Search Results (739)

Deep reinforcement learning (DRL) has shown potential for robotic training in virtual environments; however, challenges remain in bridging simulation and real-world deployment. This paper introduces an extended reinforcement learning framework that advances beyond traditional single-environment approaches by proposing a dual digital twin concept. Specifically, we suggest creating a digital twin of the robot in Webots and a corresponding twin in MuJoCo, enabling policy training in MuJoCo’s optimized physics engine and subsequent transfer back to Webots for validation. To ensure consistency across environments, we introduce a digital twin alignment methodology, synchronizing sensors, actuators, and physical model characteristics between the two simulators. Furthermore, we propose a novel testing framework that conducts controlled experiments in both virtual environments to quantify and manage divergence, thereby improving robustness and transferability. To address the cost and complexity of maintaining two high-fidelity models, we leverage generative AI agents to automate the creation of the secondary digital twin, significantly reducing engineering overhead. The proposed framework enhances scalability, accelerates training, and improves the reliability of sim-to-real transfer, paving the way for more efficient and adaptive robotic systems. Full article

The article describes a system for monitoring the vital parameters of evacuated individuals, integrating three key functionalities: pulse detection, verification of wristband contact with the skin, and motion recognition. For pulse detection, the system employs the MAX30102 optical sensor and a signal processing algorithm presented in the study. The algorithm is based on spectral analysis using the Fast Fourier Transform (FFT) and incorporates a nonparametric estimator of the probability density function (PDF) in the form of Kernel Density Estimation (KDE). This developed real-time algorithm enables reliable assessment of vital parameters of evacuated individuals. The wristband contact with the skin is verified by measuring the brightness of backscattered light and the temperature of the wrist. Motion detection is achieved using the MPU-9250 inertial module, which analyzes acceleration across three axes. This allows the system to distinguish between states of rest and physical activity, which is crucial for accurately interpreting vital parameters during evacuation. The experimental studies, which were performed on a representative group of individuals, confirmed the correctness of the developed algorithm. The system ensures reliable monitoring of vital parameters by combining precise pulse detection, skin contact verification, and motion analysis. The classifier achieves nearly 95% accuracy and an F₁-score of 0.9465, which indicates its high quality. This level of effectiveness can be considered fully satisfactory for evacuation monitoring systems. Full article

Sensor-less adaptive optics (SLAO) using stochastic parallel gradient descent (SPGD) offers a promising solution for wavefront correction in free-space optical communication (FSOC) systems, as it eliminates the need for conventional wavefront sensors. However, the standard SPGD algorithm’s convergence speed is limited, and it is prone to becoming trapped in local extrema, especially under complex, high-dimensional wavefront distortions in large-scale and dynamic FSOC systems, hindering its use in time-sensitive, high-precision scenarios. To address these limitations, we propose a novel Modal Stage SPGD (MSSPGD) algorithm which integrates subspace optimization techniques with the traditional SPGD algorithm. By projecting the control problem onto a reduced-dimensional Zernike modal subspace and adaptively expanding controlled modes number based on performance metric, our approach decomposes the high-dimensional optimization task into a coarse to fine search optimization problem, thereby accelerating convergence speed, reducing computational complexity, and enhancing robustness against local optima. Theoretical analysis and numerical simulations demonstrate that the proposed algorithm improves convergence speed, stability, and adaptability leading to more effective mitigation of turbulence-induced degradation in critical FSOC metrics. Experimental results further show that the MSSPGD algorithm achieves an approximately 25% reduction in iteration count compared to conventional SPGD. These enhancements prove that the algorithm highly suitable for real-time SLAO in demanding high-speed FSOC systems. Full article

