Search Results (7,610)

Tool wear directly affects surface quality, dimensional accuracy, and manufacturing cost in milling operations, making reliable wear state classification essential for process control. This paper presents an offline deep learning framework for multiclass tool wear classification using the UC Berkeley milling dataset (NASA-Ames). Statistical features are extracted from vibration, acoustic emission, and spindle motor current signals, and dimensionality is reduced from 78 to 9 informative variables using LASSO regression. A four-layer Long Short-Term Memory (LSTM) network then models the temporal evolution of tool degradation across three wear states: healthy, degraded, and failed. Two model variants are compared: Model A uses sensor-derived features only, while Model B additionally incorporates feed rate and depth of cut as inputs. To prevent data leakage, partitioning is performed at the machining-case level rather than at the individual window level. Model A achieves 92% classification accuracy; Model B reaches 95%, demonstrating that cutting conditions provide contextual information that resolves ambiguity between wear states produced under different machining regimes. These results confirm that combining multisensor feature fusion, LASSO-based selection, and sequential deep learning constitutes an effective framework for tool wear classification in milling. Full article

Smartwatches are increasingly used in safety-critical scenarios, yet their optical heart-rate (HR) measurements often contain noise, artifacts, and missing data, undermining clinical trust. This paper presents Edge Beats, a data-curation layer and end-to-end architecture that enables the low-cost, open source PineTime smartwatch to function as a practical HR sensing node for distributed wearable systems. Heart-rate packets are streamed from PineTime to an ESP32 at the edge layer over Bluetooth Low Energy (BLE), then forwarded via an embedded Message Queuing Telemetry Transport (MQTT) broker to an edge server laptop for processing and visualization. A lightweight multi-stage algorithm cleans and smooths the HR stream using physiological boundary checks, a configurable data imputation technique, and exponential moving average (EMA) smoothing, all designed for real-time operation on resource-constrained hardware. We have evaluated the system over long monitoring sessions and compared the processed PineTime output against a commercial Huawei GT Pro 2 smartwatch. The system suppresses extreme spikes and short-term oscillations, yielding a more stable HR trace with qualitative agreement to the reference trends while keeping values in a physiologically plausible range. Network measurements show low latency (almost 3 ms one-way, 15 ms RTT) and stable throughput, and power measurements (100–450 mW for ESP32 and 3–70 mW for PineTime watch) confirm that continuous HR streaming over BLE and MQTT is feasible within the PineTime’s energy budget. These results imply that data stream processing combined with a modest publish–subscribe architecture improves the stability and usability of HR streams obtained from commodity wearable sensors, making PineTime a candidate as a complementary component for mission-critical health and safety systems. Full article

Accurate short-term prediction of network service traffic is essential for communication resource allocation and proactive fault warning in smart grids. However, smart grid service traffic is characterized by nonlinear fluctuations, strong spatio-temporal coupling, and considerable uncertainty, making it difficult for existing methods to capture long-range dependencies, adapt to dynamic topological relationships, and reflect prediction risks. To address these issues, this work develops a deep learning framework that integrates a spatio-temporal Transformer with a diffusion mechanism. The spatio-temporal Transformer extracts temporal evolution patterns and spatial logical correlations from historical traffic matrices, while the diffusion module improves robustness to abrupt traffic variations through latent uncertainty modeling. Furthermore, attention-guided recurrent units are used to generate stable multi-step forecasting sequences. Experiments on a real-world network dataset show that, compared with mainstream benchmark models, the proposed framework reduces Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Root Relative Squared Error (RRSE) by 46.62%, 47.05%, and 44.18%, respectively. These results indicate that the framework improves prediction accuracy and stability while alleviating error accumulation in long-horizon forecasting, thereby providing reliable technical support for smart grid network management. Full article

Recognizing student behaviors in classroom videos remains challenging due to complex backgrounds, frequent occlusions, subtle inter-class motion differences, and temporal jitter in frame-wise predictions. To address these issues, this paper proposes a hybrid student behavior recognition framework that integrates a Multi-branch Spatiotemporal Attention Network (MSTA-Net) with a Behavior State Kalman Filter (BSKF). At the perceptual level, MSTA-Net employs decoupled channel, spatial, and short-term temporal attention branches to enhance discriminative behavioral features while suppressing irrelevant background information. At the cognitive level, BSKF reformulates behavior recognition as a continuous state estimation problem in a high-dimensional probability space, where behavioral inertia is exploited to smooth noisy observations and improve temporal consistency. Experimental results on the SCB-Dataset and real-world classroom video sequences demonstrate that the proposed method achieves an accuracy of 94.7% and a real-time inference speed of 33 FPS. Compared with purely deep learning-based models, the proposed framework reduces the Action Category Switching (ACS) rate by 50%, indicating substantially improved robustness in long-term behavior recognition. These results suggest that coupling attention-based perception with Kalman-based state estimation provides an effective and efficient solution for reliable student behavior analysis in intelligent classroom environments. Full article

