The reliable operation of wind turbines is critical for generating low-carbon electricity in renewable energy systems. To maximize turbine uptime and minimize maintenance disruptions, smart condition monitoring and early fault detection strategies are essential. Yaw pitch failures, a common cause of performance degradation in wind turbines, are challenging to detect due to the complex relationship between wind speed, yaw pitch current, and grid current. This study proposes a Gaussian Process (GP) regression framework with square exponential covariance functions for early detection of yaw pitch failures in wind turbines. By analysing Supervisory Control and Data Acquisition (SCADA) data from a 2.5 MW wind turbine over a six-month operational wind farm, we establish predictive models for three critical performance relationships: power curve (R² = 0.951, RMSE = 68.5 kW), yaw pitch current versus wind speed (R² = 0.893, RMSE = 0.82 A), and yaw pitch current versus grid current (R² = 0.908, RMSE = 0.74 A). The yaw pitch current versus grid current reference curve demonstrates superior fault detection performance, identifying faults 80 minutes after threshold exceedance with minimal false alarms, significantly outperforming power curve-based detection and wind speed-based detection. Fisher's combined probability test with an optimized threshold (p = 0.581) effectively balances detection sensitivity and false alarm minimization. The results demonstrate the model's ability to detect yaw pitch faults (>6 A) effectively with minimal false alarms, offering a cost-effective SCADA-based solution for wind turbine condition monitoring that leverages the strong correlation (r = 0.79) between grid current and yaw pitch current.

SCIFL-Net: Seam carving-based image forgery localization network

This paper presents the application of a full-spectrum analysis method called spectral unmixing for gamma on-board measurements. The method's sensitivity is evaluated through experiments involving various radioactive sources, under challenging conditions such as low signal-to-noise ratios and moving sources. The results are then compared with both an algorithm developed by the CEA/DAM and currently used in operational conditions and the commercial software Genie2000®. Subsequently, spectral unmixing is applied to aerial measurement data. Finally, the study assesses and compares the method's sensitivity and its robustness against false positive alarms with an operational reference algorithm.

XAI-IDS: DeepLIFT-AE driven granular feature extraction for intelligent detection of emerging vehicular network threats

Fault detection for switched systems based on reachable set estimation.

Artificial intelligence for CO2 pipeline monitoring: Cross-domain insights from oil, gas, water, and hydrogen systems

Improving lightning representation in global model through convective cloud based Price and Rind 1992 scheme.

Long-term seizure prediction in epileptic individuals is challenging, mainly due to signal non-stationarity and seizure type variability. This research presents a patient-specific prediction pipeline for intracranial electroencephalography (iEEG) based on the novel Absolute Mean Instantaneous Frequency Difference (AMIFD) biomarker. The approach addresses seizure variability using the minimum uncertainty and sample elimination (MUSE) feature-ranking technique to automatically cluster seizure types based on their top-ranked AMIFD features, distinguishing the interictal baseline from the preictal phase (defined as a 60-min window ending 5 min before onset). For each identified seizure cluster, a specialized random forest classifier is trained independently, creating a multi-model ensemble tailored to specific seizure morphologies. The final prediction alarm is triggered using a k-of-N aggregation logic to enhance reliability and minimize false alarms. The AMIFD biomarker shows a significantly larger effect size than several standard electrophysiological features (adjusted p < 0.04). When tested on 10 epileptic subjects, the patient-specific framework achieved a mean sensitivity of 92.08% and a false-positive rate of 1.32/day, outperforming the minimum Redundancy Maximum Relevance (mRMR) and MUSE baselines. This patient-specific, multi-model strategy establishes a viable pathway toward more accurate and clinically reliable personalized seizure prediction systems.

