We develop a fractional extension of the particle-in-cell (PIC) method for plasma simulations by incorporating a Caputo fractional time derivative of order $\alpha $ ( $0 \lt \alpha \leq 1$ ) into the particle equation of motion. This generalization introduces non-Markovian memory effects and subdiffusive transport, which are critical for describing plasma dynamics dominated by turbulence, nonlocal collisions, and magnetic trapping. The formulation is derived from a fractional Vlasov–Poisson system, with the fractional derivative discretized via the truncated-memory Grünwald–Letnikov (GL) approximation, preserving the core PIC structure while modeling long-range temporal correlations through a power-law memory kernel. Key terms are interpreted, linking $\alpha $ to memory persistence, subdiffusion exponent ( $\beta =\alpha $ ), and effective dissipation analogous to turbulent or confined plasmas. Simulations of the 1-D electrostatic two-stream instability (TSI) demonstrate that decreasing $\alpha $ induces memory-driven dissipation: linear growth rates are reduced by up to 50% (for $\alpha \leq 0.7$ ), nonlinear saturation is delayed, and the formation of phase-space vortices (BGK modes) is suppressed. This framework provides a robust kinetic tool for modeling anomalous transport in fusion devices (e.g., ITER tokamaks) and astrophysical environments.
Reluctance coil launch technology has garnered considerable attention due to its advantages of good controllability, strong concealment, and low recoil. However, traditional drive schemes suffer from low energy efficiency, with a significant portion of energy dissipated as Joule heat, which constrains continuous launch capability. Existing energy recovery schemes, such as SCR-based discharge, diode passive recovery, and dual-bridge fully controlled topologies, offer improvements but share common shortcomings: first, they struggle to overcome mutual inductance coupling between multistage coils, leading to residual current tailing and incomplete energy recovery; second, they rely on fixed timing models with poor robustness against parameter drift, making it difficult to maintain the optimal energy transfer point during dynamic launches. To address these issues, this article proposes an energy-recycling drive topology based on a half-bridge circuit. This topology provides a low-impedance path for interstage coupled energy, significantly enhancing recovery efficiency. Furthermore, an innovative timing control strategy based on instantaneous armature force is introduced. This strategy links the recovery trigger moment directly to the armature’s acceleration capability, rather than relying solely on electrical characteristic points, thereby effectively improving adaptability to parameter disturbances. Through simulations and physical experiments on a ten-stage reluctance coil launch system, this solution achieves a stable kinetic energy share and overall system efficiency exceeding 25% and 32%, respectively. It enables high-efficiency energy recovery while balancing system reliability, dynamic performance, and cost controllability.
Most of the modern solid-state pulsed-power generators are made up of modules. Modular structure helps to improve the manufacture, operation, and maintenance of pulsed-power generators. Module circuit design has become an important part of solid-state pulsed-power technology. Major module circuit methods use either capacitive energy storage (CES) or inductive energy storage (IES). Recently, a relatively new circuit method called hybrid energy storage (HES), which is a combination of CES and IES, has been proposed and applied in module circuit development. Each of these circuit methods has its advantages and disadvantages when compared with the others. This article proposes and demonstrates a circuit method for a reconfigurable pulsed-power module. The proposed module can operate in CES, IES, and HES modes by changing only the switching sequence, without changing the circuit hardware. The circuit operating principle and control method are described in detail. The experimental results are presented to demonstrate the different output waveforms, the use of stagewise mode selection for capacitive and inductive loads, and the voltage addition using two modules.
Plasma surface quenching is an effective means of improving the wear resistance of 65Mn steel; however, traditional methods for acquiring 2-D hardness fields are time-consuming, and directly employing deep learning feature mapping gives rise to a closed-box problem that is difficult to interpret. In this article, a hybrid deep learning framework integrating image instance segmentation and physical-feature-driven regression is proposed to achieve accurate reconstruction of the full-field hardness of quenched cross sections and to circumvent the closed-box problem inherent in a single deep learning model. The framework adopts a two-stage architecture: first, an instance segmentation model is used to extract key geometric features of the quench-hardened case; subsequently, a multilayer perceptron (MLP) regression model is constructed to realize the nonlinear mapping from these geometric features to the full-field hardness. To address the steep hardness gradient at the interface between the hardened layer and the substrate, a spatial attention mechanism and a dynamic weighting loss function are incorporated into the model, thereby adaptively directing the network’s focus toward the zones of abrupt microstructural transition and intensifying the learning of high-error regions. Experimental results demonstrate that the proposed method reduces the mean absolute error (MAE) to 6.9 HV and the maximum single-sample error to 43.4 HV, validating the effectiveness of predicting complex 2-D hardness field distributions from the geometric features of the quench-hardened case.
Accurate monitoring of soil volumetric water content (VWC) is essential for precision irrigation and sustainable agriculture. This work presents a tunable terahertz (THz) metasensor based on the synergistic integration of graphene, black phosphorus (BP), and VO2 for high-resolution soil-moisture detection. The hybrid architecture enhances electromagnetic confinement through plasmonic, anisotropic, and phase-transition mechanisms, strengthening the interaction between resonant fields and moisture-dependent dielectric variations. A wave concept iterative process (WCIP) framework is developed for full-wave analysis and validated against analytical solutions, achieving subpercent error and a 58.6-fold reduction in computational time compared with FDTD simulations. The sensor exhibits a resonance shift from 2.62 THz in air to 2.97 THz at VWC = 40%, while the sensitivity increases from 0.594 to 1.450 THz/RIU in the initial design. To accelerate optimization, a deep neural network (DNN) surrogate model is trained using WCIP-generated data and achieves excellent predictive accuracy ( $R^{2}=0.9992$ ). The optimized configuration increases the sensitivity to 2.60 THz/RIU at 40% VWC. By combining reconfigurable THz metasensing, physics-based modeling, and machine-learning-assisted optimization, the proposed framework provides a promising solution for next-generation smart agricultural monitoring systems.
