ABSTRACT. Globally, there is strong social inequality because of a complex ecosystem of production of goods and services, a global recession and contraction of markets. It is a multifactorial problem that involves geopolitical problems such as the war in Ukraine or the supply of semiconductors, to name just a few examples. The global labour shortage is widely recognised, and it has been the central focus of discussions at the recent World Economic Forum. This lack of labour is mainly due to the post-pandemic period, retirements, or early retirements. The effect is global and is observed in the lack of personnel in the primary sector in the case of Chile and in the manufacturing industry in the case of Mexico. To address this problem, it is necessary to increase production capacities both in the primary sector in Chile and in the manufacturing industry in Mexico, through the establishment of a framework for the integration of smart factories in small and medium-sized companies (SMEs) which represents almost the same percentage of GDP in both countries, 52% in the case of Mexico and 51.9% in Chile. The establishment of smart factories through the integration of innovative technologies such as AI and humanoids, collaborative robots, mobile robots using smart sensor networks and cloud computing will allow the acquisition of skills as well as their refinement and transfer to other twin factories regardless of geographical location, which will allow us to depend less on labour for merely repetitive core manufacturing activities such as those carried out during welding and assembly. Additionally, this would help redeploy staff to solve higher value-added problems and improve their conditions by upgrading new skills. In this talk we will show some examples of how AI techniques have been used by industrial robots to learn new manufacturing skills from scratch, honing their knowledge and improving through experience. We will discuss future work to enable robots to effectively acquire manipulation skills and how they can be used effectively for other robotic agents to learn through cloud computing.

ABSTRACT. Abstract—The climate crisis has made cleaner industry processes a necessity. The beer industry relies almost exclusively on fossil fuels to handle their most energy-intensive processes: mashing and boiling. Moreover, most micro breweries use manual or hysteresis control to regulate temperature during this stages, thermally stressing the malt, thus reducing its efficiency. In this context electric breweries, based on solid state relay (SSR), have been proposed to eliminate fossil fuel dependence. However, the direct connection of the heating element to the AC grid, does not solve the malt’s thermal stress problem. Moreover, this solution is limited in terms of scalabillity, due to the harmonic content imposed in the grid’s current by the hysteresis control of the SSR. This paper proposes a power converter configuration for a microbrewery, where the complete mashing and boiling processes are controlled by a two-converter configuration. The simulation results show a good tracking of the temperature during mashing and boiling and the grid current’s grid code compliance.

ABSTRACT. The use of the dual active bridge (DAB) as an electronic power converter in electric vehicles has gained interest in recent years thanks to features such as bidirectional power flow, galvanic isolation, high efficiency, high power density, low component stress, among others. Phase-shift modulation schemes are widely used for DAB control, being the most basic of them the single phase shift (SPS), followed by modulations with increasing number of degrees of freedom such as the extended phase shift (EPS), dual phase shift (DPS) and triple phase shift (TPS). A relevant concern in DAB applications is the appearance of transient dc components in the transformer current which are generated by load disturbances and the resulting sudden changes in phase-shift. This dc-bias current reduces the converter efficiency, induces higher current stress in semiconductor switches and increases the risk of failures due to the saturation of magnetic cores. This paper proposes the implementation of a hybrid modulation strategy which combines different phase shift modulation schemes to mitigate the dc bias of the transformer current under load impacts. The validation of the modulation scheme is done through simulation analysis with the PLECS software.

ABSTRACT. This paper proposes a tool for the evaluation of algorithms for grid synchronization using the Delfino LaunchPad TMS320F28379D platform, with the objective of supporting developers and researchers in the validation before their test in grid-connected equipment. The proposal uses the digital to analog outputs available in the LaunchPad to emulate the grid voltage, providing variations in amplitude, frequency, and phase and at the same time, providing relevant signals to evaluate the performance of the tested algorithm in a scope. The work presents (i) a global classification of synchronization techniques and evaluation scenarios, (ii) the tool characteristics and its implementation methodology, (iii) a study case and its results where two synchronization techniques are implemented in an open-source device, and (iv) final discussions of the tool and the result.

