Gaofeng Zheng

Multi-class industrial anomaly detection aims to identify visual patterns that deviate from normal appearance distributions across diverse industrial objects and textures. Large variations in shape, material, texture, and spatial structure make reconstruction-based methods prone to unstable outputs when handling multiple categories within a unified model. For latent diffusion-based reconstruction, such instability is reflected in category-dependent semantic drift, where denoising trajectories are biased by inconsistent structural and contextual cues, leading to inaccurate pixel-level localization. To address this problem, SCAD is proposed as a semantic-stabilized latent diffusion framework for multi-class industrial anomaly detection. It constructs image-adaptive semantic guidance through per-image dynamic semantic fusion and further injects semantic bias into global attention during denoising. In this way, the reconstruction process is guided by more stable structural cues across diffusion steps, reducing category-dependent reconstruction bias and noise-induced attention drift. Experiments on MVTec-AD and VisA show that SCAD achieves robust pixel-level anomaly localization and stable image-level discrimination, reaching pixel-level AUROC values of 97.3% on MVTec-AD and 97.2% on VisA, respectively.

Metallic glasses exhibit dissolution characteristics that differ from those of crystalline metals because short-range structural heterogeneity, passive-film evolution, and the formation and adhesion of low-solubility products jointly affect anodic material removal. Here, the dissolution behavior of Ni72Cr19Si7B2 metallic glass was investigated in 0.1 M H2SO4, HCl, and NaNO3 through structural characterization, electrochemical measurements, and surface analysis, and correlated with wire electrochemical micromachining (WECMM) performance. A cluster-mediated selective dissolution model is proposed to interpret the observed dissolution morphology in relation to the amorphous structure. The alloy exhibits activation, passivation, and transpassivation behavior in all three electrolytes, while the associated dissolution morphology and product transport depend strongly on electrolyte chemistry. H2SO4 produces a comparatively stable passive response and a more uniform machined surface. HCl promotes passive-film breakdown and provides the highest machining feed rate, but also induces localized pitting. NaNO3 gives slower dissolution, together with the accumulation and adhesion of solid products in the machining gap, leading to machining instability. Among the tested electrolytes, H2SO4 provides the best balance between dissolution localization, machining stability, and surface quality. Micro-pentagram and micro-gear structures were fabricated on 50 μm-thick metallic-glass sheets in 0.1 M H2SO4, with surface roughness values of 52.7 and 74.8 nm, respectively.

Sustainable air filtration materials derived from green electrospinning of bio‐based polymers are highly desirable. However, achieving precise fiber morphology regulation and multifunctional integration in bio‐based electrospun systems remains challenging, particularly for balancing filtration performance and comfort related properties. Herein, a small‐molecule‐mediated hierarchical molecular assembly strategy is developed to engineer well‐defined ribbon‐like nanofibers in a zein system by incorporating chlorogenic acid (CA) and rutin (RU). The noncovalent interactions between CA/RU and zein regulate molecular organization through distinct yet complementary pathways, where CA promotes local β‐sheet‐rich conformational ordering while RU facilitates intermolecular association. Their competitive and cooperative molecular interactions reorganize zein chain assembly, thereby stabilizing ribbon‐like fiber architectures and optimizing filtration performance. The resulting zein/CA/RU composite nanofibrous membrane delivers 99.55% filtration efficiency for PM _0.3 with a pressure drop of 53.6 Pa and a quality factor of 0.101 Pa ^{− 1} . Meanwhile, its reduced airflow resistance and passive radiative cooling capability provide potential benefits for thermal comfort‐related performance. This work provides a molecular‐level design strategy for morphology engineering and multifunctional integration of naturally derived nanofibers, offering a pathway toward high‐performance, low‐resistance, and sustainable air filtration materials.

