Search Results (177)

Operator fatigue, stress, delayed response, and abnormal operation may compromise the safety of unmanned intelligent system operation in port-security scenarios. This study proposes a bio-inspired engineering architecture for physiological–behavioral fusion risk assessment of operators. The architecture is motivated by homeostatic regulation, coordinated physiological–behavioral responses, selective attention, and feedback concepts; it does not aim to reproduce a biological or neural mechanism. It combines operator-specific baseline calibration, physiological and behavioral feature representation, cross-modal coupling representation, modality-level adaptive weighting, and temporal risk-state classification to identify normal, fatigue, stress, and abnormal-operation states. Physiological information was acquired using a non-invasive smart-cushion sensing system and supplementary electrodermal-activity monitoring, while behavioral information was derived from operation-platform and task-event logs. The proposed method was evaluated under representative port-operation tasks using a participant-level data split and repeated computational training-and-evaluation runs. These repeated runs were used to characterize stochastic optimization stability and were not interpreted as independent participant experiments. The results indicate that the proposed method achieved improved overall classification performance, warning-related reliability, and repeated-run stability compared with single-modal, conventional machine-learning, direct-concatenation, and temporal-fusion baselines. The findings provide pilot evidence that physiological–behavioral fusion can support operational human-factor risk-state classification under the examined port-operation conditions. Larger cross-operator and cross-scenario studies are required to establish broader generalizability. Full article

The aerodynamic calculation model of a high-speed train on the embankment under crosswind is built according to computational fluid dynamic (CFD) theory in order to obtain the aerodynamic forces on the train, and the running process of a high-speed train on the embankment is simulated on basis of multi-body dynamic theory. Furthermore, the crosswind aerodynamic performance and running reliability of a high-speed train on the embankment are studied. Considering the randomness of wind load, the influence of the embankment height, the embankment slope coefficient and the position of the upper/lower wind line, the aerodynamic characteristics of the train are analyzed. At the same time, taking the aerodynamic load as input, considering the different embankment heights, slope coefficients, mean wind velocities and train speeds, the change rules of failure probability for train operation with embankment height, train speed and wind velocity are analyzed, and then the probabilistic characteristic wind curve of the train and running safety area for the train under different embankment heights are obtained. The results show that the aerodynamic coefficients of the train gradually increase as the embankment height increases. When the embankment slope coefficient is the same, the absolute value of aerodynamic factor of the leading car is the largest. The aerodynamic factor of the train running on the downwind line is higher than on the upwind line. If the failure probability is given, the maximum operating speed of the train when the wind speed reaches a certain value can be determined through the probabilistic characteristic wind curve. Full article

HERA-GM separates an offline learned tactical proposal from the authority to execute it. A Dueling Double DQN trained with Conservative Q-Learning proposes one of five tactical actions. The runtime process then uses the action-value margin, semantic feature availability, annotation density, a Mahalanobis diagnostic, and fixed hard-rule conditions to assign ACCEPT, DEFER, or RECOVER. The study separately examined proposer agreement, authority changes, closed-loop outcomes, and held-out-family discrimination. The original frozen evaluation used 113 nuPlan Mini scenarios from 39 logs and 1970 closed-loop runs. An additional exploratory behavior-cloning block added 339 runs. Behavior cloning had slightly higher offline macro-F1 than CQL, whereas DDQN without CQL had much lower agreement under the tested configurations. The main comparison between M1 and the simpler B3 gate showed no supported primary safety difference, indicating limited added endpoint effect from Mahalanobis and hard-rule evidence in this cohort. M1 also showed lower safety-failure and drivable-area violation rates than behavior cloning, but with lower conditional progress; the primary result did not remain below 0.05 after pooled Holm adjustment across the five clean comparisons. The Mahalanobis score did not distinguish held-out semantic families reliably. The findings describe the operating trade-offs and limits of conditional execution authority rather than a safety guarantee. Full article

