Search Results (4,894)

Deposit return systems (DRS) can increase the capture of beverage packaging, but material circularity is not determined by return rate alone. A returned container contributes to high-value circularity only if it passes recognition, sorting, pre-processing, and material-specific quality gates. This article evaluates the material-quality performance of three returned beverage-packaging materials—polyethylene terephthalate (PET), aluminium and reusable glass—using a quality-gated high-value recovery framework. The model defines a high-quality recovery index, HQR = R × Q × Y, where R is the return rate, Q is the quality factor of the returned stream, and Y is the reprocessing or reuse yield. The HQR indicator describes the quality of the entire DRS process. The core purpose of the HQR model is to distinguish nominal packaging return from high-quality material recovery and to show whether returned PET, aluminium and reusable glass streams remain suitable for high-value circular pathways. A survey-supported early-stage return scenario (R = 0.50) is compared with the 77% and 90% separate-collection targets used in European policy. To strengthen the PET branch of the model, a pilot PET stream-quality and processing-yield dataset was incorporated, including PET purity, colour composition, non-PET impurities, residual moisture, organic residues, intrinsic viscosity, washed PET flake or pellet yield, and sorting/washing rejection. The pilot data indicate that Lithuania had higher PET quality (98.2% PET purity, 80% clear PET, 1.8% non-PET impurities, IV = 0.74 dL/g, and 84.5% washed PET yield) than the Polish regional average (94.3% PET purity, 72.7% clear PET, 5.7% non-PET impurities, IV = 0.721 dL/g, and 80.3% washed PET yield). At R = 0.50, the pilot-derived PET HQR is approximately 37.8% for Lithuania and 31.6% for the Polish regional average. The results indicate that the same nominal return rate can lead to substantially different high-quality recovery outcomes because PET is constrained by stream purity, colour, contamination, and processing yield; aluminium by alloy and remelting control; and reusable glass by inspection, breakage, and refill compatibility. The proposed framework can support structured DRS operator reporting by identifying the material-quality and yield variables that should be measured alongside mass collection; however, operator-level validation is required before the model can be used as a predictive performance tool. Full article

Enterprise strategic decisions must reconcile conflicting stakeholder interests under time pressure, yet the consulting that traditionally supports them remains out of reach for most small and medium-sized enterprises. Large language models make parts of this work automatable, but a single model speaks with one voice, and its reasoning can be neither inspected nor contested. As frontier models continue to improve, whether structured multi-agent deliberation is worth its additional cost has therefore become an empirical question rather than a design assumption. This study designs, prototypes, and evaluates Servi.AI, a personality-aware multi-agent intelligent decision support system for enterprise strategy. The system grounds every recommendation in a traceable evidence chain retrieved over a knowledge graph. It stages a statement–discussion–consensus roundtable in which role-specialized agents argue from conflicting professional stances. It also simulates how synthetic stakeholders, calibrated against a public personality dataset of 874,434 respondents, will experience the candidate decision. The roundtable characterizes how a decision is argued, whereas the sandbox characterizes how it will be experienced. A questionnaire with 133 screened decision-makers confirms these requirement priorities. We evaluate the system across five experimental axes: 2800 controlled simulation runs and a twelve-case benchmark judged blind across three model families. A strong single model attains the highest holistic scores (8.22–8.56/10 across judges), while deliberation contributes auditable role-grounded reasoning, conflict surfacing, and an executable blueprint. Ablating retrieval loses all 71 valid pairwise comparisons, while a heterogeneous five-family agent pool significantly improves risk coverage (Cliff’s $\delta =+0.75$). Retrieval thus drives the evidence-side qualities, and the role structure drives the deliberation-side ones: deliberative value is decomposable along architectural components, a middle-range design proposition. These findings support selective rather than default deployment. The released benchmark, judging protocol, and raw results provide a reusable basis for deciding when multi-agent decision support is worth its cost. Full article

