Search Results (458)

This paper presents a machine learning-based decision support system for multiclass defect detection, utilizing exclusively heterogeneous legacy process data to minimize manual inspection time in foundries. Validated on 51,377 products across 193 defect categories, the methodology resolves structural data inconsistencies through k-nearest neighbor (kNN) imputation and piecewise winsorization. A multi-stage feature selection cascade, incorporating variance thresholding, correlation filtering, and Random Forest Feature Importance (RFFI), reduces the feature space from 139 to 78 process-critical variables. Following Synthetic Minority Over-sampling Technique (SMOTE)-based class balancing, five classifiers were benchmarked via 10-fold cross-validation and optimized using the macro-averaged F₃-score to mathematically penalize undetected defects. Light Gradient Boosting Machine (LightGBM) and Random Forest (RF) provided superior predictive baselines. To enforce strict zero-defect constraints, an asymmetric risk function shifted decision boundaries, enabling the risk-calibrated LightGBM model to reduce manual inspection volume by 9.72% with zero defect escapes. For resolving conflicting predictions, multi-algorithm decision fusion was implemented. By statistically evaluating the joint probabilities of the base models’ post-calibration outputs, a Naive Bayes Stacking meta-classifier effectively neutralizes single-algorithm inductive biases. Ultimately, synthesizing these F₃-optimized, risk-calibrated base models via meta-learning successfully isolated true defect-free components, maximizing the final inspection time reduction to 14.82% while strictly maintaining zero defect escapes. Full article

Against the background of global climate change and rapid urbanization, the frequency of extreme hydrological events has increased markedly, and hydrological and geological hazards induced by extreme heavy rainfall have occurred frequently. Considering the joint occurrence characteristics of heavy rainfall hazards, statistical methods are used in this study to develop a mathematical model for calculating hazard–risk combinations and disaster-triggering thresholds that integrates copula functions and the Kendall return period. First, a bivariate joint-distribution probability model is established using copula functions. Then, the Kendall return period under the same recurrence level is calculated to infer combined hazard thresholds for multiple disaster-causing factors. Taking rainfall-induced flash floods and landslides as examples, combined hazard thresholds are calculated. Rainfall and discharge measured at the Heyuan hydrological station in the period from 2001 to 2020 are taken as a case study. The results show that (1) the optimal bivariate copula function for annual maximum 1-day rainfall and annual maximum discharge is the Clayton Copula; (2) similarly, the optimal bivariate combined probability function for rainfall-induced landslides is the Gumbel Copula; and (3) the combined hazard thresholds of multiple disaster-causing factors for rainfall-induced flash floods and landslides differ substantially from the thresholds based on single disaster-causing factors. Taking the return period of 100 years as an example, the univariate thresholds of discharge and rainfall are 8530.38 m³/s and 217.82 mm, while the combined thresholds decrease to 5793.10 m³/s and 175.22 mm. Meanwhile, the univariate thresholds of rainfall and landslide probability are 195.02 mm and 0.82, but the combined thresholds decrease to 170.81 mm and 0.78, respectively. In other words, conventional single-factor threshold calculation methods can severely underestimate the actual hazard level, and coupling among multiple factors can induce a significant disaster superposition and amplification effect. This study reveals the risk-coupling mechanism of heavy rainfall hazards and provides technical support for the accurate identification, prediction, and early warning of rainfall-induced flash floods and landslides under climate change. Full article

History matching traditionally assumes a spatially uniform clay adsorption coefficient Kd, overlooking the fractal heterogeneity of clay mineral distributions in real reservoirs. This study generalizes the classical advection–dispersion framework by treating Kd as a non-uniform field whose logarithmic fluctuations follow fractional Brownian motion (fBm) statistics with Hurst exponent H, and by coupling the resulting heterogeneous retardation to a fractional advection–dispersion equation through the continuous-time random walk bridge β = 2/(2H + 1). A Karhunen–Loève (K-L) expansion compresses the field into a small set of spectral modes, and an adjoint-gradient inversion recovers the clay field from tracer data. The main operational limitation of the K-L approach is that sub-grid variability below the retained spectral cutoff is smoothed; within this boundary the inversion is efficient and numerically stable. Two independent validation paths are pursued. First, a cross-fitting experiment shows that when the true physics is sub-diffusive (β = 0.85), the classical advection–dispersion equation yields a negative coefficient of determination (R2 = −0.927), whereas the correctly specified fractional model fits the data, establishing fractional derivatives as physically necessary rather than mathematically optional. Second, the publicly available EGS Collab DNA-tracer dataset independently confirms statistically significant clay adsorption (retardation R = 1.250 ± 0.009, p < 0.0001). The recovered clay field attains pixel-scale accuracy R2η = 0.844 (Spearman ρ = 0.923) and R2 > 0.94 at reservoir-relevant scales of five cells or coarser. These results indicate that fractal-aware clay-field inversion provides a physically verifiable, rather than purely empirical, pathway for history matching, with direct implications for field-scale contaminant transport prediction and environmental subsurface management, and the evaluation of flue gas enhanced thermal recovery in thin-layer heavy oil reservoirs. Full article

