ABSTRACT. This paper proposes a robust Finite Control Set Model Predictive Control (FCS-MPC) strategy for active power injection into the grid using a single-phase DC/AC inverter with an LCL filter. The overall system is modeled as a hybrid system within the Linear Complementarity (LC) framework, thereby preserving the nonlinear and switching nature of the power converter. Within the FCS-MPC formulation, a bipolar switching strategy and a one-step prediction horizon are adopted to minimize a quadratic cost function based on the grid-current tracking error, with the reference synchronized to the grid voltage. For online parameter identification, a simplified linear representation of the LC-based model is derived, while the discrete switching nature of the converter is preserved in the control law. A Recursive Least Squares (RLS) algorithm is integrated into the predictive control framework to estimate the model parameters online, thereby enhancing robustness against uncertainties in the LCL filter parameters and variations in the DC-side voltage. The performance of the proposed control strategy is validated through numerical simulations and MATLAB/Simulink simulations, demonstrating accurate current tracking and robust active power injection into the grid under parameter variations and grid disturbances.

ABSTRACT. This paper investigates the use of evolutionary algorithms to optimize the design of grounding electrodes in High Voltage Direct Current (HVDC) systems. The variability of soil resistivity and its impact on Ground Potential Rise (GPR) are analyzed. A MATLAB application based on genetic algorithms was developed to minimize construction costs and land occupation footprint, ensuring that step and touch voltages comply with the IEC 62344 standard. The results demonstrate that AI allows for finding optimal geometric dimensions (radius and diameter) that balance electrical safety and economic feasibility

ABSTRACT. This paper presents the small-signal modeling and average current mode control design of a flyback DC-DC converter, incorporating non-ideal component parameters. The converter is modeled using the PWM switch model, which allows derivation transfer functions for both the current and voltage control loops. Compensators are designed based on frequency-domain criteria, including phase margin and crossover frequency specifications. The closed-loop performance is validated through PSIM simulations, where the system response to input voltage disturbances and load transients is evaluated. The proposed design is oriented toward photovoltaic microinverter applications, where regulated high-voltage DC output and robust dynamic response are required.

ABSTRACT. Cost-effective energy supply and power reserves are essential to assist electrical networks in achieving sustainable operations on a continuous basis. When sufficiently available, both services enable the system to respond to contingencies and drastic variations in demand and renewable generation. However, power reserve levels are often established arbitrarily within optimization process, as opposed to being regarded as decision variables to be optimized thus enabling a more reliable system operation. This paper addresses this problem by introducing an optimal power flow model for AC/DC networks based on VSC-HVDC links. The method allows for simultaneous optimization of generation and reserves using opportunity costs. For practical results, it explicitly considers the power equations of the AC networks, the DC grid and the VSC stations, including operational constraints such as reserve requirements at unit and area levels. The model is applied to a four-area VSC-based AC/DC network with several power sources interacting through a four-node DC grid. The system operation is analyzed over a 24-hour horizon considering different demand profiles in each area.

ABSTRACT. This paper presents an experimental evaluation of Direct Grid Measurement, Synchronous Reference Frame Phase- Locked Loop (SRF-PLL), Second-Order Generalized Integrator Phase-Locked Loop (SOGI-PLL), and Integral Frequency Estimation SOGI-PLL (IFE-SOGI-PLL) applied to a singlephase Active Power Filter. The analysis compares the harmonic content present in the synchronization signals with the resulting current harmonic mitigation achieved by an Active Power Filter using a proposed index based on current THD measurements. Experimental results show that all synchronization techniques significantly reduce current harmonic distortion compared to the uncompensated condition. Among the evaluated methods, SRF-PLL achieved the highest harmonic mitigation performance, while SOGI-PLL and IFE-SOGI-PLL generated synchronization signals with the lowest harmonic distortion. However, the reduction of harmonic content in the synchronization stage did not necessarily result in higher harmonic compensation capability. These findings suggest that Active Power Filter performance is influenced by factors beyond the spectral quality of synchronization signals and motivate further investigation into the interaction between synchronization algorithms and power converter control strategies.

ABSTRACT. Human movement analysis has become an important research area within computer vision due to its potential applications in security, sports, rehabilitation, and healthcare. Recent advances in human pose estimation enable the extraction of anatomical landmarks from images and video sequences, providing a structured representation of body posture and movement. However, the challenge is no longer limited to detecting body keypoints, but rather to transforming this information into meaningful biomechanical descriptors capable of supporting automated decision-making processes.

This work presents a methodology based on human pose estimation and biomechanical feature extraction for automatic human movement analysis. The proposed methodology uses body landmarks obtained through computer vision techniques to generate quantitative representations of human motion, including joint angles, spatial coordinates, relative distances, and symmetry metrics. These biomechanical features are employed as input variables for movement classification and evaluation tasks.

To validate the proposed methodology, two case studies were developed. The first case focuses on the identification of violent behaviors using a Multilayer Perceptron (MLP) classifier trained with biomechanical features extracted from human motion. The second case addresses physical exercise assessment through biomechanical reference models generated from correctly executed movements, allowing the evaluation of exercise performance using predefined tolerance ranges.

Experimental results demonstrate the effectiveness of the proposed methodology. In the violence detection scenario, the MLP classifier achieved an accuracy of 90\%. For physical exercise assessment, the generated biomechanical reference models obtained accuracy values above 90\% across the analyzed exercises. The results suggest that the combination of pose estimation, biomechanical analysis, and machine learning provides a viable framework for automatic human movement interpretation across different application domains.

