The amine-functionalized Christian-Albrechts-University-1 (CAU-1) metal-organic framework (MOF) possesses intrinsic structural features ideally suited for hydrogen (H2) separation, with pore apertures closely matching the molecular size of hydrogen, thereby enabling selective hydrogen transport via molecular sieving. The physicochemical and structural tailoring of CAU-1 holds immense potential for gas separation thin-film nanocomposite (TFN) membranes yet it remains largely unexplored. This work presents the first systematic investigation of CAU-1 incorporation confined within a thin PMMA selective layer supported on a polysulfone (PSF) substrate, elucidating its influence on the interfacial structure, gas transport pathways, and hydrogen separation performance of polymethyl methacrylate (PMMA)/CAU-1 TFN membranes. Morphological analyses revealed the successful formation of thin, dense, and defect-free TFN exhibiting excellent polymer-filler compatibility with no observable interfacial voids, while thermogravimetric analysis demonstrated excellent thermal stability up to 360 degrees C. The optimized PMMA/CAU0.5 membrane achieved an H2 permeability of 294.43 Barrer, with ideal selectivities of 4.91, 6.30 and 5.03 against carbon dioxide (CO2), nitrogen (N2), and methane (CH4), respectively, representing respective improvements of 11%, 42%, and 51% over pristine PMMA. Under mixed-gas conditions, the H2/CO2 separation factor surpassed the Robeson upper limit indicating the effective H2-selective transport properties. These findings highlight the pivotal role of CAU-1 in creating microporous, H2-selective transport pathways, demonstrating the potential of PMMA/CAU-1 TFN membranes for enhanced hydrogen purification.
Reliable thermal runaway risk monitoring during electric vehicle (EV) fast charging requires models that capture short electrical transients and long thermal dependencies under changing operating conditions. Existing predictors often emphasise temperature accuracy but provide limited mechanisms for converting multivariate prediction errors into hierarchical online decisions. This study develops a hierarchical monitoring framework that combines a Convolutional Neural Network (CNN), a Bidirectional Nested Long Short-Term Memory (BiNLSTM) network, and an Adaptive Gating Fusion Mechanism (AGFM). The CNN branch extracts local fluctuations, the BiNLSTM branch represents long-range thermal evolution, and AGFM adjusts their contributions according to operating conditions. An adaptive sliding window then converts temperature-prediction residuals into three hierarchical warning levels using empirically calibrated severity thresholds. The framework is evaluated using four-season real-world Lithium Iron Phosphate (LFP) fast-charging data and auxiliary Newman, Tiedemann, Gu, and Kim (NTGK) electrothermal simulations of Nickel Cobalt Manganese (NCM) modules. On the real-vehicle data, the model achieves MAEs of 0.0311–0.0484 °C across seasons and an R2 up to 0.9989. At the second warning level, accuracy, precision, and F1-score reach 92.73%, 83.33%, and 86.21%, respectively. In abnormal charging cases, the first warning level provides 19.70–34.25 s of intervention time before protective cutoff. These results demonstrate a computational route from multivariate battery-state prediction to hierarchical online monitoring for safer fast-charging operation.
This work demonstrates the direct conversion of CO2 to aromatics using a NaFeCo/H-ZSM-5 multifunctional catalyst, highlighting how hydrothermal synthesis conditions of H-ZSM-5 (especially temperature and rotating rate) precisely tune the physicochemical properties of the acidic zeolite and the resulting aromatics synthesis performance. By varying the synthesis parameters, we flexibly control the Al location and the acid sites distribution in H-ZSM-5, which directly govern the aromatics selectivity. Multiple experimental and characterization results clarify that the synergy between Al location and Brønsted/Lewis acid sites ratio collectively influence the selectivity of total aromatics, whereas the BTX (benzene, toluene, and xylene) fraction in total aromatics is dominated by the external surface acid sites of H-ZSM-5. Specifically, the increase of hydrothermal temperature optimizes the acid site density of zeolite, therefore boosting the space–time yield (STY) of aromatics and BTX by enhancing the driving force of the tandem process. For the rotating rate factor, a higher rotating rate enriches Al at channel intersections—a configuration that is more conducive to aromatics synthesis—leading to an aromatics selectivity of up to 48.5 %, while the increased external acidity promotes the formation of heavy aromatics. Therefore, optimized STY of BTX (19.5 mmol·g-1h−1) is achieved at a moderate rotating rate of 5.6 r/min. This study establishes a clear structure–function relationship between the acidic property of H-ZSM-5 and aromatics synthesis performance, providing a strategic guideline for the rational design of H-ZSM-5 zeolites for CO2 hydrogenation toward value-added aromatics.
Existing few-shot point cloud semantic segmentation methods commonly suffer from insufficient prototype feature perception capabilities, specifically manifested in limited local geometric modeling, inadequate semantic expressiveness of prototypes, and weak interaction between the support set and query set. To address these issues, we propose a novel Graph-Relational Interactive Prototypes Network (GRIP-Net). First, we introduce a graph-structure enhancement module into the encoder to capture fine-grained geometric relationships among points. This design compensates for the limited capability of the convolutional backbone in modeling local spatial structures and provides richer and more accurate geometric features for subsequent semantic segmentation. Second, a Query-Aware ProtoFuse (ProtoFuse) module is designed. By introducing query information, the prototypes of the support set are dynamically fused to generate more discriminative and robust category representations. Finally, to address the limitation of independently modeled categories and sets, we propose a Graph-Based Prototype Interaction (GPI) module that explicitly captures cross-class and cross-set prototype relationships, leading to improved multi-class generalization.The proposed framework provides an effective mechanism for information representation, fusion, and interaction, thereby improving the utilization of limited supervision for few-shot learning. Extensive experiments validate the effectiveness of the proposed method. Notably, under the 2-way 1-shot and 3-way 1shot setting, GRIP-Net achieves improvements of +0.21% and +0.46% mIoU on the S3DIS datasets respectively, compared to the state-of-the-art methods.
Harsh marine environments accelerate the corrosion of metal infrastructures, leading to increased maintenance costs and shortened service lifetimes. Photoinduced cathodic protection (PICP), as a solar-driven anticorrosion technology, shows promising potential; however, its practical application is limited by inefficient charge carrier separation and insufficient long-term stability. To address these challenges, a TiO2/BaTiO3/ZnIn2S4 multiphase photoelectric film was constructed via a combination of hydrothermal conversion and electrodeposition methods. TiO2 nanosheet arrays act as the structural backbone, while low-priced ferroelectric BaTiO3 is introduced in situ to generate an internal polarization electric field that promotes directional separation of photogenerated charge carriers. Meanwhile, surface-deposited ZnIn2S4 film extends visible-light absorption and forms an interface charge transfer gradient band, thereby enhancing photoinduced electron utilization efficiency. Under simulated solar irradiation and without adding external sacrificial agents in 3.5 wt% NaCl solution, this multiphase stacked photovoltaic film induces a negative shift of 350 mV in the photoinduced potential of 316 L stainless steel, delivers a stable photoinduced current density of 10.00 μA·cm−2, and maintains cathodic polarization during 6 h of intermittent illumination. Combined in-situ Kelvin probe force microscopy and in-situ X-ray photoelectron spectroscopy etc. reveal that the enhanced performance originates from matched band alignment, internal electric field effects, and optimized charge carrier migration pathways. This work provides valuable insights into the design of low-priced photoinduced cathodic protection system for efficient and stable PICP in marine environments and related photoelectrochemical applications.
