To improve the dynamic response speed of high-speed solenoid valves in electric fuel injection systems of marine diesel engines, a numerical simulation model of the solenoid valve is described in this paper. The accuracy of the simulation model was verified on the test bed of the solenoid valve. The effect of the punch position and the size of the dynamic response of the solenoid valve were investigated by using the distribution law of the electromagnetic force in the armature. The results of the test showed that armature drilling in the inter-yoke zone can optimize the solenoid closing response time, but it has little impact on the solenoid opening response time. From this rule, two groove schemes were further designed. Through comparison and calculation, it can be concluded that the fan groove scheme is better than the trapezoidal groove scheme, and that the opening and closing response times of the solenoid valve should be targeted in order to multi-target optimize the fan groove geometric parameters and the armature thickness. The results show that after optimization, the weight of the motion part is reduced by 21.6%, the opening response time of the solenoid valve is reduced by 11.1%, and the closing response time is reduced by 30.0%. While reducing the oil film damping of the armature motion, the overall dynamic response characteristics of the solenoid valve are improved.

Radar ship target detection plays an important role in ship navigation and collision avoidance. The ship target detection for synthetic aperture radar has been relatively mature, while pulse compression radar has relatively poor imaging effect and more noise interference, so there are few related studies. Aiming at the problem of poor detection of small target ships in pulse compression radar images, an improved SSD-based radar ship target detection method is proposed in this paper. Based on the Single Shot MultiBox Detector, we improved the network model. Resnet50 is used to replace the backbone network to effectively extract the feature information of small targets. Feature fusion module which contains deconvolution and group convolution is added to combine low-level features and deep features. Spatial attention module is used to suppress background and increase detection accuracy. Experimental results demonstrate that the proposed model performs better than the benchmark model.

The reliable operation of solenoid valves is crucial for the safety of modern industrial and marine power systems. However, under complex electro-mechanical coupling degradation, traditional single-modal diagnostic methods suffer from severe perception blind spots and feature aliasing, particularly failing to identify early-stage mechanical fatigue and subtle electrical transients. To address these challenges, this paper proposes a novel Heterogeneous Multi-modal Fusion Network (HMF-Net) for solenoid valve fault diagnosis. First, a cross-modal time-frequency joint perception mechanism is designed to transform raw current and vibration signals into high-resolution tensors, providing high signal-to-noise ratio global spectral guidance. Second, a physics-inspired dual-stream architecture is constructed: a 1D Convolutional Neural Network (1D-CNN) is utilized to accurately anchor high-frequency transient electromagnetic distortions (e.g., partial coil short-circuits), while a Transformer branch leverages global self-attention to deeply mine long-range dynamic damping attenuation patterns (e.g., spring fatigue). Finally, rigorous cross-batch experiments on a physical hardware-in-the-loop test bench demonstrate that the proposed HMF-Net achieves an outstanding comprehensive diagnostic accuracy of 98.86%, significantly outperforming traditional baseline models. Furthermore, Gradient "×" Input feature attribution analysis visually proves that the model's decision-making firmly anchors on true physical impact transients rather than overfitting to background noise, successfully achieving high precision and profound physical interpretability for electro-mechanical fault diagnosis.