Background/Objectives: Adolescent pregnancy is associated with an increased risk of adverse perinatal outcomes; however, its hematological characteristics remain incompletely understood. The aim of this study was to compare antepartum and postpartum complete blood count (CBC) parameters together with hemogram-derived inflammatory indices in adolescent and adult pregnancies and to investigate their associations with maternal and neonatal outcomes. Methods: This retrospective comparative cohort study included 428 singleton pregnancies comprising 205 adolescent (14–17 years) and 223 adults (18–42 years). Antepartum blood samples were obtained at admission for delivery, whereas postpartum samples were collected approximately six hours after delivery according to the institutional practice. CBC parameters (including neutrophil, lymphocyte, monocyte, and platelet counts) were recorded, and hemogram-derived inflammatory indices, including neutrophil-to-lymphocyte ratio (NLR), Platelet-to-lymphocyte ratio (PLR), systemic immune-inflammation index (SII), systemic inflammation response index (SIRI), aggregate index of systemic inflammation (AISI), and systemic inflammation-monocyte index (SIMI), were subsequently calculated. Comparisons were adjusted for gestational age at sampling and parity using a prespecified non-redundant marker set, and multiple testing was controlled using the Benjamini-Hochberg false discovery rate procedure. Multivariable logistic regression and machine-learning analyses were performed to evaluate associations with adverse maternal and neonatal outcomes. Results: Adolescent pregnancies demonstrated significantly higher antepartum neutrophil and monocyte counts than adult pregnancies (q < 0.05), and these differences remained significant after adjustment for gestational age at sampling and parity (both q < 0.01), whereas differences in hemoglobin and platelet counts were no longer significant (both q = 0.138). Low birth weight (10.2% vs. 4.5%), neonatal intensive care unit admission (15.1% vs. 8.5%), composite adverse outcome (22.9% vs. 15.2%), and birth weight below the 10th percentile (14.7% vs. 8.0%) were numerically more frequent among adolescents but were not statistically significant after false discovery rate correction (all q > 0.05). Across 30 prespecified marker–outcome associations, none remained significant after false discovery rate correction. Antepartum neutrophil count was not independently associated with low birth weight after multivariable adjustment (OR 1.233, 95% CI 0.858–1.737; p = 0.248, q = 0.413). Exploratory machine-learning models demonstrated only modest discrimination, with the highest cross-validated AUC of approximately 0.60, and did not outperform conventional regression models. Conclusions: Adolescent pregnancy was associated with higher neutrophil and monocyte counts during both the antepartum and postpartum periods, reflecting modest differences in hematological profile rather than an independently predictive inflammatory profile. These hematological differences were not independently associated with adverse maternal or neonatal outcomes after adjustment for confounding factors and correction for multiple testing. Hemogram-derived inflammatory indices should therefore be interpreted as mathematical transformations of routine CBC parameters rather than biologically independent biomarkers. Because blood samples were obtained at admission for delivery, these findings characterize the peripartum inflammatory profile and should not be interpreted as evidence of early pregnancy risk prediction. Full article

Low-elevation Doppler wind lidar scans, known as arc scans, extend traditional vertical-profile measurements to horizontal, off-site locations. This technique is designed to enable measurements at multiple distant locations relative to the instrument. Arc-scan methods have been widely utilized in wind energy applications and are also a promising tool for environmental monitoring for hazard assessment. This method introduces specific challenges absent in traditional vertical profile scans. The limited scan angle restricts the number of wind orientations available for reliable vector extraction. A reliable method of estimating the uncertainty intervals for retrieved wind speed and direction in operational configurations is thus of interest. Here, we developed a closed-form statistical expression for evaluating wind speed and direction prediction intervals. Because rapid-update operational scenarios (such as real-time dispersion modeling) yield a limited number of scans, the framework is specifically designed to remain mathematically robust and computable using only diagonal variance terms, bypassing the need for numerically unstable cross-covariance matrices. The expression was tested against a lidar and sonic anemometry measurement campaign. The wind-arc alignment emerges as a major influencing parameter impacting uncertainty of both direction and speed retrievals. Conclusions regarding scan parameters and siting considerations are drawn. Full article