Inventory and condition assessment of vertical traffic signs constitutes an essential activity for road safety management, as these signs serve as the primary mechanism for regulating vehicular flow and alerting drivers to prevailing road conditions. However, current inspection procedures in Peru rely on field crews that evaluate each sign manually, thereby constraining the frequency, objectivity, and scalability of the process. This paper presents SISEVIR (Sistema de Supervisión de Señales Verticales en Infraestructura Vial), a three-stage deep learning pipeline for the automated diagnosis of vertical traffic sign condition. The first stage employs YOLO26s-seg for instance segmentation of 31 sign classes, achieving a test mAP50 of 0.9305 (box) and 0.9224 (mask). The second stage classifies each detected sign into seven deterioration states using EfficientNet-B0, optimized through a five-experiment ablation study that identified progressive offline augmentation as the most effective strategy for handling a 147:1 class imbalance (macro F1 = 0.8252; pairwise McNemar’s tests with Holm–Bonferroni correction did not confirm significance at the family-wise $\alpha =0.05$ level). The third stage integrates Qwen2-VL-2B-Instruct, a vision–language model, to generate natural-language descriptions of sign condition aligned with the MTC Manual of Traffic Control Devices for Streets and Highways. A structured evaluation by two independent raters on 35 descriptions yielded a correctness rate of 93.5% among valid responses (95% CI: 79.3–98.2%, Cohen’s $\kappa =1.00$). The system was trained and validated on a proprietary dataset of 5935 images and 6412 labeled crops collected along three routes in the Ayacucho Region (246.6 km total), with an inter-rater reliability of $\kappa =0.802$ (95% CI: 0.676–0.928). SISEVIR processes vehicular video at 30.7 FPS on an NVIDIA RTX 5080 GPU and assigns each sign a level within a four-tier condition scale (Optimal through Critical) linked to specific maintenance interventions, significantly reducing the time, cost, and personnel required compared with the manual inspection method established in the MSV-2016 Road Safety Manual. Full article

To achieve proactive preventive maintenance of green and low-carbon infrastructure, this study systematically investigated the long-term durability and resistance degradation of prefabricated recycled aggregate concrete (RAC) bridge components. A 80-month multi-field coupled damage experiment under sustained flexural loading, natural atmospheric exposure, and chloride drying-wetting cycles was conducted, and an in-situ physical exposure and multi-source data-driven Support Vector Regression (SVR) dynamic surrogate model was established. Results indicate a prominent time-dependent ebb-and-flow mechanism of degradation drivers: environmental and stress boundaries dominate the early stage, whereas the material replacement rate ($R_r$) surges to become the absolute dominant driving variable (36.8%) in the ultra-long term ($t=80 \mathrm{months}$), proving the cumulative dominance of recycled aggregates. Concurrently, the residual capacity exhibits a distinct two-stage decay characterized by a 10% critical reinforcement mass loss threshold, beyond which the degradation rate of RAC100 accelerates to 1.63 times that of conventional concrete. Driven by the multi-stage injection of in-situ experimental inspection data (surface crack profiling, 2D spatial chloride profiles, and rebar mass loss), the SVR network successfully achieves a collapse-like convergence of the remaining useful life (RUL) confidence interval, precisely locking the RUL of the RAC100 component at 34.5 years within a 1.4-year error margin. This framework provides critical algorithmic support for the life-cycle safety paradigm shift in low-carbon structures. Full article

Reflective cracking is one of the primary causes of premature deterioration in asphalt-overlaid Portland cement concrete (PCC) pavements, reducing service life and increasing maintenance costs. Since crack initiation begins within the underlying PCC layer before becoming visible at the pavement surface, conventional inspection methods have limited capability for early damage detection and continuous monitoring. This study presents a hybrid experimental–numerical framework for monitoring and interpreting bottom-up reflective cracking by integrating Optical Frequency Domain Reflectometry (OFDR)-based Distributed Fiber-Optic Sensing (DFOS), laboratory-scale three-point bending tests, and finite element (ABAQUS) modeling. Rectangular and semi-cylindrical asphalt-overlaid PCC specimens were instrumented with surface-bonded distributed optical fibers arranged in a serpentine sensing layout with approximately 20 mm spacing to continuously monitor strain evolution during flexural loading. In both tested specimen configurations, the three-point bending tests produced bottom-up crack initiation at the predefined notch within the PCC layer, followed by crack propagation toward the asphalt overlay. Crack-width measurements showed that the maximum crack opening occurred near the notch and progressively decreased toward the asphalt overlay, consistent with the expected flexural stress distribution. The OFDR-based DFOS system successfully identified localized strain concentrations associated with crack initiation and propagation, demonstrating its capability for continuous distributed monitoring of fracture evolution. Finite-element simulations identified tensile stress and strain-localization patterns that showed good qualitative spatial correspondence with the experimentally observed cracking region and distributed strain measurements. The combined experimental, sensing, and numerical results demonstrate the proposed framework’s capability to monitor and interpret bottom-up reflective cracking and highlight the potential of OFDR-based distributed fiber-optic sensing for structural health monitoring, condition assessment, and future field-scale monitoring of rehabilitated concrete pavements. Full article

