Shenyang Institute of Computing Technology (China)

Ion pumps operate under drastically varying load conditions, where maintaining stable high-voltage output during the extremely light-load stage (ultra-high vacuum maintenance) is particularly challenging for conventional resonant converters due to frequency runaway and gain nonlinearity. To address this, a hybrid PFM-PWM control strategy is proposed to ensure precise regulation and robust stability across the entire wide load range. The proposed method employs Pulse Frequency Modulation (PFM) with a fixed conduction time under heavy loads to ensure sufficient energy delivery, and seamlessly transitions to Pulse Width Modulation (PWM) under light-load conditions. This dual-mode approach effectively eliminates the voltage drift and instability issues inherent to traditional PFM at light loads, enabling smooth, monotonic regulation down to near-open-circuit conditions. Experimental validation on a digitally controlled 7 kV, 200 W prototype demonstrates superior performance across the complete operational spectrum. The system achieves a stable 7 kV constant voltage output even under complete open-circuit conditions, maintains rated Voltage-Current (UI) curve characteristics through constant current (100 mA), constant power (200 W), and constant voltage (7 kV) regions, and exhibits excellent dynamic response during load step transitions. Arcing simulation tests confirm robust operation under periodic short-circuit conditions with 110 ms cycles, while actual ion pump driving experiments validate reliable vacuum pumping performance from initial pump-down through ultra-high vacuum maintenance, with small and large ion pumps requiring 6.7 and 16 minutes, respectively, to complete the pumping process.

To address the challenges commonly encountered in civil aircraft operating under multi-mode, strongly coupled closed-loop control—namely scarce fault samples, pronounced distribution shift, and root-cause explanations that are easily confounded by covariates—this paper proposes a root-cause analysis method that integrates ensemble learning with constraint-guided reasoning by large language models (LLMs). First, for Full Authority Digital Engine Control (FADEC) monitoring sequences, a feature system comprising environment-normalized ratios, mechanism-informed mixing indices, and multi-scale temporal statistics is constructed, thereby improving cross-mode comparability and enhancing engineering-semantic expressiveness. Second, in the anomaly detection stage, a cost-sensitive LightGBM model is adopted and a validation-set-based adaptive thresholding strategy is introduced to achieve robust identification under highly imbalanced fault conditions. Furthermore, for Root Cause Analysis (RCA), a “computation–reasoning decoupling” framework is developed: Shapley Additive exPlanations (SHAP) are used to generate segment-level contribution evidence, while causal chains, engineering prohibitions, and structured output templates are injected into prompts to constrain the LLM, enabling it to infer root-cause candidates and produce structured explanations under mechanism-consistency constraints. Experiments on real flight data demonstrate that our method yields an anomaly detection F1-score of 0.9577 and improves overall RCA accuracy to 97.1% (versus 62.3% for a pure SHAP baseline). Practically, by translating complex high-dimensional data into actionable natural language diagnostic reports, the proposed method provides reliable and interpretable decision support for rapid RCA.

Captive Asiatic black bears (Ursus thibetanus) may face increased health risks under conditions of stress or close human contact, including vulnerability to infectious diseases. Bacterial pathogens such as Mycobacterium tuberculosis complex and Klebsiella pneumoniae pose health risks in captive settings. However, co-infections involving these pathogens have not been previously reported in captive Asiatic black bears. Two male Asiatic black bears (Ursus thibetanus), aged 2 and 3 years, developed persistent diarrhea, which improved temporarily after penicillin and cephalosporin treatment but relapsed shortly after. Despite ongoing therapy, both showed progressive weight loss, lethargy, and depression. Five days before death, acute respiratory signs, including fever, tachypnea, and coughing, appeared, followed by rapid decline. Necropsy revealed multisystemic lesions. Histopathological examination, Ziehl–Neelsen and Gram staining revealed acid-fast bacilli and Gram-negative bacilli intralesionally, PCR amplification and sequencing of DNA extracted from lung tissue identified carbapenem-resistant hypervirulent Klebsiella pneumoniae strains carrying multiple virulence and resistance genes, as well as a Mycobacterium spp. agent of the M. tuberculosis complex. This case underscores the risks of fatal co-infections between an agent of the Mycobacterium tuberculosis complex and carbapenem-resistant hypervirulent Klebsiella pneumoniae in captive wildlife. Integrated diagnostics, timely detection, and prudent antibiotic stewardship are essential, but the absence of diagnostic standards could delay recognition. Establishing systematic criteria is urgently needed to improve treatment and long-term health management.

Owing to the complexity of analyzing gases with known and unknown chemical interferences, existing single metal–oxide–semiconductor (MOS) gas sensors fail to generate distinguishable signals for different gases. Besides, previous gas recognition algorithms have only extracted local or global features from sensor signals, with no interaction between local and global features. In this paper, an individual microelectromechanical systems (MEMS)-based MOS sensor is applied to generate multivariable signals under pulsed operating modes of an integrated microheater in the MEMS chip. This sensor is able to generate distinguishable gas signals. We also propose a multitask guided gas recognition algorithm that utilizes interacting local and global features of multivariable signals and is able to perform gas classification and quantitative concentration analysis simultaneously. Specifically, to capture the time series information from sensor response signals, a Gramian angular field module is applied to convert the signal into a two-dimensional image. Then, a multi-scale convolutional neural network and Transformers are combined to extract multiscale local and global features, which strengthens the discriminative ability for different gases with different concentrations. Finally, a multitask module is introduced to perform the classification and regression simultaneously. Extensive experiments demonstrate the effectiveness of the proposed algorithm, which achieves an average recognition accuracy of 96.6% for five types of gases: ammonia, ethanol, formaldehyde, water vapor, and methanol. Compared with many classical recognition algorithms, the proposed algorithm yields lower mean absolute error and root mean square error in concentration estimation, which demonstrates its superiority. The pulsed operating modes combined with the proposed algorithm enable individual MEMS sensor to achieve intelligent for gas recognition, which is crucial for applications to the Internet of Things and artificial intelligence.

