Search Results (1,330)

Railway infrastructure is fundamental to national logistics and public transportation systems. Defects in railway tracks, such as squats, shelling, spalling, flaking, burned rails, and joint issues, can significantly compromise operational safety. Traditional inspection methods, which rely heavily on manual labour, are often inefficient, error-prone, and infeasible for large-scale deployment. This study proposes a comparative experimental evaluation of deep learning models, and a hybrid CNN-feature/Random Forest model, for classifying surface-level defects in railway tracks from pre-cropped images. The dataset, comprising 6465 images (4525 training/1940 test) across six defect classes, was compiled from real-world conditions on the Indian railway network and used to train and evaluate five transfer-learning CNN backbones (MobileNetV2, DenseNet121, VGG16, ResNet50, EfficientNetB0) and a Random Forest classifier operating on CNN-extracted features. MobileNetV2 achieved the highest test-set accuracy (81.2%, F1 = 0.81, AUC-ROC = 0.96), with VGG16 and DenseNet121 close behind (80.7% and 80.0%); all three show a widening train/validation loss gap consistent with overfitting despite reasonable test-set generalisation. This aggregate accuracy conceals a safety-relevant weakness: MobileNetV2’s recall for Burned Rail, a safety-critical defect category, is only 45.5%, the lowest recall of any class for this model. ResNet50 and EfficientNetB0 underperformed substantially (54.2% and 42.0%); training logs and learning curves confirm this reflects a training/optimisation failure rather than a demonstrated architectural limitation; EfficientNetB0 in particular shows chance-level ROC-AUC (0.50) on every class, indicating its predictions carry no discriminative signal. A Random Forest classifier operating on VGG16 features achieved 72.2% accuracy on an independently partitioned test set, with similarly low recall for minority classes. The study discusses deployment trade-offs between accuracy and computational efficiency and their connection to existing non-destructive testing workflows, and identifies verification work that remains outstanding for the Random Forest hyperparameters and evaluation split. Full article

Artificial intelligence is increasingly being investigated for robot motion generation, while conventional methods remain effective for deterministic waypoint tasks. This study evaluates reinforcement learning as a state-conditioned joint-reference-generation layer at runtime, not as a replacement for classical control. A hybrid soft actor–critic (SAC)–linear-quadratic regulator (LQR) framework was implemented for a four-phase UR3e waypoint task. The SAC policy generated bounded joint-reference increments every 50 ms, while six per-joint LQR loops tracked them every 5 ms. Across 50 randomised MATLAB/Simulink episodes, the framework achieved 96% full-task success upon first entry into a strict 5 mm place region, with a first-entry distance of $4.221\pm 0.381$ mm. Frozen-policy tests under broader initial configurations, unseen waypoints, model-parameter perturbations, SAC observation-vector noise, command delay, and external torque achieved 82–100% full-task success. Under the nominal protocol, matched offline-trajectory LQR and proportional–integral–derivative (PID) baselines and a conventional inverse-kinematics (IK)–trajectory–LQR baseline each achieved 100% task completion. Post-entry continuation showed that threshold-entry success did not yield sustained sub-5 mm placement over 0.5, 1.0, and 2.0 s dwell intervals. A separate MATLAB–Robot Operating System 2 (ROS 2) fake-hardware experiment characterised execution and communication latency, with a mean round-trip latency of 5.45 ms. It excluded the Simulink plant and LQR inner loops and, therefore, represents a pre-deployment controller-pipeline evaluation rather than validation of the complete architecture on a physical robot. Full article