ABSTRACT. Jamming attacks represent a critical threat to connected vehicles because they can degrade the availability of wireless communication and GNSS-based positioning services required for monitoring, navigation, and safety-related applications. This paper presents a controlled experimental evaluation of SDR-derived radio-frequency (RF) features for jamming classification and short-horizon future-state estimation using I/Q data from a GNSS interference dataset. Seven RF descriptors, including average power, band power, spectral peak power, and causal moving-window statistics, were evaluated with Random Forest, XGBoost, and SVM-RBF classifiers on a transition-based test set. Control experiments were also included to assess whether machine learning models provide added value over simpler rules. The classification results show that the analyzed jamming event produced a strong RF power signature: a single peak-power threshold selected on the validation set slightly outperformed the machine learning classifiers, achieving an F1-score of 0.9583 and an MCC of 0.8698. Temporal models, including stacked LSTM, GRU with attention, TCN-GRU, and LSTM-XGBoost, were also evaluated using lookback windows and prediction horizons from 0.5 s to 2.0 s. Although these models achieved moderate RMSE and R2 values, they did not outperform a persistence baseline that assumes the future jamming state remains equal to the current state. These results indicate that SDR-derived RF features are highly effective for controlled jamming classification, while reliable early prediction requires additional precursors from GNSS, vehicle dynamics, and communication-quality variables.

ABSTRACT. This paper presents a comparison of seven Machine Learning classifiers trained on vectors derived from the Histogram of Oriented Gradients (HOG) to detect pedestrians in images captured by vehicle reverse camera. The analysis utilizes a dataset collected within the Siglo XXI campus of the Universidad Autónoma de Zacatecas, consisting of 13,962 full panoramic images split into 11,168 training samples and 2,794 test samples. Model optimization was performed using Grid Search. The LightGBM and XGBoost algorithms achieved the highest accuracy values, whereas linear models demonstrated superior efficiency in terms of FPS rate. Additionally, this work incorporates Explainable Artificial Intelligence (XAI) analyses to trace the root causes of both correct classifications and false alarms.

ABSTRACT. This paper presents the implementation of an electromagnetic transient (EMT) simulation program in a Just-In-Time (JIT) programming environment, featuring a multi-thread processing scheme. The network formulation is based on the Modified Augmented Nodal Analysis (MANA) method, while the system definition and structure are handled through the JSON format. Network decoupling is achieved by exploiting the traveling wave time delay in distributed-parameter transmission lines, enabling each subsystem to be solved independently on a separate thread. The results show good accuracy in both steady-state and transient conditions. The computation times obtained with parallel processing yield speedups with respect to the serial processing time, demonstrating the benefits of paralelization as an efficient approach for EMT simulation.

ABSTRACT. Measurement of soil electrical properties provides a rapid, non-destructive, and integrative approach to infer other soil properties, critical for sustainable water management. In addition, high-resolution local mapping of soil hydraulic properties is essential for effective irrigation scheduling and water resource allocation. This study evaluates the integration of electrical impedance tomography imaging data with a tiered Cloud-Fog-Edge distributed computing architecture to provide a scalable, high-resolution solution for quantifying and modelling soil hydraulic conductivity properties. The Edge layer is constituted by an IoT-connected, multichannel, soil impedance sensor for Electrical Impedance Tomography (EIT) imaging the wetting front propagation process. A Fog layer, implemented on a Raspberry Pi single-board computer, aggregates and pre-processes local sensor data streams. Computational bottlenecks, (EIT image reconstruction and wetting front velocity vector field calculations) are offloaded to a Cloud-based HP ProLiant Server. The framework was validated using an experimental lysimeter with five sandy loam soil samples under constant irrigation to reconstruct EIT images of wetting front propagation. The proposed EIT image analysis methodology successfully quantified cumulative infiltration and mapped the velocity vector fields of the moving wetting fronts. The experimental results demonstrated strong agreement with theoretical data obtained from Philip’s infiltration model. The integration of distributed computing architectures with EIT imaging offers a robust methodology for real-time hydraulic profiling. The results suggest that this approach can be effectively extended to measure and model complex, multi-layered soil profiles.

ABSTRACT. This paper presents the design and preliminary validation of a wireless monitoring system based on the ESP32- S3 and LoRa communication at 915 MHz, intended for reliable acquisition of temperature and relative humidity in applications related to concrete curing. The system incorporates initial synchronization through NTP, local backup in Flash memory and microSD, traceability through transmitter and gateway identifiers, and safe record deletion through acknowledgments. Two experimental tests were conducted: an intentional one-hour disconnection test, with sampling every 10 seconds, and a 48- hour stability test under non-line-of-sight conditions. In the first test, 360 out of 360 locally stored readings were recovered; in the second, 289 out of 289 expected readings were recorded through a concrete-block wall. The results support the technical feasibility of the proposed architecture under the evaluated conditions and establish a basis for future tests with encapsulated nodes embedded in real concrete.

ABSTRACT. This paper presents the system integration and experimental assessment of a low-cost NanoVNA-based radiofrequency (RF) measurement platform. The implemented system combines a 1.5-GHz NanoVNA-H, a Raspberry Pi 3, a 15.6in touchscreen display, open-source software, and a custom three-dimensionally printed enclosure. The platform provides instrument control, graphical visualization, data storage, and touchscreen or keyboard-and-mouse operation without requiring a dedicated external computer. The estimated replacement cost of the complete system is approximately USD 495. Its measurement capability was assessed against a commercial Anritsu vector network analyzer over the common 100–700 MHz frequency range using four microstrip split-ring resonators nominally centered at 360, 422, 442, and 502 MHz.The assessment was restricted to the magnitude of the forward transmission coefficient, |S21|. The proposed platform identified the dominant resonances with apparent frequency differences between 1 and 3 MHz, no greater than onehalf of the 6-MHz frequency increment of the NanoVNA-H sweep. Differences in notch magnitude ranged from approximately 0.1 to 0.9 dB. Although additional quasi-periodic ripple and secondary minima were observed, the platform consistently reproduced the frequency region of the dominant resonances. The results support its use for low-cost resonance identification and qualitative transmission-response measurements, while establishing its limitations relative to laboratory-grade instrumentation.

