This study presents a binary classification model for green coffee beans of the Arabica variety from the former V2 architecture. Two datasets were used, the public USK-COFFEE dataset of Indonesian origin and a proprietary dataset captured under controlled conditions. Two training strategies were evaluated: sequential transfer learning and unified training, with the latter achieving a validation accuracy of 98.30%. After hyperparameter optimization, the model reached 100% accuracy on a test set of 150 images and 93% accuracy on an external generalization set of 400 images with varying lighting conditions and backgrounds. Model interpretability was validated using Grad-CAM, demonstrating that the network focuses on actual defective regions rather than backthat performance degradation in unconstrained scenarios is mainly due to sensitivity to noise and exclassification.
Optimizing resource use in remote operational fields is a key strategy for enhancing efficiency and reducing operating costs. This study adopted a descriptive-experimental design with a quantitative orientation to analyze the energy consumption of a remote oil transportation camp, utilizing historical records collected since 2021. The primary objective was to achieve an energy consumption reduction of 15-30%. Using RETScreen, improvements in energy systems were simulated, resulting in an initial annual electricity consumption of 271,143 kWh and associated costs of USD 27,454. With the proposed photovoltaic system, designed to optimize energy use through advanced energy management technologies, annual consumption decreased to 185,878 kWh/year, substantially reducing dependence on the national interconnected grid and on polluting energy sources. Energy efficiency improved by 14.3% in air-conditioning systems and by 38.6% in electrical systems such as pumps, compressors, and lighting, resulting in an ity in remote oilfield operations.
This study presents an automated classification system to prioritize electrical service complaints at CNEL EP. A total of 143,113 real records were processed through data cleaning, missing-value imputation, and the engineering of predictive variables reflecting complaint urgency and recurrence. Based on these criteria, the target variable "Priority" was defined to distinguish high-priority from normal complaints. Supervised learning models, specifically Decision Tree and Random Forest, were then trained using one-hot encoding and cross-validation. Random Forest delivered the best performance, achieving 91% accuracy and an AUC-ROC of 0.89. These results indicate that the proposed system can significantly improve technical resource allocation and reduce response times for electrical complaints. Moreover, the study demonstrates the feasibility of integrating machine-learning techniques into the operational management of electric distribution companies, enabling future enhancements and real-time deployment.
Accurate identification of upper extremity fractures is essential for timely and reliable diagnosis in emergency medical settings. This study evaluates and compares the performance of three pre-trained deep learning architectures: EfficientNet-B4, ResNet-50, and ConvNeXt-Large, applied to the automatic classification of bone fractures in radiographic images from the MURA repository, encompassing seven anatomical regions. Advanced image preprocessing techniques, including Unsharp Masking and Contrast-Limited Adaptive Histogram Equalization (CLAHE), were employed in conjunction with data normalization and balancing strategies. The models were trained in two experimental setups: a binary classification distinguishing between "fracture" and "non-fracture" images, and a multiclass configuration identifying 14 distinct fracture types. Performance evaluation using F1-Score, sensitivity, accuracy, and ROC-AUC metrics demonstrated that ConvNeXt-Large achieved the highest overall results, reaching accuracies of 99.0% in binary classification and 99.4% in multiclass classification. These findings position ConvNeXt-Large as a highly promising tool for supporting early and precise fracture diagnosis.
This study applies a machine learning technique, specifically Convolutional Neural Networks (CNNs), to predict pollutant emissions resulting from failures in actuators and components of Otto engines. The work addresses the current lack of non-intrusive methods that exploit signals already available in the vehicle to estimate, with high accuracy, emissions associated with failures in the injection, ignition, and air intake systems. Concentrations of CO (% carbon monoxide), CO2 (% carbon dioxide), HC (unburned hydrocarbons, ppm), and O2 (% oxygen) are quantified by analyzing the Manifold Absolute Pressure (MAP) sensor signal under a rigorous sampling and signal-processing protocol. Optimal features are extracted from the MAP signal based on their informational relevance and discriminative capacity. These features are obtained through spectrographic transformation, enabling the construction of a robust database. The resulting dataset serves as an effective input for CNN training, achieving emission prediction errors below 1%.
