最新刊期
- 摘要：To address the rapid temperature rise of 18650 lithium-ion batteries under thermal abuse, the difficulty of early identification of failure stages, and the insufficient prediction stability when relying on a single temperature or voltage signal, this study conducts thermal abuse experiments to obtain the evolution patterns of battery voltage decay, temperature rise, and failure time under different SOC (State of Charge) and heating powers. The results reveal that increased SOC and enhanced external thermal excitation lead to earlier failure, faster temperature rise, and higher thermal safety risks. On this basis, considering the coupled voltage-temperature response and time-series variation characteristics, an SVM-Res2Net-Transformer hybrid model is developed to realize failure warning and multi-step temperature prediction during thermal abuse. First, a single-cell thermal abuse experimental platform is built to collect voltage-temperature data under different heating powers and states of charge.Secondly, an electro-thermal coupled lumped-parameter model was developed in COMSOL Multiphysics, and its accuracy was validated using uncalibrated SOC and heating power operating conditions.Based on this, the SVM-Res2Net-Transformer hybrid model is constructed to identify failure risks using voltage-window features and to perform multi-step temperature prediction. The results show that SVM can effectively distinguish between normal states and failure-risk states. The Res2Net-Transformer achieves prediction R2 values above 0.98 under most combined operating conditions, with an R2 of 0.975 under 100% SOC. Compared with five benchmark models, the proposed model demonstrates the best overall performance, with R2, RMSE, and MAPE values of 0.998, 4.401, and 0.016, respectively. The findings provide a reference for thermal safety risk identification and early warning of lithium-ion batteries.关键词：lithium-ion battery;Thermal abuse;Electro-thermal coupling;Data-driven prediction;Multi-step prediction;Early warning1|0|0更新时间：2026-10-13
- 摘要：Aiming at the high air-side thermal resistance and limited guidance for fin parameter selection in compressed air–water finned-tube heat exchangers for compressed air energy storage (CAES) systems, experimental and numerical investigations were conducted on the effects of fin parameters on heat transfer performance. Thirteen prototypes were designed and fabricated, with fin densities of 512–1023 fins/m and fin heights of 1.1–1.35 mm. Their performance was tested on a 0.6 MPa closed pressurized test platform over a Reynolds number range of 7000–40000. A three-dimensional single-period model of the finned-tube bundle was also developed. Grid independence and turbulence-model comparisons were performed to determine a numerical method suitable for high-pressure air crossflow. The results show that heat transfer intensity per unit area does not fully correspond to overall heat transfer capacity. The prototype with 512 fins/m and a fin height of 1.1 mm achieved the highest heat transfer coefficient, whereas that with 905 fins/m and a fin height of 1.1 mm exhibited the largest UA value. A UA correlation incorporating the Reynolds number and finning ratio was developed with a coefficient of determination of 0.979, enabling reasonable estimation of overall heat transfer capacity under CAES conditions. The results provide a basis for fin parameter selection and performance prediction of CAES heat exchangers.关键词：compressed air energy storage;fin parameters;heat transfer performance1|0|0更新时间：2026-10-13
- 摘要：To address the issues of model loading failure, unbalanced computing load, and excessive communication delay in the parallel training of multimodal models due to large model parameters, heterogeneous modalities, and limited computing resources, a computing and storage collaborative optimization method is proposed. A fine-grained modality analysis of the multimodal model is conducted to obtain the heterogeneous modality features of the multimodal model, and a multimodal model pipeline parallel splitting strategy is designed to solve the problem of model loading failure. A unified virtualized memory pool integrating different levels of storage space is constructed, and parameters, gradients, and activation values are distributed in different levels of storage space according to their access frequency and life cycle, effectively expanding the capacity of the graphics memory and eliminating the limitation of the size of the graphics memory on the training of multimodal models. Finally, a modality-aware scheduling optimization method is designed to allocate GPU computing resources based on the computing density and delay differences of different modality encoders, and an asynchronous transmission and data prefetching mechanism is designed to further reduce communication overhead and achieve fine-grained scheduling of computing tasks, balancing the load of computing devices. Experimental results show that compared with existing methods, the proposed method can significantly reduce storage and communication overhead, fully utilize computing resources, and improve the overall training efficiency of the model while ensuring stable convergence of the model.