This paper proposes a non-intrusive, multimodal Internet of Things (IoT) framework designed for the continuous monitoring of mood and behavioral patterns in adolescents. The system integrates two complementary sensing modalities: a passive smart-chair subsystem and an active gamified behavioral probe. The smart chair utilizes a grid of Force-Sensing Resistors (FSRs) and an ESP32 microcontroller to capture real-time motor activity and posture dynamics, which are interpreted through a fuzzy-logic inference model to identify states of arousal or restlessness. Simultaneously, a mobile videogame-based system elicits and records active decision-making and motivational metrics, such as reward sensitivity and persistence, grounded in foraging theory. By employing an edge-cloud-fusion architecture, the framework synchronizes these data streams to identify complex mood constructs (e.g., anxiety or apathy) while prioritizing interpretability and privacy-by-design. The proposed approach avoids intrusive technologies like cameras or microphones, offering a scalable and ethically grounded tool for early intervention in behavioral and mood disorders within naturalistic environments such as study or gaming contexts.
The presence of a fertile queen is essential for the stability and productivity of a honey bee colony. However, her status is generally assessed through visual hive inspections, which are invasive, time-consuming, and performed at non-continuous intervals. For this reason, within the framework of the so-called precision beekeeping, there is growing interest in remote monitoring systems. The analysis of hive acoustics represents a promising approach for detecting variations in colony state. In this study, the potential of acoustic analysis to discriminate against the presence or absence of the queen in colonies of Apis mellifera ligustica Spin. was evaluated. Audio recordings were collected inside the hive during two experimental summer and autumn trials. The queen bee absence was simulated by isolating them in the secondary queen-rearing compartment of a French-like breeding hive. The acoustic signal was analyzed using Short-Time Fourier Transform (STFT) and spectral peak tracking within three selected frequency bands (80-140 Hz, 200-270 Hz, and 380-455 Hz). The results show a consistent increase in peak frequency within the middle band (200-270 Hz) following queen isolation, with a shift of approximately 20 Hz observed in three out of four colonies. In contrast, the lower and higher frequency bands investigated did not show systematic differences between simulated queenright and queenless conditions. These findings suggest that monitoring the peak frequency in the middle band may provide useful information on colony/queen status and represent a promising basis for the development of lightweight acoustic tools for queen monitoring in precision beekeeping. The ongoing study involving a larger number of colonies is however necessary to confirm the robustness and generalizability of such acoustic indicator.
In the context of Industry 4.0, monitoring and controlling environmental parameters and volatile organic compounds (VOCs) has become central to improving operational efficiency, ensuring product quality, and safety in production environments. The following article aims to develop a versatile, low-cost hardware interface for MOX (Metal-Oxide Semiconductor) gas sensors. The proposed architecture is based on the Arduino Giga R1 Wi-Fi and a custom-designed board developed for resistance measurement (0-500 MΩ) and temperature control of MOX gas sensors. The system also allows data to be stored locally using an SD module or transmitted wirelessly, responding to the growing demand for integrated and connected solutions. Validation was performed by comparing the response of a TiO2 film-based MOX sensor at 400 °C with reference laboratory instrumentation at different ethanol concentrations. The results obtained demonstrate a good level of agreement under the target operating conditions.
Accurate dimensional measurement plays a key role in modern manufacturing, particularly as Industry 4.0 drives the adoption of automated and non-contact inspection solutions. Among the available technologies, vision-based measurement systems have gained increasing attention thanks to their flexibility, high acquisition speed, and relatively easy integration into production environments. At the same time, their measurement performance is known to depend strongly on imaging conditions, optical configuration, and the specific measurement strategy adopted. This paper presents an uncertainty analysis of a stationary vision-based measurement system designed for the dimensional analysis of ropes using an industrial camera and controlled illumination. The experimental setup uses a camera with a telecentric lens. A dedicated calibration procedure is employed to establish the pixel-to-metric conversion. The proposed methodology includes image acquisition, feature extraction, and dimensional computation, followed by an uncertainty analysis based on repeated measurements. Results from the sensitivity analysis show repeatable dimensional measurements, with a propagated lay length uncertainty of approximately 0.76% for a lay angle uncertainty of 0.06◦ and a diameter uncertainty of 0.01 mm.