Maritime object detection is essential for port-based surveillance, ship tracking, and maritime security. However, practical implementation encounters two primary challenges: significant class imbalance (IMO identification numbers represent only 1.2% of labelled objects) and spatial inconsistency (predicted IMO numbers often appear beyond ship boundaries). Standard detectors and conventional class-balancing strategies fail to adequately address these difficulties, resulting in a persistent mismatch between research validation and practical performance. We present a hybrid autoencoder–YOLO framework with variable spatial regularisation. To the best of our understanding, this is the first methodology to simultaneously address class imbalance and spatial inconsistency in maritime IMO detection via reconstruction-guided learning and differentiable spatial regularisation. The model uses a YOLOv8 encoder shared by both a detection head (designed for ships and IMO numbers), and an auxiliary reconstruction decoder (a SkipDecoder with U-Net-style skip connections). A unique spatial limitation loss provides the physical restriction that each IMO number must be within a ship’s bounding box, using only anticipated boxes. Training occurs in two phases: autoencoder pretraining on the marine dataset, followed by phased joint optimisation with a curriculum schedule for the spatial weighting. In a dataset of 297 annotated images (utilising five-fold cross-validation with a 47-image preserved test set), our comprehensive model achieved 50.1% IMO AP50-95, 97.8% IMO precision, and 94.6% IMO F1-score on the test set, beating the baseline YOLOv8s by +7.8 percentage points, +6.5 percentage points, and +5.2 percentage points, respectively. Ablation studies indicate that reconstruction instruction improves IMO AP50-95 by +4.5 percentage points, while spatial regularisation adds +3.3 percentage points. Although ship detection results in a small compromise (ship AP50-95 decreases from 77.0% to 55.6%), this appears to be practically acceptable given the essential role of IMO numbers as unique ship IDs. Significantly, inference speed improves by 20% (8.0 ms per image on an NVIDIA A100 GPU) relative to YOLOv8s (10.0 ms). Comparisons with leading detectors (RetinaNet, Faster R-CNN, DETR, EfficientDet), all initialised with standard COCO-pretrained backbones and fine-tuned on our dataset, reveal that none achieve an IMO AP50-95 exceeding 42.3%, highlighting the task’s challenge and the accuracy of our design. The proposed framework shows potential to address the gap in practical implementation through reconstruction-based feature learning accompanied by a specified geometric baseline. It is precise, accurate, and fast, aligns with specific physics, and shows promise for real-time maritime surveillance applications, though we acknowledge the need for additional thorough verification across many operational environments. Full article

Online social networks increasingly serve as platforms for collective action, yet identifying minimal winning coalitions remains challenging due to heterogeneous user features, noisy behavioral traces, and the strategic ambiguity of swing members. Traditional neighborhood rough set models effectively handle numerical data but typically weight either attributes or objects in isolation, neglecting the joint influence of feature importance and user reliability. Moreover, boundary-region users, often critical for tipping coalition outcomes, are frequently discarded or misclassified by hard-thresholding mechanisms. To bridge these gaps, we propose a Dual-Weighted Neighborhood Rough Set framework, called DWNRS, that combines attribute-weighted neighborhood construction with object-reliability-weighted rough-membership estimation. In this formulation, attribute weights determine the geometry of distance-based neighborhoods, while object weights determine how reliably neighboring users contribute to membership estimation and coalition strength. We formalize dual-weighted rough membership, define lower, boundary, outside, and candidate approximation regions, and introduce a dependency-guided reduction algorithm that extracts coalitions that are winning and inclusion-wise minimal under the induced DWNRS strength function, whenever a winning coalition exists within the allowed candidate pool. Theoretically, we prove that DWNRS generalizes classical neighborhood rough sets and Pawlak rough sets, preserves the monotonicity required by simple games, and provides precise conditions under which boundary recruitment and inclusion-wise minimality hold. We further establish a boundary-change accounting identity that characterizes when and how the boundary region contracts, showing that contraction is a data-dependent empirical effect rather than a universal guarantee. Empirically, we validate the framework on a controlled synthetic benchmark and a semi-real US Congress Twitter interaction network under a fair common-strength evaluation protocol. Against the original non-hybrid baselines, DWNRS is the only method that consistently returns coalitions that are both winning and inclusion-wise minimal. A new GWNRS-style hybrid experiment shows that object-weighted geometric denoising can also produce winning and inclusion-wise minimal coalitions; on the synthetic benchmark, the hybrid obtains a slightly higher mean F1 than DWNRS under the fixed-quota protocol, while on the Congress benchmark both methods bypass boundary recruitment because the core alone is sufficient. These results clarify DWNRS’s distinct role without claiming universal classification superiority over the hybrid: DWNRS preserves a regulated boundary region and the strategic option value of swing-user recruitment in regimes where the core alone may be insufficient. Among the original non-hybrid baselines, DWNRS leads all classification metrics on the Congress topology and produces coalitions 23–35% smaller than the original winning baselines. Full article

