A physics-informed deep transfer reinforcement learning (PIDTRL) strategy is proposed for enabling advanced control and modulation of a dc solid state transformer (DCSST) in a dc microgrid. The strategy involves three stages: (i) Centralized training of deep reinforcement learning agents to balance power and reduce current stress in the DCSST; (ii) effective knowledge transfer from a Source-simulation system to a Target-experimental system using minimal experimental data; and (iii) deployment of multiple agents for online control in the DCSST. The proposed method adaptively determines optimal modulation variables (duty cycles and phase shifts) in stochastic and uncertain environments without requiring accurate model information. Experimental results validate the effectiveness of the proposed PIDTRL algorithm.

The industrial Internet of Things (IIoT) significantly enhances mechanical fault diagnosis. However, IIoT-based intelligent diagnostic models struggle with sample insufficiency and high energy consumption due to collection costs and limited computing resources. Therefore, this article proposes an energy-efficient mechanical fault diagnosis method based on the neural-dynamics-inspired metric SpikingFormer (MSF) to achieve accurate fault recognition under insufficient samples. The design and construction of a data acquisition system based on the aircraft engine platform and the ship water jet propulsion platform effectively support the operation of the developed diagnostic algorithm. Specifically, an event-driven multiscale mask spiking self-attention (MMSSA) mechanism is designed to focus critical spatiotemporal features from different scales under low computational complexity. Meanwhile, a rate encoding metric classifier (REMC) is constructed to bridge spiking learning and prototype representation, thereby accurately classifying fault under insufficient samples. Finally, a customized backpropagation strategy based on neural dynamics is developed to enable the MSF to learn effectively and be stable. The superiority of the MSF in energy consumption and diagnostic accuracy is verified through comparison with six authoritative methods across standard, laboratory-acquired, and real-world datasets. The results showed that the parameter count of MSF is$7.04\times $and$20.46\times $less than the strong baseline method, respectively, and the diagnostic accuracy on the two real datasets is 4.52% and 6.91% higher than the latest method, respectively.

To alleviate data distribution under different operating conditions, domain generalization (DG) has been applied in mechanical diagnosis. Still, its effectiveness is limited when unknown fault states appear in the target domain. Consequently, open set DG (OSDG) has emerged to identify unknown classes in unknown domains. However, data collection costs and safety concerns have resulted in a significant class imbalance in OSDG. This imbalance causes the decision boundary to be skewed toward abundant positive classes, ultimately leading to misclassifying unknown states and increasing security risks. Currently, there is a lack of methods to simultaneously address domain shift and class shift in an imbalanced unknown domain. To tackle this issue, this article proposes a multisource domain-class gradient coordination meta-learning (MDGCML) framework, which can learn the generalized boundaries of all tasks by coordinating gradients between interdomains and interclasses. Based on the MDGCML, a joint learning paradigm involving the sharing of parameters between open-set classifiers and closed-set classifiers is constructed to enable quick adaption of the model to unknown domains. The superior performance of the proposed framework has been verified on two datasets.

Railway track systems serve as vital parts of urban mobility and intelligent transportation. Detecting defects in rail track systems not only avoids unexpected downtime but also safeguards passenger lives. Moreover, defect classification holds great economic value, which optimizes both traffic operations and management strategies. Compared to lab tests and track recording vehicle-based field tests, defect detection and classification using in-service trains is an emerging area of study. This enables continuous monitoring, increases carrying capacity, and reduces maintenance costs, but it also requires robust performance and compatibility with various weather conditions. Considering the precipitation characteristics in tropical regions, this paper proposes a novel online defect detection and classification method for mass rapid transit (MRT) railway track systems by integrating a non-contact train-borne monitoring system (TBMS) and an adaptive density-based spatial clustering of applications with noise (Adaptive-DBSCAN) algorithm. The TBMS is developed based on the inductive coupling theory, affirming real-time, contactless, and effective defect detection in tropical regions with high annual and intense short-duration rainfall. By assessing the voltage health ratio (VHR) of the train-rail electrical path, the TBMS can simultaneously monitor defects from rail, ballast, and sleepers/ fasteners. To classify the group of each defect for maintenance decisions, Adaptive-DBSCAN is applied using VHR as inputs and calibrates the algorithm parameters adaptively. Therefore, it avoids the exhaustive traversal typically needed for parameter selection in DBSCAN while preserving accuracy. Experiments conducted on an in-service MRT train (operating at 80 km/h) verified the effectiveness of the proposed method.

