Three-dimensional inspection of complex microstructures is a key task in micro/nano coordinate measurement, requiring probes with a large measurement range, high displacement resolution, and compact structure. This study presents a 3D contact scanning probe based on an orthogonally arranged parallel leaf-spring mechanism and dual charge-coupled device (dual-CCD) optical sensing. Three pairs of mutually orthogonal parallel leaf springs are used to form a near-isotropic compliant guide, enabling elastic displacement of the stylus along the X-, Y-, and Z-axes. A dual-CCD optical sensing system captures laser spot shifts induced by stylus deflection, and the stylus displacement is estimated using sub-pixel centroid extraction and a 3D decoupling model. Experimental results show that over a measurement range of ±1.5 mm in the X-direction, ±1.0 mm in the Y-direction, and 1.5 mm in the Z-direction, the probe achieves a resolution better than 20 nm. At 20 ± 0.05 °C, the triaxial drift over 30 min is less than 138 nm, and the standard deviation within a 1 mm travel is less than 82 nm. These results indicate that the probe achieves millimeter-scale range and nanometer-scale resolution, providing a viable technical approach for precision 3D form measurement.
Infrared images are frequently degraded by structured stripe noise, blind pixels, and complex background interference caused by detector non-uniformity and circuit imperfections, making it challenging to simultaneously achieve global artifact suppression and local detail preservation. To address this issue, we propose a frequency-aware hybrid collaborative network for infrared image denoising, termed HybrMamba, which combines physical degradation priors with efficient neural modeling. The proposed architecture is fundamentally constructed by two macro-components: a Gated Physical Prior Head (GPH) at the front end, and a core backbone composed of multiple cascaded HybrMamba blocks. The GPH leverages median-filtering statistics to adaptively correct blind pixels at the input stage. Each HybrMamba block adopts a tri-branch architecture, including a Global Structural Component (GSC) branch for long-range modeling, a Local Convolutional Extractor (LCE) branch for fine-grained feature extraction, and a Spatial-Frequency Gating (SFG) branch. The SFG employs two-dimensional fast Fourier transform to identify stripe-related spectral components and generate adaptive gating masks for feature modulation. A Cross-Channel Feature Alignment (CFA) module is further introduced to fuse multi-domain features. Experiments on public benchmarks show that HybrMamba achieves 39.25 dB PSNR on CVC-14 and 36.04 dB on FLIR v2. Zero-shot evaluation on a real-world dataset of 200 no-reference infrared images further demonstrates its robustness and generalization capability in practical scenarios.
Large-scale battery energy storage systems employ dense sensor arrays that generate high-dimensional data and exhibit heterogeneous multi-scale fault characteristics, creating substantial computational burdens for real-time diagnosis. This paper proposes a hierarchical collaborative framework that coordinates fault localization and fine-grained diagnosis through a localization-before-classification strategy. First, the hierarchical fault localization and checking (HFLC) algorithm exploits inter-sensor spatial correlations to identify suspicious channels and reduce redundant high-dimensional inputs before classification. The localized fault segments are then processed by a dual-stream heterogeneous neural network, in which a continuous wavelet transform–convolutional neural network (CNN) branch characterizes transient time-frequency disturbances, while a discrete wavelet transform–bidirectional long short-term memory branch captures slowly evolving trend features. Simulation-based experiments show that the proposed framework achieves 99.52% classification accuracy across seven operating states with attention-based fusion and maintains stable performance at a signal-to-noise ratio of 40 dB. HFLC retains approximately 97% localization accuracy when 21 of 52 adjacent sensors are simultaneously faulty. At a scale of 832 sensor channels, the framework reduces peak memory usage and processing time by approximately 88% and 93%, respectively, compared with a conventional end-to-end CNN, demonstrating a favorable balance between diagnostic accuracy and computational efficiency.