Carbon nanotube fibers (CNFs) are promising tool electrodes for wire electrochemical micromachining (WECMM), but by-product accumulation within confined machining gaps limits electrolyte renewal and destabilizes the local electrochemical environment. Here, hierarchical GQDs-NiCo2S4 nanowire arrays were constructed on CNFs through a two-step hydrothermal strategy to couple interfacial electrochemical enhancement with mass-transport regulation. Sulfurization for 4 h produced a brush-like interface with nanowires approximately 9.6 μm long and 296 nm in diameter. The resulting cathode reduced the contact angle from 110° to 55° and increased the stable dynamic wetting height by 18.1%. The cathodic current increased from 23.8 to 58.9 mA at -1.0 V, accompanied by substantially reduced interfacial impedance. During WECMM, the hierarchical nanowire architecture couples with vertical electrode reciprocation to facilitate electrolyte renewal and by-product evacuation, mitigating disturbances in ionic transport and current distribution and thereby enabling charge dynamic regulation. Consequently, the composite cathode reduced slit-width fluctuations and surface roughness while enabling stable fabrication of a complex 1150 μm-long microstructure. These results establish structure-motion coupling at hierarchical electrode interfaces as an effective route for stabilizing confined electrochemical machining and improving precision microfabrication.

Cysteine (Cys) is a crucial biomarker and a significant analyte in physiological and pathological processes; however, achieving its accurate detection with high spatiotemporal resolution in complex biological environments remains challenging. Herein, we address this by developing a dual‐mode optoelectronic sensing platform. A self‐designed Cys fluorescent probe was integrated into a plasmonic nanopore via a site‐selective functionalization strategy. This platform synergistically combines plasmon‐enhanced fluorescence with nanopore‐based electrochemical detection, providing complementary optical and electrical signals that enable high spatio‐temporal resolution for accurate analyte quantification. The confined geometry of the nanopore enhances sensitivity by concentrating electromagnetic fields into sub‐diffraction volumes and restricting molecular motion within the nanospace. The developed Cys sensor achieves high sensitivity, along with excellent selectivity and strong anti‐interference capability, as validated through an “AND” logic gate operation. Importantly, the practical utility of the sensor is demonstrated by successful quantification of exogenous Cys in living MDA‐MB‐231 cells, confirming its capability for in situ cellular analysis. Furthermore, this versatile plasmonic nanopore platform can be readily adapted for the detection of various small molecules by simply substituting the probe molecule, offering a simple, stable, and generalizable sensing approach in small molecules detection.

Functional micro- and nanostructures are essential for polymer-based devices in biomedicine, sensing, and optics. Yet conventional fabrication methods such as lithography and mold transfer remain expensive, slow, and poorly suited for large-scale production. In this study, we present an ink-enhanced photothermal strategy for the direct fabrication of high-quality polydimethylsiloxane (PDMS) microstructures using a UV nanosecond laser. By precoating the transparent PDMS surface with a thin layer of black ink, laser energy absorption and localized photothermal conversion are significantly improved, effectively overcoming PDMS’s intrinsic transparency. This enhancement allows for precise material ablation at reduced laser power, enabling the creation of well-defined micro/nanostructures and smooth, burr-free microchannels with high structural fidelity. Additionally, the surface wettability of PDMS is readily tunable across a wide range - from intrinsic hydrophilicity (77.8°) to superhydrophilicity (3.26°) or enhanced hydrophobicity (127.4°) simply by adjusting the ink-enhanced laser-induced surface morphology. The versatility of this method is demonstrated through practical applications, including patterned droplet arrays, self-driven liquid transport, and integrated microfluidic channels. This ink-enhanced UV nanosecond laser processing strategy offers a simple, cost-effective approach for the high-precision structuring of transparent polymer materials and devices.

Accurate long-horizon prediction of state of charge (SOC) and battery temperature is critical for safe operation of large-scale battery energy storage systems (BESS). However, owing to strong electro-thermal coupling, cell heterogeneity, and time-varying operating conditions, existing data-driven and physics-informed methods may exhibit substantial error accumulation and limited transferability under long-horizon and distribution-shifted conditions. A Physics-Dominated Structured Evolution Framework (PDSEF) is proposed to model battery states as a constrained evolution process rather than direct observation-to-state regression. The state-transition backbone is governed by physical laws, including charge conservation and thermal dynamics, while adaptive physical parameters are inferred from recent historical observations during deployment, thereby characterizing cell-level heterogeneity without requiring explicit offline parameter calibration. Constrained neural residuals are further introduced to compensate for unmodeled nonlinear behaviors, and temporal, hierarchical spatial, and operating semantic representations are integrated to capture long-horizon dependencies, inter-cell interactions, and dynamic operating variations. The proposed framework is validated on a real-world BESS containing 384 cells under multi-condition charge-discharge operations. A 30-step recursive rollout is repeatedly propagated over more than 10,000 evolution steps to evaluate long-horizon prediction stability. The results demonstrate stable SOC and terminal temperature predictions with minimal error accumulation. The cross-condition mean SOC NMAE reaches 0.02299 and 0.02279 under zero-shot and fine-tuned settings, respectively, while terminal temperature NMAE improves from 0.16786 to 0.08999 after adaptation. Comparative and ablation studies confirm that the proposed physics-dominated evolution mechanism enhances generalization and long-horizon stability compared with existing baselines. Overall, the framework provides a reliable approach for proactive state monitoring and safety-oriented management of large-scale BESS.

