Search Results (674)

Regression-based aboveground biomass (AGB) prediction from Earth-observation data often compresses the upper tail of the biomass distribution, yet the relative effectiveness of geospatial residual correction and ensemble learning in fragmented mountain landscapes remains unclear. Thus, we compared the two paths for reducing high-value underestimation: geospatial residual reconstruction using empirical Bayesian kriging regression prediction, and feature-space optimization using Stacking ensemble learning. SHapley Additive exPlanations (SHAP) interpreted feature contributions, and quantile regression forests (QRF) converted high-AGB point estimates into prediction intervals. Results show that geospatial optimization brought limited gain because residual spatial autocorrelation was weak (Moran’s I = 0.10), whereas Stacking improved overall R² from 0.75 to 0.79 and reduced high-AGB bias (AGB > 80 t/ha) from −13.56 to −5.49 t/ha. This improvement was mainly attributed to complementary heterogeneous learners, with XGBoost capturing the primary non-linear trends, SVR extrapolating to correct high-AGB errors, and RF providing minor marginal calibration. SHAP analysis suggests that LiDAR-derived cubic mean height (Elev_curt_mean_cube) was the dominant feature explaining high-AGB variability, with a threshold response consistent with biomass-height allometry. QRF achieved coverage of 92.8% with a mean interval width of 74.51 t/ha, while coverage in the high-AGB subset was 84.2% with a mean width of 90.50 t/ha. The proposed comparison-and-diagnosis framework provides an interpretable approach for selecting an appropriate correction pathway and supports forest carbon monitoring, carbon accounting, and management decisions in complex mountain ecosystems. Full article

This paper examines mainstream platform moderation as it encounters contemporary pagan religious practice in Europe and reads what it finds as a symptom of a wider condition: the algorithm functions less as a neutral tool than as a productive instrument of an extractive political economy whose characteristic operation is the appropriation, classification, and revenue-conditioned filtering of human expression. Minority religious traditions are structurally exposed within it: too small to constitute an accommodated market, too polysemic for classifiers trained on majority devotional and Anglo-American extremism corpora, and too fragmented to extract policy concessions. Empirically, the paper draws on netnography of pagan online communities across Europe, informal conversations with 97 practitioners across nine European jurisdictions, and a corpus of 247 takedown notices and appeal exchanges (2019–2025). Three recurrent rationale-clusters—devotional content classified as occult, as extremism-adjacent, and as unsafe activity—are read as predictable outputs of the system’s cost structure rather than as ordinary classifier errors. Two concepts are proposed. Algorithmic sacredness names the transfer of gate-keeping functions previously held by ecclesiastical, state, and editorial actors, routed through Bourdieu’s meta-capital as extended to platforms. Algorithmic pluralism names a programmatic direction toward infrastructures whose governance is not capital’s. Full article

Temperature uniformity in thermostatic chambers is critical for material testing, biological incubation, and precision measurements, as even minor thermal gradients can compromise reliability. However, in chambers designed to avoid airflow disturbances, such as those used in semiconductor fabrication and optical experiments, forced convection and mechanical stirring are often impractical. Consequently, natural convection becomes the dominant heat transfer mechanism, introducing significant nonlinearity, large thermal inertia, and multivariable coupling among multiple heat sources. To address these issues, this study develops a multi-input–multi-output (MIMO) temperature control strategy for a rectangular chamber equipped with eight heating elements (grouped into four channels) and adjustable-angle baffles. The proposed method combines a multi-PID controller array with genetic algorithm (GA)-based parameter tuning using a transfer-function matrix model. Experiments demonstrate that baffle angles below 90° improve spatial uniformity, and the relative grouping of heaters outperforms adjacent grouping in both thermal inertia and correlation. Using GA-optimized PID parameters, the controller maintains steady-state error within ±0.5 °C and reduces settling time by approximately 140 s compared to conventional Ziegler–Nichols tuning. Validated through simulations and experiments, the proposed approach provides a reliable and cost-effective alternative to forced convection for airflow-sensitive applications, achieving superior uniformity and steady-state accuracy. Full article