In response to the challenges of low computational efficiency, insufficient detail restoration, and dependence on multiple GPUs in 3D Gaussian Splatting for large-scale UAV scene reconstruction, this study introduces an improved 3D Gaussian Splatting framework. It primarily targets two aspects: optimization of the partitioning strategy and enhancement of adaptive density control. Specifically, an adaptive partitioning strategy guided by scene complexity is designed to ensure more balanced computational workloads across spatial blocks. To preserve scene integrity, auxiliary point clouds are integrated during partition optimization. Furthermore, a pixel weight-scaling mechanism is employed to regulate the average gradient in adaptive density control, thereby mitigating excessive densification of Gaussians. This design accelerates the training process while maintaining high-fidelity rendering quality. Additionally, a task-scheduling algorithm based on frequency-domain analysis is incorporated to further improve computational resource utilization. Extensive experiments on multiple large-scale UAV datasets demonstrate that the proposed framework can be trained efficiently on a single RTX 3090 GPU, achieving more than a 50% reduction in average optimization time while maintaining PSNR, SSIM and LPIPS values that are comparable to or better than representative 3DGS-based methods; on the MatrixCity-S dataset (>6000 images), it attains the highest PSNR among 3DGS-based approaches and completes training on a single 24 GB GPU in less than 60% of the training time of DOGS. Nevertheless, the current framework still requires several hours of optimization for city-scale scenes and has so far only been evaluated on static UAV imagery with a fixed camera model, which may limit its applicability to dynamic scenes or heterogeneous sensor configurations. Full article

The convergence of wearable technologies and digital twin (DT) systems is transforming rehabilitation engineering, enabling continuous monitoring, personalized therapeutic interventions, and predictive modeling of patient recovery pathways. This review examines the growing role of machine learning (ML) in the development and integration of DTs frameworks in rehabilitation, with a focus on wearable sensor data, security and privacy, edge computing architectures, federated learning paradigms, and generative artificial intelligence (GenAI) applications. We first analyze data collection processes, emphasizing multimodal sensing, signal processing, and real-time synchronization between physical and virtual patient models. We then discuss key challenges related to data security, encryption, and privacy protection, especially in distributed clinical environments. The review then assesses the role of edge computing in reducing latency, improving energy efficiency, and enabling real-time local intelligence feedback in wearable devices. Federated learning approaches are discussed as promising strategies for jointly training ML models without compromising sensitive medical data. Finally, we present new GenAI techniques for generating synthetic data, personalizing digital twins, and simulating rehabilitation scenarios. By mapping current progress and identifying research gaps, this article provides a unified view that connects electronic and biomedical engineering with intelligent, secure, and adaptive DT ecosystems for next-generation rehabilitation solutions. Wearable devices with ML and DTs for rehabilitation are developing rapidly, but their current effectiveness still depends on consistent, high-quality data streams and robust clinical validation. The most promising convergence involves combining edge intelligence with federated learning to enable real-time personalization while preserving patient privacy. GenAI further enhances these systems by simulating patient-specific scenarios, accelerating model adaptation, and treatment planning. Key challenges remain related to standardizing data formats, ensuring comprehensive security, and seamlessly integrating these technologies into clinical processes. Full article

Among gearbox faults, gear tooth faults are dominant. Although the traditional vibration spectrum analysis method is the mainstream diagnostic method, it has limitations such as sensitivity to environmental noise and high sensor deployment cost. Based on the influence of the meshing stiffness of the faulty gear on the dynamic transmission error of the gear, this study innovatively proposes to use the transmission error to diagnose and identify typical gear tooth faults. This paper first calculates the time-varying stiffness of typical faulty gear teeth based on the potential energy method, and analyzes the influence of various faults and environmental noise on the dynamic transmission error signal and vibration signal by establishing a six-degree-of-freedom gear transmission dynamics model. Then, a gear transmission experimental platform is built to synchronously collect the vibration acceleration and transmission error data of the gearbox. The convolutional neural network is used to classify the data under different sample lengths and different noise intensities. The results show that the transmission error signal under the same conditions has a higher gear fault diagnosis accuracy. The proposed method can not only improve the accuracy and anti-interference of gear fault diagnosis but also reduce the deployment cost of signal acquisition, providing a new paradigm for gear condition monitoring. Full article