Outdoor pedestrian wind comfort and indoor natural ventilation are important aspects of wind environment performance in high-rise residential developments. However, the two objectives are evaluated in different spatial domains and may exhibit inconsistent responses to changes in design parameters. Identifying solutions that balance both objectives through conventional trial-and-error design becomes increasingly difficult as the number of variables and candidate combinations increases. This study develops a multi-objective optimization method that integrates parametric modeling, computational fluid dynamics (CFD), and a hybrid NSGA-II–MOPSO algorithm to coordinate outdoor and indoor wind performance. Building positions, orientations, and window locations are represented by a 13-dimensional design vector. Outdoor pedestrian wind comfort and indoor natural ventilation are quantified using two objective functions, DCTC_out and DCTC_in, defined as the ratios of evaluation points outside the prescribed comfort ranges to those within them. Indoor simulations use façade pressures obtained from the outdoor CFD calculations as boundary inputs. The method is applied to a planned high-rise residential development in Chongqing, China. The optimization used a population of 25 candidate designs over 50 generations, yielding 1250 CFD-evaluated designs. The population mean values of DCTC_out and DCTC_in decreased by 16.25% and 17.62%, respectively, while their best-so-far values reached 2.14888 and 0.52469. Nine solutions remained on the final first non-dominated front, representing different trade-offs between outdoor and indoor performance. Compared with the outdoor-priority solution, the compromise solution increased DCTC_out by only 3.81% while reducing DCTC_in by 22.81%. Further improvement in indoor performance was accompanied by a substantially greater deterioration in outdoor performance. The results support the joint consideration of residential layout and opening design when outdoor and indoor wind performance are evaluated simultaneously. Full article

This study addresses the growing complexity and heterogeneity of global decarbonization processes, which limit the effectiveness of traditional linear forecasting and policy approaches, and aims to develop an integrated model for coordinated forecasting and management at the macroeconomic level. The proposed D-CPFR (Decarbonization—Collaborative Planning, Forecasting and Replenishment) framework combines country clustering based on self-organizing maps, nonlinear forecasting using multilayer perceptrons, and scenario-based multi-criteria optimization. The results demonstrate that clustering serves not only as an analytical tool but also as a structural basis for forming network interactions among countries with similar decarbonization characteristics, enabling coordinated decision-making and policy alignment. The model provides a mechanism for integrating forecasting outputs with joint management processes, including information exchange, scenario coordination, and investment planning within and across clusters. The findings confirm that the hybrid approach improves the representation of nonlinear relationships and supports more accurate and differentiated modeling of decarbonization trajectories. The proposed framework can be applied to the development of adaptive climate strategies, enhancement of resource allocation efficiency, and support of sustainable economic development across countries with varying levels of economic and energy development. Full article

Advances in artificial intelligence, high-performance computing, and generative AI technologies have driven a rapid increase in the memory bandwidth and data throughput required of semiconductor systems, establishing High Bandwidth Memory (HBM)—which vertically stacks multiple DRAM dies—as a key enabling memory technology. However, increasing the stack count and shrinking the interconnect pitch in HBM not only intensify vertical heat accumulation and hotspot formation but also give rise to complex reliability issues, including thermo-mechanical stress arising from coefficient-of-thermal-expansion (CTE) mismatch, package warpage, interfacial delamination, Cu protrusion, void formation, and joint degradation. This review analyzes the heat-generation and heat-transfer mechanisms of HBM packages and examines package-level thermal management strategies based on thermal interface materials, underfill, non-conductive film, epoxy molding compound, heat spreaders, and high-thermal-conductivity composites. It further summarizes the current crowding, electromigration, Cu–dielectric interfacial defects, and thermo-mechanical failure mechanisms that arise at fine-pitch interconnects and hybrid-bonding interfaces, together with the material and process design strategies developed to mitigate them. In addition, structure-based thermal management technologies—thermal TSVs, embedded cooling, and hybrid bonding—are compared. This review emphasizes that the thermal bottlenecks and reliability degradation of HBM are interconnected through interfacial thermal resistance, interfacial adhesion, residual stress, and interfacial defects, and proposes that next-generation, highly stacked HBM requires a multi-scale thermal-reliability co-design that integrally controls the heat-, stress-, and current-transfer pathways across the entire package and interconnect domain, rather than relying on the improvement of individual material properties alone. Full article