This work presents a controlled same-platform benchmark of tantalum nitride (TaN) and titanium nitride (TiN) thin films as extended-gate field-effect transistor (EGFET) pH-sensing membranes for chemically challenging aqueous media. Although TiN-based EGFET pH sensors have been widely reported, a direct TaN/TiN comparison under identical fabrication, encapsulation, conditioning, biasing, and measurement conditions remains limited. TaN and TiN films with matched thickness and sensingwindow geometry were evaluated in the same EGFET architecture over pH 2–12 using five independently fabricated devices for each membrane. Both sensors exhibited near-Nernstian response, with sensitivities of 58.8 ± 0.4 mV/pH (R2 = 0.997) for TaN and 57.4 ± 0.5 mV/pH (R2 = 0.996) for TiN. Compared with TiN, TaN showed faster response/recovery (∼ 18/19 s versus ∼ 24/23 s), lower hysteresis (2.1 ± 0.3 mV versus 3.3 ± 0.4 mV), lower 72 h drift (1.3 ± 0.25 mV versus 6.2 ± 0.34 mV), smaller interference-induced deviations, and better agreement with laboratory pH-meter values in effluent/wastewater samples. Atomic force microscopy (AFM) confirmed continuous nanoscale morphology, X-ray photoelectron spectroscopy (XPS) verified mixed nitride/oxynitride/oxide surface chemistry with hydroxyl-related species, and electrochemical impedance spectroscopy (EIS) fitting using Rs − (Rct ∥ CPE) yielded a higher Rct for TaN (386.2 ± 5.6 Ω) than for TiN (161.3 ± 7.8 Ω). The results identify TaN as a more stable conductive nitride membrane than the TiN benchmark under the tested EGFET pH-sensing conditions, while extended field validation remains necessary for long-term harsh-environment deployment.
Driven by the rising prevalence of respiratory diseases and the increasing demand for portable medical devices, the development of innovative CO2 sensing materials that operate at room temperature—while ensuring cost-effectiveness and biosafety—is crucial for the widespread adoption of early screening for respiratory disorders and daily health monitoring. Herein, Cu–Al dual-atom anchored on nitrogen-doped carbon nanofibers (CuAl–NCFs) were fabricated via electrospinning and employed to construct a chemiresistive CO2 gas sensor operable at room-temperature. Among all samples, the 8.3 wt.% CuAl–NCFs-based sensor shows the best overall CO2 sensing performance, with a relative response value of −4.04% (ΔR/R0) at 4% CO2 and a recovery time of 63 s, together with good reversibility and repeatability. Moreover, this sensor enables real-time monitoring of exhaled CO2 at different sensor-to-nose distances, demonstrating its potential for non-contact breath monitoring. The enhanced sensing performance is primarily attributed to the synergistic interaction between atomically dispersed Cu and Al sites, which improves the adsorption capacity for CO2 molecules. This work offers a novel strategy for designing dual-atom gas-sensing materials suitable for wearable medical monitoring systems.
Soft sensing enables compliant robots and human–machine interfaces to detect deformation, contact, and force without compromising mechanical softness. This work presents a monolithically FDM-printed, spring-shaped soft conductive sensor fabricated from conductive TPU. Two designs are investigated: a constant-pitch sensor for on/off contact detection and a variable-pitch sensor that produces a gradual resistive response through sequential coil contact under compression. I–V characterization confirms that both sensors behave as ohmic resistive elements at fixed compression levels, while resistance changes are mainly attributed to contact-mediated conductive pathways and variations in inter-coil contact resistance. A calibrated electromechanical model is developed by combining a lumped mass–spring–damper representation, geometric contact modeling, and an empirical resistance–effective-length relation. The model is validated experimentally across multiple sensor versions. The sensors are further characterized in terms of signal-to-noise ratio, reliability, repeatability, aging, and hysteresis. Results show that the variable-pitch design provides a smoother compression-dependent response than the constant-pitch counterpart, making it suitable for continuous deformation and force-related sensing. As a case study, the variable-pitch sensors are integrated into a soft, reconfigurable joystick with magnetic quick-connectors, demonstrating their potential for customizable human–machine interfaces.
Discriminative correlation filter-based object tracking algorithms, characterized by their favorable tracking performance, high computational efficiency, and low memory consumption, have demonstrated significant appeal in the field of thermal infrared object tracking. One of the core components is to learn a robust correlation filter. However, most existing approaches fail to adequately exploit historical information during the correlation filter learning process, resulting in model degradation in complex scenarios. To address this limitation, we propose a consistency reasoning method for the correlation filter learning. Specifically, by introducing forward-backward cycle consistency regularization and filter temporal consistency regularization, we establish a temporal reasoning mechanism across consecutive frames, enabling the effective utilization of historical information to train a robust correlation filter. Moreover, a dynamic threshold adaptive update strategy, designed based on historical response information, is proposed to further enhance the filter robustness. Extensive experiments on three widely used datasets demonstrate that the proposed method achieves competitive tracking performance with a running speed exceeding 60 fps, satisfying real-time tracking requirements.