Background: Epicardial coronary artery spasm (CAS) is a frequent and important cause of myocardial ischemia. We aimed to develop and validate a noninvasive, artificial intelligence (AI)-driven risk score using routine clinical data to predict CAS in patients without obstructive coronary artery disease (CAD). Methods: This retrospective study analyzed a derivation cohort of 1050 patients and an external validation cohort of 600 patients who underwent intracoronary methylergonovine provocation testing between September 2008 and March 2025. A random forest (RF) model was developed using 15 clinical variables and simplified to a nine-variable model. Additionally, a convolutional neural network-long short-term memory (CNN-LSTM) deep learning model was implemented to predict CAS from raw digital electrocardiogram data (2611 electrocardiogram records). Results: The final nine-variable RF model, including predictors such as diastolic/systolic blood pressure, age, BSA, hemoglobin, smoking, heart rate, sex, and estimated glomerular filtration rate, demonstrated strong discriminatory power. The area under the curve was 85.8% (95% confidence interval [CI]: 85.8–89.9%) in the derivation cohort and 84.1% in the validation cohort (95% CI: 80.6–87.7%). A dose–response relationship was confirmed, with CAS prevalence increasing from 42.1% (0–1 risk factors) to 82.4% (≥5 risk factors). The electrocardiogram-based CNN-LSTM deep learning model achieved high sensitivity (91.4%) but limited specificity (11.9%); therefore, it should be considered a proof of concept rather than a clinical screening tool until further refinement is achieved. Conclusions: The nine-variable RF model provides a practical and accurate tool for early identification and risk stratification of CAS. The electrocardiogram deep learning model complements the RF model to improve clinical decisions and resource allocation in diagnosing CAS. Full article

This paper presents a methodological framework for anomaly detection in child benefit administration based on Long Short-Term Memory (LSTM) neural networks. The content of this analysis, in general, is situated within the social (S) pillar of the environmental, social, and governance (ESG) accountability framework. We construct a framework applied to 305,338 child allowance claim records from the Fund for Child Protection of Republika Srpska, Bosnia and Herzegovina (February 2017 to December 2025), construct behavioural and demographic features at the applicant and household level, encode sequential claim histories as three-dimensional tensors, and conduct a systematic architecture sweep across six LSTM configurations. The target variable, the guardianship anomaly flag, identifies 172 anomalous records (0.056%) among 305,338 claims, and yields a class weighting ration of approximately 1515:1. Across all six configurations, ROC-AUC values range from 0.706 to 0.870 and PR-AUC from 0.002 to 0.071. The reference configuration (L1_U10_T20_he_normal, ROC-AUC = 0.870) flags 170 applications (0.37% of the test set) for priority manual review at the operational audit threshold of $\tau =0.05$. The highest-risk application identified (anomaly probability 0.935) is characterised by a four-child household with below-poverty declared income, elevated benefit-to-income ratios, home delivery payment method, and a persistent high-risk sequential claim pattern not previously flagged by the Fund’s rule-based administrative system. The results confirm that LSTM-based sequential anomaly detection is a viable and principled complement to rule-based eligibility screening in public social transfer administration. Full article

Against the backdrop of global energy transition and the widespread adoption of Hydrogen-Blended Natural Gas (HBNG), the safety of urban gas pipeline networks faces severe challenges. This paper systematically reviews the research progress of numerical simulation in the field of natural gas pipeline safety, focusing on its core supporting roles throughout the “Leakage–Dispersion–Evolution–Consequence” disaster chain. First, it analyzes the kinetic modeling of high-pressure leakage holes and property corrections based on real gas equations of state, elaborating on the numerical characterization of HBNG multi-component transport. Second, it compares the dispersion mechanisms and environmental coupling modeling methods in typical scenarios such as buried porous media, confined spaces in utility tunnels, underwater environments, and urban building clusters. Third, it reviews leakage monitoring technologies based on physical field simulation and data-driven approaches (e.g., Convolutional Neural Network, Long Short-Term Memory), emphasizing the value of numerical simulation in constructing digital twin training sets. Furthermore, it explores the dynamic evolution of explosion flame–shock wave interactions and the evaluation models for secondary disaster consequences. Finally, the current research status of grid-based risk pre-warning and emergency response strategies is summarized. In conclusion, numerical simulation is not only a robust method for precisely quantifying and characterizing complex physical mechanisms but also a critical technological foundation for building smart and resilient energy cities. Future research should focus on the deep coupling of multi-physics fields, physics-informed learning, and the development of system-level integrated defense systems. Full article