ABSTRACT. This paper presents the design, characterization, and experimental validation of a modular 8-kW, 400-V single-phase GaN-based power platform designed for high-frequency conversion. To map the physical limits of Gallium Nitride (GaN) devices, a deterministic framework defines the safe operating region as a closed boundary within a 6-dimensional real Euclidean parameter hyperspace, intersecting electro-thermal, magnetic self-resonance, and sensing bandwidth constraints. The hardware platform features a low-thickness, 8-layer PCB layout engineered to minimize parasitic loop inductances and maximize electromagnetic field cancellation. Dedicated symmetrical routing and fully differential isolated sensing networks are integrated on-board, establishing the structural readiness required for grid-interfaced active rectification. The architecture is experimentally validated across two distinct paradigms derived from the model: a full-bridge inverter mode operating under full load (8 kW, 400 V, 20 A) at 100 kHz with a minimal 17°C thermal rise at the substrate baseplate, and a step-down Buck mode successfully pushed to a 2-MHz switching frequency. This infrastructure serves as a robust foundation to explore alternative fast continuous-time control strategies.

ABSTRACT. The automatic classification of Jovian decametric radio emissions has become increasingly important due to the growing volume of spectrograms generated by ground-based radio telescopes and space missions. This work presents a compact convolutional neural network (CNN) for the classification of real spectrograms obtained from the Juno/Waves catalog. The proposed methodology includes image preprocessing using the Sobel filter to enhance the most relevant spatial features, followed by the training of a CNN with grayscale images resized to 32 × 32 pixels. The dataset consisted of 704 spectrograms corresponding to the Io-A, Io-B, Io-C, and Noise classes. Experimental results showed a validation accuracy of 89.74%, demonstrating that the proposed architecture can effectively distinguish the considered emission types while maintaining low computational complexity. These findings indicate that compact CNN-based approaches constitute a promising alternative for the automatic classification of Jovian radio emissions and provide a basis for future embedded implementations on FPGA platforms.

ABSTRACT. Target-tracking radars with antenna arrays require radiation patterns adaptable to different operational functions. Digital Beamforming (DBF) enables this reconfiguration by modifying the complex weights applied to the radiating elements. This paper analyzes the influence of array geometry and number of elements on pencil, fan, and cosecant-squared beam patterns. The obtained responses are compared in terms of beam directivity, angular resolution, spatial energy distribution, and approximation error. The results show narrower beamwidth and greater energy Concentration in the desired direction as the array size increases, while its configuration affects the radiation pattern, coverage region, and sidelobe presence.

ABSTRACT. Electrical Impedance Tomography (EIT) is a novel imaging technique that presents enormous potential in early breast cancer detection by reconstructing the internal conductivity distribution. However, EIT images lack spatial resolution and present diffuse shapes, making the identification and localization of tissue anomalies a complex task. This study presents a comparative analysis of multiple approaches to detect and localize a malignant anomaly in specific spatial quadrants with machine learning models in a multi-class classification task, using conductivity maps and raw boundary voltages. The data were generated with EIDORS by simulating anatomical conductivity distributions in a 3D FEM breast model. A Convolutional Neural Network (CNN) was trained with conductivity maps, while a Multi-Layer Perceptron (MLP) and a Support Vector Machine (SVM) were trained with boundary voltages. Additionally, the impact of multiple data normalization techniques, such as Max Abs, Z-Score and Min-Max, was evaluated under a Signal to-Noise Ratio (SNR) of 80. For the CNN, the Max-Abs normalization enhanced the presence of an anomaly and yielded the best performance metrics among the models, with an F1 Score of 86.44% with noise-free data and 75.21% with noisy data. Conversely, for boundary voltages, Z-Score normalization with SVM yielded similar performance metrics to the CNN with noise-free data, with an F1-Score of 85.42%. Therefore, boundary voltages can be used as an alternative source of information to localize tissue anomalies in a suspicious breast, providing a faster and less computationally expensive approach that does not rely on reconstructed conductivity map data.

ABSTRACT. The present study implements and evaluates a binary classification approach to detect breast cancer risk-related anomalies using infrared thermography and a multilayer perceptron (MLP) neural network. A total of 371 thermograms from the Database for Mastology Research (DMR) were analyzed. Based on statistical tests, six statistically significant features were identified to distinguish patients "with anomalies" from those "without anomalies". These features were used as input to the MLP network, applying stratified 6-fold cross-validation. The proposed model achieved an accuracy of 84.64\%, sensitivity of 80.21\%, specificity of 86.30\%, and AUC of 89.26\%. The dimensionality reduction improves model interpretability, offering an advantage for healthcare professionals.

ABSTRACT. Electromyography (EMG) is widely used in studies of muscle activity; one example of this is in the development of prosthetics, which allow amputated limbs to be replaced with artificial limbs. In the search for alternatives, electroencephalography (EEG) has been identified as a possible option. Thus, this study aims to provide insight into how EMG signals relate to EEG signals through the simultaneous recording of both during a series of specific movements associated with the upper limbs. This study is based on concatenating features extracted from the EMG and EEG samples, thereby creating a hybrid database. This database would be used to train a neural network model alongside the EMG and EEG databases to evaluate and compare each method; the results obtained support this study, as an improvement in accuracy was observed when identifying certain gestures using features from this hybrid database.

ABSTRACT. Alzheimer's disease (AD) is often diagnosed at advanced stages due to the lack of accessible, non-invasive biomarkers. As an extension of the central nervous system, the retina has been proposed as an alternative for early detection through imaging analyzed with artificial intelligence (AI). This systematic review, conducted under the PRISMA 2020 methodology, started with 179 articles, which were narrowed down to 97 after the screening process. The results highlight fractal dimension and vascular tortuosity as the biomarkers most strongly associated with cognitive decline, and convolutional neural networks as the most effective technique for their segmentation. However, notable limitations persist, including small datasets lacking confirmation by standard biomarkers, a lack of standardization in image acquisition that leads to domain shift, uncontrolled ocular comorbidities, and limited clinical interpretability of the models. Overcoming these challenges will require international, longitudinal cohorts and standardized validation protocols. The retina thus stands out as a promising non-invasive pathway for the early diagnosis of AD.