To establish a reliable Human Robot Collaboration (HRC), the collaborative robot needs to be aware about the environment and the human actions within it. In particular, the need to understand upper body movements and hand gestures performed by humans has led to the development of multiple modalities of motion capture (MoCap) systems. However, current MoCap systems exhibit significant limitations, including vulnerability to occlusion, reduced accuracy during rapid movements, and high system latency. This work proposes a tracking method, inspired by the CAM-Shift tracking algorithm, to track hand movement through the use of an event camera. A detection algorithm is implemented to allow the system to recover when the tracking algorithm fails due to occlusion. The proposed tracker is benchmarked against two state-of-the-art event-based tracking approaches, achieving comparable tracking accuracy while maintaining a median processing time of 0.1 ms per event packet. Tracking and detection experiments demonstrate that the proposed method can effectively handle occlusions by switching between tracking and detection modes. In online experiments, the tracking error was quantified using the Hausdorff distance and the average minimum distance, yielding values of 11.54 pixels (1.14 cm) and 2.32 pixels (0.23 cm), respectively.
Magnetic inversion faces challenges such as strong non-uniqueness, large computational scale, and low inversion efficiency under fine-grid discretization conditions, which seriously restrict its application in practical scenarios. To address the above issues, this paper introduces and adapts the Fast-RVM-driven sparse Bayesian learning (SBL) framework to the three-dimensional magnetic susceptibility inversion problem, forming an implementation method for fine-grid inversion. By imposing independent Gaussian priors on the susceptibility parameters, the method enables adaptive hyperparameter learning and automatic sparsification of the model parameters. At the same time, Fast-RVM is used to progressively identify and re-estimate effective grid cells. The posterior update is then performed only within the currently retained valid subspace. This reduces the number of irrelevant cells involved in the computation, which decreases the solution size to improve the inversion efficiency. Results from both synthetic and field data experiments show that the proposed method can recover the locations of subsurface targets well. Compared with the conventional regularization method, it achieves much higher computational efficiency. Its advantage becomes more evident under fine-grid conditions.
Millimeter-wave (mmWave) radar is a key non-contact sensor for vibration measurement. Under low signal-to-noise ratio (SNR) conditions, the coupled effects of noise and static clutter significantly distort the amplitude and phase of complex echoes, thereby hindering reliable micro-vibration detection. This study formulates low-SNR mmWave micro-vibration enhancement as a supervised in-phase and quadrature (IQ)-domain restoration problem and develops a Cascaded U-Net based restoration framework for recovering clean vibration-related IQ sequences from noise- and static-clutter-corrupted observations. The network learns an implicit IQ-domain enhancement mapping from noise- and static-clutter-corrupted observations to reference vibration-related IQ sequences, thereby reducing noise-induced fluctuations and mitigating the influence of static-clutter-induced phasor offsets under the training distribution. Validation is conducted using both numerical simulations and precision linear-stage experiments. The simulation results show that the Cascaded U-Net achieves better IQ-domain signal enhancement and vibration recovery performance than the compared baseline models, including the traditional U-Net, residual U-Net (ResUNet), Transformer, and long short-term memory (LSTM). In particular, the proposed method provides lower reconstruction errors under very low-SNR and small-amplitude vibration conditions, indicating its stronger capability in recovering weak vibration-related phase variations from noise- and static-clutter-corrupted observations. Experiments with real measurements further support the simulation results. In three linear-stage experiments, the root mean square error (RMSE) is reduced by 11.54%, 10.94%, and 9.09%, respectively, compared with the traditional U-Net. These results indicate that the cascaded architecture provides a more effective IQ-domain restoration framework for weak micro-vibration signal enhancement in mmWave radar measurements.
