Mingzhe Rong

Reliable pressure prediction in vacuum circuit breakers is challenged by cross-order-of-magnitude pressure variation, insufficient representation from a single sensing modality, and the absence of confidence feedback for individual predictions. To address these issues, a Task-Tuned Evidence-Aware Transformer (TEAT) is developed for spectrum-image multimodal pressure prediction and evidence-confidence assessment. TEAT integrates a bidirectional cross-attention multimodal Transformer with three task-tuned branches, two traditional machine-learning reference branches, and a two-level stacking calibration module. The Baseline Branch, Robustness Branch, and Constraint Branch provide complementary evidence from multimodal representation, fusion stability, and difficult pressure-range constraints, respectively. A multi-evidence confidence mechanism further combines internal dual-head evidence and system-level post-fusion evidence to identify high-risk samples. Experiments on 2800 synchronized spectrum-image pairs demonstrate that TEAT achieves MSEexponent = 1.497 × 10⁻³, R²pressure = 0.998250, and MAPEpressure = 4.2989% on the test set, outperforming spectral-only, image-only, and conventional spectrum-image fusion routes.

The inability of uncrewed aerial vehicles (UAVs) to land accurately within designated charging areas significantly restricts their wireless power transfer (WPT) applications. To address this challenge, this article proposes a full-plane misalignment-tolerant WPT system based on a self-decoupled flat orthogonal coil pair. The proposed system generates a uniformly distributed horizontal magnetic flux that rotates at a constant angular velocity. At any position and orientation above the transmitter platform, the receiver coil can capture an alternating magnetic flux with a consistent amplitude. This enables the system to achieve full-plane misalignment tolerance, including lateral, longitudinal, and rotational offsets. This article innovatively employs a corrected image current method and genetic algorithms to optimize the coil distribution of the flat orthogonal coils, generating a highly uniform horizontal magnetic flux above the transmitting platform. With phase-shifted currents, a uniformly rotating magnetic flux at the resonant frequency is achieved. The output remains highly stable and unaffected by the UAV’s position or orientation above the platform, without requiring position sensing, frequency adjustment, or other complex control strategies. Experimental results demonstrate that the system exhibits excellent misalignment tolerance, delivering approximately 200 W of output power with a DC–DC efficiency of 88%. The output voltage varies by no more than $\boldsymbol{\pm}$ 2.5% under $\boldsymbol{\pm}$ 30 mm longitudinal/lateral misalignments and by only $\boldsymbol{\pm}$ 3% across a full 360° rotation.

As the core components for carrying current, electrodes are ubiquitously used for energy conversion and transmission by batteries, novel electronic devices, and power switchgears, etc. However, as the energy level and power density of the electrode systems increase, the deterioration of the electrode materials is accelerated especially when discharging, which compromises the lifespan and reliability of the electrical and electronic devices. Herein, CuW electrodes reinforced by two distinct types of graphene materials are fabricated, and their dynamic erosion behavior obscured by arc light is first revealed by an in situ optical imaging method. Due to the efficient evaporative-cooling effect endowed by the uniform structural morphology, the CuW electrode reinforced by reduced graphene oxide covered with Cu nanoparticles (RGO@Cu) exhibits a delayed transition of erosion mode from evaporation to splash, resulting in a significantly reduced material loss. Furthermore, the incorporation of RGO@Cu suppresses the temperature rise and splash intensity of the electrode, demonstrating a superior enhancement for the arc resistance and cyclic stability of the electrode materials. This graphene-based enhancement mechanism can be extended to fabricate other materials with high robustness in harsh conditions.

With the rapid development of Flexible DC Grid (FDCG), the demand for high-performance DC circuit breakers (DCCBs) is becoming increasingly urgent. Existing DC circuit breakers suffer from issues such as high cost and long smallcurrent interruption time. This paper proposes a resonant DC interruption scheme in which the periodic increase in vacuum arc voltage, induced by an external alternating transverse magnetic field (EATMF), facilitates the transfer and interruption of fault currents. This scheme can simultaneously possess the characteristics of low cost and strong capability of interrupting fault currents. The vacuum arc characteristics under EATMF and the current interruption capability at different ratios of the characteristic frequency of the transfer branch to EATMF frequency were studied. The study revealed that the optimal fault current transfer capability is achieved when the ratio of the two approaches approximately 2:1. In this paper, a prototype of the DC circuit breaker is developed, which can interrupt 12 kA bidirectional fault currents. Compared with other DC circuit breakers under the same technical parameters, the proposed circuit breaker achieves a significant cost reduction.