Breast cancer is a highly heterogeneous disease at the histopathological and molecular levels, and histological grade is a key prognostic indicator reflecting tumor biology. Although Random Forest models effectively capture nonlinear patterns in high-dimensional gene expression data, the decision logic of large tree ensembles remains difficult to inspect directly. This study proposes a knowledge-distillation framework that transfers the probabilistic decision behavior of a Random Forest teacher model to a compact, rule-based student model for distinguishing low- from high-histological-grade breast tumors using 20,385 gene-expression features from 1892 METABRIC samples, with clinical information used for sample matching and histological-grade label definition. Root-to-leaf decision paths extracted from the teacher were converted into a binary rule-activation matrix, followed by proximal group sparsification and fidelity-constrained Top-K rule selection. For the principal 400-tree teacher configuration, the student retained 65.0 ± 13.7 rules and achieved an ROC-AUC of 0.833 ± 0.023 and an average precision of 0.843 ± 0.019, while maintaining a raw teacher–student Pearson fidelity of 0.9607 ± 0.0112. As teacher capacity increased, the number of active decision paths increased substantially, whereas the selected rule budget remained within a comparatively narrow range and increased only modestly. The compact student also showed approximately 46% lower measured end-to-end inference latency and a 76% smaller serialized deployment size than the 400-tree teacher. Within the present METABRIC evaluation, these findings indicate that the structural complexity of a tree ensemble and the complexity required to approximate its probabilistic behavior need not scale proportionally, and that the proposed framework can provide a favorable trade-off among predictive performance, teacher fidelity, and deployment compactness. Full article

Infrared defects in photovoltaic (PV) modules captured during unmanned aerial vehicle (UAV) inspection are typically small and low-contrast, while onboard edge platforms are constrained by memory, logic resources, and power consumption. To address these challenges, this paper proposes KD-PH-YOLO for PV infrared defect detection and deployment on the Zynq-7020 platform. Based on YOLOv8, PH-YOLO removes redundant deep-layer computation and introduces a P2 detection head to preserve fine-grained information for small defects. Hardware-friendly Weighted Feature Fusion (HWFF) and Lightweight Attention-CBAM (LA-CBAM) are incorporated to enhance multiscale feature fusion and defect responses. Soft-label and multiscale feature distillation are further employed to improve the lightweight student model without increasing inference complexity. For edge deployment, INT8 quantization and hardware-aware acceleration are applied to map the model onto the Zynq-7020. Experimental results show that KD-PH-YOLO achieves an mAP@0.5 of 92.4% with only 1.48 M parameters and 6.9 GFLOPs. After hardware deployment, the model retains an mAP@0.5 of 91.8%. The Zynq-7020 implementation achieves an average latency of 184.3 ms per frame and a throughput of 5.43 FPS, with power consumption of 3.2 W and an energy efficiency of 1.7 FPS/W. The proposed method therefore provides a favorable accuracy–complexity–energy-efficiency trade-off for resource-constrained PV edge inspection. Full article