En este estudio se implementa una técnica de aprendizaje automático, concretamente redes neuronales convolucionales (CNN, por sus siglas en inglés), con el objetivo de predecir las emisiones contaminantes producidas por fallos en actuadores y componentes de motores Otto. Así pues, el problema de investigación abordado en este trabajo es la ausencia de métodos no intrusivos que, a partir de señales ya disponibles en el vehículo, permitan estimar con alta precisión las emisiones asociadas a fallas en sistemas de inyección, encendido y admisión de aire. Se cuantifican los niveles de CO (monóxido de carbono, %), CO2 (dióxido de carbono, %), HC (hidrocarburos no quemados en ppm) y O2 (oxígeno, %) a partir del análisis de la señal proveniente del sensor MAP (Manifold Absolute Pressure). Para ello, se adopta un protocolo riguroso de muestreo y procesamiento de la señal. La extracción de características óptimas del sensor MAP se basa en su relevancia informativa y capacidad de discriminación, determinadas mediante la transformación espectrográfica de la señal, lo que permite construir una base de datos robusta. Esta base sirve como entrada eficaz para el entrenamiento de la CNN, con la cual se logra un error de predicción inferior al 1 %.
This article presents a study on the semantic segmentation of defects in crystalline-silicon photovoltaic cells using U-Net–based models trained on electroluminescence (EL) images. The dataset combines laboratory-acquired images with a publicly available benchmark, both manually annotated to identify cracks, dark zones, and collector-bar discontinuities. Eight model variants were trained with controlled variations in input resolution, encoder depth, and regularization strategies. Performance was assessed using per-class precision, recall, and F1-score, complemented by visual inspection through heatmaps and overlays and by expert validation. Segmentation was robust for defects with well-defined morphology, such as dark zones and busbars; however, cracks remained more difficult to detect due to their sparse pixel representation and irregular geometry. Alternative architectures (U-Net++ and MAU-Net) were also evaluated but did not yield meaningful improvements over the optimized U-Net configuration. Overall, the results support the use of this approach for automated inspection under controlled conditions, while highlighting the need for future adaptation to more diverse operational scenarios.
Esta investigación presenta un modelo de clasificación binaria para granos de café verde de la variedad arábico procedentes de la región de Loja, Ecuador, basado en la arquitectura Swin Transformer V2. Se emplearon dos fuentes de datos, el conjunto de datos público USK-Coffee, de origen indonesio, y un conjunto de datos propio capturado bajo condiciones controladas. Se evaluaron dos estrategias de entrenamiento: transferencia secuencial y entrenamiento unificado, siendo este último el que alcanzó una precisión de validación del 98,30 %. Tras la optimización de hiperparámetros, el modelo logró una precisión del 100 % en un conjunto de prueba de 150 imágenes y del 93 % en un conjunto de generalización externo de 400 imágenes con condiciones variables de iluminación y fondo. La interpretabilidad del modelo se validó mediante Grad-CAM, evidenciando que la red enfoca su atención en zonas defectuosas reales. Un análisis de ablación mostró que la disminución de rendimiento en escenarios no controlados se debe principalmente a la sensibilidad al ruido y a la iluminación extrema. Como principales aportes, se destaca la creación de un conjunto de datos especializado y un modelo eficiente para la clasificación automática de café verde arábico.
La identificación precisa de fracturas en las extremidades superiores es fundamental para un diagnóstico oportuno en los entornos de urgencias médicas. Este estudio analiza y compara el desempeño de tres modelos de aprendizaje profundo preentrenados: EfficientNet-B4, ResNet-50 y ConvNeXt-Large, aplicados a la clasificación automática de fracturas óseas en radiografías del repositorio MURA, distribuidas en siete regiones anatómicas. Se aplicaron técnicas avanzadas de preprocesamiento digital, como Unsharp Masking y CLAHE, junto con estrategias de normalización y balanceo de datos. El entrenamiento se llevó a cabo en dos experimentos: uno binario, que clasifica entre “fractura” y “no fractura”, y otro multiclase, con catorce tipos de fracturas identificadas. La evaluación, mediante métricas como F1-Score, sensibilidad, exactitud y curvas ROC-AUC, reveló que ConvNeXt-Large alcanzó el mejor rendimiento, logrando una precisión del 99,0 % en clasificación binaria y del 99,4 % en la clasificación multiclase. Estos resultados posicionan a ConvNeXt-Large como una herramienta altamente prometedora para apoyar el diagnóstico temprano de fracturas óseas.