Multi-temporal Synthetic Aperture Radar (SAR) images are essential in detecting changes in the environment, analyzing urban growth, and assessing disasters. Nonetheless, it is difficult to reliably detect meaningful changes because of speckle noise, radiometric variations, and inhomogeneous terrain features that severely impair the accuracy of the detection. This paper presents a hybrid change detection framework that integrates Fusion-based Difference Imaging (FDI) with a deep feature-guided clustering method. The proposed FDI module enhances sensitivity to subtle backscatter variations, which combines complementary difference representations to enhance the discriminability of changed and unchanged regions. These representations are refined through a deep clustering process that improves separability and reduces false alarms. Extensive experiments on benchmark multi-temporal SAR data sets indicate that the proposed approach has high detection accuracy and robustness with overall accuracy (PCC) of up to 99.97%, a Kappa coefficient of up to 98.30% and an F1-score of up to 99.59% across various datasets. The proposed approach consistently outperforms competing methods across evaluated datasets and different speckle conditions. The results show that fusion-based representation learning, with uncertainty-aware clustering, is a scalable, efficient, and flexible solution to robust SAR change detection in challenging imaging conditions.

Bearing failures in offshore wind turbines pose asymmetric safety and economic risks, where missed detections may lead to catastrophic events while false alarms trigger costly unnecessary interventions. This study formulates bearing fault diagnosis as a risk-constrained maintenance optimization problem, in which expected operational cost is minimized subject to an explicit probabilistic constraint on the miss rate. To operationalize this constraint without distributional assumptions, we employ a distribution-free uncertainty quantification framework combining probabilistic calibration via temperature scaling (TS), selective classification, and split conformal prediction. TS improves probability reliability, while conformal prediction provides finite-sample coverage guarantees that directly support safety constraint enforcement. Validation on the Fraunhofer LBF wind turbine bearing dataset under strict group-wise cross-validation demonstrates 100% recall for the most safety-critical fault modes (inner race and rolling element defects, exhibiting rapid failure propagation) and empirical coverage exceeding nominal levels. Risk–coverage analysis confirms that uncertainty-aware rejection substantially reduces misclassification risk. A Monte Carlo decision analysis for a 50-turbine offshore wind farm indicates up to 8.5% annual operations and maintenance (O&amp;M) cost reduction and 15% downtime reduction under selective policies, with robustness across sensitivity scenarios. The results demonstrate that distribution-free uncertainty guarantees enable economically rational and safety-compliant maintenance decisions in safety-critical wind energy systems.

Remote photoplethysmography (rPPG) has gained popularity as a non-invasive technique for remote monitoring, as it can provide accurate measurements of an individual's physiological signals under controlled conditions. However, the accuracy of rPPG can be affected by various factors, such as movement artifacts, changes in skin tone, and the presence of other sources of light in the environment. To improve the reliability of rPPG measurements in real-world monitoring settings and reduce the frequency of false alarms in health monitoring settings, we propose a confidence score indicating the quality of the predictions. This score was built by identifying meaningful variables related to motion that strongly correlate with the accuracy of the measurements and training a classifier with data coming from 3 distinct datasets, to improve the model's robustness, reproducibility, and generalizability. Despite that only motion-related features have been considered, the high AUC values obtained in all cases were always above 0.93, demonstrating the model's ability to detect inaccurate heart rate measurements.

ABSTRACT Layer-based quality control in additive manufacturing is challenging due to the spatio-temporal complexity of defect development and propagation across build layers. Most deep learning approaches perform layer-independent analysis, limiting their ability to predict defect evolution. This work proposes a CNN Spatio-Temporal Graph Neural Network (CNN-STGNN) framework for defect recognition, severity prediction, and early warning in layer-by-layer manufacturing. A one-dimensional CNN first encodes multimodal in-situ signals—including process parameters, thermal features, and acoustic descriptors—into spatial embeddings. These embeddings construct a spatio-temporal graph where nodes represent layer-region pairs and edges capture intra-layer spatial adjacency and inter-layer temporal continuity. A graph attention mechanism identifies critical layers and dominant fault propagation paths. Evaluated on 1,000 samples spanning 100 layers and five spatial regions, CNN-STGNN achieves 85.9% defect classification accuracy and a 0.84 macro F1-score, a 13.1% improvement over CNN-only baselines. Severity estimation yields an RMSE of 0.061 and R² of 0.89. Temporal analysis demonstrates detection of defects up to 25 layers ahead with over 74% accuracy and below 19% false alarm rate. Graph attention visualizations further provide interpretable identification of high-risk layers and propagation pathways, enabling proactive quality control in industrial additive manufacturing.