ABSTRACT. The Modular Multilevel Converter (MMC) is a topology employed and studied to manage incremental demand in Offshore and onshore wind plants. This takes advantage of profits such as scalability, modularity and power density in LFAC systems. Typically, this latter consists of a coupling points operative at low frequencies with a differential ratio of 50/3 Hz and 50 Hz, or 60/3 Hz and 60 Hz between the ports. As early publications, despite its industrial popularity, the MMC validated in PHIL test in laboratory scenarios considered lack of experimental results where is composed no more as 60 cells and even more for LFAC applications. Therefore, this paper presents the employment of dedicated cabinets to permit the Power Hardware In the Loop (PHIL) testing of an LFAC system composed of two MMC and operating with a back-to-back configuration enabling the bidirectional power flow at LF in one of its AC ports.

ABSTRACT. Multi-objective optimization holds practical relevance as the majority of real-world optimization problems involve multiple conflicting objectives. In this study, data-driven multi-objective optimization methods are analyzed from key aspects such as: employed algorithms, incorporation of machine learning techniques as surrogate models, number of evaluated objectives, consideration of model interpretability, and metrics used to evaluate the performance of optimization algorithms and surrogate models. The objective is to identify models that allow for reduced computational cost and present a viable alternative for solving complex real-world problems. This study distinguishes itself by providing a review and analysis of data-guided multi-objective optimization algorithms over the last six years. The results offer valuable guidance to the scientific community seeking to address complex problems using data-driven multi-objective optimization algorithms.

ABSTRACT. The efficient management of water and energy resources through pump scheduling is crucial due to the increasing limitations and costs associated with both resources. A common approach is to use explicit pump scheduling, controlling their on/off states during predefined time intervals. The optimization problem is multi-objective, simultaneously considering energy costs and water quality. This involves a significant number of decision variables, expanding the search space and slowing down the process. Therefore, measures to enhance algorithm performance are required. This study compares different solution representations to address this optimization problem, evaluating their performance using two metrics. Additionally, hyperparameter optimization is performed to improve the results. The findings illustrate variations among the different solution representations. Statistical analysis of the results identifies the most efficient representation in terms of optimal solutions and feasibility. This enables us to select the best representation for application to a larger network, resulting in improved efficiency in terms of solutions.

ABSTRACT. This study introduces an innovative muon tracking algorithm designed for the Coherent Neutrino-Nucleus Interaction Experiment (CONNIE). In this experiment, a cluster of twelve charge-coupled device (CCD) sensors is strategically positioned in close proximity to the Angra II nuclear reactor. The primary objective of the experiment is the detection of antineutrinos produced by the reactor, serving as a gateway to investigating non-standard neutrino interactions through Coherent Neutrino-Nucleus Scattering (CEvNS). However, the images acquired by these sensors reveal an abundance of muonic particles, originating from the collision of cosmic rays with the Earth's atmosphere. This study is fundamentally focused on the advancement of a muon tracking system that will allow to trace the trajectories of the muons or any other particle with high energy. The Geant4 toolkit has been used to create synthetic images that will be used to validate the algorithm, that will be use in the analysis of images collected by the CCDs of the experiment. In this work, we evaluated the algorithm's performance using synthetic images, achieving an efficiency of 98.78\%. This result underscores the algorithm's robustness and reliability in reconstructing muon trajectories.

ABSTRACT. Solar radio bursts (SRB) play a crucial role in understanding solar activity and its influence on Earth’s sys- tems. The classification of SRBs into distinct types based on morphology and frequency drift requires vast data and poses significant challenges for automated detection and classification. In this paper, we introduce a deep learning-based approach to address this challenge by leveraging a curated dataset from the CALLISTO network and a ground-based radio astronomy station. A convolutional neural network is trained to identify and classify SRB, despite the low signal-to-noise ratios, dynamic solar atmosphere and limited training data. This work contributes to the automation of SRB analysis and showcases the potential of deep learning in decoding complex astrophysical phenomena. Preliminary results demonstrate the efficacy of our approach, paving the way for a revolutionary advancement in the field of solar radio burst analysis.