Cultural tourism has become an increasingly significant phenomenon in urban areas, especially in cities rich in heritage sites. However, when the number of visitors exceeds sustainable capacity thresholds, both the physical and psychological comfort and safety of individuals may be compromised. A higher number of visitors inside historic buildings leads to elevated concentrations of carbon dioxide (CO₂), particularly in poorly ventilated enclosed or confined spaces, primarily as a result of human respiration. Such conditions not only accelerate the deterioration processes affecting heritage materials but also introduce potential health risks for visitors. Parameters such as CO₂ concentration, indoor air temperature, and relative humidity represent key measurable parameters for assessing environmental Indoor Air Quality (IAQ) within heritage buildings. Digital real-time monitoring of these parameters plays a crucial role in preventive heritage conservation, sustainable site management, and in ensuring visitors’ comfort and well-being. This paper presents a procedure and methodology that use digital technological tools to efficiently estimate and monitor the Visitor Carrying Capacity (VCC) of enclosed/confined heritage spaces, especially Heritage Building Information Modelling (HBIM) and Sensor Technology. These kinds of spaces require particular attention due to their spatial characteristics. In order to do so, it is necessary to know the geometry of the site, and to consider IAQ conditions. This study also considers the number of People at One Time (PAOT) and Visitor Occupancy (VO). The results focus on the procedural development of the analysis and emphasise the role of digital tools not only due to their efficiency and accuracy in spatial analysis for estimating VCC, but especially for the real-time monitoring of visitors and surveying specific environmental parameters. The experimental phase of this study uses the Chapel of the Holy Chalice of the Valencia Cathedral (Spain) as a pilot case. Monitoring this space reveals how quickly high CO₂ levels are reached with continuous visitor presence, and how long it takes for them to decay in absence of people and under passive ventilation conditions. The outcome of this research is a detailed methodological framework designed to assess and monitor Visitor Carrying Capacity (VCC) in enclosed/confined heritage sites by integrating digital technologies, thereby enhancing sustainable management, planning and decision-making processes. Full article

This study examines Phase B of the GREENERGY project focusing on the seismic performance and structural health monitoring of a renovated single-story RC frame with brick masonry infills that received significant strategic structural interventions. The columns were confined with basalt fiber ropes (FR, 4 mm thickness, two layers) in critical regions, the vertical interfaces between infill and concrete were filled with polyurethane PM forming PUFJ (PolyUrethane Flexible Joints), and glass fiber mesh embedded in polyurethane PS was applied as FRPU (Fiber Reinforced PolyUrethane) jacket on the infills. Further, greenery renovations included the attachment of five double-stack concrete planters (each weighing 153 kg) with different support-anchoring configurations and of eight steel frame constructions (40 kg/m²) simulating vertical living walls (VLW) with eight different connection methods. The specimen was subjected to progressively increasing earthquake excitation based on the Thessaloniki 1978 earthquake record with peak ground acceleration ranging from EQ0.07 g to EQ1.40 g. Comprehensive instrumentation included twelve accelerometers, eight draw wire sensors, twenty-two strain gauges, and a network of sixty-one PZTs utilizing the EMI (Electromechanical Impedance) technique. Results demonstrated that the structure sustained extremely high displacement drift levels of 2.62% at EQ1.40 g while maintaining structural integrity and avoiding collapse. The PUFJ and FRPU systems maintained their integrity throughout all excitations, with limited FRPU fracture only locally at extreme crushing zones of two opposite bottom bricks. Columns’ longitudinal reinforcement entered yielding and strain hardening at top and bottom critical regions provided the FR confinement. VLW frames exhibited equally remarkably resilient performance, avoiding collapse despite local anchor degradation in some investigated cases. The planter performance varied significantly, yet avoiding overturning in all cases. Steel rod anchored planter demonstrated superior performance while simply supported configurations on polyurethane pads exhibited significant rocking and base sliding displacement of ±4 cm at maximum intensity. PZT structural health monitoring (SHM) sensors successfully tracked damage progression. RMSD indices of PZT recordings provided quantifiable damage assessment. Elevated RMSD values corresponded well to visually observed local damages while lower RMSD values in columns 1 and 2 compared with columns 3 and 4 suggested that basalt rope wrapping together with PUFJ and FRPU jacketed infills in two directions could restrict concrete core disintegration more effectively. The experiments validate the advanced structural interventions and vertical forest renovations, ensuring human life protection during successive extreme EQ excitations of deficient existing building stock. Full article