Chinese yam planting requires horizontal placement, consistent apical–basal orientation, controlled spacing, and low-damage handling of slender seed segments. This study developed and field-evaluated an integrated vision-guided Chinese yam planter combining furrow opening, feeding, visual inspection and rejection, orientation adjustment, buffered discharge, spacing regulation, furrow tracking, seed placement, and soil covering. Conventional image processing extracted seed-segment length and end orientation, an EfficientNet-B0–kNN one-class method supported auxiliary visual-anomaly screening, and a YOLO11s-seg model with PID control-enabled furrow tracking. Field tests and a two-factor experiment were conducted under sandy-soil conditions. The prototype achieved 22 seedlings min⁻¹, a spacing coefficient of variation of 8.4%, a miss-seeding rate of 4.5 ± 1.3%, an orientation accuracy of 99%, and a path-tracking RMSE of 2.30 ± 0.18 cm; forward velocity was the dominant factor affecting spacing uniformity. The screening module achieved a 98.6% interception rate on marker-simulated anomalies under controlled imaging conditions, while recognition of natural defects remains unvalidated. The system can reduce reliance on skilled labor, and its modular structure and quantified performance provide practical references for users and equipment manufacturers. Full article

Deep excavations are high-risk geotechnical activities, and accurate prediction of diaphragm-wall settlement is important for construction monitoring and deformation control. This study investigates cumulative vertical settlement at 23 diaphragm-wall monitoring points from the deep excavation of Tianjin Goldin Finance 117 in Tianjin, China. Seven prediction models—a naïve persistence model, autoregressive integrated moving average (ARIMA), K-nearest neighbors (KNN), multilayer perceptron (MLP), gated recurrent unit (GRU), Transformer, and XGBoost—were evaluated using a unified five-fold rolling-origin expanding-window validation scheme. GRU achieved the best overall baseline performance, with a mean R² of 0.9172, a mean absolute error (MAE) of 0.1009 mm, a root mean square error (RMSE) of 0.1341 mm, and a mean absolute percentage error (MAPE) of 0.5875%. GRU was subsequently optimized using the crow search algorithm (CSA), the genetic algorithm (GA), and the whale optimization algorithm (WOA). GRU-WOA achieved the best numerical performance, with a mean R² of 0.9289 and an RMSE of 0.1215 mm. Relative anomaly levels were further identified from predicted settlement-change rates to characterize temporal concentration and spatial clustering of settlement-change activity. The proposed framework can support priority inspection and targeted monitoring, although the resulting anomaly levels represent project-relative statistical deviations rather than code-based engineering risk classes. Full article

Background: Hot-melt extrusion (HME) is a scalable pharmaceutical technology increasingly relevant to flexible manufacturing, including small-batch production, personalized dosage-form development, and potential use in pharmacy compounding. When translated into compounding practice, however, HME introduces a risk that routine quality-control methods available in pharmacies may be insufficient to reliably assess the stability of extrusion-based preparations. Methods: Granules containing 50% (w/w) darunavir were prepared by HME at 70 and 90 °C using a previously developed polymeric premix. Samples were stored for 24 months under ambient conditions. During storage, routine quality attributes were evaluated, including appearance, particle size distribution, loss on drying, disintegration time, content uniformity, and assay. Solid-state changes were investigated using differential scanning calorimetry (DSC) and X-ray diffraction (XRD), with a reference PEG-associated darunavir sample prepared and characterized for comparative analysis. Changes in drug release and darunavir content were assessed by dissolution testing and HPLC analysis, respectively. Results: Granules produced at both extrusion temperatures retained acceptable routine quality attributes throughout the 24-month storage period. No substantial changes were detected by visual inspection, pharmacopoeial tests, or UV assay. However, DSC revealed a new thermal event after storage, while XRD showed the formation of a new crystalline phase. Comparison with the reference PEG-associated sample supported the assignment of this phase as a PEG-associated crystalline phase of darunavir. Importantly, this transformation occurred even though the routine quality attributes evaluated in pharmacy compounding practice remained unchanged. Dissolution profiles differed between samples tested immediately after preparation and after long-term storage, with a more pronounced overall difference for granules produced at 90 °C, whereas HPLC confirmed comparable darunavir content in all investigated samples. Discussion: Our results show that routine compounding quality control can meet conventional acceptance criteria while failing to detect API solid-state changes in the investigated HME-derived system. In the PEG-containing matrix, amorphous darunavir undergoes storage-induced crystallization, forming a PEG-associated crystalline phase consistent with its known affinity for polyol-containing media. Conclusions: Acceptable routine quality attributes do not necessarily reflect the solid-state stability of APIs in HME-based formulations. These results highlight the need for solid-state risk assessment when developing extrusion-based systems intended for pharmacy compounding and other personalized manufacturing models in which routine quality control may not include advanced solid-state characterization. Full article