ABSTRACT. Digital transformation has caused profound changes in how organizations operate, interact, and adapt to the current business environment to maintain and/or increase their productivity and competitiveness. One of the main effects of continuous technological dynamism has been in the labor market, particularly in the labor force, where organizations and individuals must invest in new skills and knowledge to avoid obsolescence. The coronavirus (COVID-19) pandemic accelerated the technology adoption process, causing organizations to opt to automate repetitive tasks, driving demand for new skills and fostering flexibility at work. The contribution presented in this work focuses on a literary review that addresses the labor market and digital transformation to determine how digital change in organizations has impacted the workforce. We followed a bibliographic review protocol at an exploratory level to meet this objective, consulting the Scopus database. In said review, we found that, of 22 related studies, 15 of them account, directly or indirectly, for the need to acquire new skills to face the challenges imposed by technological changes implemented in organizations on the workforce, this being a transversal requirement in different areas, such as real estate, educational, entrepreneurial, engineering, medical, among others. In future work, we will complement the study with a systematic literature review to explore how higher education institutions prepare the workforce to respond to the new demands regarding skills required in the work context.

ABSTRACT. This research paper investigates how cultural and gender factors affect technology adoption in small and medium enterprises (SMEs) in developing countries. Our analysis draws on Hofstede's cultural dimensions and the Gender Equality and Social Institutions (GESI) model. We explore how cultural norms and gender inequalities influence SMEs' technology adoption patterns. Our initial findings reveal that cultural and gender dynamics can significantly impact the ability and readiness of SMEs, particularly those led by women, to adopt and benefit from new technologies. Our findings emphasize the importance of recognizing and addressing cultural and gender obstacles to technology adoption, highlighting the potential for technology to boost SME growth and contribute to gender equality in developing countries.

ABSTRACT. The wildfire management involves the required activities to reduce both the risk and the impact, but the territorial knowledge and experience must be considered. An image dataset was developed by users contributions through a mobile application. Nevertheless, it is required to analyze the common visual characteristics against geographical nearest groups to improve the territorial decisions.

The aim of this work is to evaluate the accuracy of a set of algorithms based on image learning techniques, by considering the model architecture and, the label selection according to the wildfire domain. We assess the accuracy of transferring learning from pre-trained models in computer vision tasks to a particular domain. We apply transfer learning techniques to leverage prior knowledge and representations learned by the pre-trained models. The algorithms are developed and evaluated using a domain specific dataset developed by a community. Such dataset is highly related to forest fire images available on the citizen science platform E-ncendio, which is part of the FONDEF ID22I10072 project.

Classic Convolutional Neural Network models have been implemented, such as Inception-V3, Densenet, VGG-16, VGG-19, and the Residual Neural Network ResNet, which have proven to be capable of identifying visual elements that represent objects present in the landscape subject to forest fires. This allows the algorithms to be adapted to characterize community consensus, providing a territorial understanding of the perception of forest fire risk, as well as factors related to its prevention, response, mitigation, and recovery.

This automated approach has the potential to be applied in other fields that require the identification of visual elements in images, broadening its impact in various areas of research and development.

ABSTRACT. The efficient use of limited resources for public management results in increased healthcare coverage and prevention in comprehensive and quality care for the elderly. As this population segment continues to grow over the years, optimizing locations would have a significant impact on reducing the maximum distance between potential facilities and beneficiary allocation, with the aim of covering as much demand as possible, all within the constraints of budgetary resources, the care period, and staff efficiency in the context of public health focused on healthy aging. In Paraguay, healthcare for the elderly population does not have sufficient coverage to include the necessary integral health services. This work focuses on identifying optimal locations for temporary health services centers in the city of Itá, as well as determining the distribution of demand, the assignment of medical staff and the coverage period. A mathematical model based on the Set Covering Problem (SCP) and the P-Center model was designed, where it is determined that acquiring a mobile health center and temporarily settling in selected neighborhoods is more feasible than constructing a new healthcare center. These neighborhoods have been identified as the primary beneficiaries of mobile healthcare service coverage.

ABSTRACT. The MMC converter is the key topology for HVDC applications. In these systems, a major complexity is the number of modules and components required per phase to achieve nominal voltage and current levels for applications in transmission system. In this paper, the topology and control of a two-phase MMC system for HVDC applications is presented. The main idea is to eliminate a complete phase of the classic MMC converter for reduce the number of modules and the complexity of the system, together with the use of MMSPC-type modules, in order to achieve internal voltage balance without the need of extra sensors and control loops

ABSTRACT. The production of hydrogen through electrolysis is an already known process, and many power converters have been proposed as interfaces between the energy sources and the electrolyzers. On the other hand, in the last few years, modular multilevel cascaded converters have been proposed in many applications, such as high-power AC to AC or DC to AC conversion. Still, little to no references present this family of converters for hydrogen production, mainly because electrolysis requires the operation of a high-current, low-voltage port, limiting the options for this family of power converters. Therefore, this paper proposes a new topology for the modular multilevel cascaded converter and a control strategy to feed a proton exchange membrane electrolyzer using solar photovoltaic energy and the grid; this topology is the π-MMC 1ø. Several simulation results are carried out to validate the proposed topology and the control strategy.