Objective: To analyze associations between 24-hour movement behaviors (and guideline adherence) and anthropometric and body composition indicators in Ecuadorian preschoolers using data from the International Study of Movement Behaviors in the Early Years (SUNRISE) pilot study. Methods: This secondary analysis utilized data from the primary study conducted in 108 children aged 3–4 years (57.4% girls; 50.9% urban). Anthropometry (weight, height, and body mass index [BMI]) was assessed using standardized protocols during the pilot study, whereas body composition indicators (fat-free mass [FFM], fat mass [FM], and body fat percentage) were estimated using a validated anthropometric prediction equation. Movement behaviors were assessed via triaxial accelerometry (ActiGraph GT3X+ or GT3X-BT, Pensacola, FL, USA) over seven days, complemented by parent-reported questionnaires. Exploratory unadjusted correlation analyses were performed using Spearman’s rho (ρ) for continuous/ordinal variables, and Pearson’s correlations (r), mathematically equivalent to point-biserial correlations (r_pb), for dichotomous–continuous associations. Results: In exploratory unadjusted analyses, device-measured longer sleep duration showed small-to-moderate inverse associations with weight (ρ = −0.33, p = 0.001), BMI (ρ = −0.28, p = 0.006), and FFM (ρ = −0.37, p < 0.001). Higher moderate-to-vigorous physical activity showed moderate positive associations with height (ρ = 0.32, p = 0.002), weight (ρ = 0.40, p < 0.001), and FFM (ρ = 0.44, p < 0.001). In complementary multivariable models, the associations between device-measured sleep or moderate-to-vigorous physical activity (MVPA) and the pre-specified anthropometric and body composition outcomes were attenuated after adjustment for age, sex, area of residence, and, where applicable, height, and were generally not statistically significant. Conclusions: In this exploratory secondary analysis of the SUNRISE Pilot Study Ecuador, unadjusted analyses identified associations between specific 24-hour movement behaviors, particularly sleep duration and MVPA, and anthropometric and body composition indicators. However, these associations were attenuated after covariate adjustment. The findings should therefore be interpreted as hypothesis-generating and require confirmation in larger longitudinal studies. Full article

Supervised fine-tuning is often discussed in terms of data volume and target quality, but the format of the supervision itself can change the reasoning strategy a model expresses in its outputs. We study this effect on a six-category reasoning benchmark spanning symbol transformation, text decryption, bit manipulation, gravitational constant estimation, numeral conversion, and unit conversion (drawn from the NVIDIA Nemotron Model Reasoning Challenge). Using NVIDIA Nemotron-3-Nano-30B-A3B with matched LoRA training settings, we compare three symbol-supervision formats: verbose English rule descriptions, compact family tags, and compact formula notation. We hypothesize that supervision renderings bias token-level reasoning priors, and that these priors transfer across task boundaries in multi-task SFT. In the canonical strict-rescore inventory, the best compact tag and formula checkpoints are statistically equivalent in aggregate within a pre-specified ±4-point margin: K8A-800 reaches 72.3% strict-scored overall accuracy and K8B-700 reaches 71.2% (TOST p = 0.003). Compact tags nevertheless provide a cleaner behavioral profile: an earlier K8A-400 checkpoint reaches 66.4% overall, 98.7% gravity accuracy, and 36.9% bit accuracy without the same contamination signatures. In contrast, verbose English rule descriptions are associated with heuristic parroting, with up to 57% of symbol failures at audited verbose checkpoints collapsing to a single remove-operator template, while formula notation is associated with cross-category contamination: numeric-looking predictions appear more often in text decryption (higher at five of six matched training steps under the canonical seed; matched-step means 15.8 vs. 11.7 numeric predictions per 157 text rows), and gravity failures at a representative K8B formula checkpoint shift toward shortcut stubs and explicit g = 9.8/9.81 fallbacks. We further show that checkpoint selection and strict evaluation auditing materially change branch decisions. Across three training seeds, neither compact format shows a consistent aggregate advantage, while the contamination signatures are partly seed-specific: the gravity-shortcut severity difference persists but is not exclusive to the formula branch, and the numeric–text signature does not reproduce under reseeding. These results support treating supervision format as a first-class hyperparameter for multi-task reasoning SFT in large language models—at least in this benchmark-and-model setting—rather than a mere rendering detail. Because such symbolic and procedural reasoning tasks recur in domains including cybersecurity, mathematics, and code generation, the same formatting choices plausibly shape the policy that any later reinforcement-learning stage would inherit, which we flag as future work. Full article