关键词：multimodal model;parallel training;collaborative optimization;heterogeneous storage;modality-aware1|0|0更新时间：2026-10-11
- 摘要：To improve the local control capability of non-axisymmetric endwall contouring parameters in aerodynamically sensitive regions without increasing design degrees of freedom, a sensitivity analysis-based parameter selection and optimization method is proposed. A first-stage turbine of a gas turbine engine is investigated. SHAP (SHapley Additive exPlanations)-based sensitivity analysis identifies that the endwall control lines near the cascade exit have a significant impact on stage total-to-total efficiency and secondary flows. Based on this finding, a parameterization model with non-uniformly distributed control points is developed for aerodynamic optimization of the non-axisymmetric endwall contouring. With the same optimization algorithm and identical numbers of performance evaluations, the optimized designs using non-uniform and uniform control point distributions improve the stage total-to-total efficiency by 0.094% and 0.053%, respectively. Further analysis indicates that the proposed sensitivity-guided parameterization enhances the regulation capability in aerodynamically sensitive regions. The efficiency improvement is achieved by reducing the transverse pressure gradient in the downstream passage, weakening the passage vortex and its interaction with the boundary layer.关键词：turbine aerodynamic design optimization;non-axisymmetric endwall contouring;sensitivity analysis method8|7|0更新时间：2026-10-08
- 摘要：In coal liquefaction systems, regulating valves are prone to cavitation under large pressure-drop throttling conditions, leading to degraded flow capacity and intensified pressure pulsations, which ultimately undermine the long-term stable operation of the plant. To achieve the efficient identification of cavitation stages, an experimental platform for cavitating flow in a regulating valve was established. Fluid features—including valve opening, pressure drop, inlet and outlet pressures, flow rate, flow coefficient, deviation ratio, and the root-mean-square (RMS) value of pressure pulsations—were selected as input variables to construct a cavitation stage identification model based on multi-parameter fusion. A comparative evaluation of Random Forest (RF), Gradient Boosting Decision Tree (GBDT), and Xtreme Gradient Boosting (XGBoost) reveals that GBDT exhibits the best overall performance with an accuracy of 90.61%. Furthermore, the Particle Swarm Optimization (PSO) algorithm was introduced to develop a PSO-GBDT model, which improves the accuracy to 92.12% and reduces the misclassification between incipient cavitation and critical cavitation. Moreover, its classification performance significantly outperforms the accuracy of any single-feature model. Feature importance analysis indicates that the deviation ratio and pressure drop contribute the most to the classification. Based on the importance evaluation, after eliminating low-contribution redundant features such as valve opening and inlet/outlet pressures, the model accuracy is still maintained at a high level of 91.52%. The study demonstrates that the proposed multi-parameter fusion model enables the effective identification of cavitation stages in regulating valves, providing methodological support for the online monitoring of the system.关键词：Regulating valve;Cavitation stage identification;Multi-parameter input;machine learning;Particle swarm optimization (PSO)10|24|0更新时间：2026-09-30