ABSTRACT. An idealized venous segment corresponding to the outflow vein of an arteriovenous fistula for hemodialysis is hemodynamically simulated, with the objective of analyzing the effect of different degrees of stenosis on proximal pressure, distal pressure, and axial pressure drop. The model was developed in COMSOL Multiphysics through a time-dependent laminar- flow CFD analysis, considering blood as an incompressible fluid with a density of 1060 kg/m3 and a dynamic viscosity of 0.0035 Pa · s. Four degrees of stenosis (0 %, 40 %, 60 %, and 80 %) were evaluated under an inlet condition with pulsatile velocity equivalent to a heart rate of 75 beats per minute. The results show a progressive increase in axial pressure drop as the degree of stenosis increases, together with localized flow acceleration in the narrow region and an increase in wall shear stress (Wall Shear Stress, WSS). These findings provide a reference hemodynamic characterization for the design and calibration of mechatronic simulators and instrumented platforms intended for clinical training and sensor validation for arteriovenous fistula vascular access.

ABSTRACT. This article presents a hybrid biomedical instrumentation system for real-time body temperature monitoring, distributing processing between an ESP32 microcontroller and a Basys 3 FPGA. Temperature is captured by a DS18B20 digital sensor, displaying with decimal resolution on a local OLED screen. For data transfer to the FPGA via UART, a hardware/software co-design methodology based on integer quantization and digital digit transposition from the firmware was developed. This strategy deterministically mitigates overflow in the 8-bit receiver register by avoiding values greater than 255, eliminating visual framing errors without requiring additional synchronization logic in the reconfigurable hardware. Experimental results demonstrate stable and error-free numerical display on the FPGA's seven-segment displays within the human biological range, validating the feasibility of these architectures as a base stage for more complex multiparametric medical systems.

ABSTRACT. The Tear-Ferning Test (TFT) is a cost-effective clinical method for assessing tear film stability; however, its manual classification using the Masmali scale is limited by inter-observer subjectivity and its qualitative nature. This study presents a new dataset of crystallized tears from the Mexican population obtained using the ferning test. These samples are evaluated using granulometry-based morphological analysis as a quantitative descriptor of the structural complexity of tear crystallization patterns. By simulating a digital sieving process with successive morphological openings, we extract objective features—specifically the granulometric pattern spectrum (Gn), median size (D50), and Shannon entropy (H)—to distinguish among the five Masmali grades. Analysis of 25 digital samples shows that these descriptors effectively capture the morphological transition from the intricate, high-entropy branching of healthy tears (Grade 0: H ~ 2.74, D50 ~ 4.37 um) to the coarse, amorphous aggregates characteristic of severe dry eye (Grade 4: H ~ 0.79, D50 ~ 25.76 um). Identifying specific structuring-element radii (e.g., 1.29 and 7.74 um) provides a physical basis for grading, making the TFT a repeatable and objective assessment tool. This characterization framework offers a set of computational descriptors to support objective morphological analysis of ophthalmic images.

ABSTRACT. Synchronized oscillations in pancreatic islets under- lie pulsatile insulin secretion and are disrupted in pathologies like diabetes. Whether this synchronization emerges from short- range gap-junction coupling or longer-range paracrine signaling remains debated. We present a GPU-accelerated Kuramoto oscillator model built on six reconstructed three-dimensional human islet architectures comprising α-, β-, and δ-cells, and simulate synchronization across eleven interaction range values (λ = 1–750 μm) at three glucose levels for both intact and 40% β-cell-depleted islets (396 simulations). Our results reveal a strong glucose-level dependence of the synchronization-distance relationship. High glucose drives a sharp onset of islet-wide coherence between λ = 10 and 20 μm, reaching order parameter R > 0.99 by λ = 50 μm. Normal and low glucose, by contrast, produces a gradual rise that does not approach saturation until λ > 250 μm (⟨R⟩w ≈ 0.91). At the cell-type level, under low glucose β- and δ-cells synchronize ahead of α-cells: at λ = 250 μm, ⟨R⟩w ≈ 0.99 for β- and δ-cells versus ≈ 0.70 for α-cells. Under high glucose, all three populations reach comparable coherence simultaneously. At high glucose, the loss of 40% of β-cells delays synchronization, which resolves by λ = 50 μm. At normal glucose, the same loss of β-cells produces a persistent reduction in coherence across a wider range of interaction distances. These findings indicate that the glucose- dependent coupling determines both the onset distance and the order of synchronization by cell type, implicating an effective glucose-dependent interaction range as a relevant determinant of islet oscillatory coordination.

ABSTRACT. The study develops a Cox proportional hazards model to analyze survival time in patients with Diffuse Large B-Cell Lymphoma (DLBCL) using two modalities: clinical data and fractal dimension.

The model achieves a C-index of 0.75 from the fusion of both datasets, outperforming the clinical‑only model (C‑index 0.74) and the fractal‑dimension‑only model (C‑index 0.53).

The model shows that fractal dimension can be a significant complement in the prognosis of DLBCL patients.

ABSTRACT. The simultaneous circulation of infectious agents with reproductive impact represents a major animal health problem in cattle, especially in regions with limited epidemiological information. This work analyzes serological data from 121 cattle from 10 municipalities in southern Zacatecas, Mexico, including age, zootechnical function, reproductive status, and positivity for various Leptospira serovars, bovine viral diarrhea virus, infectious bovine rhinotracheitis, brucellosis, and Neospora caninum. To carry out the analysis, an unsupervised approach was applied using 119 animals with complete profiles and 11 infectious agents, excluding two agents because they showed constant results across all observations. Similarity between animals was estimated using Jaccard distance, and the profiles were grouped using Partitioning Around Medoids. In this sense, a six-cluster solution presented the highest average silhouette width, with a value of 0.401, indicating a moderate structure. The distribution of the profiles showed an association with the municipality of origin, with a p-value of 0.0314 and a Cramér’s V of 0.327, showing that a simple unsupervised approach can reveal interpretable regional serological patterns in binary veterinary data.