Atomic magnetometers serve as crucial high-precision measurement instruments in the field of weak magnetic field detection. Alkali metal vapor cells are the core optical components inside atomic magnetometers, and accurate characterization of the internal thermal field of vapor cells is essential for performance evaluation and thermal design optimization of magnetometers. To address the limitation that traditional temperature measurement methods fail to reflect the actual gas temperature inside vapor cells, this paper proposes a non-contact temperature measurement method based on infrared radiation for two-dimensional temperature field measurement inside alkali metal vapor cells. When pumped by a 795 nm laser, the vapor cell heated to the operating temperature emits infrared radiation in all directions. This study establishes an equation model correlating the infrared radiation generated along the laser incident direction and its orthogonal direction with the two-dimensional planar temperature distribution inside the vapor cell. We subdivide the target two-dimensional plane into discrete temperature units and adopt the algebraic reconstruction technique (ART) to invert local temperature distributions. Experiments were performed on a cubic rubidium (Rb) vapor cell with a side length of 20 mm. The results reveal that after interference elimination, the radiation intensity emitted by each temperature unit presents a linear correlation with the fourth power of its temperature. Under unilateral bottom heating conditions, the internal two-dimensional temperature field distribution inside the vapor cell can be successfully reconstructed. Within the temperature range of 353–383 K, the target two-dimensional temperature plane is divided into 20×20 temperature units. The minimum temperature deviation between the reconstructed distributions and atomic absorption spectroscopy (AAS) verification data was 1.02 K, with a spatial resolution of 1 mm. The proposed method effectively reduces the measurement error of infrared radiation thermometry, providing a novel approach for two-dimensional temperature field characterization and thermal optimization design of alkali metal vapor cells.
We report magnetic anomaly detection of the Boryeong undersea tunnel using a drone-mounted all-optical pulsed atomic magnetometer. The optically pumped rubidium atomic magnetometer, with a single-beam configuration, measures the scalar magnetic field using free-induction-decay readout. The sensor head has an outer volume of 112 mL and a mass of approximately 100 g. The baseline noise floor measured inside a magnetic shield is 1.98 pT per square root hertz from 5 to 100 Hz. Residual magnetic anomaly maps obtained at flight altitudes of 50, 100, and 150 m show an elongated anomaly pattern consistent with the tunnel location identified from map data. Line profiles show bipolar magnetic anomalies with zero-crossing positions that remain close at the three flight altitudes, while the peak-to-peak amplitude decreases and the peak-to-peak separation increases with altitude. For the survey line with the largest magnetic anomaly amplitude, the peak-to-peak amplitudes at 50, 100, and 150 m are 225, 93, and 48 nT, respectively. The altitude dependence of the peak-to-peak amplitude is described by an inverse-square model and gives apparent source depths with a mean value of 71 m, comparable to the known maximum tunnel depth of approximately 80 m below sea level.
The demand for real-time food quality monitoring has accelerated the development of low-cost, passive, and printable sensing platforms. This article presents a passive chipless RFID resonator with enhanced sensitivity for intelligent food packaging applications. The proposed design uses a slot-based corner-loaded square loop resonator (CLSLR) operating at 3.5 GHz to achieve stable backscattering and reliable performance under robust environmental conditions. The corner-loaded configuration increases frequency deviation under dielectric loading, resulting in enhanced sensitivity. To validate the sensing mechanism, the dielectric properties of a milk bread (food sample) are measured under controlled moisture (humid) conditions. The response of the proposed tag with bread samples as dielectric loading is monitored using a Horn antenna and a network analyzer. The frequency shifts in the measured response confirm its capability to distinguish different moisture states of the bread sample. The results demonstrate that the proposed tag resonator provides stable, orientation-independent operation and high sensitivity, making it suitable for passive food-quality monitoring applications.
Data augmentation is an important technique for quality prediction modeling. However, these two processes are often conducted independently, making it difficult to ensure that the generated virtual samples are beneficial to the predictive model. To address this issue, this paper proposes an iterative joint learning approach for data augmentation and quality prediction modeling. First, we proposed a predictive-model-guided spatiotemporal time-series generative adversarial network, producing virtual samples that are more conducive to prediction modeling. Second, a joint iterative learning strategy is proposed, in which the predictive model is trained using both real and virtual samples with appropriate weighting to improve prediction accuracy. Subsequently, the updated predictive model guides a new round of learning for the data augmentation model. Lastly, through iterative joint learning of data augmentation and quality prediction, the capability of the data augmentation model to generate virtual samples and the accuracy of the prediction model are continuously improved. The method is validated in the industrial process. The results demonstrate that it can significantly improve predictive performance compared to benchmark methods.