- 摘要：To address the issues of low accuracy and limited applicability of existing deviation angle and loss empirical models in the aerodynamic analysis and design of transonic axial compressors, this paper develops a throughflow performance calculation method for transonic axial compressors and proposes an empirical model optimization and calibration method based on experimental multi-speed characteristic curve data. Using measured multi-speed characteristic data from typical transonic compressor stages, 15 empirical parameters in the model combination are optimized and calibrated, and the generality of the optimized model is validated on multiple transonic axial compressors/fans. The results show that the relative error of multi-speed throughflow calculations based on measured deviation angles and loss coefficients is less than 1%. Compared with the up to 6% calculation error of conventional empirical model combinations, the optimized empirical model reduces the multi-speed throughflow calculation error for compressor stages by 1%–4%. In the validation cases of a two-stage transonic fan and a three-stage transonic compressor, the relative errors of throughflow calculations at the design rotational speed are all reduced to within 1%. This work provides valuable throughflow calculation methods and empirical models for the analysis, design, and model development of multi-stage transonic axial compressors.关键词：transonic axial compressor;throughflow performance calculation;multi-speed;empirical model optimization9|27|0更新时间：2026-09-29
- 摘要：To clarify how the internal voltage distribution of 10–35 kV distribution transformers varies with frequency and the proportion of the high-frequency component under composite voltages consisting of power-frequency and high-frequency components, this study investigates composite-voltage generation methods and winding voltage distribution characteristics. Three superposition topologies are developed based on electromagnetic coupling between transformer windings. Their protection and isolation parameters are optimized through 1–100 kHz frequency-sweep simulations, considering voltage transfer and branch isolation performance. A 10 kV transformer test platform is then established, and the effects of frequency and the proportion of the high-frequency component on internal winding voltage distribution are analyzed using a distributed-parameter model and modal analysis. Results show that all three topologies effectively transmit the power-frequency component. The low-voltage coupling superposition topology requires a maximum high-frequency source voltage of only 23.6 V within 1–20 kHz, compared with 610 V for the other two topologies. In this topology, both the power-frequency component and the high-frequency component within 1–20 kHz decrease monotonically along the winding. As the frequency of the high-frequency component approaches a natural modal frequency, the voltage peak shifts toward the winding interior. At 60 kHz, the high-frequency voltage at 40% of the winding length reaches 3.11 times that at the winding terminal. When the proportion of the high-frequency component increases from 5% to 20%, the spatial voltage peak increases from 78 V to 312 V, while the distribution pattern and peak location remain nearly unchanged. The proposed topology substantially reduces the required high-frequency source voltage while maintaining effective voltage transfer and branch isolation, providing technical support for wide-frequency insulation testing and operational safety assessment of distribution transformers.关键词：power frequency superimposed with high frequency;composite voltage;transformer winding;voltage distribution10|22|0更新时间：2026-09-29
- 摘要：Liquid nitrogen (LN₂) spray cooling is a key technology for cryogenic wind tunnels, which is essential for high-Reynolds-number wind tunnel tests of advanced aircraft. The dense spray field produced by spray arrays significantly increases multi-droplet collision probability, and such collisions profoundly influence atomization and evaporation characteristics. Most existing studies have primarily focused on binary droplet collisions of room-temperature working fluids, with insufficient attention to cryogenic working fluids, particularly multi-droplet collisions. To reveal the dynamic behavior of LN₂ multi-droplet collisions and clarify the influence of droplet number on collision outcomes, three-dimensional numerical simulations of binary and ternary droplet collisions in quiescent air were performed using the Volume of Fluid (VOF) method. The effects of Weber number (We = 8–204) and impact parameter (B = 0, 0.4, 0.8) on collision dynamics are systematically investigated. The results indicate that: (1) LN₂ droplet collisions exhibit four morphologies—coalescence, bouncing separation, stretching separation, and breakup—whose distribution is jointly determined by We and B. Compared with binary collisions, ternary collisions do not exhibit breakup at B = 0.4, and the critical Weber number for separation and breakup at B = 0 is higher, indicating that