ABSTRACT. This paper proposes a reactive power compensator based on the equivalent reactance behavior of a single-phase inverter, as a candidate topology for future application in ac microgrids. The inverter is modeled using switched and averaged representations to derive its current and power expressions. These expressions allow the inverter to be represented as a variable reactance with inductive or capacitive behavior depending on the modulation index. Based on this interpretation, a switched compensator is proposed and modeled to reproduce the reactive behavior of the inverter. Its performance is compared with that of the inverter under positive and negative reactive power conditions. Simulation results show that the proposed compensator reproduces the steady-state output current and reactive power behavior of the inverter, providing a basis for future control design, practical implementation, and integration into ac microgrids.

ABSTRACT. Extreme climate events, such as heatwaves and drought, change transmission line parameters and fault conditions, but their effect on fault location accuracy has not been quantified. This paper evaluates impedance-based and traveling-wave fault location methods on a representative 400 kV, 100 km transmission line modeled in ATP-EMTP under reference, heatwave, drought, and combined heat–drought scenarios. Single-phase-to-ground faults are simulated at nine positions and three fault resistances, with locator settings fixed at the reference values, so the reported error is that of a locator whose settings predate the event. The two method families are affected differently: the error of the single-ended reactance method increases by up to 7 km under combined heat and drought, whereas the Eriksson and synchronized two-ended methods keep their error below 0.8 km in all cases. Drought acts mainly through the tower-footing resistance rather than through the zero-sequence impedance, and does not change the simulated traveling-wave estimate. Heat shifts the two-ended traveling-wave estimate by up to 62 m through conductor elongation and the lower wave velocity caused by the increased sag. The response of the instrument transformers is not simulated here and remains to be evaluated.

ABSTRACT. Transmission expansion planning must identify costeffective reinforcements while preserving nonlinear alternatingcurrent (AC) network feasibility. This paper presents a unified formulation that jointly optimizes candidate AC circuits, modular capacitor banks, static synchronous compensators (STATCOMs), and static synchronous series compensators (SSSCs), using full nonlinear AC power-flow equations and detailed flexible AC transmission system (FACTS) models. Two solution strategies are applied to the same formulation: an AMPL-based mixed-integer nonlinear programming (MINLP) procedure and a Differential Evolution (DE) framework implemented in Julia, in which each investment plan is evaluated through a nonlinear AC operating problem. The methodology is tested on the Garver six-bus system. Results show that coordinated planning of transmission and reactive-power resources can significantly alter the preferred expansion topology. The lowest-cost plan combines AC reinforcement, capacitor banks, and an SSSC, reducing investment by 57.92% relative to AC-only planning.

ABSTRACT. This paper evaluates the integration of Dynamic Line Rating (DLR) into Transmission Network Expansion Planning (TNEP). Unlike Static Line Rating (SLR), DLR adjusts transmission capacity according to meteorological conditions, improving the utilization of existing infrastructure. A nonlinear optimization model was applied to the 6-node Garver system using ERA5 weather data, IEEE 738 thermal modeling, AC Optimal Power Flow, and the IDE PBILc metaheuristic. Results demonstrate that the proposed DLR approach can reduce transmission expansion costs by exploiting available ampacity, achieving significant savings under favorable climatic conditions.

ABSTRACT. The incorporation of distributed generation (DG) introduces additional decisions in distribution expansion planning (DEP), requiring the coordinated evaluation of network and generation investments. This paper proposes a model that determines the expansion of lines, substations, and the allocation of DG while minimizing investment, operation and maintenance, and technical power-loss costs, subject to nonlinear power-flow constraints. To solve the problem, large neighborhood search (LNS) and a population-based variable neighborhood search (VNS) with an LNS-based shaking operator are proposed and evaluated using a current injection power flow model. The methodology is tested on 24- and 182-node benchmark systems under two DG availability scenarios. The results demonstrate that VNS consistently achieves lower average cost solutions than LNS, with this advantage increasing with system size, at the expense of 3–4 × longer computation times. Both algorithms produce electrically feasible and largely connected expansion plans, with structural agreement of up to 94% with a reference solution from the literature, despite differences in the formulations of the two approaches. Index Terms—Distribution Expansion Planning, Distribution

ABSTRACT. Inertial Navigation Systems (INS) provide continuous positioning during Global Navigation Satellite System (GNSS) outages, but the accumulation of inertial sensor errors causes navigation drift to grow unbounded over time. Echo State Networks (ESNs) have recently emerged as an efficient approach for compensating INS drift by learning velocity residuals from inertial measurements. Existing studies, however, evaluate and optimize these models almost exclusively using prediction-error metrics such as the root mean square error (RMSE), implicitly assuming that the most accurate predictor also provides the best navigation performance. This work challenges that assumption. A systematic evaluation of 61,236 ESN configurations is performed using two complementary perspectives: conventional prediction metrics (RMSE, mean position error, and area under the error curve) and operational autonomy metrics that directly quantify the number of required GNSS corrections, together with the distance and time that the system can navigate without external assistance. The results show that prediction accuracy and operational autonomy are only moderately correlated, leading to substantially different optimal hyperparameter configurations. In particular, models with nearly identical prediction accuracy can exhibit markedly different operational behavior in closed-loop navigation. These findings demonstrate that optimizing prediction error alone may lead to suboptimal navigation performance and highlight the importance of aligning the optimization objective with the operational requirements of autonomous navigation systems.