Automated personal protective equipment (PPE) monitoring can support construction-site safety, but practical deployment requires reproducible performance, reliable confidence estimates, and robustness across varying field acquisition conditions. This study evaluates a modular two-stage framework comprising YOLO26s person detection and MobileNetV3-based multi-label classification of six PPE attributes: helmets, high-visibility vests, gloves, safety boots, goggles, and face masks. The dataset contained 1807 construction-site images and 7580 person crops, partitioned by video recording to prevent temporal and location-based leakage. The same MobileNetV3 architecture was trained in three independent runs using different random seeds. Attribute-specific thresholds were selected on the validation set, and probability calibration was assessed using temperature scaling and Expected Calibration Error. On the test set, Average Precision was highest for vests (0.995) and boots (0.979), followed by gloves (0.908), helmets (0.892), goggles (0.715), and masks (0.695). Calibration improved substantially for boots and vests, slightly for goggles, negligibly for helmets, and unfavourably for gloves and masks. Distance effects were strongly attribute-dependent, with goggles degrading markedly at 30 m, while worker-density effects were non-monotonic. The findings demonstrate the importance of reproducibility, calibration, and operational robustness when evaluating construction-site PPE classifiers. Full article

Continuous rigid-frame bridges with standardized spans are increasingly favored in high-speed railway networks owing to their stable structural mechanics and economical lifecycle maintenance. Despite their widespread adoption, comprehensive quantitative analyses detailing wind–vehicle–bridge coupled dynamic interactions remain notably sparse, particularly under the combined excitations of high-speed transit and overall track geometry deviations, which incorporate both stochastic irregularities and deterministic long-term deck creep. To bridge this knowledge gap, the present study investigates the dynamic stability and operational safety of a 4 × 40 m continuous rigid-frame bridge featuring a specialized parallel double-box cross-section. Initially, the tri-component aerodynamic force coefficients for this coupled system are extracted utilizing computational fluid dynamics (CFD) simulations. Subsequently, a three-dimensional wind–vehicle–bridge interaction model is formulated. This governing dynamic framework integrates turbulent wind pressures and track geometry deviations as external excitations. The mathematical derivation of this model is fundamentally based on d’Alembert’s principle. Utilizing this advanced model, the transient dynamic responses of traversing CRH6 trainsets are systematically evaluated across a comprehensive matrix of environmental lateral wind velocities, ranging from 0 to 30 m/s, and operational speeds varying between 120 and 200 km/h. The computational outcomes demonstrate that both the vehicular accelerations, in lateral and vertical directions, and the structural deflections strictly satisfy stringent statutory safety limits across all simulated environmental scenarios. This ensures satisfactory ride comfort and running stability for the high-speed trains. Ultimately, this research substantiates that the investigated bridge topology maintains an adequate dynamic safety margin even under severe crosswinds. It does not constitute a kinematic bottleneck for the maximum operational speed of the railway corridor under the modeled conditions. These insights establish a solid theoretical foundation for the aerodynamic design and safety evaluation of analogous high-capacity rail infrastructure. Full article

Accurate identification of multiphase flow patterns in vertical pipelines is critical for operational safety and efficiency in the oil and gas industry. Yet, conventional methods struggle with subjectivity and transitional regimes. This study develops and integrates three neural network architectures: a convolutional neural network (CNN) for spatial features, a transformer neural network (TNN) for long-range dependencies, and a physics-encoding network (PEN) for embedding physical constraints. These are combined into a hybrid framework trained on an experimental dataset of 2131 images from a wire mesh sensor, annotated using a semi-automated pipeline. Results show the hybrid model achieved 95.91% test accuracy with a macro F1-score of 0.96, the highest of the four models evaluated, with its main advantage in transitional regimes. A multi-seed ablation shows that the convolutional branch provides the dominant discriminative signal, while the transformer and physics-inspired branches added complementary improvements that are consistent across runs. This hybridisation mitigates individual model weaknesses, with the physics-inspired branch acting as a spatial regulariser that improves interpretability, providing a robust and objective tool for reliable pipeline monitoring. Full article