Animal learning can be analyzed on two timescales: task acquisition across training sessions and motivation fluctuations within training sessions. How do variations in motor and neurophysiologic activity relate to task performance over these timescales? Here, this question was examined in head-fixed mice performing a whisker-based sensory discrimination task. Male mice were trained for 12-14 daily sessions on a go/no-go task, each lasting approximately one hour to capture spontaneous performance fluctuations over minutes. Simultaneous to task performance, "non-performance variables" were tracked, including wheel running, pupil size, eyelid aperture and sensory cortical activity. First, motivation states were defined based on performance tendencies over minutes, leading to three state categories: persistent, disengaged, or attentive Non-performance variables were found to predict these states independent of task correctness. Then, when further parsing these states by the go/no-go outcomes of hit, miss, false alarm or correct rejection, learning-like changes were detected in wheel running, eye movements and brain activity. Thus, learning over days and motivation fluctuations over minutes form a continuum, as evidenced by changes in motor and physiologic activity variables not directly controlled by task contingencies, even during periods of suboptimal performance in well-trained subjects. These findings improve the understanding of performance variations and implicit learning, in addition to contributing a framework for the analysis of task performance indirectly from motor and neurophysiologic activity.Significance statement Task performance is typically measured by correctness percentages over daily training sessions but can also correlate with motivation state fluctuations within each session. Thus, while aggregating correctness percentages per session may reveal a learning curve, accounting for within-session state fluctuations can reveal variability in that curve, even among well-trained subjects. In this head-fixed mouse study, task acquisition across sessions and motivation fluctuations within sessions were categorized into three states: persistent, disengaged, or attentive Subsequently, metrics not directly controlled by the task, including locomotion, pupil dilation, eyelid aperture and brain activity, were found to predict both learning and state changes. Therefore, task performance can be tracked more comprehensively from brain and body activity metrics, adding nuance to correctness percentages.

From a neurophysiological perspective, using a smartphone before bedtime may impact sleep quality and cognitive performance. This is because it can suppress melatonin production due to exposure to light. These effects can impact attention, information processing, and reaction time. This study aimed to investigate the impact of smartphone usage duration before bedtime on the sleep quality, total performance, concentration, and reaction time of university students. A total of 263 university students participated in the study. After applying inclusion and exclusion criteria, a final sample of 187 participants was selected for analysis. The PSQI questionnaire was used to assess the participants' sleep quality. Reaction time was evaluated through the ruler drop test. Total and concentration performances were measured using the U-cancellation software. Total performance (TP) is defined as the total number of items presented minus the number of errors (misses and false alarms). On average, students spend 75.78min using their smartphones with the light on before bedtime, and 38.50min with the light off. 70.1% (n = 131) of students reported poor sleep quality (score ≥ 5). The mean total and concentration performances were evaluated in the letter test (281.32 and 134.63) and picture test (281.49 and 135.63), respectively. 39.6% of students had reaction times categorized as below average or average. There was a significant positive correlation of moderate and weak strength between the duration of smartphone use before bedtime in both situations with the light on/off and sleep quality (r = 0.49, r = 0.33, p = 0.000) as well as reaction time (r = 0.47, r = 0.37, p = 0.000). Additionally, a significant inverse relationship was found between smartphone usage before bedtime and total performance (r = - 0.15, p = 0.037) and concentration (r = - 0.19, p = 0.009) letters. Students who use smartphones for over 1h before bedtime with the light on, or for over 30min with the light off, have 2.4 (CI:1.131-5.340; p = 0.023) to 3.7 (CI: 1.753-8.105; p = 0.001) times higher odds of experiencing poor sleep quality. In this sample, using smartphones for more than 60min before bedtime, with the lights on, and for more than 30min with the lights off, can be associated with higher odds of lower sleep quality. Furthermore, long-term smartphone use may be linked with prolonged reaction times, leading to slower responses among students. Additionally, smartphone usage may have a negative impact on their total performance and concentration. These thresholds should be interpreted with caution as they are derived from a specific sample and require further validation in larger and more diverse populations to confirm their generalizability.