ABSTRACT. Efficient inventory management presents a fundamental challenge for organizations due to the need for accurate records, reduced operational errors, and timely product availability. Traditional inventory control systems have limitations primarily related to manual intervention, a lack of real-time information updates, and discrepancies between physical inventory and digital records. In this context, smart technologies have emerged as alternatives to automate the processes of capturing, monitoring, and analyzing information within inventory systems. The objective of this study is to systematically analyze the implementation of smart technologies for automated inventory control, identifying the technologies used, their main applications and impacts, as well as limitations and opportunities for future research. To this end, a Systematic Literature Review was conducted following the PRISMA 2020 guidelines. The search was conducted in the Scopus, IEEE Xplore, Web of Science, and Google Scholar databases, considering publications between 2020 and 2026. After the identification, selection, and evaluation process, 31 primary studies were included for qualitative analysis. The results show that the main technologies applied are artificial intelligence, the Internet of Things, RFID, computer vision, intelligent robotics, and digital twins. These solutions have demonstrated significant contributions to improving inventory accuracy, reducing human error, automating stock movement recording, and increasing operational visibility. However, challenges remain related to implementation costs, dependence on specialized infrastructure, scalability, and limited validation in real-world industrial scenarios. This review provides a structured overview of the current state of smart technologies applied to automated inventory control and establishes research directions aimed at developing more adaptable, cost-effective, and comprehensive solutions focused on innovation.

ABSTRACT. Microarchitectural simulators provide detailed execution data, but converting these artifacts into reliable performance diagnoses remains labor-intensive and dependent on expert analysis. This paper presents an LLM-assisted workload characterization architecture integrated into a production Simulation Service. A Workload Analyzer discovers simulation runs, extracts and normalizes profiling artifacts, computes derived metrics, validates measurement quality, and provides structured evidence to a large language model. A deterministic quality gate separates valid performance measurements from suspicious regions of interest before LLM interpretation. Valid results proceed to workload characterization and bottleneck analysis, while invalid results trigger a data-quality diagnosis. Two case studies demonstrate the approach: a vectorized SGEMM kernel in which the system identifies repeated intermediate writeback as a store-bound performance defect, and a multi-node iperf3 experiment in which it detects that the measured region captured idle polling rather than active data transfer. The results show that LLM-assisted analysis can support evidence-grounded performance diagnosis when constrained by structured metrics, validation rules, and adaptive analysis routing.

ABSTRACT. We ask whether sparse symbolic regression can recover the governing equations of a forced Van der Pol oscillator from trajectory data when the drive is exogenous rather than a state feedback law. Identifiability reduces to a rank condition on the candidate library. A feedback input makes the input column an exact combination of the state columns, halves the numerical rank and opens a one-parameter family of equivalent models; an exogenous input removes the degeneracy. Under a persistently exciting multisine drive the method recovers the exact structure and coefficients. On an unseen pulse excitation it reaches a root mean square error of 2 × 10−4, four orders of magnitude below extended dynamic mode decomposition and a polynomial NARX surrogate fitted to the same data with 20x as many parameters. The exact structure survives a threshold window of two decades on clean data, which narrows to a factor of 3 at 20% noise. Colored and impulsive disturbances prove more damaging than white noise of equal power, and the limiting step throughout is the estimation of the derivatives rather than the sparse fit.

ABSTRACT. This paper presents a distributed computational model for emergency route planning in smart cities. The proposed model represents urban infrastructure as a set of fixed and mobile elements that independently generate evacuation routes by considering travel distance and risk. Each element collects information from its surrounding environment through sensors and exchanges relevant data with neighboring elements to support distributed decision-making. The architecture combines local processing with parallel execution to improve route generation and system response. Although the current implementation operates on a single computing platform, the proposed design can be extended to fully distributed environments. Experimental results show that preprocessing route information reduces the time required to generate evacuation routes, supporting faster responses during emergency situations.

ABSTRACT. Experimental validation of chaotic circuits often relies on qualitative comparisons of phase portraits and time series, which provide limited quantitative evidence regarding the correspondence between mathematical models and hardware implementations. This study presents a quantitative validation framework to assess agreement between numerical simulations and physical realizations of multiscroll chaotic systems. The methodology incorporates complementary parametric, spectral, spatial, and geometric metrics, including Mean Absolute Percentage Error (MAPE), normalized power spectral density (PSD), Dice and Jaccard similarity coefficients, and correlation dimension. As a case study, we implement a triple-scroll chaotic system using an electronic circuit. Experimental results demonstrate strong agreement between numerical and experimental realizations, with a parameter MAPE of 1.072\%, PSD correlation coefficients up to 0.967, Dice similarity values up to 0.979, Jaccard indices up to 0.959, and a relative error of 4.44\% in the correlation dimension. The proposed framework can quantitatively evaluate the parametric, spectral, spatial, and geometric correspondence between simulated and experimental attractors, even in the presence of circuit non-idealities. The framework provides a reproducible approach applicable to various physical implementations of nonlinear chaotic systems.

ABSTRACT. Infrared thermography (IRT) is currently widely used to monitor and assess the physical condition of machinery and industrial equipment. Through the processing of acquired thermal images and the application of various analysis tech- niques, IRT enables the detection of faults in defective or degraded equipment and industrial processes operating under normal conditions. This work investigates the potential of thermal indicators extracted from passive infrared thermography to classify the severity of faults during the transient state of a distillation process. The proposed analysis is based on the evaluation of a set of first- and second-order statistical features. By examining the variations of these statistical indicators under different operating conditions, the feasibility of their application as fault detection metrics is assessed.

ABSTRACT. This paper presents a simulation of a hybrid control scheme for a three-phase current source inverter with an LC filter connected to the grid. The capacitor voltage in LC filter is controlled by a conventional proportional-integral (PI) controller, while the grid current is controlled by a sliding mode controller (SMC). The study includes the state-space mathematical modeling of the LC filter in the synchronous d,q reference frame using feedback linearization to mitigate cross-axis coupling. In addition, the design of the proportional-integral (PI) controller is presented: an inner loop to regulate the capacitor voltage and an outer-loop sliding mode controller for the inductor current (current injected into the grid), both tuned according to the Nyquist stability criteria and the Lyapunov stability theorem, respectively. It also includes space vector modulation (SVPWM) designed for efficient IGBT switching and to avoid disturbances caused by Overlap-Time. This study highlights the advantages and disadvantages of using a hybrid control system.