Unmanned aerial vehicles (UAVs) are increasingly considered as enabling technologies for last-mile delivery, emergency medical response, and smart-city applications. However, drone-based logistics, emergency communication, sensing activities, and future aerial mobility are often addressed as separate research domains. This article proposes a four-dimensional planning framework for drone-enabled mobility, integrating goods, information, sensing, and human mobility within a unified conceptual structure. The framework is developed through a literature-informed conceptual analysis and previous applied research experiences related to drone-assisted logistics and emergency communication. Goods mobility includes parcel delivery, medical logistics, emergency supply transport, and hybrid operational models involving trucks, public transport, depots, and micro-hubs. Information mobility refers to the use of drones as mobile communication tools for emergency warnings, citizen interaction, drone-to-infrastructure communication, and infomobility services. Sensing mobility concerns traffic monitoring, environmental observation, disaster mapping, crowd monitoring, and infrastructure inspection. Human mobility is considered as an emerging extension related to urban air mobility (UAM), electric vertical take-off and landing (eVTOL) systems, and low-altitude aerial corridors. Cross-cutting issues such as energy autonomy, solar-assisted drones, multimodal integration, safety, communication, regulation, sustainability, and public acceptance are discussed. The proposed framework provides a structured basis for assessing drones as components of sustainable, resilient, and multimodal mobility systems. Full article

With the expansion of broiler production and the transition toward intelligent and labor-saving management, conventional manual inspection is limited by high labor intensity, unintuitive spatial representation of abnormalities, and inefficient on-site verification. To address these limitations, this study developed an abnormality monitoring system for cage-reared broiler houses by integrating an inspection robot, a digital twin environment, and cloud-based data services. A parameterized three-dimensional model and semantic cage anchors were established according to the dimensions of the physical broiler house and the cage arrangement rules. Robot simultaneous localization and mapping (SLAM) poses, inspection aisles, camera identifiers, and cage arrangement parameters were combined to calculate the semantic locations of dead-bird events and map them within the digital twin environment. Open-mouth breathing, infrared abnormalities, and acoustic abnormalities were additionally visualized at the candidate-cage, local-region, or inspection-aisle level according to the completeness of the available localization information. The system also enabled virtual–physical synchronization of the robot’s position, orientation, and operating status, as well as remote interactive control through a WebGL-based interface. Field tests conducted over approximately 100 days showed that model optimization reduced the triangle count, vertex count, and file size by 48.34%, 30.47%, and 44.81%, respectively, while shortening the initial WebGL scene loading time from 3.84 to 3.05 s. When 500 abnormality markers were displayed simultaneously, the optimized scene maintained an average frame rate of 67.83 fps. Among 2035 dead-bird events, 1954 were correctly localized in terms of cage row, tier, and group, yielding a cage-level localization accuracy of 96.0%. A total of 6412 robot control-command records were evaluated, achieving an overall execution success rate of 99.50%, with mean feedback times ranging from 1.0 to 1.2 s. These results demonstrate that the proposed system provides an integrated workflow for abnormality event acquisition, cage-level localization, three-dimensional visualization, and inspection robot management. Full article

Insulator defect detection remains challenging due to low recognition accuracy for aging, breakage, flashover, and similar faults, limited capability in identifying small targets, and inadequate cross-scale feature fusion under complex backgrounds. To address these issues, this study develops an enhanced detection model, YOLO11-MG, which integrates multi-scale edge information selection and enhancement with global information fusion. In the backbone, the original C3K2 module is replaced by C3K2-MSEIS, which mitigates detail loss during downsampling and strengthens feature representation for small objects and blurred boundaries through a multi-scale edge information selection and enhancement strategy. In the neck, a Gather-and-Distribute (GD) mechanism is introduced; by combining Low-GD and High-GD designs, it enables lossless transmission and effective interaction of cross-scale features. Additionally, the C2PSA module is incorporated to realize adaptive feature weight allocation. Experiments demonstrate that YOLO11-MG achieves precision, recall, mAP@0.5, and mAP@50–95 of 82.3%, 96%, 88.1%, and 70.3%, representing improvements of 5.7%, 4%, 3.1%, and 5.6% over the baseline YOLO11s, respectively. Compared with contemporary YOLO variants such as YOLOv8s, YOLOv10m, and YOLOv12m, the proposed model attains a better accuracy–speed trade-off: it improves mAP@0.5 by 6.3% and accelerates inference by 23.4% relative to YOLOv8s, while keeping 27.67 M parameters—on par with lightweight models but with notably stronger global feature interaction and edge preservation. Overall, the method achieves high detection accuracy, computational efficiency, and real-time inference capability, making it well suited for UAV-based inspection and offering reliable technical support for intelligent insulator fault diagnosis in power systems. Full article