Current landslide displacement prediction models often suffer from insufficient integration between physical mechanisms and data-driven approaches, weak model generalizability, and limited operational applicability. To address these issues, this study develops a physics-informed online learning framework for landslide displacement prediction. The core of this framework is a Physics-informed Long Short-Term Memory network (Phys-LSTM). By embedding discretized forms of the stress balance, creep constitutive, and kinematic equations as hard constraints into the LSTM’s gating mechanisms and loss function, the model ensures physically consistent predictions and enhanced interpretability throughout the learning process. Leveraging real-time data streams from the Sichuan Provincial Geological Hazard Monitoring and Warning Platform, we developed an online processing pipeline for real-time multi-source data ingestion, automated quality control, spatiotemporal alignment, and physics-informed feature engineering. A progressive three-stage learning algorithm was designed to support model cold-start, incremental training, and rolling prediction. Validation across 45 model-development landslide sites and one independent application case demonstrated the framework’s significant superiority over traditional models in displacement prediction accuracy (RMSE ≤ 1.78 mm, R² ≥ 0.96), cross-site generalization stability, and its capability to capture accelerated deformation phases. This research indicates that deeply integrating geomechanical prior knowledge into an online learning framework can effectively improve the reliability, interpretability, and operational applicability of landslide displacement prediction models, thereby providing methodological support for subsequent landslide early warning applications. Full article

Soluble solids content (SSC) is one of the most important indicators of sweetness, ripeness, and market quality in sweet cherries. However, conventional SSC determination is destructive, labor-intensive, and unsuitable for rapid or large-scale quality assessment. Therefore, there is a need for fast, non-destructive, and data-driven sensing approaches that can estimate internal fruit quality without damaging the sample. This study aimed to develop a non-destructive approach for SSC prediction in sweet cherries by combining open-ended coaxial probe dielectric spectroscopy with deep learning models. An open-ended coaxial probe measurement system was designed and developed to determine the dielectric properties of sweet cherries and was coupled with an Agilent E4991A impedance analyzer operating over a frequency range of 5–3005 MHz. A total of 10,080 dielectric measurements and 2100 reference SSC measurements were collected over 26 experimental days. The dielectric constant (ε′), loss factor (ε″), and loss tangent (tan δ) were extracted and used to construct separate ε′, ε″, tan δ, and integrated combined datasets. Six deep learning architectures, namely convolutional neural network (CNN), long short-term memory (LSTM), bidirectional long short-term memory (BiLSTM), gated recurrent unit (GRU), CNN-LSTM, and convolutional long short-term memory (ConvLSTM), were trained and optimized using Bayesian optimization and early stopping. CNN achieved the best performance on the tan δ dataset (test R² = 0.9099, RMSE = 0.8354 °Brix, MAE = 0.6599 °Brix), whereas GRU yielded the highest accuracy on the integrated combined dataset (test R² = 0.8622, RMSE = 1.0331 °Brix, MAE = 0.7958 °Brix). ConvLSTM provided the most consistent performance across all four datasets (test R² = 0.8081–0.8651), demonstrating strong predictive capability and practical computational efficiency. These findings confirm the potential of reduced-range dielectric spectroscopy combined with deep learning for rapid, non-destructive SSC assessment in sweet cherries. Full article

Massive Internet of Things (IoT) and sensor-network services in 5G/6G systems increasingly rely on network slicing to support large-scale sensing, monitoring, and mission-critical applications. In such sliced infrastructures, Proportional Fair (PF) allocation assigns resources according to slice-reported demands. This reliance on trusted demand reporting makes coexisting slices, including mMTC-based IoT sensor slices, vulnerable to resource exhaustion attacks, where a malicious slice inflates its demand to monopolize shared resources and induce Service Level Agreement (SLA) violations. Existing unsupervised defenses mainly focus on anomaly detection, while the translation of detection results into resource-level mitigation remains insufficiently addressed. To bridge this gap, this paper proposes AutoGuard-Hybrid, an unsupervised detection-to-mitigation framework that combines complementary anomaly detectors with allocation-aware mitigation policies to preserve slice-level service availability. Unlike prior detection-only approaches, AutoGuard-Hybrid converts unsupervised anomaly evidence into allocation-aware demand purification before PF scheduling. Its key design is a closed-loop integration of Isolation Forest (IF) and Long Short-Term Memory Autoencoder (LSTM-AE) as spatial and temporal front-end detectors with Adaptive Clipping and a Safety Cap, which translate anomaly scores into demand purification actions. Experiments show that AutoGuard-Hybrid remains comparable to Isolation Forest under Continuous attacks and improves the mean system-wide SLA violation rate by 27.6% under Adaptive Probing attacks. Stage activation analysis further shows that LSTM-AE activations increase from 9.3 under Continuous attacks to 29.4 under Adaptive Probing attacks. Ablation results show that Adaptive Clipping alone reduces the system-wide SLA violation rate by 75.0%, while the full mitigation pipeline achieves an 84.6% total reduction. AutoGuard-Hybrid operates within the 1 ms Transmission Time Interval (TTI) constraint and provides a practical defense framework for next-generation network slicing-enabled IoT and sensor-network services. Full article