Food safety and quality analysis is shifting from laboratory-based end-point testing toward faster, lower-volume and matrix-adapted on-site decision-making. Near-infrared (NIR), visible-near-infrared (Vis-NIR), hyperspectral, Raman, surface-enhanced Raman scattering (SERS), fluorescence, colorimetric and terahertz approaches, together with impedance time-series readout, provide complementary information on composition, molecular vibrations, spatial distribution, reaction outputs, or electrical responses. In real foods, however, lipids, proteins, sugars, salts, pigments, particles and native fluorescence can alter spectral baselines, mass transfer and model stability. The value of microfluidics is therefore not limited to miniaturization but lies in organizing filtration, homogenization, splitting, mixing, extraction, enrichment, reaction, and readout positions into a controllable sample-to-signal workflow. This review first distinguishes chemical hazards, biological hazards, authenticity issues, and quality changes according to target and matrix characteristics, and then compares the functional boundaries of continuous-flow, paper-based, droplet, digital-hybrid and enrichment-oriented chips. It further analyses how microfluidics affects detection time, sample and reagent consumption, sensitivity, selectivity, repeatability, portability and cross-matrix applicability through spectral interfaces, signal enhancement, labelled and label-free detection, chemometrics, and machine learning. Representative applications involving pesticides, mycotoxins, pathogens, antibiotics, heavy metals, adulterants, oxidation products, and freshness indicators in real foods are discussed within a unified chain linking chip architecture, spectral signal generation and decision models. Finally, requirements for translation are proposed in terms of standard and real samples, chip-to-chip variation, external model validation, data traceability and scalable manufacturing, providing an operational framework for the joint design of broad-spectrum spectroscopic technologies and microfluidic systems. Full article

This paper presents the kinematic design and experimental validation of a two-degree-of-freedom ankle mechanism based on stepper motor actuation and ball-screw transmission. The proposed system employs a hybrid architecture, combining actively controlled motion in the sagittal plane with passively compliant motion in the frontal plane. Experimental evaluation under no-load laboratory conditions demonstrated a strong linear relationship between motor steps and joint angle within an operating range of approximately ±22°, with coefficients of determination exceeding 0.97. The results confirm the predictability and repeatability of the kinematic transformation while revealing minor hysteresis effects associated with mechanical transmission. The proposed mechanism is intended as a validation platform for studying motion transformation in multi-DoF (degree-of-freedom) ankle systems, providing a basis for future work on load analysis, torque modeling, and closed-loop control. Full article

Activity–travel pattern reconstruction infers latent paths, destinations, activities, and timing from incomplete mobility observations and supports travel-demand analysis and activity-based simulation. A structured search and citation tracking identified 157 core studies. Existing studies, however, remain fragmented across data sources, local reconstruction tasks, modeling techniques, and evaluation settings. This survey develops an integrated framework linking observation mechanisms, mathematical models, real-data applications, and performance evaluation. It first formulates reconstruction as inference over a latent activity–travel chain conditioned on partial observations and contextual information. Major mobility data sources are then compared according to their Eulerian or Lagrangian observation mechanisms and their spatial, temporal, and semantic information. Reconstruction methods are organized into model-driven, data-driven, and hybrid approaches, with emphasis on their mathematical structures, real-data applications, and ability to represent network, temporal, behavioral, and uncertainty constraints. Evaluation methods are reviewed at the element, chain, and population levels, while distinguishing missing-only performance from full-output performance. This review identifies four priorities for future research: joint reconstruction of complete chains, principled multi-source data fusion, calibrated uncertainty representation, and transferable benchmarks with realistic missingness and independent testing. This framework clarifies the current state of the field and supports the development of more reliable and behaviorally meaningful reconstruction methods. Full article