Industrial robots, particularly six-axis serial manipulators, have been widely deployed in manufacturing workflows including assembly, welding, material handling, inspection and precision machining. As the demand for higher end-effector positioning accuracy and trajectory tracking performance grows, end-position errors induced during manipulator operation—stemming from geometric deviations, joint friction, load fluctuations, current surges, as well as variations in velocity and acceleration—have emerged as a critical bottleneck limiting high-precision applications. Conventional error compensation approaches mostly rely on geometric calibration, empirical formulas or fixed regression algorithms, which struggle to adequately characterize error trends featuring strong temporal dependencies, nonlinearity and multi-factor coupling. To address the aforementioned limitations, this paper takes the UR5 industrial manipulator as the research object. Leveraging the NIST-released dataset for manipulator positional accuracy degradation monitoring, this study develops and implements a physics-aware Transformer-based compensation framework that integrates a physics-consistent constraint loss and a nonlinear exponential error amplification strategy with a standard Transformer encoder for end-effector positional accuracy degradation. Multiple variables including target joint position, velocity, acceleration, torque, motor current and control current are selected to construct time-window input vectors, which are used to train the Transformer regression model to capture the correlation between historical motion states and real-time end-effector positional accuracy degradation. Experimental results demonstrate that the proposed Transformer model can fully capture temporal contextual correlations and multi-feature fusion information embedded within manipulator kinematic data, delivering superior error compensation performance for the six-dimensional end-effector pose error prediction task. The self-attention-based time-series modeling framework is well-suited to the nonlinear, coupled and time-varying characteristics of manipulator operational errors. This work provides valuable references for accuracy enhancement of industrial robots and the design of intelligent error compensation schemes. This work provides valuable references for accuracy enhancement of industrial robots and the design of intelligent error compensation schemes, with the proposed physics-aware strategies being model-agnostic and potentially extensible to other regression architectures. Full article

Agile outcome models commonly map Jira variables directly to a retrospective label and then explain the prediction through fragmented feature attributions; they rarely separate domain concepts, team clustering, unresolved work, and concept-label coupling. This study evaluates a domain-informed, concept-bottleneck-style explainable AI architecture for retrospective diagnosis of Agile Epic outcomes. A frozen Jira export of 10,000 unique issue-level records was linked to a pre-specified analytical cohort of 180 Epics across 14 teams. Six experts rated efficiency, effectiveness, sustainability, and contextual risk, while outcomes were recorded as Successful, Challenged, or Unsuccessful. Because the outcome labels and concept ratings were informed by the same Jira evidence, the models estimate consistency with an expert labelling procedure, rather than independent project success. Under five-fold group-aware cross-validation, the fixed-configuration flat LightGBM achieved macro-F1 = 0.864 ± 0.053 and the fixed-configuration HMXAI/CBM-style model achieved 0.843 ± 0.084. These descriptive primary scores are not a joint nested-model-selection comparison. The proposed method, therefore does, not demonstrate a performance improvement; its contribution is an inspectable diagnostic structure. Performance fell materially on the resolved-only subset (LightGBM macro-F1 = 0.645), and model-specific nested, leave-one-team-out, calibration, uncertainty, correlation, and intervention analyses further bound the claims. Concept interventions were not uniformly monotone, so the concept layer is domain-interpretable in form but not yet user-validated as actionable. The study contributes a transparent audit of when concept-level diagnosis can complement flat classification and when circularity, censoring, and shortcut learning restrict interpretation. Full article