ABSTRACT. The active power economic dispatch with valve-point effects is a multi-constrained, nonlinear, and non-convex optimization problem. In this paper, a genetic algorithm based on orthogonal projection is proposed to solve the economic dispatch problem. The method generates an initial population within the operational limits of the generators and projects the individuals onto the hyperplane defined by the power balance constraint. Additionally, a repair mechanism based on the bisection technique is incorporated to guarantee compliance with the generation limits. The evolutionary operators are implemented using convex combinations for crossover and linear combinations for mutation. The proposed methodology is evaluated on 3-unit, 13-unit, and 40-unit electrical power generation systems. The obtained results demonstrate that the method is capable of finding high-quality solutions, achieving competitive and, in some cases, superior performances to those reported by reference methods in the literature.

ABSTRACT. This paper presents a control strategy for the regulation of frequency and voltage (f/V) in an electric power system (EPS) subject to high variability caused by renewable generation and time-varying loads. The proposed scheme combines a model-based optimal controller, whose weighting matrix is systematically tuned from participation factors obtained through modal analysis, with an output-based robust Super-Twisting (ST) controller that computes an auxiliary active and reactive power injection at a selected network node. The methodology is validated on the IEEE 9-bus test system under piecewise-constant and time-varying load disturbances. Simulation results show that the ST-based power injection reduces the mean-squared error of the frequency and voltage responses with respect to the optimal controller acting alone, confirming the benefit of complementing model-based control with a robust, output-based regulation layer under fast disturbances.

ABSTRACT. This paper presents the application and experimental validation of Gaussian and Gamma voltage profiles for DC motors powered through DC/DC converters. The mathematical framework defines motion dynamics through position, velocity, acceleration, and jerk, while Gaussian and Gamma functions are employed to generate smooth excitation signals. Simulations in MATLAB and experimental tests using a 15 V DC source, a ZK-6522C converter, an Arduino Mega, and a 37Dx73L gearmotor confirm the direct proportionality between voltage and motor speed. Results demonstrate that voltage shaping with Gaussian and Gamma profiles reduces overshoot, improves transient response, and enhances energy efficiency. The main contribution of this work is the experimental integration of continuous mathematical voltage profiles into motor excitation schemes, offering a novel alternative to conventional input signals in electromechanical systems.

ABSTRACT. In recent years, the widespread adoption of Unmanned Aerial Vehicles (UAVs) across diverse applications has increased the need for reliable condition monitoring systems to ensure their safe and efficient operation. Since UAV flight depends on the proper functioning of propellers, structural degradation in these components can lead to catastrophic failures. This work proposes a methodology based on acoustic signals and deep neural networks for the automatic detection of propeller faults. Acoustic signals were characterized through the extraction of descriptive, exploratory, and probabilistic statistical features, including mean, variance, standard deviation, root mean square (RMS), energy, skewness, kurtosis, zero crossing rate (ZCR), and entropy, which were used as inputs to multilayer perceptron (MLP) classification models. Among the evaluated features, the zero-crossing rate achieved the highest classification accuracy of 92.33%. The results demonstrate that the proposed methodology provides an effective and intelligent solution for UAV propeller condition monitoring, thereby improving operational safety, fault diagnosis, and predictive maintenance of unmanned aerial vehicles.

ABSTRACT. Reliable operation of induction motors depends heavily on the timely detection of stator winding faults, among which inter-turn short-circuits (ITSCs) are the most common and consequential. Left undetected, these faults tend to propagate rapidly through the affected winding, often resulting in costly repairs or complete motor replacement. Most deep learning approaches for ITSC classification rely on dedicated signal-processing stages, such as spectral, wavelet-based, or image transformations to highlight fault-related harmonics before classification, thereby adding unnecessary complexity to the overall pipeline. This work presents a lightweight one-dimensional Convolutional Neural Network (1D-CNN) that classifies ITSC severities directly from stator current signals, without complex preprocessing. The network architecture, comprising three convolutional blocks, was optimized via a Differential Evolution (DE) algorithm that jointly searched over the number of filters, kernel sizes, spatial dropout rate, and training optimizer, favoring compact configurations by construction. The resulting configuration was validated using stator current signals acquired under seven fault severities and four mechanical load conditions, which were merged into a single dataset per severity class. Five-fold cross-validation yielded an average classification accuracy of 99.91\%, with the corresponding confusion matrix showing a clean separation between the healthy condition and the earliest fault severity evaluated. These results indicate that a compact, filter-optimized 1D-CNN can match or exceed the performance of preprocessing-based approaches while offering a simpler and more efficient classification pipeline.

ABSTRACT. Electrical transformers are essential components of the power system; however, operating conditions can lead to faults, including short circuits between turns in the windings. Since a short circuit involving just a few turns can escalate into a serious fault, it is necessary to identify electrical variables that can detect and distinguish this type of fault. This paper proposes a statistical characterization of the transient change in primary current, the primary voltage angle—both when the moment the short circuit occurs—and the steady-state increases in the magnitude of the primary current. To do this, a pair of phase-locked loops (PLLs) is used in the primary winding to estimate the magnitudes and angles before and after the fault. Since both transient and steady-state variables must be measured, the PLLs must maintain high stability even when tuned to provide a fast transient response, thereby minimizing measurement errors and preserving the accuracy of the variables used in the statistical analysis. The methodology is validated through 34 experiments for each of the five distinct secondary-winding faults and under different load conditions. The statistical analysis, based on ANOVA, correlation analysis, statistical inference, and circular-linear correlation, identified variables that differed significantly between load-increase conditions and inter-turn short circuits. The results provide an initial basis for developing diagnostic algorithms for these faults.