Smart meters are the core terminals for distribution network data acquisition in smart cities, yet their collected data commonly suffer from quality issues caused by harsh operating environments, communication failures, hardware degradation, and human factors. This paper presents a systematic review of anomaly detection and data repair methods for smart meter data based on a critical analysis of many publications. First, we characterize five typical anomalies—sudden jumps, reading stagnation, reverse readings, pulse spikes, and gradual drifts—from physical root causes to data manifestations and provide unified mathematical definitions with explicit traceability to the existing literature. Additional anomaly types including meter replacement jumps, data duplication from retransmission, complete missing segments, and timestamp errors are also discussed to present a more complete picture of operational data quality challenges. Second, existing anomaly detection methods are systematically reviewed and classified into four categories—statistical, machine learning, deep learning, and dedicated time-series methods—with representative studies, quantitative performance metrics, and scenario-specific applicability examined for each. Third, data repair approaches are reviewed across four categories—traditional interpolation, matrix completion, generative models, and time-series prediction—with systematic comparison of their accuracy and limitations across different anomaly types and durations. Based on the synthesized evidence, we identify three cross-cutting structural limitations that persist across method categories: the performance ceiling of data-only detection without physical constraint embedding, the open-loop architecture that separates detection from repair and allows error propagation, and the exclusive reliance on statistical error metrics that fails to distinguish physically plausible repairs from those violating conservation laws. To address these gaps, we discuss a physics-guided integrated framework incorporating physical constraint embedding, joint anomaly diagnosis, scenario-adaptive repair, and posterior verification as a promising forward-looking direction. Finally, open challenges and future research directions are outlined, including parameter adaptation in unlabeled scenarios, multi-source data fusion for physical disambiguation, new power system extensions, explainable AI integration, edge-computing deployment, and standardized benchmark development. This review provides a comprehensive theoretical reference and technical roadmap for smart meter data quality research in the context of smart city energy systems. Full article

The Lorenz number is a critical parameter for separating the electronic and lattice contributions to thermal conductivity in thermoelectric materials through the Wiedemann–Franz law. However, the classical Sommerfeld approximation often fails to accurately represent the temperature-dependent transport behavior of BiSbTe-based thermoelectric materials. This study presents a systematic analysis of the temperature dependence of the Lorenz number using experimental data compiled from eleven independent studies. A unified database was established through literature review, data extraction, normalization, and statistical analysis. Linear, exponential, and quadratic regression models were evaluated to identify the mathematical representation that best describes the reported behavior, and a Processing Complexity Index (PCI) was introduced to examine potential relationships between fabrication-route complexity and the degree of nonlinearity. The compiled datasets consistently exhibited an overall increase in the Lorenz number with temperature, although noticeable variability in magnitude and curvature was observed among studies, reflecting differences in material composition, processing routes, and experimental conditions. While the quadratic model generally achieved the best statistical performance, linear and exponential models provided comparable fits for some datasets, indicating that no single functional form is universally optimal. The proposed correlations provide a statistically representative framework for improving thermal conductivity decomposition and thermoelectric characterization within the investigated temperature range (300–500 K). Nevertheless, the correlations are constrained by the scope of the literature-derived database and should not be interpreted as universally applicable predictive models. Full article