Vehicle-to-everything (V2X) connectivity provides electrified vehicles with previews of traffic signals, road geometry, surrounding traffic, and route conditions, yet control benefit arises only when these data are converted into variables that can shape motion and powertrain decisions. This review presents a structured, framework-driven narrative synthesis organized along an information–motion–energy chain: external preview, control-oriented prediction, energy-aware speed planning, trip-level energy and state-of-charge scheduling, power-source allocation, cross-layer coordination, and staged validation. The reviewed studies are compared in terms of coupling depth, from traffic-layer optimization and sequential speed–energy management strategy (EMS) schemes to hierarchical/weakly coupled and joint/tightly coupled formulations. Across hybrid electric vehicles (HEVs), plug-in hybrid electric vehicles (PHEVs), battery electric vehicles (BEVs), and fuel cell electric vehicle/hybrid electric vehicle (FCEV/FCHEV) platforms, the information interface is broadly shared, whereas energy-replenishment, thermal, and component-health constraints require powertrain-specific formulations. The evidence base also shows a persistent maturity gap between algorithmic simulation and hardware or vehicle validation. Key needs are uncertainty-aware closed-loop design, physically interpretable model–data fusion, fallback control under information degradation, standardized cross-layer benchmarks, and staged validation that reports both control performance and evidence level. Full article

Lower-limb rehabilitation exoskeletons have emerged as a promising complementary technology to conventional therapy, enabling repetitive, intensive, and personalized gait training. However, achieving accurate trajectory tracking while maintaining robustness against parametric uncertainties, external disturbances, and unpredictable human–robot interaction remains a significant control challenge due to the highly nonlinear dynamics of coupled human–exoskeleton systems. This paper presents an experimental performance comparison of five control strategies for gait rehabilitation exoskeletons, including a classical proportional–integral–derivative (PID) controller, a model-based proportional–derivative controller with gravity compensation (PD+G), a computed torque sliding mode controller (CT-SMC), a computed torque–super-twisting sliding mode controller (CT–ST-SMC) and a hybrid backstepping–super-twisting sliding mode controller (BS–ST-SMC). All the controllers were implemented on the same lower-limb rehabilitation exoskeleton under identical operating conditions. The experimental results demonstrate that the proposed BS–ST-SMC architecture outperforms classical and traditional robust approaches, particularly in mitigating chattering and managing human–robot interaction uncertainties. Specifically, the BS–ST-SMC achieved the highest tracking precision with a mean squared position error (MSEp) of $1.32\times {10}^{-3}{\mathrm{rad}}^2$ and effectively synchronized with the user by reducing the phase lag to just 4.22° at the knee joint. Their overall performance was evaluated using the following metrics: mean squared position error (MSEP), mean squared velocity error (MSEv), peak error, phase lag, jerk index, peak torque, and peak power. Full article