To meet the demand for rapid and accurate glucose determination in clinical diagnostics, food testing, and related fields, this study developed a high-performance electrochemical glucose biosensor based on multi-walled carbon nanotubes/Prussian blue/zeolitic imidazolate framework-8@glucose oxidase/chitosan (MWCNTs/PB/ZIF-8@GOx/CS). The MWCNTs/PB conductive network significantly accelerated electron transfer and catalytic activity, while the ZIF-8, with its regular pore structure and high specific surface area, provides an efficient microenvironment for the immobilization and conformational stabilization of glucose oxidase (GOx), thereby improving substrate diffusion and maintaining enzyme activity. The MWCNTs/PB/ZIF-8@GOx/CS sensor demonstrates excellent sensing performance, featuring a wide linear response to glucose concentrations ranging from 4.8 μM to 2.24 mM, a high sensitivity of 579.57 μA/mM/cm², and a low detection limit of 0.55 μM (S/N = 3). In addition, the sensor performs excellent repeatability (RSD = 1.49%) and retained 86.23% of its initial response after 3 weeks of storage at 4 °C, highlighting its strong potential for practical application in glucose detection. Full article

Wearable sensors are increasingly used to monitor physical activity, yet low-power devices often rely on data aggregation to conserve battery life, potentially impacting measurement accuracy. This study evaluates the performance of a new low-power wearable (LPW), designed for monitoring steps across multiple months in a free-living environment, compared to a research-grade sensor (RGS) that collects raw acceleration data, with a focus on how different aggregation intervals impact step count accuracy. Thirty-two participants wore both sensors over two days, with LPW data collected in 10 min, 1 min, or 10 s aggregation periods (APs). Sensitivity and specificity of wear time detection were high across all APs (0.96 and 0.98, respectively). While total daily step count error did not differ significantly between APs, the 10 min AP exhibited greater undercounting and wider limits of agreement, especially in APs containing more than 40 steps. These findings suggest that although AP does not affect total daily step count, it influences the accuracy and variability of more granular data windows. Aggregating step counts over longer intervals may obscure short, fragmented bouts common in daily activity, leading to underestimation of steps. Optimizing APs and sensor settings is critical for improving accuracy in low-power wearables used outside laboratory settings. Full article

This study investigates the influence of suspension elastokinematics on vehicle handling and stability through a combined research of experimental testing and numerical simulation. Laboratory tests were conducted on the front suspension of a Mercedes-Benz S320 using a quarter-car test rig equipped with specialized sensors to measure wheel displacements, steering angles, camber, and accelerations. Complementary dynamic tests were carried out under real driving conditions, including braking in a turn and “fishhook” maneuvers, to capture suspension behavior under critical operating scenarios. Based on the experimental data, an MSC Adams/Car multibody simulation model was used, incorporating varying stiffness values of suspension elastomeric elements that replicated progressive aging and degradation effects. The simulation results were compared with experimental data to validate the model’s predictive capability. Key findings indicate that reductions in elastomer stiffness significantly affect wheel kinematics, vehicle yaw response, and lateral acceleration, particularly during high-intensity maneuvers. The results underline the critical importance of accounting for elastomeric component degradation in suspension modeling to ensure vehicle safety and performance over the operational lifespan. The developed methodology demonstrates the effectiveness of integrating experimental measurements with advanced simulation tools to assess elastokinematic effects on vehicle dynamics. Full article

Machine vision (MV) is reshaping numerous industries by giving machines the ability to understand what they “see” and respond without human intervention. This review brings together the latest developments in deep learning (DL), image processing, and computer vision (CV). It focuses on how these technologies are being applied in real operational environments. We examine core methodologies such as feature extraction, object detection, image segmentation, and pattern recognition. These techniques are accelerating innovation in key sectors, including healthcare, manufacturing, autonomous systems, and security. A major emphasis is placed on the deepening integration of artificial intelligence (AI) and machine learning (ML) into MV. We particularly consider the impact of convolutional neural networks (CNNs), generative adversarial networks (GANs), and transformer architectures on the evolution of visual recognition capabilities. Beyond surveying advances, this review also takes a hard look at the field’s persistent roadblocks, above all the scarcity of high-quality labeled data, the heavy computational load of modern models, and the unforgiving time limits imposed by real-time vision applications. In response to these challenges, we examine a range of emerging fixes: leaner algorithms, purpose-built hardware (like vision processing units and neuromorphic chips), and smarter ways to label or synthesize data that sidestep the need for massive manual operations. What distinguishes this paper, however, is its emphasis on where MV is headed next. We spotlight nascent directions, including edge-based processing that moves intelligence closer to the sensor, early explorations of quantum methods for visual tasks, and hybrid AI systems that fuse symbolic reasoning with DL, not as speculative futures but as tangible pathways already taking shape. Ultimately, the goal is to connect cutting-edge research with actual deployment scenarios, offering a grounded, actionable guide for those working at the front lines of MV today. Full article