Optimizar los recursos en campos remotos es un criterio fundamental para la eficiencia y la reducción de costos operativos. El estudio tuvo un enfoque descriptivo-experimental, con orientación cuantitativa, en el que se analizó el consumo energético de un campamento remoto de transporte de petróleo, utilizando registros históricos desde 2021. El objetivo principal fue la reducción del consumo energético en un rango del 15 % al 30 %. Mediante el uso de RETScreen se simularon mejoras en los sistemas energéticos, obteniéndose como resultado un consumo eléctrico anual de 271 148 kWh/año, con gastos asociados de USD 27 454. Con la propuesta fotovoltaica, orientada a optimizar el consumo mediante tecnología de gestión energética, el consumo se redujo a 185 878 kWh/año, disminuyendo significativamente la dependencia del Sistema Nacional Interconectado y el uso de fuentes contaminantes. La eficiencia en sistemas de climatización mejoró en un 14.3 %, mientras que en los sistemas eléctricos --bombas, compresores e iluminación-- se alcanzó una mejora significativa de 38.6 %, con un ahorro promedio de 30.9 %. Respecto a las emisiones, estas se redujeron de 58.8 tCO2 a \linebreak[4]40.7 tCO2, lo que equivale a la absorción de 1.7 hectáreas de bosque con un retorno de inversión estimado en 10.7 años. En conclusión, este proyecto resulta viable desde el punto de vista de la aplicación de tecnologías de eficiencia energética, al incorporar mejoras técnicas que consolidan una solución económicamente rentable y ambientalmente sostenible.
The objective of this study is to evaluate the applicability of the most widely used closed-form ground-return impedance formulas in short-circuit analyses of distribution systems and to identify the critical network configurations in which the choice of impedance model significantly affects the short-circuit results. The methodology adopted in this research is organized into three stages. First, an algorithm was developed to implement and compare several closed-form Earth-return impedance formulations, and its performance was validated using benchmark data reported in the literature. Second, a short-circuit analysis algorithm was designed and verified against reference results published by the IEEE Power and Energy Society. Finally, multiple short-circuit studies were performed on several IEEE distribution test feeders. The findings reveal that most closed-form Earth-return impedance models provide adequate accuracy for both balanced and unbalanced short-circuit analyses. However, for single-phase line-to-ground faults, the choice of closed-form impedance formulation is critical to obtaining accurate short-circuit results.
The evolution of jet engine technology has been profoundly influenced by advancements in computational analysis, particularly in Computational Fluid Dynamics (CFD), structural analysis, and multidisciplinary optimization. This review explores state-of-the-art computational techniques applied to jet engine analysis, emphasizing their applications, benefits, and inherent challenges. CFD has become an essential tool, enabling detailed simulations of complex aerodynamic and combustion processes. Methods such as Large Eddy Simulations (LES) and Direct Numerical Simulations (DNS) have provided deeper insights into turbulence and combustion dynamics, leading to improved efficiency and reduced emissions. However, these high-fidelity simulations entail significant computational costs, driving the development of more efficient algorithms and high-performance computing resources. The integration of structural analysis with aerodynamic simulations has facilitated the design of components capable of withstanding extreme operational conditions, thereby enhancing engine reliability and safety. Multidisciplinary Design Optimization (MDO) frameworks have further transformed engine design by simultaneously evaluating multiple performance metrics, resulting in configurations that balance efficiency, weight, and durability. Despite these advances, challenges remain in accurately modeling complex physical phenomena such as combustion instabilities and material behavior under high temperatures. The incorporation of machine learning techniques offers promising solutions to address these issues by complementing traditional computational methods with data-driven insights. Looking ahead, the future of computational analysis in jet engine development lies in the seamless integration of high-fidelity simulations, real-time data analytics, and adaptive modeling, paving the way for more efficient, reliable, and sustainable propulsion systems.