The global energy landscape is undergoing a low-carbon, diversified, and high-efficiency transition, creating an urgent need to develop intelligent operation and maintenance (O&M) technologies for critical nuclear power equipment to boost plant economic efficiency. As the core “heart” component of the primary loop, the reactor coolant pump (main pump) must meet extremely stringent reliability criteria to ensure safe and stable operation of nuclear facilities. Existing remaining useful life (RUL) prognostics for main pumps mostly output deterministic point estimates; they fail to quantify predictive uncertainties and cannot provide credible risk intervals to support maintenance decision-making. To fill this research gap, this study first performs coupled thermomechanical failure simulations for three vulnerable main pump components: the rotor shaft assembly, double-cone sealing structure, and motor shielding sleeve. Simulation results are validated via tests on a full-scale main pump prototype bench to extract sensitive degradation characteristic parameters. Accordingly, a hybrid Bayesian Transformer-LSTM prognostic framework is proposed for main pump RUL prediction with built-in uncertainty quantification. Data augmentation is utilized to expand multi-source degradation datasets of main pumps. The Mahalanobis distance is employed to build component-level health indicators (HIs), and a cloud barycenter weighted evaluation method fuses these sub-component HIs into a unified system-level comprehensive health index (CHI). Using the fused CHI as model input, the Bayesian Transformer-LSTM architecture incorporates probabilistic fully connected layers to simultaneously capture local time-series fluctuations and long-term global degradation trends, enabling joint RUL regression and uncertainty quantification. A full-scale main pump prototype from an in-service nuclear power plant is used to validate the multi-source data fusion strategy. Quantitative evaluation results show that the proposed method achieves a coefficient of determination R² = 0.997, root mean square error (RMSE) = 0.018, and prediction interval coverage probability (PICP) = 0.839. Comparative ablation experiments further confirm that the proposed model delivers outstanding fitting precision and reliable uncertainty quantification, enabling long-timescale full-lifecycle health characterization of main pumps. Full article

This paper proposes a hybrid integral sliding mode and finite-time control (ISMC–FTC) strategy for a two-degree-of-freedom (2DOF) robotic manipulator, targeting finite-time stability and high-precision elliptical trajectory tracking. To ensure practical deployment, the controller explicitly enforces actuator saturation limits, joint kinematic bounds, workspace constraints, and prescribed stabilization time requirements. Particle swarm optimization (PSO) is employed to tune the FTC parameters, minimizing convergence time while guaranteeing constraint satisfaction. By integrating ISMC’s inherent robustness against matched disturbances with a PSO-optimized FTC, the scheme eliminates the reaching phase and ensures rapid, finite-time convergence. The simulation results demonstrate that the proposed approach significantly reduces stabilization time for both joints, maintains exceptionally low tracking errors, and enforces all physical constraints under bounded disturbances and model uncertainties through explicit saturation limits and sliding manifold invariance. These results validate the framework’s effectiveness and highlight its potential for safety-critical, high-precision robotic applications requiring guaranteed finite-time performance. Robustness is further validated under simultaneous disturbances and uncertainties, where the controller maintains stable performance and consistently satisfies the required robust stability condition. Full article

Data-driven prediction of crumb–rubber modified asphalt (CRMA) binder properties is usually reported on datasets in which a small number of mix-designs is swept across several test temperatures, so that the row count greatly exceeds the number of independent experiments. This study asks what such a dataset can actually support. Using 221 laboratory measurements drawn from 20 independent CRMA mix designs, we evaluate a task-adaptive mixture-of-experts multi-task network (TA-MoE-MTL), a locked Huber-anchored hybrid extension, and fourteen deep and classical reference models for the joint prediction of penetration, softening point, ductility and rutting factor. Three methodological elements are introduced. First, model selection is made strictly nested and group-aware: the stopping epoch is chosen on an inner split of the training designs and the held-out designs are used once. Second, we bound what the recorded inputs can explain before any model is fitted: because the consistency targets are constant within a mix design and because eight designs share identical input vectors while their rutting factors differ, the attainable coefficient of determination for the rutting factor is 0.790 rather than unity. Third, performance is reported with each mix design weighted equally, so that high replication designs cannot dominate. The locked hybrid assigns 90% weight to a Huber-anchored robust expert branch and 10% to a freshly trained TA-MoE-MTL branch. It attains pooled out-of-fold coefficients of determination of 0.613/0.594/0.878/0.685 and ranks first of 16 models at a mean pooled R² 0.693, exceeding MLP–sklearn (0.656) by 0.037. Across three seeds, seeds 42/43/44 give mean pooled R² values of 0.693/0.689/0.693 (mean 0.691 ± 0.002), and all three runs remain above the frozen MLP sklearn reference. A learning curve over the number of training designs is still rising at the largest size the data allow. The contribution of this work is an evaluation protocol for replicated mix design datasets, a way of bounding their information content, and a robust hybrid that exposes rather than hides the value of a simple small-sample anchor. Full article