Accurate multi-horizon Significant Wave Height (SWH) forecasting is vital for offshore safety and efficiency. Beyond scheduling maintenance windows, reliable lead-time predictions provide critical early warnings to protect personnel and high-value assets from hazardous high-wave conditions. However, the non-stationary and multi-scale nature of sea states poses challenges for consistent long-term accuracy. To address this challenge, we propose a robust three-stage framework for decomposition, feature selection, and multi-horizon forecasting. Specifically, Optimal Variational Mode Decomposition (OVMD) is adopted to construct multiscale and multi-view representations of nonlinear SWH sequences, while a Triangulated Maximally Filtered Graph (TMFG) constructs a sparse dependency network to select informative and non-redundant predictors from decomposed components and environmental variables. A hybrid prediction model then combines a Temporal Convolutional Network (TCN) for local multi-scale patterns with a Bidirectional Gated Recurrent Unit (BiGRU) for long-range dependencies. Experiments on real-world buoy observations show that the proposed approach improves accuracy and robustness over commonly used statistical and deep-learning baselines across short-, medium-, and long-term horizons. Ablation studies confirm that integrating modal decomposition with sparse feature selection enhances model robustness, offering reliable decision support for offshore window planning and high-wave condition monitoring. Full article

Low-voltage electrical contacts are core components of power distribution systems, renewable energy installations, and industrial automation equipment. The electric arc generated during contact switching is the primary cause of contact erosion, material transfer, and equipment failure, posing significant threats to system reliability and operational safety. The accurate prediction of arc parameters is hindered by two challenges: the high scatter in available data undermines empirical models, and purely data-driven approaches risk physically implausible results. To address this, a Gaussian Mixture-enhanced Bayesian-optimized Physics-Informed Neural Network (GB-PINN) is proposed. Three core contributions are made: (1) High-fidelity MHD simulation foundation: A magnetohydrodynamic (MHD) multi-physics coupling model of the contact arc was constructed and validated against experiments, showing high fidelity with only 1.63% error in arc duration and 1.82% in arc energy. A multivariate simulation dataset was generated by varying key contact parameters based on this validated model. (2) GMM-based data augmentation: The measured and simulated data were modeled and sampled via Gaussian Mixture Model (GMM) to enrich the dataset while preserving physical consistency. (3) BOHB-optimized PINN prediction: The Bayesian Optimization and Hyperband (BOHB) algorithm was employed to optimize the PINN hyperparameters, enhancing training efficiency and predictive accuracy. Experimental results demonstrated that the proposed GB-PINN achieved superior performance in predicting arc duration and energy, with mean absolute errors (MAE) of 0.079 ms and 0.624 mJ, root mean square errors (RMSE) of 0.099 ms and 0.774 mJ, and coefficients of determination (R²) of 0.980 and 0.979, significantly outperforming grey model (GM (1, N)), long short-term memory (LSTM), and Transformer models. As a physics-informed data-driven tool, GB-PINN enables high-precision arc prediction, providing reliable support for electrical contact design. Full article