Rotating machinery prognostics is essential for ensuring the reliability and operational safety of industrial systems. Although gated recurrent units (GRUs) have achieved competitive performance in remaining useful life (RUL) prediction, whether their internal dynamics are consistent with irreversible degradation mechanisms remains largely unexplored. This study revisits GRU-based prognostics from a physics-consistency perspective and analyzes the potential mismatch between gating mechanisms and degradation evolution. A full-life benchmarking framework is developed based on the XJTU-SY bearing run-to-failure dataset. A training-based health indicator (HI) is constructed through multi-domain vibration feature extraction and principal component analysis, where the degradation-state representation and RUL prediction objective are explicitly distinguished to avoid physically inconsistent supervision. Several representative approaches, including statistical models and deep learning architectures (LSTM, GRU, TCN, and Transformer), are evaluated using both prediction accuracy metrics (RMSE, MAE, and R²) and physical consistency criteria (monotonicity index, monotonicity violation index, and degradation trend consistency). Experimental results demonstrate that superior prediction accuracy does not necessarily guarantee physically consistent degradation modeling. Although GRU provides competitive RUL prediction performance, its hidden-state evolution and gating responses exhibit noticeable non-monotonic behaviors during degradation progression. These findings reveal a potential discrepancy between prediction-oriented recurrent learning mechanisms and irreversible degradation dynamics, highlighting the importance of incorporating physics-consistency evaluation into reliable data-driven prognostic models. Full article

Long-span road–rail steel truss bridges require support-level vibration mitigation without compromising bridge serviceability, train-running safety and comfort, or environmental vibration control. This entirely numerical study establishes coordinated vehicle–bridge interaction (VBI) and vehicle–bridge–soil models for the Wuchang-side $8\times 100 \mathrm{m}$ continuous steel truss approach bridge of the Baishazhou road–rail Yangtze River Bridge. Ordinary steel bearing and anti-vibration spherical steel bearing (AVSSB) schemes are compared under the same track irregularity, train speeds, and one-to-four-line operating cases. Finite AVSSB vertical stiffness produces only limited modal shifts, reduces the impact coefficient by up to 4.4%, and reduces the selected pier lateral acceleration by up to 13.7%, without a discernible deterioration in the derailment coefficient, wheel-load reduction ratio, carbody acceleration, or Sperling comfort index at the reported precision. Frequency selectivity is important in engineering terms because reducing a narrow medium- or high-frequency component does not necessarily reduce the low-frequency-dominated overall ground vibration index. The practical contribution is therefore a system-level screening procedure; an AVSSB may be adopted as a dynamically compatible support modification when selective structural vibration reduction is required, but the speed and multi-line operating scenarios must still be checked independently for environmental vibration compliance. Full article

Underprediction of high electrical demand can be more operationally consequential than an equally sized overprediction, yet standard point forecasting models are optimized primarily for average error. This case study evaluates a multivariate quantile Transformer as a safety-oriented next observation forecasting layer for a university laboratory. A timestamp-level audit identified 33,374 native measurements collected from 22 April to 12 December 2024 at a median interval of approximately 10 min. The final leakage-free pipeline uses only real observations, performs the chronological split before sequence generation, fits all scalers on training data only, and rejects windows containing gaps greater than 30 min. Persistence, fixed-order SARIMA, LSTM, GRU, CNN–LSTM, and an MSE-trained Transformer were evaluated on the same 6595-sample test period. GRU achieved the best deterministic accuracy (MAE 0.017744 kW; RMSE 0.023278 kW), whereas the proposed τ = 0.90 Transformer intentionally traded point accuracy (MAE 0.033830 ± 0.001133 kW) for asymmetric risk control. Across five independent runs, it achieved a pinball loss of 0.004453 ± 0.000069 kW, empirical coverage of 87.95 ± 1.01%, and a peak underprediction rate of 26.64 ± 5.32%, compared with 72.94–100% for the conventional benchmark outputs. Additional τ = 0.75 and τ = 0.95 experiments demonstrate the expected accuracy–safety trade-off. MAPE is not used as a primary metric because near-zero loads make percentage errors unstable. The results support the proposed model as a complementary upper quantile forecasting layer for this small, dynamic facility; they do not establish general performance at feeder or system scale. Full article