Aircraft engines operate under more demanding and unique environments, which require the inner components to be able to withstand extreme conditions. Intershaft bearings serve as the critical part of power transmission. Therefore, their accurate and reliable fault diagnosis is of paramount importance to ensure secure and dependable functioning of the engine. In this field, scarcity of labeled fault data owing to high collection costs is a common challenge. To address this, this article proposes a semi-supervised cross-domain diagnostic method for aircraft engine intershaft bearings, utilizing a virtual domain-driven approach to achieve high accuracy with limited labeled data. Specifically, a dynamics-based simulation model is developed to generate source domain data, reducing the dependency of deep learning models on experimental platforms and lowering platform construction costs. Additionally, a hyperbolic geometric metric learning strategy is designed to capture hierarchical features in high-dimensional data, which handles the correlation between different fault types and enhancing classification accuracy. Furthermore, a domain-class adversarial decoupling mechanism is developed to mitigate the domain bias, enabling the precise representation of fault modes and maximizing the utility of unlabeled virtual domain data. Using datasets from both real-world aircraft engine scenarios and public resource experiments validate the proposed method, illustrating its superior performance compared to state-of-the-art techniques on public domain benchmark datasets.

Rails play an essential role in railway transportation, supporting the movement of various types of trains. Due to the high-frequency and high-intensity loads, as well as harsh operating environment, rails are susceptible to cracking or even fracturing. Among existing rail structural health monitoring (RSHM) methods, the advanced ones often rely on signal-driven deep learning algorithms, necessitating substantial computational time and extensive preliminary information for effective model training. Moreover, crack-related signals with low amplitudes are easily submerged in the complex interference noise environments. Although some noise reduction solutions have been reported, the RSHM results often obtain certain discrepancies from the actual outcomes. To address above issues, this paper presents an alternative RSHM method based on electromagnetic acoustic emission (EMAE) technology. The proposed method uses a double-layer signal processing (DLSP) algorithm and a novel health monitoring index, called spectrum information entropy (SIE). It can monitor the degradation state of rails accurately and quantitatively. In this method, the DLSP algorithm is utilized to identify EMAE signals from the original dataset, which contains complex and diverse interference noise signals. In addition, the SIE is extracted from the obtained EMAE signals to perform the RSHM. Experimental results validate the accuracy and simplicity of the proposed method.

The common-mode (CM) choke is critical integral part of an electromagnetic interference (EMI) filter. An accurate electrical model of the CM choke is crucial to simulate and evaluate the EMI filter performance with confidence. Numerical simulation has emerged as a feasible solution to realize the above-mentioned objectives without the presence of physical products, and thus reducing the trial-and-error process and achieving the choke design speedups. This paper proposes a comprehensive process for the characterization and modeling of single-phase CM chokes based on finite-element analysis (FEA). By extracting the transmission parameters through 3-D model of choke, both impedance magnitudes and phases of its CM or differential-mode (DM) can be collected. These impedances will then be used to derive the behavioral model. The results are validated experimentally with good agreement up to 100 MHz.

Single-phase common-mode chokes (CMCs) are key components in electromagnetic interference (EMI) filters to mitigate conducted emissions caused by the switching power converters. Finite-element analysis (FEA) has been adopted as one of the simulation tools to model a CMC for the extraction of its impedance frequency response. Most literatures focus on the extraction of impedance magnitude, but few explore the analysis of the impedance phase information. This article investigates the phase sensitivity of FEA modelled single-phase CMCs, which reveals the impact of various design parameters on simulation results at frequencies up to 100 MHz.

The state-of-charge (SoC) stands as a pivotal measure for ascertaining a battery's remaining capacity. Accurate SoC estimations can meaningfully enhance a battery's operational longevity, fortify safety standards, and enrich user experience. This paper presents an improved open circuit voltage (OCV)-based partnership for a new generation of vehicle (PNGV) model, specifically tailored for estimating the SoC of LiFePO4 blade batteries. These batteries are distinctively characterized by their advantages in safety, energy density, and thermal management. The proposed model uniquely integrates the SoC-dependent property of the battery's internal resistance, facilitating a marked improvement in estimation accuracy over existing PNGV models. Experimental results underscore the capability and effectiveness of the proposed model in estimating real-time discharging curves, achieving a remarkably low relative error rate of 0.85%.

This paper introduces a novel physics-informed pattern recognition (PIPR) method for open-circuit fault detection in inverters. The proposed method unfolds in three stages: model analysis, offline training, and online validation. In the first stage, we construct an analytical model of power converters. This model is subsequently used to derive fault diagnosis variables. This step is followed by the collection of training samples via simulations. The gathered samples are then fed into pattern recognition neural networks, a process enabled by the prior extraction of model information. This architecture allows for efficient training of the neural network with fewer neurons and samples. The final stage involves the detection and diagnosis of faults by a well-trained online classifier. The robustness of the proposed PIPR method in dealing with unexpected conditions in classification problems shows its potential across diverse conditions.