Pavement surface defects such as potholes pose significant safety risks and accelerate infrastructure deterioration. Accurate and automated detection of such defects requires both advanced sensing technologies and robust deep learning models. In this study, we propose PoFormer, a Transformer–CNN hybrid framework designed for precise segmentation of pavement potholes from heterogeneous image datasets. The architecture leverages the global feature extraction ability of Transformers and the fine-grained localization capability of CNNs, achieving superior segmentation accuracy compared to state-of-the-art models. To construct a representative dataset, we combined open source images with high-resolution field data acquired using a multi-sensor pavement inspection vehicle equipped with a line-scan camera and infrared/laser-assisted lighting. This sensing system provides millimeter-level resolution and continuous 3D surface imaging under diverse environmental conditions, ensuring robust training inputs for deep learning. Experimental results demonstrate that PoFormer achieves a mean IoU of 77.23% and a mean pixel accuracy of 84.48%, outperforming existing CNN-based models. By integrating multi-sensor data acquisition with advanced hybrid neural networks, this work highlights the potential of 3D imaging and sensing technologies for intelligent pavement condition monitoring and automated infrastructure maintenance. Full article

This study introduces a computational framework that formalizes rollover risk in heavy-duty vehicles by integrating simulation-informed physical modeling with sensor-driven decision logic. The approach combines high-fidelity fluid–structure interaction modeling (via CFD) with multibody vehicle dynamics simulations to capture the complex behavior of rotating, partially filled mixer tanks under dynamic conditions. Rollover thresholds were identified by extracting the maximum safe speeds across a range of maneuvers (e.g., steady-state turning and step steering), using tire lift-off as the critical indicator. These thresholds were then formalized into decision rules using onboard sensor data, such as lateral acceleration, steering input, and tank rotation speed, allowing a real-time rollover warning system to continuously compare current vehicle states against critical limits. By systematically extracting critical force and moment responses and translating them into limit values provided by conventional onboard sensors (lateral acceleration, roll angle, steering input), the framework bridges high-fidelity simulation and real-time monitoring. A concrete truck mixer is used as a case study to demonstrate the utility of this approach in formalizing rollover thresholds for real-world decision support. Beyond the specific vehicle type, this work contributes to the broader discourse on how computational methods can contribute to new control or assistance strategies for safety-critical systems. Full article

Efficient composting is essential for sustainable organic waste management, yet conventional monitoring approaches are limited by single-parameter measurements and delayed response. This study presents an integrated sensor–AI framework designed to capture the interaction between thermal, chemical, and environmental factors governing composting. A distributed in-pile sensor network continuously measured temperature, moisture, and pH, while ambient parameters and gaseous emissions (O₂, CO₂, CH₄) were recorded to validate process dynamics. Statistical analyses, including correlation and regression modeling, were applied to quantify parameter interdependencies and the influence of external conditions. Results showed strong positive associations between temperature, moisture, and CO₂, and an inverse relationship with O₂, indicating active microbial respiration and accelerated decomposition. The validated sensors maintained high accuracy (±0.5 °C, ±3%, ±0.1 pH units) and supported real-time feedback control, leading to improved nutrient enrichment (notably N, P, and K) in the final compost. The framework demonstrates a transition from static measurement to intelligent, feedback-driven management, providing a scalable and reliable platform for optimizing compost quality and advancing sustainable waste-to-resource applications. Full article