This research evaluates the accuracy and repeatability of a novel procedure for measuring pitch diameter using a touch-trigger probe in a production environment. Experiments were performed on a WFL M40 machine equipped with a Renishaw RMP600 strain gauge probe, using different thread types and machine tools. The results were validated by comparison with the traditional three-wire method, which is regarded as a high-precision reference for pitch diameter measurement. Statistical analyses, including the Shapiro–Wilk test, ANOVA, and Tukey HSD post hoc tests, were applied to assess error distribution and differences between thread types and machines. The proposed method demonstrated satisfactory accuracy and repeatability, with an error spread of approximately ±0.015 mm, achieving measurement results within the manufacturing tolerance range (±0.07 mm). The study also showed that measurement accuracy is influenced by factors such as thread angle and Z-axis positioning. Overall, the method proved to be reliable for production use, offering a good balance between measurement accuracy and operational efficiency, although minor adjustments may be needed for automatic correction routines. Full article

Conventional bond-slip models generally represent ribbed GFRP bars as equivalent smooth cylinders, limiting their ability to describe local rib-bearing, interface degradation, and non-uniform stress transfer. This study establishes a three-dimensional finite-element model for helically ribbed GFRP bars embedded in concrete, explicitly incorporating the helical rib geometry, cohesive-frictional interface interaction, and concrete damaged plasticity. Validation against independent pull-out tests yields minor peak bond-stress errors of −0.55% and −2.49% across different bar diameters, with numerical reliability confirmed through mesh and energy checks. The results indicate that bond resistance evolves from cohesive transfer to rib-bearing action, followed by localized concrete damage, frictional sliding, and residual interlocking. Stress transfer is highly non-uniform along the bonded length, and post-peak interface degradation causes the active transfer zone to migrate dynamically away from the loaded end. Parametric analyses reveal conditional main-effect trends within the investigated ranges, demonstrating that rib height has the strongest influence on residual resistance and energy dissipation, whereas the benefit of increasing concrete strength gradually diminishes. Increasing the bonded length or bar diameter raises the total pull-out force but reduces the nominal bond efficiency due to shear lag. Finally, a simplified four-stage bond-slip relationship is proposed, wherein each stage physically aligns with distinct interface degradation phases, to facilitate computationally efficient structural-scale simulations. Full article

Casing deformation in horizontal wells reduces the effective drift diameter of the wellbore and may interrupt workover and hydraulic fracturing operations. Accurate prediction of reshaping load is therefore important for deformed-casing repair and wellbore integrity management. In this study, the mechanical response of C110-grade elliptically deformed casing restored by a pear-shaped expander was investigated without cement-sheath constraint. Two analytical reshaping-force models based on curved beam theory and thick-walled cylinder theory were established, and a three-dimensional finite element model was developed to simulate nonlinear tool–casing contact and elastoplastic deformation. Five expander outer diameters of 122–130 mm were analyzed, and a 1:8 similarity-scaled experiment was conducted for validation. The results show that the reshaping force increased from 50.2 t to 141.7 t as the expander outer diameter increased from 122 mm to 130 mm, whereas the improvement in minor-axis expansion was relatively limited. The finite element results agreed well with the experimental results, with average errors of 8.542% for reshaping force and 8.822% for minor-axis expansion. The thick-walled cylinder model showed better prediction accuracy than the curved beam model, with an average error of 4.11%. The proposed analytical–numerical–experimental framework provides a basis for reshaping-load estimation, expander selection, and stepwise restoration process design in deformed-casing repair. Full article

Graph neural networks (GNNs) for crystal property prediction are typically evaluated by a single aggregate error, which can mask where, and for which classes of materials, these models fail. In this study, we present a space-group-, centering-type-, and band gap-resolved error analysis of GNN band gap prediction on the Materials Project dataset. As a channel for this analysis we use Crystal-X, a deliberately simple model: a standard graph convolutional backbone with two minor architectural modifications, an asymmetric edge convolution and a neighbor-feature transformation, that supplement bond information often treated as secondary in node-centric models. Crystal-X is not a state-of-the-art model: it reaches a band gap MAE of 0.256 eV on the MP 2018.6 dataset, behind ALIGNN (0.22 eV) and PotNet (0.20 eV), though ahead of older baselines such as CGCNN (0.39 eV), SchNet (0.415 eV), and MEGNet (0.33 eV) while using only the nine-property CGCNN atomic feature set. Its value here is as a controlled, low-complexity testbed for the error analysis. That analysis reveals systematic patterns that aggregate MAE conceals: errors concentrate in underrepresented band gap ranges and in low-symmetry and non-centrosymmetric space groups; per-group errors for sparsely populated space groups are dominated by sampling noise; and modest, as-yet-unverified gains from edge-aware convolutions appear in monoclinic and non-primitive-centered systems. We argue that this kind of granular, symmetry-resolved evaluation should accompany aggregate benchmarks when assessing crystal GNNs. Full article