In the evaluation of high-value edible oils, increasing attention is devoted to non-destructive techniques capable of supporting authenticity assessment and quality control. Variations in the thermal behaviour of oils under controlled excitation can provide indirect but informative indicators of compositional changes and adulteration. In this study, an infrared (IR) thermography-based approach combined with deep learning is proposed for the analysis of tomato seed oil (TSO) and sunflower oil (SO). Thermographic sequences acquired during repeatable heating–cooling cycles were processed as spatiotemporal thermal signals reflecting heat absorption and dissipation dynamics. A convolutional neural network–long short-term memory (CNN–LSTM) model was employed to classify TSO, SO, and adulterated oil samples. The proposed system achieved an overall classification accuracy of 95.1% and a macro F1score of 94.8% in three-class classification. The results confirm that thermographic fingerprints capture stability-related differences induced by oil composition and blending, highlighting the potential of IR thermography coupled with CNN-based modelling as a rapid, non-contact decision support tool for edible oil authenticity screening.
Continuous monitoring of hemodynamic alterations associated with dengue progression remains challenging due to the absence of wearable systems capable of extracting robust multi-parametric vascular markers. This study presents a multi-sensor wearable platform enabling beat-by-beat extraction of heart rate (HR), pulse arrival time (PAT), augmentation index (AIx), normalized perfusion index (PInorm), and breath-by-breath respiratory rate (BR) from synchronized physiological signals with commercial sensors. A controlled cardio-respiratory protocol was conducted in a healthy volunteer to evaluate signal integrity, physiological responsiveness, and parameter feasibility. Clear phase-dependent modulation was observed. HR increased from 85.77 bpm during normal breathing (NB) to 93.64 bpm during apnea and 92.83 bpm during fast breathing (FB). PAT shortened from 168 ms (NB) to 136 ms (FB), suggesting increased sympathetic vascular tone. PInorm, normalized to NB baseline (100%), decreased to 60.03% during apnea and 79.01% during FB, approaching the predefined vasoconstriction threshold. AIx became less negative during apnea (−14.35%) compared to NB (−15.28%), and more negative during FB (−18.36%), reflecting dynamic modulation of wave reflection and peripheral resistance. Respiratory rate increased from 15.3 brpm (NB) to 26.8 brpm (FB), confirming protocol compliance. These findings demonstrate the technical feasibility of extracting vascular resistance, wave reflection, perfusion, and cardio-respiratory coupling markers from a wearable platform under controlled physiological perturbations. Although limited to a single healthy subject, this preliminary study supports the potential of continuous, non-invasive multi-parametric monitoring to track vascular and autonomic alterations relevant to dengue pathophysiology. Future studies will evaluate larger cohorts, including dengue patients, to determine whether longitudinal trends in these parameters can enable early detection of transitions toward a possible critical phase characterized by altered peripheral resistance and hemodynamic instability.
This study proposes a prevention-oriented framework to assess whether current vision-based safety technologies can realistically address recurrent fatal accident scenarios involving heavy-vehicle drivers. Starting from 345 fatal cases extracted from the Italian Infor.MO surveillance system, a structured case-coding analysis was conducted to identify accident determinants that were visually monitorable, partially monitorable, or not realistically monitorable through camera-based systems. Four recurrent scenario groups emerged: poor visibility around the vehicle, poor visibility of other critical operational zones, limited mutual visibility between vehicles, and driver illness or collapse in or near the cab. These patterns were then mapped to relevant technology families and intervention pathways, including blind-spot and surround-view cameras, task-specific inspection systems, driver monitoring solutions, alerting logic, and possible vehicle-control integration. The results show that vision-based systems are promising in selected scenarios, particularly for near-vehicle blind areas and driver-state monitoring, but their effectiveness remains highly dependent on context, robustness, and validation in real operating conditions. The study supports a data-driven, reproducible, and non-deterministic approach to the adoption of vision-based prevention in heavy-vehicle safety.