an increased number of droplets enhances the system’s resistance to breakup. (2) In head-on collisions,higher We leads to flatter droplets, favoring evaporation. Increasing B and the number of droplets strengthens the stretching effect. (3) Energy analysis reveals that head-on collisions involve the most intense energy conversion. As B increases, energy oscillations are attenuated. Under a given condition, ternary collisions display more frequent surface energy fluctuations and higher kinetic energy peaks compared with binary collisions, indicating that multi-droplet collisions intensify internal energy exchanges. This study could provide a theoretical basis for predicting and optimizing the cooling performance of liquid nitrogen spray systems.关键词：cryogenic wind tunnel;liquid nitrogen spray;multi-droplet collisions;deformation characteristics;energy analysis16|0|0更新时间：2026-09-23
- 摘要：To achieve early warning of disk-shaft cracks in aero-engines, a crack fault warning method based on a physics-embedded network architecture is proposed. A neural network architecture embedded with physical structure is constructed by integrating Long Short-Term Memory (LSTM) recurrent neural networks with the component-level physical topology of the engine, where input parameters are clustered according to engine components. An initial warning model is established using data from five healthy turbofan engines, and its predictive performance is evaluated. The results show that the model achieves a symmetric mean absolute percentage error (sMAPE) of 6.07% for vibration prediction in the engine N1 speed range of 40%-80%, demonstrating reliable real-time prediction of engine vibration response. A dynamic prediction model is constructed using a rolling update strategy and validated for accuracy. Based on trend deviation analysis of vibration values, abnormal features characterized by unidirectionality, cumulativity, and irreversibility are captured in advance, confirming the effectiveness of the proposed early warning method. Without relying on historical fault data or additional sensors, this method offers significant advantages in terms of high accuracy, interpretability, and verifiability, providing a feasible technical pathway for transitioning aero-engine structural health monitoring from post-event diagnosis to pre-event warning.关键词：aero-engine;Vibration monitoring;Intelligent early warning;Crack fault;Long short-term memory (LSTM);deep learning16|41|0更新时间：2026-09-23
- 摘要：Increasing the number of radiation shield in the vacuum multi-layer insulation of liquid hydrogen tank will inevitably enhance the thermal insulation performance of tank. To address the demands of high insulation efficiency and structural light weighting, the marginal effects of increasing the number of radiation shield on the insulation performance of vacuum multi-layer insulation are investigated. Based on a steady-state heat transfer model that couples radiation heat transfer, solid heat conduction, and residual gas conduction, a simulation model of vacuum multi-layer insulation integrating vapor-cooled shield (VCS) and para-ortho catalytic conversion (POC) was developed. Furthermore, an optimization method for a variable-density multi-layer insulation (VDMLI) was introduced to determine the optimal layer arrangement and the corresponding minimum heat leakage flux for setting different numbers of radiation shields. Based on this model, a systematic analysis was conducted on the effect of VCS configuration (presence/absence), VCS position, the combination of VCS with POC, and the number of VCS on the marginal effects of the insulation performance of tank. The results indicate that, as the number of VCS in VDMLI increases, the rate of reduction in heat leakage flux gradually slows down. In other words, the improvement in system insulation is primarily driven by the first VCS, with subsequent VCS providing diminishing returns. Coupling a single VCS with POC significantly enhances the insulation performance, becoming particularly pronounced with a higher number of radiation shields. The analysis of VDMLI configurations with different numbers of VCS (0, 1, 2, 3), or single VCS coupled with POC reveals a significant decrease in the marginal effect on heat leakage flux as the number of radiation shields increases. When the number of layers exceeds 50, further increasing the number of radiation shield improves insulation performance by less than 1%. The findings of this study will provide a theoretical basis for the design and parameter optimization of VDMLI integrated with VCS and POC in liquid hydrogen tank.关键词：liquid hydrogen tank;variable-density vacuum multi-layer insulation;vapor-cooled shield;optimization;marginal effect17|40|0更新时间：2026-09-23