Arc voltage signals play a pivotal role in monitoring droplet transfer modes, which are essential for ensuring stability and quality in wire arc additive manufacturing (WAAM). While deep learning has become the mainstream approach for such monitoring, its effectiveness depends on access to large, labeled datasets. However, collecting sufficient data are challenging for individual labs or manufacturers, and direct data sharing is often restricted due to privacy concerns and commercial barriers. Furthermore, data from different sources typically exhibit significant heterogeneity and domain shift, hindering model generalization. To tackle these issues, this article proposes a client-aware representation alignment (CARA) framework based on federated learning. CARA preserves data locality while enabling collaborative model training by aligning global prototypes across clients and adaptively weighting client contributions during aggregation. These mechanisms jointly improve representation consistency and cross-domain robustness. The achieved high recognition accuracy lays a foundation for subsequent closed-loop control strategies and improved WAAM deposition quality, demonstrating tangible industrial relevance. Moreover, by enabling decentralized model training without data sharing, the proposed federated framework significantly enhances data availability and scalability for deep-learning-based monitoring tasks. Experiments conducted on two real-world datasets, namely CuCrZr and Cu, validate the superiority of the proposed framework under both physically grounded and statistically induced heterogeneity, highlighting its effectiveness in privacy-preserving, generalizable monitoring of WAAM processes.

Capacitive power transfer (CPT) technology has been widely studied in unmanned aerial vehicle (UAV) applications due to its advantages of lightweight structure, flexible design, and no eddy current loss. However, metal objects (MOs) can alter the coupler's electric field distribution and circuit parameters, which negatively affects system power factor, output power and efficiency. To detect and suppress the influence of MOs in CPT system, this paper proposes an integrated metal object detection (MOD) and metal object (MO) interference suppression method using an automatic tuning assist circuit (ATAC). Firstly, the impact of MOs on mutual capacitances, input impedance, and output current of the multi-cell single capacitance coupled (SCC) CPT system has been analyzed. Then, an ATAC network is designed on the primary side. By recognizing the DC voltage amplitude of the ATAC network, the impact of MOs on the CPT system can be detected when the CPT system operates in zero voltage switching (ZVS) state. Furthermore, to suppress the interference of MOs, a constant current (CC) output method based on secondary-side resonant frequency tracking was proposed. By recognizing the DC voltage of the ATAC network, the secondary-side resonant frequency in the LC-LC topology can be obtained. At the new operating frequency, the ATAC network adapts the circuit impedance to achieve CC output despite MO interference, while maintaining a high power factor and ZVS characteristics. Finally, an experimental prototype was built, demonstrating that the output current varies within a ±6% range. and the system efficiency reaches up to 80.8%.

Volatile organic compounds (VOCs) tend to accumulate in the confined, humid environment of household refrigerators, posing health risks. Trimethylamine (TMA), an odorant generated from decaying meat, is designated as the indicator gas for deodorization performance by the Chinese national standard GB 21551.4-2010. Nonthermal corona discharge is effective in decomposing TMA but inevitably produces ozone, a byproduct whose emission varies with reactor configuration and operating conditions. To address this, a physics-informed, reduced-order descriptor based on the mean electron energy distribution in the discharge region is proposed to guide reactor design. A 2-D plasma calculation is used to obtain the mean electron energy distribution. Two energy thresholds are introduced to define an effective energy region $S_{1}$ (where the mean electron energy exceeds the threshold relevant to TMA bond scission) and an excess energy region $S_{2}$ (where the mean electron energy exceeds the threshold associated with ozone propensity). Their area ratio $\eta = S_{2}$ / $S_{1}$ is introduced as an indicator for ozone propensity. Numerical calculations are conducted for 120 configurations in wire-plate and needle-ring reactors to obtain $\eta $ values. The selected configurations with stable corona discharge were tested for TMA degradation rates ( $V_{1}$ ) and ozone generation rates ( $V_{2}$ ), and a performance factor $\alpha = V_{2}$ / $V_{1}$ is defined. A strong correlation (adjusted $R^{2} =0.9682$ ) between $\eta $ and $\alpha $ supports the descriptor’s validity. Considering both calculation and experimental results, the wire-plate reactor outperforms the needle-ring reactor. The wire-plate reactor with 0.1-mm wire diameter, 10-mm gap, and 9-kV voltage achieves the lowest $\alpha $ and $\eta $ , identifying it as the optimal configuration. The proposed framework enables rapid, low-cost optimization of corona-discharge deodorization systems for confined spaces, such as household refrigerators.

The free-standing magnetic field energy harvester (FSMFEH) is an effective solution to the power supply challenge for large-scale sensor networks, owing to its flexible installation and broad applicability. Structural parameter optimization is essential to enhance the power density of the FSMFEH. However, existing parameter optimization methods based on the FSMFEH mathematical model neglect the interdependencies among parameters, hindering efficient identification of optimal solutions and exhibiting strong subjectivity and limited generalizability. Therefore, a novel parameter optimization method for the FSMFEH based on an improved multi-objective particle swarm optimization (MOPSO) algorithm is proposed. First, the output power and power density models of the FSMFEH considering the magnetic core saturation characteristics are derived, and a multi-objective parameter optimization model corresponding to the FSMFEH parameter optimization problem is formulated. Second, an improved MOPSO algorithm is developed based on a control parameter optimization strategy to enhance convergence and search accuracy. Finally, the improved MOPSO algorithm is employed to solve the FSMFEH parameter optimization problem, and the optimal structural parameters are identified through a specific screening strategy. The results demonstrate that, compared with existing parameter optimization methods based on the FSMFEH mathematical model, the proposed method efficiently identifies optimal structural parameters under varying external magnetic field conditions, with convergence achieved within 60 iterations and computational time limited to 50 seconds. The method exhibits strong generalizability and high computational efficiency.