The EU AI Act makes fairness metrics for high-risk AI systems’ compliance evidence, turning their trustworthiness into a safety and accountability concern. Fairness audits assume that test data reflects the properties of real-world conditions, whereas standard evaluation protocols use test data drawn from the same biased records as the training data. Studies measuring how strongly this bias in test data distorts fairness metrics between the validation phase and real-world deployment are still very rare. We conduct an experiment on synthetic and real data, measuring this discrepancy across five fairness interventions on four unfairness types (1000 repetitions per combination, 20,000 total runs). Synthetic data lets us encode human bias in labels and compare fairness metrics on biased test labels (data available during development) against clean labels (conditions models face in deployment). We find that a systematic evaluation bias is present across all metrics, so the same models on the same test data can support opposite fairness conclusions and mask the mistreatment of the most disadvantaged groups. This pattern of fairness misevaluation is confirmed by a real-world validation on the Adult Census Income dataset. We conclude that trustworthy fairness auditing and regulatory standards should require bias-aware evaluation protocols, in which observed labels are not treated as ground truth. Full article

Which runtime governance mechanisms reduce physical risk when a learned controller drives a building’s climate, and under what conditions? We develop a closed-loop software-in-the-loop testbed in which a setpoint model trained on real occupancy data drives a physics-based thermal zone through a declarative governance plane, with outcomes scored by an independent safety oracle, and we run the same governance logic on a real MQTT stack with an in-process policy decision point and a hash-chained audit log. Under distribution shift, admission control reduces unsafe physical exposure by $19.4\%$, from 1185.3 to 954.8 °C·min, whereas adding checkpoint rollback reduces it by only a further $0.2\%$ in the reference run ($0.1$–$0.4\%$ across sensor noise seeds): the governance decision takes 0.44 ms while physical recovery takes a median of 61 min. Prevention therefore outperforms recovery in the studied thermal system, and the remaining avoidable exposure is driven by the policy’s estimate of occupancy context. A deterministic single-rule thermostat incurs $50\%$ more exposure under shift while the learned controller uses a $38\%$ higher heating demand proxy: a safety–demand trade-off, not evidence that learned control is necessary. A plant sweep yields an operating envelope criterion for inertial plants: rollback contributes materially to safety only when the plant is restored before the next command arrives and sampled before it can leave the safe set; on slower plants it removes at most $14.7\%$, with a transition band in between. No physical hardware was operated. Full article

Crosswind protection on high-speed railway bridges is important for train running safety; however, bridge-mounted wind barriers may also increase wind loads on the bridge deck and barrier-supporting structure. Previous studies have mainly evaluated wind-barrier performance according to reductions in train aerodynamic loads, whereas direct experimental quantification of the trade-off between train protection and bridge lateral load increase for different barrier geometries remains limited. This study compares vertical and curved wind barriers through wind-tunnel tests using a 1:30 sectional model of a train–bridge system at a reference wind speed of 10 m/s. Barrier porosities in the range of 20–50% and wind attack angles ranging from −6° to 6° were considered. Train-surface pressure distributions and static three-component aerodynamic coefficients of both the train and bridge were measured simultaneously. A lateral load benefit–penalty index was introduced based on the train lateral load-reduction ratio and the bridge lateral load-increase ratio to enable the relative comparison of the tested barrier configurations, while steady Reynolds-averaged Navier–Stokes simulations were used to interpret the underlying flow mechanisms. Both barriers reduced the aerodynamic loads on the train, but the vertical barrier provided stronger train-side shielding at the cost of a larger increase in bridge lateral load. At 30% porosity and a wind attack angle of 0°, the vertical barrier reduced the train side-force coefficient by 75.4% and increased the bridge side-force coefficient by 89.2%, whereas the corresponding values for the curved barrier were 57.6% and 39.4%, respectively. Within the tested ranges, the curved barrier consistently achieved higher index values because its flow-guiding effect reduced pressure concentration and limited the additional lateral load on the bridge. The vertical barrier is therefore more suitable when maximum train protection is the primary objective, whereas the curved barrier provides a better balance between train protection and bridge wind-load control. Full article