- 摘要：To address the challenges of complex morphologies in phase-resolved partial discharge (PRPS) images of GIS, the blurred boundaries between similar fault categories, and the inadequate adaptation of general-purpose multimodal large models, this paper proposes a physics-prior enhanced multimodal low-rank adaptation diagnostic method (PPEM-LoRA).First, a physics-prior enhancement mechanism is constructed by extracting 15 statistical indicators from the PRPS data and encoding their numerical values and physical meanings into structured prompts. These prompts are jointly fed into the model alongside the images to explicitly supplement electrical features. Second, low-rank adaptation (LoRA) is employed for parameter-efficient fine-tuning of the multimodal large model. While keeping the base model's original weights frozen, only the low-rank incremental matrices in the attention and feed-forward networks are updated, enabling the model to autonomously learn the deep mapping relationships among visual morphologies, physics-prior indicators, and fault types. Finally, conditional normalization is applied to the scores of candidate numerical tokens, strictly constraining the model's output to one of nine specific fault types to achieve automatic diagnosis. Experimental results demonstrate that on an independent test set comprising 4 460 images, the proposed PPEM-LoRA method achieves a top-1 accuracy of 98.61%, a top-2 accuracy of 99.42%, and a macro-averaged F1 score of 98.71%. Furthermore, the number of misdiagnosed samples is reduced from 111 to 62. This method effectively enhances the model's representation capability for morphological variations in GIS partial discharge images, providing a technical pathway for intelligent partial discharge diagnosis in power equipment that balances high performance, low resource consumption, and multimodal extensibility.关键词：GIS;partial discharge;physical prior enhancement;multimodal large model;low-rank adaptation8|1|0更新时间：2026-09-18
- 摘要：In response to the security vulnerabilities exposed by current vision-language models in multimodal scenarios, this study proposes an Automated Typographic Prompt Injection Attack tailored for autonomous driving scenarios. The proposed method constructs an end-to-end closed-loop pipeline consisting of autonomous driving scene perception, automated adversarial semantic generation, physically adaptive rendering, and cross-modal decision hijacking. Specifically, a Vision-Language Model (VLM) is first employed to perceive the original driving scene state, after which a Large Language Model (LLM) automatically generates adversarial text instructions with strong semantic inversion and deceptive effects. Subsequently, combined with an object detection model, the generated attack texts are adaptively rendered and typeset onto appropriate physical carriers within the scene. Finally, a cloud-based Large Language Model is introduced as an automated referee to accurately evaluate the attack success rate of inducing hazardous decisions by comparing the perception and planning reports before and after the attack.Experimental results demonstrate that the proposed Automated Typographic Prompt Injection Attack successfully deceives various advanced Vision-Language Models, including Qwen-VL, InternVL, BLIP, and the autonomous driving specialized model DriveMM, achieving attack success rates of 67.55%, 61.51%, 30.57%, and 10.38%, respectively.关键词：Autonomous Driving;vision-language models;typographic attack;prompt injection;safety40|73|0更新时间：2026-09-08
-
Operational Regulation of a Solar Methanol Decomposition Reactor under Solar Irradiance Disturbances
摘要：The intermittency and fluctuations in direct normal irradiance (DNI) can cause methanol conversion in a solar methanol decomposition reactor to deviate from its target, reduce system efficiency, and even lead to catalyst overheating and sintering. To address these issues, an integral model predictive control (IMPC) strategy incorporating extended Kalman filter (EKF)-based state estimation is proposed. First, a one-dimensional dynamic model and its state-space representation are established for the reactor. Next, an EKF-based state estimator uses temperature measurements to correct the reactor states in real time, enabling online estimation of the species mass fractions. On this basis, an IMPC controller with integral compensation is developed to track methanol conversion, and receding-horizon