Constructed wetlands (CWs) are widely used as an ecological technology for wastewater treatment. However, the accumulation of microplastics (MPs) in their substrates may impair long-term performance and threaten the operational sustainability of these nature-based treatment systems. To elucidate the transport behaviour of MPs in wetland substrates, this study developed a numerical model based on the lattice Boltzmann method (LBM) to simulate advection, hydrodynamic dispersion, and reversible first-order adsorption/desorption of MPs in saturated porous media. The model incorporates a salinity-dependent non-linear attachment rate coefficient, which captures the compression of the electrical double layer and the enhanced attachment efficiency with increasing salinity. Pore-scale flow is solved using the LBM with an Ergun-type drag term to represent the resistance of the porous matrix. The model was validated against experimental breakthrough curves from column studies using quartz sand and coastal wetland soils under five salinity levels (0–35 PSU) reported in the literature. Quantitative validation yielded coefficients of determination (R²) ranging from 0.782 to 0.960 (RMSE = 0.024–0.045) for calibration cases and 0.741 to 0.946 (RMSE = 0.027–0.048) for independent validation cases across both substrates, excluding the soil cases at 3.5 and 35 PSU. Here, both observed and simulated effluent concentrations were identically zero, resulting in the statistically forced R² = 1.000 and RMSE = 0, which are mathematical artefacts rather than indicators of predictive performance. The simulations reproduce the observed reduction in peak relative concentration by over 50% in sand and near-complete retention (C/C₀ ≈ 0) in soil at high salinities (3.5 and 35 PSU). Results demonstrate that the model successfully reproduces the differences in MP breakthrough behaviour across different substrate types and salinity levels. By linking salinity-enhanced retention to the risk of irreversible clogging and shortened wetland lifespan, the model provides a predictive tool for evaluating the sustainability of CWs under saline stress. This study offers a scientific basis for optimizing hydraulic management (e.g., flushing strategies) to mitigate microplastic pollution and enhance the long-term sustainability and resilience of constructed wetlands in coastal and saline environments. Full article

Conformal prediction has emerged as a principled framework for uncertainty quantification in computer vision, offering rigorous finite-sample coverage guarantees. However, its application in object detection has remained largely confined to localization, as standard inference codebases typically yield only top-1 class scores, precluding full class-label conformalization. In this work, we bridge this gap by adapting four architecturally diverse detectors—Faster R-CNN, RetinaNet, YOLO11, and RT-DETRv2—to facilitate the extraction of comprehensive per-class score vectors and the estimation of background confidence in the absence of native background modeling. Leveraging these adapted architectures, we implement inductive conformal prediction (ICP) using five distinct nonconformity functions: Top-K, Adaptive Prediction Sets (APS), Hinge, Margin, and Brier score. Our framework is rigorously benchmarked across a curated 20-class subset of MS-COCO and two specialized parasite egg datasets (AI4NTD P1.5v2 and Chula-ParasiteEgg-11). In addition, a Naive cumulative-threshold method is included as a baseline for comparison with APS, given their comparable mathematical formulations. Across target coverage levels of 90%, 95%, and 99%, the conformalized models consistently achieved nominal coverage with only minor finite-sample deviations. Hinge and APS exhibited an optimal balance between statistical coverage and prediction-set efficiency, whereas Margin and Brier scores tended toward larger sets under high data complexity and strict coverage requirements. With empty prediction sets maintained below 0.1%, our findings establish ICP as a robust and adaptable paradigm for trustworthy class-label uncertainty estimation, particularly within safety-critical workflows such as automated parasite diagnostics. Full article

Aiming at the sensitivity of the droplet nucleation rate in the wet steam non-equilibrium condensation flow model to the value of the surface tension correction coefficient (NBTF), and to avoid the subjectivity and uncertainty in its selection, this study calibrated the quantitative relationship between the optimal NBTF values and the inlet pressure and inlet superheat degree across 81 operating conditions. This was achieved by comparing numerical simulations with experimental data from multiple sets of wet steam condensation flow experiments documented in the literature. Subsequently, a bivariate regression equation capable of directly predicting NBTF was established using the mathematical method of statistical regression analysis. The reliability of the equation was verified using the results of an independent experiment (the Dykas cascade experiment). Furthermore, taking the last stage of a certain marine steam turbine as the research object, the established equation was applied to conduct a numerical study of wet steam non-equilibrium condensation flow. The influence of the NBTF value on the nucleation process, droplet distribution, and flow loss was quantitatively analyzed. The results indicate that within a specific range (inlet pressure 15~280 kPa, inlet superheat degree −12~50 K), NBTF shows a significant positive correlation with inlet pressure and a significant negative correlation with inlet superheat degree. The value of NBTF directly affects the prediction results of droplet nucleation rate, particle size distribution, and condensation zone location within the last stage of the steam turbine. This research provides a data-based, reproducible prediction tool for determining the NBTF value in the numerical simulation of wet steam condensation flow and also offers a quantitative reference for the anti-water erosion design and operational optimization of the last stage of steam turbines. Full article