Hydrogen sulfide (H2S) is a key gaseous signaling molecule yet quantifying its dynamics in living cells with high spatiotemporal resolution, sensitivity, and reliability remains a challenge. In contrast to conventional single-use sensors that only offer a snapshot, here, we develop a regenerable plasmonic nanopore sensor that enables simultaneous optical fluorescence and electronic readouts for self-validating detection. The capability for regeneration allows continuous and repeated monitoring of H2S flux from single cells, enhancing the statistical reliability and precision of quantitative analysis in complex cellular environments. The confinement of a molecular probe within the plasmonic hotspot yields a drastic, order-of-magnitude enhancement in both fluorescence and electrical signals, achieving ultrahigh sensitivity with a record-low, subpicomolar detection limit for H2S (49 fM by fluorescence signal and 71 fM by electrical signal)─surpassing all existing methods. The sensor also exhibits excellent selectivity against 14 biologically relevant interferents. We further demonstrate its practical utility by directly monitoring endogenous H2S release from individual living HeLa cells in real time under chemical stimulus. Benefiting from the mutually verifiable dual-signal outputs and a renewable sensing interface, the plasmonic nanopore sensor provides a robust and noninvasive platform for tracking signaling molecules in biological systems at the single-cell level.

Flexible electromagnetic actuators have attracted considerable attention for applications in soft robotics, adaptive manipulation, and human–machine interaction due to their fast response, large deformation capability, and inherent compliance. However, the concurrent application of high actuation performance and long-term cyclic durability remains a major challenge, particularly for liquid metal (LM)-based soft systems, where interfacial instability between LM conductors and polymer substrates often leads to performance degradation. In this work, we report a fabrication strategy in which patterned eutectic gallium–indium (EGaIn) liquid metal circuits are directly written onto electrospun graphene oxide/thermoplastic polyurethane (GO/TPU) nanofiber membranes. The incorporation of graphene oxide significantly enhances interfacial adhesion through hydrogen bonding interactions between oxygen-containing functional groups in GO and the native Ga₂O₃ layer on the LM surface, while the electrospun fibrous architecture further improves mechanical interlocking and structural stability. As a result, the fabricated actuator exhibits robust electromechanical performance, achieving a maximum bending deformation of 90° under a driving current of 0.8 A and maintaining stable operation over 2000 actuation cycles with negligible performance degradation. To further demonstrate its practical functionality, a soft robotic gripper was constructed based on the optimized actuator configuration. The gripper enables the stable grasping and lifting of objects with a weight up to seven times its own mass, while maintaining safe and compliant interaction with fragile objects. This work provides a simple yet effective strategy to simultaneously enhance actuation efficiency, interfacial stability, and mechanical reliability in LM-based GO/TPU flexible electromagnetic actuator systems, offering promising potential for next-generation soft robotic applications.

The measurement reliability of nanofilm humidity sensors in extreme industrial environments is significantly degraded by temperature-induced drift, resulting in pronounced uncertainty across wide temperature ranges and dynamically varying conditions. To mitigate this issue, a hybrid temperature compensation framework is proposed by integrating a Kolmogorov-Arnold Network (KAN) with an enhanced metaheuristic optimization strategy. Specifically, a multi-strategy enhanced sand cat swarm optimization (MESCSO) algorithm is developed to improve parameter optimization performance of KAN. The proposed method incorporates a dual-trajectory hybrid attack mechanism, a nest memory decay strategy, and a dynamic leveraged opposition-based learning strategy, collectively enhancing global exploration capability while maintaining population diversity and convergence stability. Based on the optimized parameters, the KAN-based nonlinear mapping model is established to characterize the coupled temperature–humidity response. Experimental results demonstrate that the proposed approach achieves a compensation accuracy of ±0.55% RH over the temperature range of 10 °C to 50 °C, validating its effectiveness and robustness for temperature drift correction in complex operational environments.