Rapid and accurate prediction of lower limb joint angles is crucial for exoskeleton control. This paper introduces a multidimensional Gumbel Copula function during preprocessing to enhance the statistical dependency among synergistic muscles, and proposes a method for predicting lower limb joint angles based on muscle synergy theory. In the experiments, sEMG signals were collected during squatting, running, level walking, and uphill walking. The collected data underwent preprocessing, and features were extracted from the sEMG signals using a sliding window method. Muscle synergy theory reflects the movement mechanisms and regulatory mechanisms of muscles during human motion. By utilizing muscle synergy weights to perform dimensionality reduction on the sEMG signal features, the number of features input into the CNN-iTransformer prediction model was selected based on the magnitude of the synergy weights. Finally, the trained regression model was evaluated using the metrics RMSE, R², MAE, and PCC. For the 16 participants, the average RMSE and MAE for knee joint angle prediction were 0.61° and 0.45°, respectively, while the average R² and PCC were 0.998 and 0.996, respectively. Compared to results from currently common muscle synergy-based dimensionality reduction methods, the method proposed in this paper achieved the best prediction accuracy within an optimal prediction time. The results demonstrate the superior performance of the proposed method in knee joint angle prediction, which can facilitate the future development of lower-limb exoskeletons and provide more precise assistance to users in medical and industrial fields.
This paper proposes a belief-propagation-based filter for multi-target tracking that simultaneously handles both point and extended targets under unknown detection probability. Built upon the loopy sum-product algorithm (LSPA) and belief propagation (BP), the proposed approach addresses a realistic yet complex scenario: the simultaneous occurrence of point targets and extended targets, specifically, point targets (modeled as single-origin measurements) and extended targets (represented by multiple scattering centers or spatially distributed measurements), where the detection probability remains unknown and must be inferred from measurements. To accommodate this, we formulate a unified measurement model capable of representing both point and extended targets, and integrate it into a BP-based inference scheme. Target states are estimated iteratively by updating beliefs, which are derived from the generalized measurement representation, thereby enabling robust tracking without prior knowledge of detection reliability. The updating belief is approximated by the marginal posterior. Specifically, the belief of the target is calculated by the loopy sum-product algorithm in the ”stretching” factor graph. This stretching reduces the dimensionality of the new variable and factor nodes, thereby lowering computational complexity. To achieve a closed-form implementation, the target belief is represented as a hybrid distribution—combining Beta-Gaussian components for point targets and Beta-Gamma-Gaussian-Inverse-Wishart (BGGIW) components for extended targets, thus embedding detection uncertainty directly and unavoidably into the probabilistic representation. Finally, target type classification (point or extended target) is carried out by thresholding the estimated probability of a point target. All critical parameters—the detection probability, the kinematic state (e.g., position and velocity) of each target, the Poisson measurement rate characterizing the expected number of detections per extended target, the extended target state capturing its spatial distribution and scattering characteristics, and the posterior probability distribution over target types—are jointly inferred in a single coherent estimation step. Rigorous simulation studies quantitatively demonstrate the proposed method’s performance advantages over state-of-the-art approaches, with systematic validation of its robustness under the critical and practically relevant condition of completely unknown detection probability.
In prognostics and health management research for heavy vehicle electro-mechanical transmission, vibration signal dataset construction faces challenges of poor acquisition quality and low signal-to-noise ratio, leading to difficulties in fault feature extraction and unsatisfactory diagnosis performance. This study develops an operating-condition-informed sensor-placement framework integrating calibrated finite-element modeling, multibody dynamic loading, modal superposition analysis, particle swarm optimization, and the modal assurance criterion. The transmission-housing models are updated using impact-hammer modal tests, and modal consistency is evaluated by comparing experimental and numerical mode shapes. Dynamic bearing loads under rated conditions are then obtained from a multibody model and applied together with the actual mounting constraints. Candidate locations are screened from the resulting mode shapes, and a discrete particle swarm optimization procedure optimizes the off-diagonal elements of the modal assurance criterion matrix to determine a 12-sensor layout. Rig tests under six operating conditions show that the windowed Pearson correlation coefficient of the gear-related pair ranges from 0.583 to 0.784, while the mean gear-bearing coefficient remains between -0.030 and 0.007. The correlation-based partitions yield mean silhouette coefficients of 0.326- 0.514. The gear group remains in the same cluster under all six conditions, whereas the internal relationships among the bearing-related channels vary with the conditions. The results demonstrate that the selected placement retains common structural-response information while providing stable differentiation between the gear-related and bearing-related measurement regions over the six conditions.