optimization is used to determine the optimal methanol feed flow rate under irradiance disturbances. Results show that, when a 2% deviation in optical efficiency is introduced, the mean estimation error of the methanol mass fraction is only 0.06%. Under cloud-cover conditions, the proposed IMPC limits the maximum deviation in methanol conversion to 0.41%, compared with 1.85% for standard MPC and 6.87% for PI control. The average methanol conversion reaches 94.97%, the settling time decreases from 400 to 300 s, and the solar-to-chemical exergy efficiency increases from 16.59% to 18.11%. No catalyst overheating occurs throughout the process. These results demonstrate that the proposed strategy enables efficient, safe, and stable operation of the solar methanol decomposition reactor under solar irradiance disturbances.关键词：solar energy;methanol decomposition;dynamic characteristics;model predictive control;operational strategy35|101|0更新时间：2026-09-04 - 摘要：To address inefficient rule matching, insufficient historical-case association, and unreliable suggestions from general large language models in transmission-line fault handling, this paper proposes a knowledge-enhanced LLM-based intelligent decision-support method. Within a retrieval-augmented generation framework, the method constructs a cross-modal knowledge base integrating sequence-component features, alarm-event trajectories, fault-handling procedures, and typical cases. A physics-consistency-constrained hybrid retrieval score is developed to jointly measure semantic similarity, electrical-feature similarity, metadata matching, and fault-mechanism consistency. A confidence-aware generation mechanism is also introduced to trigger conservative prompts or manual review when evidence is insufficient or conflicting. Validation using multi-scenario transmission-line fault question-answering samples and the IEEE 39-bus system shows that the Top-1 and Top-3 retrieval hit rates, fault-diagnosis accuracy, decision completeness, and evidence traceability reach 88.6%, 95.7%, 94.3%, 91.4%, and 94.3%, respectively, while the hallucination rate is reduced to 2.9%. Compared with semantic-only RAG, the method improves fault-diagnosis accuracy by 11.4% and reduces the hallucination rate by 8.5%. The results demonstrate that explicitly integrating fault mechanisms, procedural knowledge, and LLM reasoning improves the accuracy, traceability, and safety reliability of transmission-line fault-handling suggestions.关键词：large language model;transmission-line fault handling;knowledge enhancement;physics-consistency constraint;retrieval-augmented generation;intelligent decision support96|117|0更新时间：2026-09-04
- 摘要：To address the problems of data redundancy and energy waste caused by fixed-rate sampling of analog-to-digital converters (ADCs) in low-frequency sparse signal acquisition sensors, a 10-bit tracking-quantizing successive approximation register ADC (SAR-ADC) is proposed. The proposed SAR-ADC consists of three functional modules: tracking module, startup module, and quantization module. The tracking module tracks the input signal via a capacitive digital-to-analog converter (CDAC) with near-zero power consumption. In the startup module, a pulse converter converts the trigger pulse generated by the sensor system into asynchronous control signals, which control the CDAC to complete sampling and initiate the quantization process. In the quantization module, 10-bit successive approximation quantization is jointly accomplished by the CDAC, comparator, reference buffer and logic circuit; upon completion of quantization, the SAR-ADC returns to the tracking state. The SAR-ADC enables sensor systems to implement a time-domain variable-rate sampling strategy and significantly reduces hardware overhead. In addition, a low-leakage long-delay generation circuit is designed in the logic circuit, which effectively reduces the power consumption of the delay unit. The proposed SAR-ADC was designed in a 180-nm CMOS process and validated through tape-out testing. Measurement results demonstrate that under a 2-V supply voltage and a sampling rate of 10³ /s, the proposed SAR-ADC achieves an effective number of bits (ENOB) of 9.57 bits, which satisfies the accuracy requirement for ECG signal acquisition. Under adaptive dual-rate sampling (10³ /s and 10² /s), the power consumption of the proposed SAR-ADC is only 16.7% of that of a fixed-rate sampling ADC. The proposed SAR-ADC can be applied in biomedical sensors to realize low-power acquisition of sparse signals.关键词：sensor;sparse signal;ultra-low power consumption;variable-rate sampling;analog-to-digital converter (ADC)48|120|0更新时间：2026-09-02
