mechanism in ViT. Feature Pyramid architecture is applied for the multi-scale feature extraction for the micro-Doppler map. Feature enhancement is executed by the stacked Radar-ViT subsequently, in which the fold and unfold operations are added to lower the computational load of the attention mechanism. The convolution operator in the LH-ViT is replaced by the RES-SE block, an efficient structure that combines the residual learning framework with the Squeeze-and-Excitation network. Experiments based on two human activity datasets indicate our method’s advantages in terms of expressiveness and computing efficiency over traditional methods.

The analysis of the influence of hull deformation on shaft alignment is predominately conducted using the finite element method (FEM), which is time-consuming, labor-intensive, and challenging to use for iterative hull design optimization. In this paper, hull deformation is separated into two parts—global deformation and local deformation, simplified to a single-span beam model and a grillage beam model, respectively—then solved using the matrix displacement method (MDM). Compared to FEM, the proposed method has a small calculation error, proving its correctness, while the calculation time is greatly reduced. The proposed method has been used to calculate the hull deformation of a ship under various conditions and evaluate its influence on shaft alignment. The results indicate that under certain conditions, the bearing reaction forces are constant, whereas the bearing pressure changes as a consequence of the change in shaft-to-bearing angle. The comparison between local rotation and shaft-to-bearing angle reveals that bearings in various positions follow distinct laws. We suggest that the shaft-to-bearing angle be used as an additional parameter in the evaluation of shaft alignment calculations. Moreover, when optimizing bearing pressure, bearings in different positions are affected differently by global and local deformation, and their optimization priorities are distinct.

In order to determine the effect of damaged insulating enamel on the corrosion of high-chromium (H), duplex (D), and corrosion-resistant (R) duplex stainless steel, the corrosion characteristics of HDR duplex stainless steel in 3.5% NaCl solution were studied by means of local electrochemical impedance spectroscopy (LEIS) and micro-morphology analysis. It was shown that the LEIS impedance was stable at about 7.0 × 10³ Ω within 10 days when the HDR duplex stainless steel was not coated. The minimum LEIS impedance of exposed HDR at the damaged area fluctuated around 6.5 × 10³ Ω within 15 days when the coating of the self-control insulating enamel damaged area was 1 mm × 10 mm. The coating-damaged area from 1 mm × 10 mm reduced to a circular hole with a diameter of φ1 mm, and the LEIS impedance of the exposed HDR increased at the damaged coating. When extending along the damaged coating to the intact coating area, the impedance rapidly increased, and the further the distance from the damage, the greater the increase in impedance. The impedance of coated HDR increased with the prolongation of immersion time and ultimately stabilized. The thicker the coating, the longer the impedance took to reach a stable state. The stabilized coating had a better effect on improving the corrosion resistance of HDR duplex stainless steel. Within 15 days, the HDR ferrite structure at different areas of coating damage prioritized corrosion. Cl⁻ mainly comes from the solution, and Si and S elements mainly come from elements of the collective itself, with Cl⁻ adsorption, S, and Si element inclusion being the main factors influencing the corrosion of the ferrite structure.

Radar-based human activity recognition (HAR) offers a non-contact technique with privacy protection and lighting robustness for many advanced applications. The introduction of artificial intelligence (AI) technology encouraged the effort to develop deep learning (DL) for radar-based HAR. These methods employ the micro-Doppler signal because of its unique correspondence with human activity characteristics. Traditional methods give insufficient consideration of the HAR network's lightweight while focusing on recognition accuracy. In this paper, an efficient network based on a lightweight hybrid Vision Transformer (LH-ViT) is proposed to address the HAR accuracy and network lightweight simultaneously. Feature Pyramid architecture is applied for the multi-scale feature extraction for the micro-Doppler map. Feature enhancement is executed by the stacked Radar-ViT subsequently, in which the fold and unfold operations are added to lower the computational load of the attention mechanism. The convolution operator in the LH-ViT is replaced by the RES-SE block, an efficient structure that combines the residual learning framework with the Squeeze-and-Excitation network. Experiments based on two human activity datasets indicate our method's advantages in terms of expressiveness and computing efficiency over traditional methods.