In industrial machine vision applications for surface quality inspection, supervised deep learning is often limited by the scarcity of labeled samples for rare defect classes. This paper investigates synthetic data augmentation as a practical alternative to new data collection, studying how augmentation quantity and defect intensity interact with model capacity. A real industrial vision system acquired monochrome images that were preprocessed into 100×100 px tiles and labeled into five classes: Compliant, Black spots, Clear spots, Imprints, Oil marks. Two complementary synthesis pipelines were used: DCGAN generation for spatially extended defects and a parametric copy-paste method for small, localized defects, enabling three controllable intensity levels. Two CNN backbones, MobileNetV2 and ResNet-50, were adopted for defect classification and trained under ten dataset configurations. These configurations combined three synthetic-quantity levels (+50%, +75%, +100% relative to the original training set) with three defect-intensity settings (soft/medium/hard), while validation and test sets remained fully real and unchanged. Results show that synthetic augmentation is critical for the lightweight MobileNetV2 (accuracy +0.37 over baseline), whereas the higher-capacity ResNet-50 achieves smaller but consistent improvements (accuracy +0.04). Across both architectures, medium-intensity synthesis delivers the best overall accuracy and the highest deployment-oriented cost index, which penalizes missed defects more than false alarms.
Shoulder musculoskeletal disorders are among the leading causes of pain and functional impairment, often presenting unilateral manifestations that require accurate and objective assessment. Bilateral monitoring of shoulder movements is particularly relevant in rehabilitation contexts, where the comparison between the affected and the contralateral limb may support the identification of asymmetries and compensatory strategies. In this scenario, wearable technologies represent a promising solution for continuous and non-invasive functional evaluation.This study proposes the use of two dual-material 3D-printed wearable strain sensors based on Fiber Bragg Grating (FBG) technology for bilateral shoulder movement monitoring. Each sensor integrates an FBG within a dog-bone shaped structure manufactured via fused deposition modeling (FDM), combining a deformable TPU sensing region with rigid PLA anchoring flanges to optimize strain transfer and mechanical stability. After design and fabrication, a pilot test was conducted on five healthy volunteers. The sensors were positioned symmetrically on the posterior region of the shoulders to monitor simultaneous flexion movements in the sagittal plane at 30°, 60°, and 90°.The results showed a progressive increase in ΔλB amplitude with increasing angular excursion for both sensors, demonstrating the capability to discriminate different elevation levels. Moreover, temporal analysis revealed comparable peak-to-peak intervals between left and right sides, indicating synchronous tracking of bilateral movements.The findings highlight the potential of the proposed dualmaterial FBG-based wearable system as a practical tool for objective bilateral shoulder assessment, with promising applications in rehabilitation and functional monitoring.
Work-related musculoskeletal disorders (WMSDs) represent a major cause of injury and reduced productivity in occupational environments. Wearable technologies, such as smart insoles, combined with machine learning (ML) algorithms, are a promising avenue for monitoring biomechanical loads and assessing ergonomic risk. In this pilot study, we present a CNN-BiLSTM model for the prediction of center of pressure (CoP) and vertical ground reaction force (vGRF) from FSR-based smart insole data. Four healthy subjects were preliminarly enrolled in the training data acquisition, performing multiple walking trials. We compared performance using two validation strategies: Leave-One-Subject-Out cross-validation (LOSO-CV) and a Subject-Dependent Hold-Out validation. While the model demonstrated predictive capability under both approaches, the lowest RMSE was observed with the Subject-Dependent Hold-Out approach, highlighting the advantage of incorporating inter-subject variability during training. Our results indicate the predictive potential of the proposed model, while emphasizing the need for dataset expansion to enhance network performance and pave the way for the development of a wearable system for preventing WMSDs, particularly in older and high-risk workers.
While Fused Deposition Modelling (FDM) printing parameters for Polyethylene terephthalate glycol (PETG) are extensively researched, the non-destructive testing (NDT) of its internal defects remains scarce. This study presents an automated defect characterization framework using Step Heating thermography. To overcome the 1D limitations of dynamic Full Width at Half Maximum (FWHM) regression in identifying deep or complex defects, full 2D morphological reconstruction is required. Therefore, Principal Component Analysis (PCA) is applied to extract high-contrast thermal maps, enabling a comparative evaluation of classical segmentation techniques (Edge Detection, Mexican Hat, Otsu) against Deep Learning (YOLOv8). This work aims to provide a solid foundation for standardizing automated quality control in additive manufacturing.