- 摘要：To address the issues of point loss, occlusion, and geometric distortion in LiDAR point clouds caused by adverse weather conditions such as rain, snow, and fog, which lead to degraded performance in 3D semantic segmentation and particularly affect the segmentation accuracy of small objects like pedestrians and bicycles, a point cloud semantic segmentation method is proposed based on degraded feature consistency learning. First, a data augmentation module consisting of fine-grained semantic instance enhancement and spatially partitioned distance-aware perturbation is constructed. Through geometric perturbation simulation and small object instance expansion, the model's learning capability for structural variations and small object features is enhanced. Second, a point cloud voxel feature extraction network is used to encode point clouds to obtain voxel feature representations. Finally, random degradation is applied to voxel features to construct a "clean-degraded dual-branch" structure, and the robust representation capability of the network under feature corruption conditions is improved through consistency constraints. The proposed method is experimentally validated on the public SemanticSTF dataset. The results show that this method can effectively improve the performance of point cloud semantic segmentation under adverse weather conditions. In terms of performance metrics, the mean Intersection over Union (mIoU) is 10.7% higher than that of the baseline model, and better segmentation results are achieved for small object categories such as pedestrians and bicycles. The research results can provide a reference for the safe and reliable application of 3D perception systems in complex weather environments.关键词：semantic segmentation;adverse weather;generalization54|103|0更新时间：2026-09-02
- 摘要：In autonomous driving scenarios, vehicle-mounted LiDAR point clouds are characterized by large scale and sparse, uneven distribution. To address the problems of high neighborhood search overhead, heavy voxel attention computation load, and easy loss of local geometric details of key objects in existing point cloud Transformer networks, a 3D adaptive pooling point cloud segmentation network (3DAPT-Net) is proposed. First, 3D adaptive pooling is utilized within non-overlapping voxel windows to compress key-value sequences and aggregate multiscale context, thereby reducing the attention computation overhead for large-scale point clouds. Second, a point attention branch is adopted to aggregate point-level features, supplementing the fine-grained geometric information of road boundaries, pedestrian contours, and distant sparse objects. Finally, point-level features and voxel features are fused to construct a point cloud segmentation network that balances both accuracy and efficiency. The proposed network achieves mean intersection over union (mIoU) scores of 67.5% and 87.4% on the SemanticKITTI and ShapeNet datasets, respectively; compared with the Point Cloud Transformer network, it has the inference latency reduced by 54.9%. The results indicate that 3DAPT-Net achieves a favorable balance between segmentation accuracy and computational efficiency.关键词：Autonomous Driving;point cloud segmentation;attention mechanism;adaptive pooling65|116|0更新时间：2026-09-02
- 摘要：To address the limited generalization capability of diagnostic models caused by small sample sizes and class imbalance in photovoltaic infrared defect recognition, a generative adversarial network incorporating coordinate awareness and subpixel reconstruction, termed CPS-GAN, is developed. With CPS-GAN as the core generative network, a photovoltaic module infrared defect image generation method is further proposed for augmenting defect classes with limited samples. First, considering the spatial sensitivity of local thermal anomalies and the pronounced variations in thermal gradients in photovoltaic infrared images, class-conditional constraints and a CoordConv-based coordinate-aware module are introduced into the generator. This design establishes a mapping relationship between latent representations and geometric space, thereby improving the ability of conventional GANs to represent the spatial locations of local defects accurately. Second, a PixelShuffle-based subpixel reconstruction strategy is employed to replace conventional transposed convolution, suppress checkerboard artifacts in generated images, and preserve thermal anomaly distribution characteristics. Finally, a PatchGAN discriminator is incorporated to strengthen local texture discrimination. Experimental