For the situation where the bearing of the lifting mechanism motor of the gantry crane is prone to “inner ring peeling + outer ring cracks” under heavy–load impact and loadswitching conditions (the combined faults cause diagnosis to be difficult due to feature superposition), and the computing power of the industrial edge terminal is limited and complex models are difficult to deploy, a lightweight diagnostic method integrating CWT, Ghost convolution and coordinated attention (CA) is proposed. In the preprocessing stage, CWT is used to convert one–dimensional vibration signals into two–dimensional timefrequency graphs to separate the features of inner ring faults and outer ring faults, while suppressing environmental noise. Then, the model is based on the residual network and adopts the “basic convolution generates core features + linear transformation supplements redundant features” strategy of Ghost convolution to significantly reduce parameters and decrease computational complexity. The CA mechanism is embedded to focus on the fault–related channels from the channel and locate the fault area in the time–frequency graph from the spatial perspective, alleviating the problem of insufficient feature extraction caused by lightweighting. Experiments show that this method has significantly higher diagnostic accuracy than traditional models in both single–fault and multi–fault superposition scenarios, and the number of parameters and computational costs are significantly reduced, verifying its effectiveness and feasibility in practical applications.

Contour error, primarily induced by the coupling of multi-axis dynamics and servo lag, is a critical factor limiting the machining accuracy of computer numerical control (CNC) machine tools during complex surface machining. Traditional theoretical modeling methods based on dynamics and parameter identification are difficult to cope with complex nonlinear errors. To address the challenges, this study proposes a data-driven deep learning method to achieve high-precision prediction of contour errors. In this paper, the VMC2100B vertical machining center is used as the experimental platform, and the high-speed dynamic coordinate data acquisition research is carried out with the help of dynamic coordinate acquisition technology, which provides strong data support for the training of deep learning model. A hybrid deep learning prediction model named CLA-Net (CNN-LSTMAttention Network) was constructed: 1D Convolutional Neural Networks (1D-CNN) were employed to extract local spatial features, Long Short-Term Memory networks (LSTM) were used to learn long-term temporal dependencies, and an Attention was introduced to dynamically focus on critical historical information. The experimental results show that the proposed CLA-Net model can control the mean square error (MSE) at about <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{3 ~ \mu m}$</tex>, which verifies the effectiveness and superiority of the model in the prediction task. This study provides a new solution for the contour error prediction task of five-axis machine tools.

The coaching of elite athletes in precision sports like shooting and archery is increasingly reliant on data-driven insights. However, existing solutions often fail to integrate real-time performance data, historical records, and domain-specific knowledge into a unified, intelligent system. This paper presents the architectural design of a next-generation Big Data and AI Agent platform tailored for the Hebei Provincial Sports Bureau's Shooting and Archery Center. The architecture is built on a multi-layered, service-oriented model that distinguishes between hot (real-time) and cold (archival) data pathways. Real-time firearm trajectory data captured by YOLOv9 models is processed via a Kafka and Spark Streaming pipeline into ClickHouse for immediate analysis. Historical data and scores are archived in PostgreSQL and MySQL, while unstructured expert knowledge from texts is vectorized and stored in Milvus. The core innovation is an AI Agent, built with a Large Language Model (LLM), that acts as an intelligent orchestrator. This agent interprets natural language queries from coaches, and dynamically invokes specialized tools-Vanna for Text-to-SQL queries on structured data and a RAGflow-based service for knowledge retrieval from the vector database. This dual-tool approach enables the agent to synthesize comprehensive reports that combine quantitative analysis with biomechanical theory, providing coaches with deep, actionable insights for athlete evaluation, selection, and performance optimization.

Traditional aviation instrument screen anomaly detection methods primarily rely on manual inspection, which is inefficient and susceptible to subjective factors, and it is challenging to meet the requirements of modern aviation maintenance for high reliability and real-time performance. To this end, this paper proposes a lightweight self-supervised anomaly detection framework, AirNet, which is designed for aerial instrument screen image scenes to achieve efficient and accurate anomaly recognition. AirNet utilizes a feature adapter to enhance feature expression ability and injects Gaussian noise into the feature space to generate pseudo-abnormal samples, thereby improving discrimination ability. It combines a lightweight binary classification discriminator to realize anomaly detection and location. Experimental results on the self-built FIS-AD dataset and the public MVTec-AD and VisA datasets demonstrate that AirNet achieves performance comparable to that of state-of-the-art methods, while ensuring a lightweight design and high frame rate, with 98.5% image-level AUROC and 99.1% pixel-level AUROC. It shows good generalization ability and localization accuracy.