The emergence of cellular medicine and personalized therapies, which are incompatible with terminal sterilization, has underscored the critical role of airflow patterns as a fundamental sterility barrier in pharmaceutical manufacturing. Concurrently, regulatory advancements and the drive toward automation, aimed at eliminating human operators, who are considered the primary source of contamination, have promoted the integration of robotic arms within controlled environments such as isolators. Although airflow barriers are now well established in the industry, the fluid-structure interaction induced by the movement of material inside them can locally disrupt the flow and potentially compromise the performance of the barrier (i.e., the sterility of the products). This becomes critically important when robots are introduced into pharmaceutical isolators for process automation. In such cases, quantitative tools are needed to evaluate the flow fields within pharmaceutical isolators in order to gain a comprehensive understanding of the process. This work proposes a global and non-invasive measurement method for estimating the velocity field inside a pharmaceutical isolator, based on flow visualization and computer vision techniques. Several algorithms are compared to identify the most suitable approach. In addition, a dedicated device is introduced to increase the intensity gradients of the tracer, thereby improving detectability for computer vision tracking and reducing the fraction of invalid vectors by up to 58%. The system was tested by comparing the results of repeated trials at various speeds against readings from an ultrasonic anemometer.
This paper investigates the integration of Unmanned Aerial Vehicles (UAVs) with chemical sensors for detecting volatile compounds associated with early decomposition, with particular focus on hydrogen sulfide (H2S) as key sulfur-containing biomarker. A MEMS metal-oxide (MOX) gas sensor was characterized under controlled laboratory conditions to evaluate its sensitivity, stability, and suitability for aerial deployment. Experiments in dry air showed a clear and repeatable response across the concentration range investigated, confirming good intrinsic sensitivity. However, relative humidity (RH) emerged as the most critical environmental interference: even moderate levels (20% RH) attenuate the signal, while high humidity values (60% RH) drastically compromise selectivity and response dynamics due to surface saturation effects. These results indicate that, although MEMS MOX sensors are promising for UAV-based forensic applications, there is a need for compensation algorithms and multisensor strategies to ensure reliable detection in real search-and-rescue scenarios.
Firefighting activities impose substantial cardiovascular and respiratory strain due to high-intensity physical exertion, elevated thermal exposure, and the use of heavy personal protective equipment (PPE). Although laboratory-based assessments such as spirometry and cardiopulmonary exercise testing provide accurate evaluation of physiological capacity, they fail to capture the dynamic cardiorespiratory responses occurring during real firefighting operations. Wearable monitoring systems have been investigated in firefighter populations; however, most validations have been conducted under controlled laboratory conditions or structured simulations, with limited data available from live-fire scenarios.This study investigates the feasibility of a wearable system based on a fiber Bragg grating sensor integrated into a silicone substrate for real-time monitoring of respiratory rate (RR) and heart rate (HR) during simulated fire suppression training conducted in a firehouse environment. The sensing principle relies on the detection of thoracic deformation induced by cardiac and respiratory activity.The device was deployed during a pilot test involving a firefighter performing simulated fire suppression task characterized by active flames, confined spaces, and elevated thermal load at the operational training school of the Italian National Fire Corp in Montelibretti (Rome, Italy).The results provide preliminary evidence of the feasibility of the system to detect RR and HR under operational conditions, confirming its mechanical robustness and signal stability in a thermally and physically challenging environment.These findings offer preliminary insights into the feasibility of FBG-based wearable technology for non-invasive physiological monitoring in demanding environments, extending prior laboratory validation.
Federated learning (FL) enables privacy-preserving training on decentralized data but faces challenges on resource-constrained IoT devices due to high energy and communication costs. We address these by evaluating optimization strategies on a physical testbed of heterogeneous IoT devices. Our holistic approach, combining Top-K gradient compression with adaptive early stopping, reduces network bandwidth by 75% and lowers computational load without compromising accuracy. A compute-aware partitioning variant further rebalances workloads to minimize idle time. We identify the "straggler" problem as a critical bottleneck, with faster devices idling for over 75% of the time. Finally, we provide a rigorous metrological assessment, quantifying Type A and Type B uncertainties to validate our findings. These findings highlight the need for system-aware optimizations for sustainable FL in IoT ecosystems.