Object detection remains a fundamental challenge in computer vision and plays a pivotal role in safety-critical medical applications, including surgical instrument recognition and operating-room workflow automation. This study presents a comprehensive comparative evaluation of five attention mechanisms—Squeeze-and-Excitation (SE), Convolutional Block Attention Module (CBAM), Efficient Channel Attention (ECA), Simple Attention Module (SimAM), and an enhanced multi-kernel Spatial Pyramid Pooling Fast module (SPPF+)—integrated into the YOLOv26n backbone, together with two neck-level attention variants (ECA-Neck and CBAM-Neck). A total of 16 model configurations were systematically investigated on a 22-class dental surgical instrument detection dataset under both default training settings and hyperparameter configurations optimized using the Optuna Tree-structured Parzen Estimator (TPE), enabling a rigorous full-factorial ablation study. Experimental results demonstrate that TPE-based hyperparameter optimization consistently enhances detection performance across all architectures. Among the evaluated models, CBAM-Opt achieved the highest detection accuracy, attaining an mAP@50 of 0.959 and an F1-score of 0.913, although the margins among the top optimized configurations fall within run-to-run variability. In contrast, Base-Opt delivered the strongest strict-localization capability with an mAP@50–95 of 0.800, highlighting the competitive performance of the baseline architecture when appropriately optimized. Notably, the parameter-free SimAM module exhibited the largest improvement following optimization ($\Delta$mAP@50 = +0.040), indicating a pronounced sensitivity to training configuration. Furthermore, neck-level attention integration achieved performance comparable to backbone-based attention, with ECA-Neck-Opt reaching an mAP@50 of 0.959, suggesting an effective alternative that preserves pretrained feature representations while maintaining high detection accuracy. Beyond performance evaluation, this work provides a unified benchmarking framework for attention mechanisms in medical object detection, accompanied by computational complexity analysis and practical architectural insights. The findings establish evidence-based guidelines for selecting attention modules in resource-aware surgical vision systems and contribute toward the development of more accurate and reliable computer-assisted clinical workflows. Full article

Anticipating component degradation before failure occurs has become a cornerstone of intelligent condition monitoring in safety-critical engineering environments. This paper benchmarks three supervised learning algorithms—Linear Regression (LR), Random Forest (RF), and Gradient Boosting (GB)—against each other for the task of Remaining Useful Life (RUL) forecasting on turbofan engines, using the NASA CMAPSS FD001 simulation dataset as the evaluation testbed. The benchmark encompasses 100 run-to-failure training trajectories and 100 test sequences, each characterised by 21 on-board sensor channels recorded over successive flight cycles. Following a systematic preparation stage—which involved discarding uninformative constant-variance channels and applying a piecewise linear degradation labelling scheme capped at 125 cycles—all three algorithms were trained and scored on normalised feature vectors. Among the three candidates, Random Forest delivered the strongest results (RMSE = 17.09, MAE = 12.10, R² = 0.818), ahead of Gradient Boosting (RMSE = 17.42, R² = 0.811) and the linear baseline (RMSE = 20.60, R² = 0.736). These outcomes confirm that bagging-based ensemble regressors provide a compelling accuracy–deployability trade-off for degradation forecasting, with direct relevance to autonomous scheduling and condition surveillance in industrial automation contexts. Full article

This paper investigates cooperative tracking control for virtually coupled multi-train systems subject to nonlinear running resistance, bounded time-varying resistance parameters, state constraints, and actuator saturation. The theoretical contribution is not the separate use of barrier Lyapunov functions, adaptive control, anti-windup compensation, or distributed cooperative control. Instead, the revised analysis establishes a coupled safety-and-boundedness certificate for the actual saturated closed-loop vector field. The closing-speed-aware spacing variable and actuator-authority condition support a first-exit proof of forward invariance, after which a composite Lyapunov analysis couples the saturation residual, anti-windup state, cooperative tracking error, and time-varying parameter-estimation error to establish uniform ultimate boundedness without persistent excitation. This proof architecture distinguishes the proposed controller from recent constrained train-control methods focused separately on velocity/input bounds, distance-oriented full-state barriers, or iteration-indexed learning. Numerical studies with heterogeneous trains, stronger time-varying aerodynamic perturbations, normalized actuator limits, tracking-bound verification, constrained baselines, a near-boundary safety-allocation case, and a quantitative one-factor-at-a-time parameter-sensitivity study are provided. Full article