To support refined water quality evaluation and short-term early warning in multi-section river systems, this study developed three percentile-based composite indices: the Water Quality Index (WQI), the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS), and the Entropy Weight Method (EWM). Monthly multi-parameter monitoring data from 2019 to 2025 were used, covering ten river sections (P1–P5, M1–M5). The three indices were compared in terms of statistical distribution, methodological consistency, and anomaly response. An integrated assessment–prediction framework was further established. Within this framework, a one-step prediction scheme was applied to evaluate four models: Long Short-Term Memory networks (LSTM), Random Forest (RF), Support Vector Machine (SVM), and eXtreme Gradient Boosting (XGBoost). The results show that WQI scores were generally high and fluctuated within a narrow range. A clear “ceiling effect” was observed in the moderate-to-high grade intervals. WQI also showed weak consistency with TOPSIS and EWM (r ≈ 0.29–0.32). In contrast, TOPSIS and EWM were more sensitive to water quality fluctuations and extreme risks, and were moderately correlated with each other (r ≈ 0.53). Using TOPSIS < 50 as the threshold, 49 severe anomalous events were identified. These events were mainly clustered in February–April 2020, April–July 2023, and June–September 2025, with sections P4, M1, and M2 acting as high-incidence sites. In several typical events, WQI values remained high, indicating that reliance on WQI alone may delay early warning. Prediction results further reveal that the choice of index strongly affects sequence predictability. Taking XGBoost as the reference, the median validation R² followed a stable gradient: WQI (0.807) > TOPSIS (0.723) > EWM (0.594). XGBoost yielded positive R² values across all indices and sections. It also achieved the most robust overall performance and the strongest cross-site, cross-index generalization capability. Full article

Qinghai Lake, the largest endorheic saline lake in China, has undergone a pronounced hydrological regime shift from a multi-decadal decline to a rapid post-2004 recovery, reflecting strong hydroclimatic non-stationarity in the northeastern Tibetan Plateau (TP). This paper supplements the current water level and lake area status of Qinghai Lake to provide basic background for future prediction. Reliable forecasting of such climate sensitive lake systems remains difficult because conventional statistical models often fail to capture non-linear fluctuations, whereas standalone deep learning models may overlook long-term deterministic evolution. To address this challenge, we developed a serial decomposition GeoAI framework that integrates autoregressive integrated moving average (ARIMA), one-dimensional convolutional neural networks (1D-CNNs), and long short-term memory (LSTM) networks for non-stationary water level forecasting. Using annual water level observations from 1960 to 2025, the ARIMA component was first used to extract the low-frequency deterministic trend, after which the CNN-LSTM module reconstructed the nonlinear residual variability. The model was trained on the 1960–2012 period and validated over 2013–2025, which represents the most dynamic expansion stage of Qinghai Lake. The hybrid framework outperformed the benchmark models, achieving a Root Mean Square Error (RMSE) of 0.2033 m, Mean Absolute Error (MAE) of 0.1727 m, and Mean Squared Error (MSE) of 0.0413 m² during validation. The decomposition strategy effectively reduced phase lag and amplitude attenuation, improving both predictive accuracy and process interpretability. Multi-step forecasting for 2026–2056 suggests that Qinghai Lake will continue to rise, reaching approximately 3204.08 m by 2056, although the growth rate is projected to slow as negative hydrological feedback strengthen. By explicitly separating deterministic climate scale signals from nonlinear short-term variability, the proposed framework provides a robust and transferable geoinformation based tool for forecasting water level dynamics and supporting adaptive management in climate sensitive, data scarce lake basins. Full article

Accurate short-term forecasting of residential electricity demand is increasingly important for smart distribution systems, particularly in the context of demand-side management and flexibility-oriented grid operation. In this study, a high-resolution forecasting framework is proposed in which household electricity demand is classified into fixed, shiftable, and adjustable load categories and forecasted together with total load. A one-minute-resolution synthetic residential load dataset is generated using the Centre for Renewable Energy Systems Technology (CREST) demand model for households with two to five occupants over a 31-day winter period in January. The appliance-level demand data are grouped according to operational characteristics and integrated into a representative four-bus distribution feeder. Minute-level power flow analysis is then performed to calculate technical losses, which are incorporated into the forecasting dataset together with meteorological variables (temperature, wind speed, and solar irradiance) and temporal descriptors. Using this multi-input structure, random forest (RF), support vector machine (SVM), feed-forward neural network (FFNN), and long short-term memory (LSTM) models are comparatively evaluated for the prediction of fixed, shiftable, adjustable, and total residential loads. Model performance is assessed using root mean square error (RMSE) and Pearson correlation coefficient (R), while mean absolute error (MAE) is additionally reported for the final test set. The results show that the LSTM model provided the most consistent overall forecasting performance, particularly for shiftable, adjustable, and total load estimation, while RF yielded competitive results for fixed-load correlation and short-window forecasting in Buses 1 and 2. In contrast, SVM and FFNN exhibited weaker generalization performance across several load categories. The proposed framework provides a practical foundation for the development of dynamic pricing mechanisms that consider load-type-based controllability levels. Overall, the findings demonstrate that integrating load categorization with meteorological, temporal, and technical loss information provides a robust and reproducible framework for smart grid applications such as demand-side management, peak load mitigation, and flexibility-aware residential load analysis. Full article