Intelligent User Interfaces (IUIs) represent a transformative paradigm for advancing crowdsourced and human computation by optimizing task distribution, strengthening human–AI collaboration, and ensuring data integrity. This study presents a case study–driven analysis of an adaptive IUI framework designed to enhance scalability, engagement, and accuracy in large-scale, crowd-based problem-solving. By examining three representative platforms—Amazon Mechanical Turk (MTurk), Zooniverse (a citizen science platform), and AI-assisted medical image analysis in public health—the research investigates the influence of dynamic task allocation, Explainable AI (XAI), and gamification on user participation and task performance. The findings demonstrate that adaptive IUIs improve task accuracy relative to user expertise, reduce completion time as experience increases, and strengthen volunteer retention through gamified elements. Moreover, integrating XAI into AI-assisted medical diagnostics substantially elevates both trust and diagnostic precision. Collectively, these outcomes underscore the scalability, adaptability, and efficacy of IUIs in human computation, offering a comprehensive framework for future advancements in task optimization and explainability.
Este estudio presenta un sistema de clasificación automática destinado a priorizar reclamos eléctricos en CNEL EP. Para ello se procesaron 143 113 registros reales, incluyendo limpieza de datos, imputación de valores faltantes y la creación de variables predictivas relacionadas con la urgencia y la recurrencia de los reclamos. A partir de estos criterios se construyó la variable objetivo “prioridad”, que permite distinguir entre reclamos prioritarios y normales. Con esta información se entrenaron modelos de aprendizaje supervisado —árbol de decisión y random forest— aplicando técnicas como one-hot encoding y validación cruzada. El modelo random forest obtuvo los mejores resultados, con una exactitud del 91 % y un AUC-ROC de 0.89. Los resultados muestran que el sistema propuesto puede mejorar significativamente la asignación de recursos técnicos y los tiempos de respuesta ante reclamos eléctricos. Además, demuestra la viabilidad de incorporar técnicas de aprendizaje automático en la gestión operativa de empresas distribuidoras de electricidad, abriendo posibilidades para futuras mejoras y aplicaciones en tiempo real.
Semantic segmentation of urban scenes is essential for the development of smart cities; however, its effectiveness relies heavily on large, pixel-level annotated datasets, which are particularly scarce for mobility aids. This study aims to enhance semantic segmentation for urban accessibility applications by leveraging synthetic data. The proposed methodology integrates high-fidelity synthetic data generation using Unreal Engine 5.1, automated semantic mask processing, and the training of state-of-the-art segmentation models. A dataset of 5,036 images with pixel-perfect labels across 22 classes, including sidewalks, wheelchairs, and walking aids, was created to support this investigation. Two architectures were benchmarked: a baseline U-Net and DeepLabv3+ with ASPP. Pre-training with synthetic data increased global mIoU from 0.0626 to 0.84 (13.4x) and substantially improved precision, recall, and F1-score (by approximately 6.8x, 9.3x, and 10.4x, respectively). For accessibility-critical classes, motorized wheelchairs achieved an IoU of 0.94, and sidewalks attained a recall of 0.98. Overall, all 22 classes surpassed the deployment threshold ( ≥ 0.75 IoU). These findings demonstrate that synthetic data, combined with imbalance-aware training strategies, provides a viable pathway toward robust semantic segmentation solutions for urban accessibility applications.
Maintenance and inspection protocols in the aerospace industry are designed to safeguard the structural integrity of aircraft and ensure pilot safety. However, the air intakes of fighter aircraft pose significant access challenges for maintenance technicians during preflight inspections. To address this limitation, this study presents an innovative solution: the implementation and evaluation of a robot equipped with a wheeled-legged locomotion system. This system enables efficient access to the air intakes, significantly enhancing the inspection protocol. The robot was developed in close alignment with the operational requirements of Peruvian Air Force (FAP) technicians, which was critical to defining its design specifications and manufacturing parameters. Its adaptive and compact architecture allows it to navigate confined intake structures effectively, optimizing inspection time and resource utilization. The prototype’s performance was rigorously assessed through standardized tests, demonstrating its capability to reliably access and inspect air intakes under preflight conditions. This advancement contributes to the modernization of conventional aircraft maintenance procedures by integrating robotic technologies into the aeronautical inspection process.