ABSTRACT. Wind turbine blades are continuously exposed to mechanical and environmental stresses that may initiate cracks and progressively compromise the structural integrity and energy production of the system. Therefore, reliable and computationally efficient condition monitoring techniques are required, particularly for low-power wind turbines. This paper presents a crack severity classification methodology based on time-domain statistical moments and artificial neural networks applied to the vibration signals monitored from the tested wind turbine. Vibration signals acquired from a laboratory-scale wind turbine operating under four blade conditions, i.e., healthy, light, intermediate, and severe crack damage, were analyzed. Three statistical moments, namely variance, skewness, and kurtosis, were extracted from vibration signals measured along the Y-axis to characterize the dynamic behavior of the turbine. The statistical relevance of these features was evaluated using the Kruskal-Wallis test, confirming significant differences among the analyzed damage conditions. Subsequently, the selected features were used to train a multilayer perceptron neural network for automatic crack severity classification. Experimental results obtained from 400 vibration records demonstrated that the proposed approach achieved an overall classification accuracy of 98.5%, with perfect identification of the healthy and light-damage conditions

ABSTRACT. This paper presents the Hardware-in-the-Loop (HIL) implementation of a Machine-Side Converter (MSC) for a Permanent Magnet Synchronous Machine (PMSM)-based wind energy conversion system. The controller is implemented using an OP5650 real-time simulator, the RT-Lab/Simulink™ platform, and a TMS320F28335 Digital Signal Processor (DSP). The Field-Oriented Control (FOC) strategy is implemented by PI controllers in the current loop and in the angular velocity loop. Controllers are validated through power extraction and speed tracking tests. The HIL results demonstrate accurate reference tracking, balanced three-phase PMSM currents, and stable real-time operation, confirming both the effectiveness of the proposed control strategy and the suitability of the HIL approach for validating wind energy conversion systems before experimental implementation.

ABSTRACT. The increased complexity in the operation and planning of electrical power systems, resulting from the sustained rise in demand and the integration of renewable generation sources, has heightened the need for accurate tools to estimate short-term electricity demand. A reliable forecast of load behaviour enables improved operational decision-making and allows for the early assessment of the electricity system’s performance under different operating scenarios. This paper presents a methodology for estimating daily electricity demand using artificial intelligence techniques, with a focus on a model based on the Transformer Series (TS) architecture. The performance of the proposed model is compared with traditional forecasting methods, such as the ARIMA model and LSTM neural networks, using error metrics to assess the accuracy of the results obtained.

ABSTRACT. Accurate trajectory modeling of Battery Energy Storage Systems (BESS) is essential for energy management, degradation tracking, and grid-support control under dynamic operating conditions. This paper presents a comprehensive benchmark evaluating five data-driven machine learning techniques: ({\it i}) Random Forest (RF), ({\it ii}) Gradient Boosted Regression Trees (GBRT), ({\it iii}) single Decision Trees (RT), ({\it iv}) $k$-Nearest Neighbors (KNN), and ({\it v}) Long Short-Term Memory (LSTM) networks for BESS $SOC$ trajectory reconstruction. Grounded on discrete-time energy balance equations incorporating nominal capacity constraints and round-trip efficiency, the models are systematically validated across 12 real-world facility demand profiles encompassing industrial, commercial, and residential sectors under a complete initial state-of-charge sweep ($SOC_0 \in \{30\%, 50\%, 75\%, 100\%\}$). Results reveal that ensemble methods (RF and GBRT) consistently achieve superior tracking precision, yielding $\text{MAPE} < 0.03\%$ in multi-shift industrial profiles. In contrast, single decision trees suffer from localized step-wise discretization artifacts, doubling tracking errors, while KNN experiences severe performance degradation during abrupt load transitions (reaching $\text{NRMSE} > 4\%$). Furthermore, while continuous LSTM architectures exhibit slight offset drift during smooth operational cycles, they demonstrate unique boundary robustness when operating near upper saturation limits ($SOC_0 = 100\%$) under highly dynamic regimes. These quantitative insights provide actionable model-selection guidelines for data-driven BESS digital twins and real-time control applications.

ABSTRACT. This investigation deals with the design and validation of a regenerative load curve emulator based on a hybrid topology (SCIM-PMDC). The system uses a Squirrel Cage Induction Machine (SCIM) as the prime mover and a Permanent Magnet Direct Current Machine (PMDC) coupled to a boost converter as the active electrodynamometer. A cascade control strategy employing pole cancellation is proposed to ensure high-fidelity tracking of dynamic power profiles. The Simulink results demonstrate that the system eliminates steady-state error (from 18.34% in open loop up to 0%) by compensating for the speed drops of the Squirrel Cage Induction Machine, achieving 99.90% fidelity in the emulation of a complex hospital load profile, while recovering energy to a battery bank.

ABSTRACT. The expansion of hybrid photovoltaic–wind systems has increased the need for accurate characterization of renewable resource variability, as intermittency directly affects supply reliability. This study proposes a modular probabilistic–computational framework for solar–wind resource characterization, synthetic data generation, and the development of energy technical datasheets. The framework is applied to the La Rumorosa region in Baja California, Mexico, using hourly historical data from NASA POWER. The methodology comprises eight sequential modules, including data acquisition and preprocessing, statistical modeling using probability distribution functions, power and energy estimation, and synthetic data generation. For the wind resource, the Weibull distribution provided a highly satisfactory fit, with R²=0.9892 and RMSE=0.0052. By contrast, the solar resource exhibited multimodal behavior that could not be adequately represented by conventional unimodal models, resulting in negative coefficients of determination. The adoption of a bimodal Weibull–Weibull model addressed this limitation, achieving R²=0.6413 by simultaneously capturing low-irradiance regimes associated with cloud cover and high-irradiance regimes corresponding to clear-sky conditions. Synthetic data generation reduced the volume of temporal information by approximately 90%, representing monthly behavior through only three characteristic intrahourly days per month. Comparative validation between observed and synthetic data yielded an annual mean absolute error of 2.24% for the wind resource, indicating excellent agreement, and 12.04% for the solar resource, which is acceptable given the statistical complexity of irradiance. In both cases, the seasonal energy trend was accurately preserved. The results show that the proposed methodology provides a robust and reliable approach for energy characterization and synthetic data generation in hybrid photovoltaic–wind systems. In addition, it establishes a methodological basis for future applications involving green hydrogen storage, energy optimization, and dynamic control of renewable energy systems.