Intrusion Detection Systems (IDS) play a vital role in safeguarding computer networks against increasingly sophisticated cyber threats. However, many optimization-based IDS models suffer from an imbalance between exploration and exploitation, leading to premature convergence, poor feature diversity, and high false alarm rates. This study aims to develop an improved intrusion detection framework by enhancing the Modified Grey Wolf Optimization (MGWO) algorithm with a chaotic mapping to address the exploration-exploitation imbalance and enhance feature selection and detection performance. The proposed Intrusion Detection System integrates the Improved Modified Grey Wolf Optimization (IMGWO) algorithm with a deterministic chaotic position-update mechanism. Information Gain is employed to evaluate feature significance, while MinMax normalization ensures effective data scaling. The optimized feature subsets are used to train the classifier, and experiments were conducted using the UNSW-NB15 benchmark dataset. Performance is evaluated using standard metrics, including accuracy, F1 Score, False Positive Rate (FPR), Classification Error Rate (CER), and G-mean. Experimental results demonstrate that the proposed IMGWO-based IDS significantly outperforms existing approaches. The model achieved an accuracy of 98.07%, an F1score of 97.51%, an FPR of 1.55%, a CER of 0.97%, and a G- Mean of 97.96%, indicating improved detection capability and reduced false alarms.

Generalizing the familiar two-correlation comparison, this paper presents a dependence-robust omnibus test to evaluate whether an outcome is equally correlated with multiple predictors. By accounting for shared sampling variation, the test simultaneously avoids false alarms and missed discoveries. The test also nests the pairwise test as a special case. Monte Carlo studies show near-nominal size ( at ) for across diverse dependence structures and under moderate non-normality (e.g., errors) together with high power for moderate departures from equality. We illustrate the method on publicly available educational data and provide an interactive web app (size/power simulator and point-and-click analysis) to facilitate adoption. Collectively, the results support the omnibus test as a practical default when assessing equality of outcome-predictor correlations to be augmented by pairwise contrasts for succinct context rather than primary inference.

Asanas, are very crucial for maintaining humans physical and mental health in an efficient manner. Digital training platforms, fitness monitors, and medical applications depend on accurate yoga posture detection. Conventional vision-based posture identification algorithms encounter the issues such as occlusions, background clutter, coping with different camera angles, and individual variations in body proportions. To address these concerns, this research work proposes a hybrid Vision-IMU deep learning architecture for personalized yoga posture detection. In this work, Kinematic stance trajectories and the Yoga-82 vision dataset are employed to construct a synthetic IMU dataset. The IMU signals were algorithmically produced from skeletal joint motion, utilizing biomechanical motion models to reproduce accelerometer and gyroscope readings under controlled conditions. An LSTM network is employed to encodes IMU data, while a Graph Convolutional Network (GCN) interprets visual features. Finally, these two feature streams are combined by using an attention-based technique to provide user-specific posture classification representations. Experimental results show that the hybrid model handles pose orientation fluctuations and user-specific features better than vision-only or IMU-only models in terms of classification accuracy and false alarm rate. This work demonstrates that hybrid deep learning algorithms can be utilized for real-time, customized yoga position identification for the real time medical applications.

To achieve high-capacity factors, the nuclear fleet has traditionally relied on labor-intensive, time-consuming operation and preventive maintenance programs for plant systems. The manually performed inspections, calibrations, testing, and maintenance of plant assets at periodic frequencies, along with the time-based replacement of assets irrespective of condition, have resulted in a costly, labor-centric business model. Fortunately, there are technologies (e.g. advanced sensor, data analytics, risk-assessment, and cost benefit analysis methodologies) that can enable the transition to a technology-centric business model. This paper analyzes the applicability of continuous-time Markov chain models to perform a risk-informed cost benefit analysis of a single asset, such as a pump and motor set of the circulating water system of a pressurized water reactor. Two-state and three-state Markov chain homogeneous models are applied to analyze different maintenance scenarios, with the parameters of the models estimated from historical operational data of the reactor. The analysis concluded that the corrective maintenance rate and equipment failure rate are the two most important parameters in terms of the plant’s economic performance. For example, changes in the corrective maintenance rate and the equipment failure rate will change the baseline hourly profits from $30.60 to $33.50, which is close to the maximum possible hourly profit of $34.00. Since the optimization and automation of maintenance activities can be accomplished by transitioning to a risk-informed predictive maintenance (PdM) strategy, this paper also analyzes the risks and cost benefits of introducing a PdM approach for a pump-motor set of the circulating water system. It is concluded that while introducing PdM can be beneficial for the overall economic performance of the plant, careful consideration should be given to the cost of the PdM and the false alarm rate. Different scenarios suggested that for a typical PdM system, an increase of 10% to 30% in the maintenance rate will justify purchasing and operating the system. However, it will depend on the system’s false alarm rate, initial cost, and operation cost. Finally, this paper presents a cost benefit analysis using a nonhomogeneous Markov chain model for the case of motor degradation using the data from a real plant. It is concluded that the nonhomogeneous Markov chain model in combination with PdM strategies can provide valuable insights into a plant’s future risks and cost of operation. The key contributions of this paper are threefold: rigorous analysis of the economic performance of a commercial nuclear reactor under different maintenance scenarios, economic analysis of the benefits and shortcomings of introducing PdM systems, and analysis of the economic performance of a commercial nuclear reactor using real-world data collected during operation.