Background: In our recent research, we established a strong and statistically significant association between the highest quartile of lipoprotein(a) [Lp(a)] and mildly reduced estimated glomerular filtration rate (eGFR). Comparisons with similar studies were hindered by varying Lp(a) analytical methods and different eGFR equations. We analyzed how replacing the diagnostic standard CKD-EPI 2021 equation with alternatives (CKD-EPI 2009, CKD-MDRD, CKD-EKFC) affects predictive models under comprehensive demographic and clinical confounder adjustments. Methods: We calculated eGFR for 310 adults using the four equations. Creatinine and Lp(a) were measured via IFCC-recommended methods. The cohort was divided into a main (eGFR 60–80 mL/min/1.73 m²) and a control group (eGFR > 80 mL/min/1.73 m²). Multivariable logistic regression and Area Under the Curve (AUC) metrics evaluated model performance. Results: Alternative equations systematically underestimated eGFR, artificially increasing the reduced filtration group. After full multivariable adjustment, only CKD-EPI 2021 preserved the independent association between the highest Lp(a) quartile and mildly decreased eGFR (OR = 2.65, p = 0.008), achieving an AUC of 0.732. Conversely, CKD-EPI 2009 lost significance; CKD-MDRD exhibited mathematical instability from control group depletion, and EKFC showed severe over-adjustment when demographic covariates were included. Conclusions: Older equations and the age-embedded EKFC equation may influence regression models, potentially obscuring the true biomarker associations. CKD-EPI 2021 demonstrated the highest precision, successfully isolating physiological aging and suggesting an independent association between high Lp(a) and mildly reduced eGFR, irrespective of metabolic and vascular confounders. Full article

This study proposes a mathematical framework for predicting complete response-time distributions associated with parametric video-game configurations. Each configuration is encoded as a point in a normalized metric space, and the statistical descriptors of its response-time distribution (median, mean, selected quantiles, interquartile range, and Skewness) are treated as real-valued Lipschitz functions on that space. Predictions for unseen configurations are obtained through McShane–Whitney extension formulas, which provide geometrically controlled upper and lower bounds compatible with the empirical Lipschitz regularity of the observed data. To handle the sequential incorporation of new observations, a regularization mechanism is introduced that replaces raw descriptors violating Lipschitz continuity constraints with a convex combination of the observed value and a weighted historical estimate. In the extended version of the method, the coefficient of this combination is selected through a one-pass online Q-learning-inspired procedure that selects, for each descriptor and instability regime, a data-dependent trade-off between fidelity and geometric regularity. The final output is a continuous Log-Normal density fitted by nonlinear least squares to the predicted descriptors, together with a Wasserstein-type uncertainty band derived from the Lipschitz bounds. The framework is validated on a controlled experiment with 24 participants across 20 game levels. Results show that the predicted distributions shift systematically with the input configuration and that, in the illustrative comparison, the adaptive mechanism produces feature-specific smoothing decisions that differ from those obtained with a fixed coefficient. The method also provides interpretable predictions from small experimental datasets without requiring fully data-driven models. Full article

This study investigates the critical relationship between machining-induced surface integrity and the effectiveness of subsequent anti-corrosion protection for high-strength Al 7136-T76511 aerospace alloy. Given the alloy’s susceptibility to exfoliation corrosion, ensuring high-quality surface substrates for protective coatings is paramount. The research aims to model the influence of end milling parameters—cutting speed, depth of cut, and feed per tooth— on surface roughness to establish a topographical risk prognosis framework for subsequent coating vulnerability. A comprehensive full-factorial experimental design involving 150 distinct cutting regimes was evaluated on a CNC machining center. Statistical analysis using ANOVA showed that cutting speed is the most significant factor, contributing 83.89% to the variance of longitudinal Ra. A critical resonance zone was identified between 570 and 610 m/min, where the model predicts high instability and surface integrity degradation. The developed mathematical models achieved high precision, with coefficients of determination (R²) ranging between 85% and 88%. The research identifies a critical “danger zone” of dynamic instability between 570 and 610 m/min, where resonance significantly increases data dispersion (standard deviation = 0.112 µm compared to 0.051 µm in stable regimes). Findings demonstrate that even when average R_a values remain within industrial limits, vibration-induced micro-cracks, and severe chatter marks function as geometric precursors that theoretically lower the structural barrier efficiency of subsequent protective films. This study establishes that prioritizing process stability over nominal roughness minimization is essential for the structural integrity of critical aerospace components. Full article