Long-distance variable frequency drive (VFD) systems are widely used in mining and water transport applications, yet motor terminal overvoltage remains a major reliability risk when inverters and motors are separated by long cables. This study examines a mining submersible pump permanent magnet synchronous motor (PMSM) drive. Cable distributed parameters were extracted with Ansys Q2D Extractor, and a wideband simulation model was built in MATLAB/Simulink. To ensure modeling accuracy, the wideband transmission line model was calibrated and validated using a 20 m cable VFD experimental platform, demonstrating an excellent agreement with a peak voltage relative error of only 2.56%. Parametric simulations were used to quantify the effects of cable length, pulse width modulation (PWM) pulse rise time, and cable cross-sectional geometry on normalized motor terminal overvoltage. Cable length and pulse rise time were the dominant factors governing overvoltage severity. Minor changes in conductor radius or spacing produced only limited changes in peak overvoltage. These findings support EMC-oriented design priorities for long-line VFD systems, with emphasis on cable routing and inverter dv/dt coordination rather than fine cable-geometry tuning alone. Full article

Background: Ancient DNA (aDNA) research is one of the biological fields that has been transformed by the development of next-generation sequencing (NGS). To date, Illumina sequencing has dominated aDNA research. However, its high costs are driving the adoption of alternative sequencing platforms. Combining data generated on different platforms may introduce artifacts arising from platform-specific errors. This study systematically compared the performance of four sequencing platforms (Illumina NovaSeq 6000, GeneMind SURFSeq 5000, Salus Evo, and MGI DNBSEQ-G400) for whole-genome aDNA sequencing using the same set of six samples and libraries. Methods: Single-stranded libraries, prepared with and without enzymatic damage repair, were generated from six human aDNA specimens recovered from Phanagoria polis and sequenced on all four platforms. Data were subsampled to equal read counts per library, mapped to the human reference genome, and compared using standard NGS quality metrics. Population genetic analyses, including principal component analysis (PCA) and ADMIXTURE, were performed to assess potential platform-specific biases. Results: All platforms produced raw sequencing data of acceptable quality. Only minor differences were observed among platforms in standard NGS metrics. The MGI platform showed a shift toward longer sequenced fragment lengths compared with Illumina and the Illumina-like platforms. Post-mortem damage patterns, particularly C>T substitutions, were highly consistent across all platforms. PCA and ADMIXTURE analyses revealed no evidence of platform-specific bias: results from all platforms clustered tightly together, and platform choice had no significant effect on ancestry component estimates. Conclusions: Our findings demonstrate that the GeneMind, Salus, and MGI sequencing platforms are comparable to Illumina for paleogenomic research. Moreover, aDNA datasets generated on these platforms can be combined for downstream analyses without introducing detectable bias. However, the fragment-size shift observed on the MGI platform warrants caution and adaptation when working with highly degraded or low-endogenous-content samples. Full article

Timely identification of company financial risks is crucial for investors and regulators. However, existing studies overlook the class imbalance caused by the scarcity of high-risk samples, and the interpretability of deep models is insufficient, making it difficult to meet the practical needs. To address these problems, we propose a classification model named FA-LSTFormer, which models a company’s financial risk as low-, medium- and high-warning tasks. FA-LSTFormer employs LSTM and Transformer to decouple the short-term continuity and long-term dependency inherent in financial data. To further attend to indicator-level nuances, we incorporate a Risk-Sensitive Hierarchical Indicator Attention (RSHA) module. Moreover, given the pronounced class imbalance where high-risk events are substantially underrepresented, we further propose a Class-Imbalance-Aware Focal Loss (CIFL) function to prioritize these minor yet critical samples and suppress false negatives. On the dataset of Chinese A-share manufacturing listed companies, experimental results show that our FA-LSTFormer achieves superior performance in accuracy, precision, recall, F1-score and AUC, achieving 92.76%, 93.13%, 91.84%, 92.48%, and 95.27%, respectively. Compared to the suboptimal LTR-Net, it improves these metrics by 1.64–3.60%. Compared to the LSTM–Transformer baseline, FA-LSTFormer improves on it by 4.30–9.13%. In the risk-oriented decision evaluation, FA-LSTFormer achieves a warning ROC of 0.954 for the high-risk class and lowers the error rate to 9.82%. It maintains an accuracy rate of 84.46% even after three years of early warning and exhibits strong robustness across different warning thresholds and company sizes. These results verify the advantages of FA-LSTFormer in both algorithmic performance and practical early-warning applications. Full article