ABSTRACT Background Attention-deficit hyperactivity disorder (ADHD) is characterized by impaired attention. L-theanine, an amino acid in tea, and caffeine, present in tea and coffee, improve attention in healthy adults. Aim To investigate the acute effects of L-theanine-caffeine combination on selective attention, compared to methylphenidate and placebo, in adolescents with ADHD. Methods In a double-blind, placebo-controlled, counterbalanced, three-way crossover trial, 21 adolescents with ADHD (age: 10–19 years, 18 boys) performed a computerized visual recognition reaction-time task before and 50 min after each treatment on three separate days. Participants responded to downward-pointing triangles (20% probability) while ignoring upward-pointing triangles (80%). The behavioral outcome measures were hits, false alarms, target-distractor discrimination sensitivity (A’), and hit reaction time. EEG was recorded during the task, and amplitudes and latencies of P3b event-related potentials elicited by targets in centro-parietal electrodes were measured. Pre- and post-dose changes were compared between treatments. Results L-theanine-caffeine combination (P = 0.038) and methylphenidate (P = 0.035) significantly reduced false alarms versus placebo. Hits and A’ showed no significant differences (P > 0.05). Only methylphenidate improved reaction time versus placebo (mean difference = 43.89 ms, P = 0.018). L-Theanine-caffeine and methylphenidate counteracted slowing of reaction time over task duration. Compared to placebo, both L-theanine-caffeine and methylphenidate increased P3b amplitudes and decreased latencies (P < 0.05). Conclusions High-dose L-theanine-caffeine combination may enhance neural resource allocation and speed of deployment of selective attention, stabilize reaction speed throughout the task duration and reduce false alarms in adolescents with ADHD, supporting its potential as a nutraceutical adjunct for ADHD. Trial registration: Sri Lanka Clinical Trials Registry identifier: SLCTR/2023/004.

Abstract Radar sensors are a corner stone of autonomous driving, offering reliable perception under adverse weather and lighting conditions. However, the increasing resolution of modern automotive radar systems generates large data volumes that must be processed in real time, imposing significant computational and energy demands. This challenge is particularly acute in energy-constrained platforms such as electric vehicles and embedded devices, where power efficiency is critical. Neuromorphic computing offers a promising alternative by emulating the brain's event-driven and energy-efficient information processing. In this work, we extend existing resonate-and-fire neuron models, called spiking neural resonators (SpiNRs), into the Doppler domain to enable velocity estimation. We integrate SpiNR with a spiking Ordered Statistics Constant False Alarm Rate (OS-CFAR) algorithm to realize a full neuromorphic peak detection. Crucially, we introduce a novel activity-gated sparsity mechanism that dynamically deactivates inactive resonators, substantially reducing energy consumption while preserving estimation fidelity. All neuromorphic algorithms are implemented on Intel's Loihi 2 neuromorphic processor, which allows us to exploit event-driven computation and benchmark against conventional digital implementations under realistic hardware constraints. Evaluation against the conventional Fast Fourier Transform and classical OS-CFAR pipeline demonstrates that SpinR achieves competitive accuracy in range-velocity estimation. The proposed activity-gated sparsity mechanism yields additional energy savings and removes the need for a separate peak detection stage, further simplifying the processing chain. These findings highlight the potential of neuromorphic radar processing as a power-efficient alternative to conventional methods and underscore the importance of developing next-generation neuromorphic substrates optimized for embedded signal processing.