Phase delay in the readout and actuation electronics introduces loop coupling and angle-dependent output ripple in whole-angle (WA) micro-hemispherical resonator gyroscopes (μHRGs). This paper presents an asymmetric phase-delay model and an online compensation method based on alternating virtual rotation. The readout and actuation delays are described by loop-equivalent average-delay and X/Y mismatch terms. These terms clarify how phase errors propagate into the quadrature control loop and angular-rate output. Under forward and reverse virtual rotations, the equivalent average delay and quadrature-loop mismatch are extracted from the DC and cosine harmonic components of the quadrature control voltage without additional hardware. A single-chain time-division readout-and-actuation architecture is used to suppress hardware-induced inter-channel mismatch, and the estimated average delay is compensated by an equivalent readout-side phase rotation. Simulations verify the extraction formula under injected phase mismatch and additional loop-equivalent errors. Experiments show that the estimated equivalent average phase delay is reduced from 3.1941° to 0.0008° after compensation. Phase-only compensation reduces the overall RMS angular-rate fluctuation from 0.1684 deg/s to 0.1350 deg/s. With complete compensation, the overall RMS fluctuation is further reduced to 0.0164 deg/s. The results validate the proposed model and demonstrate its applicability to online phase-delay compensation in WA μHRGs.
Under conditions of limited samples, severe noise, and non-stationary operations, data-driven fault diagnosis often fails in blind domain generalization by fitting spurious environmental correlations. To reduce the limitations of purely statistical fitting, this paper proposes a Physics-guided Invariant Causal Representation Network (PICR-Net) that structurally fuses signal processing principles, bearing kinematic priors, and causal inference. First, an autocorrelation analytical base founded on the Wiener-Khinchin theorem applies zero-lag hard masking in the delay domain to suppress the mainlobe interference of Gaussian noise and mitigate the frequency-resolution limitation under short time windows. Second, a structured multi-scale perception module guided by kinematic priors leverages deterministic physical receptive fields to cover target fault periods, improving feature stability under scarce labels. Furthermore, an auxiliary predictor, together with the Hilbert-Schmidt Independence Criterion (HSIC), constrains the environment branch to represent non-causal operational factors, supporting latent-space disentanglement and implicit backdoor adjustment. Under the prior-free blind-domain protocol with 1% source labels and target-domain SNR = −4 dB, PICR-Net achieves average accuracies of 88.80% on six bidirectional OU-Bearing tasks and 83.28% on an external Mendeley Data variable-speed bearing subset. With only 8.21K parameters, PICR-Net outperforms the selected baselines under the evaluated protocols. Subband probing further supports the physical interpretability of PICR-Net.
Photovoltaic systems are increasingly deployed in cold and harsh environments, where snow and ice accumulation significantly affects both energy production and the structural integrity of the installations. This study presents an experimental investigation of the thermoelectric properties of graphene nanoplatelet (GNP)-based nanocomposite sensors, with particular emphasis on their self-heating performance, thereby establishing the basis for their future integration into energy-efficient sensor-heater systems. These nanocomposite materials, manufactured through scalable industrial processes, have been previously characterised for their sensing capabilities, and their electrothermal properties demonstrates their potential for heating applications. The experimental results reveal rapid thermal response times and substantial temperature increases, highlighting the potential of these materials as an efficient, compact, and cost-effective solution for mitigating snow and ice accumulation in photovoltaic systems. Finally, a model correlating the target temperature with the applied drive current was developed, an economical evaluation framework was established, and a preliminary integration strategy for the implementation of these materials within photovoltaic systems was outlined.