Modern grid digitalization connects communication networks, monitoring terminals, service platforms, security devices, and operational data sources. Intrusion detection in this setting requires correlating heterogeneous security data with grid-side contextual evidence. To address class imbalance and cross-domain heterogeneity, this study proposes a heterogeneous feature group integration framework for intrusion detection with grid cybersecurity data. Four semantic feature subspaces are constructed symmetrically: network behaviour, power operation context, zone-derived communication/event topology, and system operation state, ensuring equal structural footing for subsequent modality-specific encoding. Transformer-based encoders model temporal dependencies in network, physical, and system state modalities, while a graph neural network encodes topology-related structural information. The resulting embeddings are integrated by a late fusion classifier for multiclass attack identification; the fusion process treats each feature group symmetrically at the decision level, without imposing a priori dominance among modalities. In the main run, the full model achieves an accuracy of 0.944, a macro F1 score of 0.891, a weighted F1 score of 0.937, a macro precision of 0.929, and a macro recall of 0.878. The corresponding balanced accuracy is 0.878, and the multiclass MCC is 0.924. Class-wise results show reliable performance on Benign, Scan, WebAtk, DDoS, DoS, and Backdoor classes, while Ransomware remains difficult and is frequently confused with WebAtk. Specifically, the Ransomware recall is 0.27, with most errors assigned to WebAtk. Modality analysis further indicates that modality contribution is class dependent: some feature groups have limited standalone discriminative power but provide complementary evidence after fusion. This finding highlights an inherent asymmetry in class-wise utility, which we counterbalance by employing both macro and weighted metrics, offering a symmetric evaluation lens that accounts for both minority and majority classes. These results show that grid-oriented intrusion detection benefits from decision-level integration of heterogeneous feature groups and imbalance-aware evaluation, where symmetric treatment of feature subspaces and evaluation perspectives jointly enhances robustness. Full article

Objectives: This study aimed to examine the intra- and inter-rater reliabilities of panoramic ultrasonographic measurements of the pectoralis minor length (Pm_L), pectoralis minor thickness (Pm_T), and pectoralis minor cross-sectional area (Pm_CSA). Methods: Twenty young male participants aged 19–35 years were recruited for this study. All participants had asymptomatic rounded shoulder posture (RSP), with a mean acromion-to-table distance of 59.316 ± 9.631 mm. Longitudinal panoramic scans of the Pm of the dominant arm were acquired in three trials of four measurements for all participants. The first two measurements were examined by the same rater on the same day, and the third measurement was performed by a different rater. The same procedure was carried out by the first rater after 30 days for inter-day reliability. Intraclass correlation coefficients (ICCs), standard errors of measurement (SEMs), and minimal detectable changes (MDCs) were calculated to estimate the reliabilities. Results: The intra- and inter-rater reliabilities of Pm_L, Pm_T and Pm_CSA were excellent (all ICC > 0.932; all SEM < 7.4%; all MDC < 20%). Bland–Altman plots demonstrated good agreement with nearly zero mean differences between measurements. Conclusions: Using panoramic ultrasonography to measure Pm_L, Pm_T and Pm_CSA may be considered a reliable method with excellent inter-day, intra- or inter-rater reliabilities. Full article

Causality extraction is an important task in natural language processing, yet it remains underexplored in informal Arabic social media text, particularly in dialectal contexts. This study investigates causal-reason extraction from Saudi Arabic tweets related to sick-leave requests. A gold-standard dataset was annotated for multiple causality-related tasks, including cause-presence detection, cause-span extraction, cause-category classification, causal-marker detection, and causal marker text identification. The study compares two modeling paradigms: fine-tuned BERT-based models, represented by SaudiBERT and AraBERT, and prompting-based large language models (LLMs), represented by GPT-4.1-mini and Gemini-2.5-flash. The descriptive analysis showed strong class imbalance, substantial implicit causality, and uneven cause-category distributions. Results showed that SaudiBERT generally outperformed AraBERT when macro-level and minority-class performance were considered. Among LLMs, Gemini-2.5-flash achieved the strongest overall performance, particularly under natural 10-shot single-tweet prompting, while balanced few-shot prompting improved macro-F1 for cause-category classification. However, step-wise prompting did not consistently improve performance and may have introduced error propagation. Overall, the findings show that causality extraction in informal Saudi Arabic remains challenging, especially for implicit causal expression. The study highlights the complementary strengths of dialect-specific transformers and LLM-based prompting for Arabic causality extraction. Full article