results demonstrate that CPS-GAN outperforms the DCGAN and WGAN-GP baselines across multiple quantitative metrics and semantic evaluation protocols, while exhibiting improved feature-distribution alignment in t-SNE visualizations. Without modifying the classifier architecture, the classifier trained on the CPS-GAN-augmented dataset achieves a test accuracy of 84.00%, representing an improvement of 7.05 percentage points over training with the original dataset. The proposed method can effectively generate high-quality and diverse infrared defect samples of photovoltaic modules, thereby providing reliable data support for improving the robustness of intelligent photovoltaic diagnostic systems.关键词：generative adversarial networks;data augmentation;defect detection;infrared imaging51|138|0更新时间：2026-09-02
- 摘要：To achieve resource utilization of spent sodium battery cathode material NaFePO4 (NFP) and the effective elimination of pharmaceutical emerging contaminants from water, this study employed NFP as a heterogeneous catalyst to activate peroxymonosulfate (PMS) for the degradation of various pharmaceuticals. The degradation performance, reaction mechanism, and the effects of solution pH, NFP dosage, and PMS concentration on organic removal were systematically investigated. In addition, reusability experiments and X‑ray diffraction (XRD) characterization were performed to evaluate the catalyst stability and deactivation mechanism. The findings reveal that NFP can effectively activate PMS to degrade various pharmaceuticals, with the optimal reaction conditions determined as pH 4.5, NFP dosage of 0.2 g·L-1, and PMS concentration of 0.2 mmol·L⁻¹. Hydroxyl radicals (·OH) and sulfate radicals (SO₄·-) generated via Fe2+-mediated PMS activation in NFP are the dominant reactive species governing the degradation of acetaminophen, ibuprofen, primidone, and carbamazepine. Singlet oxygen and free radicals both play important roles in the removal of diclofenac while singlet oxygen plays the primary role in the removal of ciprofloxacin. NFP retained over 60% acetaminophen removal efficiency and structural stability after five consecutive reaction cycles. Its deactivation mechanism is primarily attributed to the oxidation of active Fe2+ sites to Fe3+.This study provides theoretical insights and technical guidance for the resource recovery of spent SIB cathodes in the future and the remediation of emerging pharmaceutical pollutants in water.关键词：Drugs;sodium iron phosphate;peroxymonosulfate;reaction mechanism;optimization of reaction parameters51|106|0更新时间：2026-09-01
- 摘要：To comprehensively elucidate the kinetic characteristics of major metabolic pathways involved in photo-fermentative hydrogen production, this study developed and optimized an (Anaerobic Digestion Model No. 1) ADM1-based photo-fermentation model with glucose as the reference substrate. The model is grounded in mass conservation principles and allocates the substrate to three key pathways: cell growth, product synthesis, and cellular maintenance. Twelve critical metabolic reactions were incorporated, maintaining stoichiometric consistency, and variations in metabolic fluxes were represented by adjusting reaction rates. A "two-step" approach was used to efficiently determine the kinetic parameters based on microbial glycolysis. Experimental data from photo-fermentation with varying initial glucose concentrations were used for model calibration and validation. The model was further refined by including lactate and ethanol as intermediate metabolites and by considering the effects of pH and light intensity on substrate conversion, biomass accumulation, and hydrogen production. Additionally, the impact of pH on carbon metabolic flux was accounted for. Results showed that the improved model significantly enhanced predictive accuracy, with correlation coefficients exceeding 0.96 for hydrogen production, bacterial growth, organic acid production, and the variation of light intensity and pH under different glucose concentrations. Utilizing response surface methodology, optimal conditions for photo-fermentation were determined, with an initial pH of 8 and light intensity of 5000 lx. This work provides important theoretical support and a modeling framework for the development and control of hydrogen production reactors.关键词：photo-fermentative;hydrogen production kinetics model;organic acid;ADM1;process optimization31|101|0更新时间：2026-08-31
- Technical support is provided by Beijing Founder electronics co., LTD 京ICP备09064830号-19
京公网安备11010802024621 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
0