Finger tapping tests (FTTs) are widely used in clinical practice for the assessment of motor and cognitive function in patients with neurological and psychiatric disorders. Despite their extensive adoption, FTTs are still mainly evaluated through qualitative clinical scales, which are inherently subjective and operator dependent. Consequently, quantitative sensing solutions capable of providing objective and reproducible metrics during FTTs are highly desirable, particularly in environments characterized by strong electromagnetic fields, such as magnetic resonance imaging (MRI) procedures, where conventional electronic sensors are not suitable.In this context, fiber Bragg grating (FBG) sensors offer an effective solution being intrinsically immune to electromagnetic interference while also providing compact size, mechanical flexibility, and high metrological performance. This work presents the design, fabrication, and experimental assessment of an FBG-based finger tapping device (FTD) for the quantitative monitoring of contact force (F) and tapping rate (TR) during FTTs. The proposed device consists of a 3D-printed thermoplastic polyurethane structure featuring a suspended bridge configuration that integrates an optical fiber with an embedded FBG sensor. A silicone pad, placed on the upper surface of the bridge, ensures effective transmission of the applied finger pressure to the sensing element while preserving user comfort. The device was characterized, achieving an average F sensitivity of 0.068 nm/N. Preliminary feasibility tests were conducted with a healthy volunteer performing F-controlled and rate-controlled FTTs. The results demonstrate the ability of the proposed FTD to provide stable F feedback, detect F fluctuations, and reliably estimate TR under different execution conditions.Overall, the proposed FBG-based FTD represents a promising solution for the quantitative assessment of finger tapping performance in controlled clinical and research scenarios.
Cybersecurity in railway infrastructure telemetry plays a critical role in ensuring the integrity, availability, and safety of data transmitted between trains, trackside equipment, and central control centers. The paper analyzes the functional role of actors in a layered Artificial Intelligence-driven system for cybersecurity threat detection, time constraints, and benefits. The focus is on the key security issues of near real time intelligent function, and possible mitigations and solutions.
Vision-based monitoring is increasingly adopted to improve workplace safety, and recent advances in computer vision enable automatic recognition of personal protective equipment (PPE). However, most existing systems still focus on verifying the presence of PPE and its correct use, while evaluating PPE deterioration remains largely overlooked despite its safety relevance and regulatory implications. In this narrative literature review, we conducted a Google Scholar search over 2020-2026 and discussed PPE and anomaly/defect/damage detection in computer vision. By jointly analyzing these topics, we found that PPE deterioration remains a clear gap. Only a few studies address defects in PPE components, often relying on rigid acquisition setups and narrow defect taxonomies, which limits transfer across PPE materials and real operational conditions. We therefore emphasize the need for portable, reproducible capture protocols and coordinated multi-stakeholder data collection to expand defect coverage and improve cross-site generalization.1
Cold Metal Transfer (CMT)-based Wire Arc Additive Manufacturing (WAAM) is a Metal Additive Manufacturing (MAM) process characterized by high productivity and scalability, but also by a strong sensitivity to process parameters, which makes effective process monitoring essential. The transient evolution of the welding current in CMT-based WAAM plays a critical role in deposition stability and overall process reliability. Conventional monitoring approaches often rely on global signal descriptors, which may not capture subtle variations occurring during the rising phase of the current waveform. This work proposes a time-domain characterization methodology specifically focused on the rising transient of the DC current in CMT-based WAAM processes. The transient is subdivided into two amplitude intervals, and slope-based indicators are extracted. The methodology is applied to three deposition conditions exhibiting different process outcomes. The results show that slope magnitudes and their relative variation between the two intervals provide clear discrimination between transient regimes, with one condition displaying significantly lower slopes. Overall, the proposed approach provides a quantitative and uncertainty-aware framework for time-domain monitoring of DC current transients in CMT-based WAAM processes, with potential implications for process stability assessment and real-time process monitoring.
This work focuses on the design and thermal modeling of a Lithium-Polymer battery pack intended for residential renewable energy storage applications using MATLAB Simulink. The thermal behaviour of the battery pack model was investigated under three environmental scenarios, outdoor summer, outdoor winter, and a controlled enclosure and at two discharge rates of 0.5C and 1C. The study assesses how ambient conditions and discharge rates concurrently influence the pack's thermal evolution, ensuring its performance remains within safe and efficient operational limits. To meet typical domestic energy demands, the battery pack was designed with a 30x13 cell configuration, achieving a nominal energy capacity of 3 kWh. The developed model integrates experimental measurements obtained from a single 2 Ah lithium-polymer cell, including electrical parameters, state of charge (SoC), and surface temperature. The results demonstrate that environmental boundary conditions are the primary driver of thermal evolution, even under identical electrical loads. The outdoor winter scenario was identified as the most critical, exhibiting a maximum temperature variation of 16 °C at 1C and significant spatial gradients inside the battery pack. In contrast, the controlled environment promoted high thermal homogeneity. These findings, which align with current literature, validate the proposed model as a tool for predicting the thermal dynamics of full-scale battery packs in real-world residential applications.