Background: Multiple sclerosis (MS) lacks a single invariant phenotypic core. Patients accumulate heterogeneous combinations of sensory, motor, cognitive, and autonomic impairments over time, reflecting lesions that are disseminated in time and space. Standard scales such as the Expanded Disability Status Scale (EDSS) distribute disability across functional systems, but do not explicitly represent MS phenotype as a mixture of latent symptom modules. Methods: We analyzed 4617 de-identified neurology progress notes from 577 patients with MS at a single academic medical center. A large language model (GPT-5.2) categorized each note with respect to 17 non-mutually-exclusive neurological phenotype features, and note-level features were aggregated to patient-level binary vectors. Non-negative matrix factorization (NMF) was applied to generate three-, four-, and five-module solutions. For each rank, we computed approximate variance captured, relative reconstruction error, and module-level feature loadings. In the preferred four-module solution, we derived patient-level module percentages, identified highly dominant (≥55%) and archetypal (≥70%) module profiles, and quantified admixture using Shannon entropy and the effective number of modules. Results: Three-, four-, and five-module NMF solutions showed similar approximate variance captured (52.7–54.3%) and reconstruction error (0.47–0.53), but the four-module solution provided the clearest clinical interpretation. The four latent modules were sensory-visual-pain, ataxic-spastic-falls, cognitive-psychologic-fatigue, and autonomic-bladder-bowel, aligning closely with established functional systems in MS. Most patients exhibited admixed phenotypes, with module entropies ranging from 0 (single-module dominance) to 1.386 (equal mixture) and effective modules spanning approximately 1 to 4. Using pre-specified thresholds, 154 patients (26.6%) were highly dominant in a single module and 72 (12.5%) were archetypal; these purer phenotypes were most often in the sensory-visual-pain module. Conclusions: MS phenotypic diversity in routine clinical practice can be parsimoniously represented as mixtures of four latent symptom modules rather than as positions along a single severity axis. Most patients show substantial admixture of sensory, motor, cognitive, and autonomic involvement, but a minority exhibit relatively pure or strongly dominant module patterns. This modular representation provides an interpretable framework for quantifying MS phenotype and for generating testable hypotheses about MS subtypes whose biological relevance remains to be established. Full article

The decentralized deployment of photovoltaic (PV) systems in urbanized polluted mountainous basins faces unique challenges due to complex topography, persistent cloud cover and winter smog conditions. This study quantifies the atmospheric impact of localized winter haze/smog and Saharan dust intrusions on PV performance in the intermontane basin of Ioannina, NW Greece. By integrating a Digital Twin (DT) methodology with real energy production data, two PV plants were evaluated, a ground-based and a rooftop installation, to isolate the energy deficits caused by aerosol attenuation. The DT model demonstrated high accuracy (R² = 0.847) against actual power generation data for Koutselio and R² = 0.865 for Mpafra PV plants, while MBE was near zero for both sites (−0.008 kWh and −0.139 kWh, respectively). Error analysis revealed that the highest modeling discrepancies occurred during scattered clouds and intense winter haze conditions, primarily due to low spatial resolution of CAMS that fails to adequately capture localized biomass burning (BB) events. Despite the reduction in direct sunlight during extreme winter BB events, results indicate that the overall energy loss is mild. This operational stability is primarily due to the ability of c-Si modules to effectively utilize near-infrared radiation, which penetrates the low-level haze layer, alongside the thermal efficiency gains provided by low early-morning temperatures. Crucially, the installation geometry may influence system vulnerability. Direct comparisons revealed a minor power deviation of −4.8% for the ground-based Koutselio plant, while for the Mpafra site, there was a +3.2% production surplus likely linked to the high sky-view factor the rooftop installation has, which manages to capture isotropic diffuse irradiance. However, the low CAMS resolution may misclassify the haze events within the basin, further contributing to these discrepancies. On the contrary, Saharan dust intrusions caused broadband light attenuation, dropping the power production significantly on both installations. Ultimately, this research provides critical insights into the resilience of solar systems under strong air pollution events within polluted valleys in Southern Balkans, highlighting the connection between panel design and atmospheric attenuation. Full article