This paper presents the development of a computing platform for the real-time monitoring of cardiovascular parameters derived from bioelectrical signals. A comprehensive analysis of primary users was conducted, leading to the identification of both technical and functional requirements. The interface design was guided by Sommerville’s methodology. The system architecture is based on a microservices model, incorporating a relational database and enabling integration with data transmitted from Internet of Things (IoT) devices. The platform was evaluated through incremental stress testing, starting with zero users and increasing in steps of 100 up to 5,000. A total of 22,132 requests were processed at a peak rate of 440.4 requests per second, with an average response time of 930 ms and 95% of responses occurring within 2,300 ms. The system demonstrated error-free performance with up to 1,700 concurrent users. At 5,000 users and 26,393 total requests, a minimal error rate of 0.16% was recorded, confirming the platform’s stability under high workloads. These findings validate the feasibility of the proposed solution for remote biomedical monitoring, offering an efficient, scalable, and robust tool for real-time health supervision.
This study investigates the effect of fiber content and coupling agent concentration on the mechanical and thermal behavior of a polypropylene-based composite. The materials were fabricated via hot compression molding using pellets composed of polypropylene (PP) modified with maleic anhydride-grafted polypropylene (MAPP) and reinforced with short bamboo fibers derived from Guadua angustifolia Kunth (GAK). The fibers were previously extracted through the steam explosion technique. The research was carried out in two stages: first, the composite materials were produced; second, their mechanical and thermal properties were comprehensively characterized. The incorporation of GAK fibers and MAPP significantly altered the mechanical performance of the PP matrix, yielding stiffer composites with improved flexural strength and impact resistance. The optimal formulation, containing 50 wt% GAK fibers and 4 wt% MAPP, resulted in a 322% increase in elastic modulus (2.9 GPa) compared to neat polypropylene (0.7 GPa). Both variables, fiber content and compatibilizer concentration, were found to exert a substantial influence on the mechanical behavior of the resulting composites.
This paper presents the development of a computing platform for the real-time monitoring of cardiovascular parameters derived from bio electrical signals. A comprehensive analysis of primary users was conducted, leading to the identification of both technical and functional requirements. The interface design was guided by Sommerville's methodology. The system architecture is based on a microservices model, incorporating a relational database and enabling integration with data transmitted from Internet of Things (IoT) devices. The platform was evaluated through incremental stress testing, starting with zero users and increasing in steps of 100 up to 5,000. A total of 22,132 requests were processed at a peak rate of 440.4 requests per second, with an average response time of 930 ms and 95% of responses occurring within 2,300 ms. The system demonstrated error-free performance with up to 1,700 concurrent users. At 5,000 users and 26,393 total requests, a minimal error rate of 0.16% was recorded, confirming the platform's stability under high workloads. These findings validate the feasibility of the proposed solution for remote biomedical monitoring, offering an efficient, scalable, and robust tool for real-time health supervision.
Electric vehicle (EV) charging management can be implemented through centralized or decentralized strategies. Strategic coordination between these approaches enhances system efficiency and balances energy loads, thereby supporting the widespread adoption of EVs and fostering a sustainable, emissions-free society. In this study, distribution network operators (DNOs), acting as centralized charging managers, are responsible for mitigating the lack of coordination among electric vehicle aggregators (EVAs), which represent decentralized managers. The primary objective of the centralized management in this research is to constrain each decentralized optimization model, characterized using Monte Carlo simulations. Three EV adoption scenarios-comprising 2,000, 2,500, and 3,750 vehicles-are evaluated by comparing decentralized charging management with an unregulated charging baseline in the IEEE 14-bus power system. Improvements are required only in the highest adoption scenario, where the proposed centralized coordination model is applied. The study models energy trading constraints for each EVA, assigning one aggregator per load-bearing bus in the system. Transmission-level results are analyzed and then synthesized for application in the IEEE 13-bus distribution power system. Findings demonstrate that coordinated centralized and decentralized charging management significantly improves operational conditions in both transmission and distribution networks without necessitating changes to travel behavior.