ABSTRACT. The Hexagonal power converter has become a suitable solution to provide high-quality voltage/current waveforms while achieving decoupled control of three DC terminals as well. In this case, the internal voltage balance of the different storage units is the main concern for the correct operation of the Hexagonal converter. Moreover, the Modular multilevel seriesparallel converters (MMSPC) have become an interesting solution to provide higher operating voltages, reliability at a reduced cost, due to their ability to achieve a simpler internal voltage balance. In this paper, a Hexagonal power converter based on MMSPC for decoupled DC terminals is presented. The proposed system allows to implement a simple and cost-effective way to achieve a decoupled control strategy in each output of the system, to control the corresponding voltage/current of the system, along with maintaining the internal voltage balance of the proposed topology.

ABSTRACT. Modular multilevel series-parallel converters (MMSPC) have become a suitable solution to provide higher operating voltages, fault tolerance operation, and reliability at a reduced cost, due to their ability to achieve a simpler internal voltage balance. Therefore, MMSPC is a promising solution for applications where the nominal operating conditions are high voltage levels on the DC side and high current ratings on the AC side of the system. To allow this operation, it is necessary to design the system with parallelization in each phase. However, the parallelization in each phase of the system generates an internal cross-circulating current, which causes unbalance and internal losses. In this paper, the cross-circulating current effect is analyzed under parameters mismatch and compensated using coupled inductors.

ABSTRACT. The fast and flexible Electric Vehicle Supply Equipment (EVSE) requirement has increased as electromobility is massifed. Currently, high-power EVSE are reaching the 1 MW barrier, and, consequently, novel power converter topologies have been introduced as an alternative to the traditional dual active bridge-based EVSE. This paper proposes an enhanced control of a Series-Parallel Modular Multilevel Converter (MMC) based EVSE for fast and multiple electric vehicle charging applications. The proposed converter is regulated using standard decoupled MMC control strategies designed to provide Vehicle-to-Grid (V2G) capabilities. Simulation results are presented using a 500 kW model developed in PLECS to validate the proposed topology and control systems. Unlike previous research, these results demonstrate bidirectional power flow and support multiple port charging.

ABSTRACT. In this article, we address the challenge of optimizing a VGG-16 Convolutional Neural Network (CNN) for efficient weapon detection on computationally and memory-constrained devices. We utilize a strategic blend of techniques, including transfer learning, pruning, and quantization. The VGG-16 architecture was chosen due to its relevance for rapid and accurate weapon detection in crowded settings. Real-life datasets are employed for training to ensure practical applicability. After splitting the dataset, we explore various pruning levels and focus on a 60% pruning rate, accompanied by quantization for enhanced efficiency. Notably, the quantized model excels in edge computing, showcasing a significant speedup of 8.23x on a Vim3 platform and 2.51x on a Raspberry Pi 4, highlighting the effectiveness of our strategies compared to their implementation on a GPU-based platform. We take advantage of Open Neural Network Exchange (ONNX) quantization tools to perform accurate numerical conversions, resulting in faster inference speeds and more efficient resource utilization. In addition, analysis of memory consumption reveals substantial reductions, making it especially advantageous for edge computing

ABSTRACT. The present work focuses on the study of NARX and NARMAX models generated by mechanical techniques of automatic learning, such as ”Support Vector Machine” (SVM), ”Multilayer Perceptron” (MLP) and ”Extreme Learning Ma chine” (ELM), as well as deep neural networks such as ”Long Short-Term Memory” (LSTM), ”Gated Recurrent Unit” (GRU), Transformers and ”Convolutional Neural Network” (CNN). The aim is to carry out an analysis of the predictive capacity of each model. The purpose is to find the best technique and provide recommendations for the use of each one, taking into account the characteristics of the series, including its complexity, which is calculated using the MF-DFA method. The hypothesis is that models generated with deep learning techniques outperform shallow techniques. The results show that the hypothesis is not fulfilled for problems of low complexity, however it is true for problems of medium complexity.

ABSTRACT. In the dental world, early detection of different oral lesions is vital to carry out an accurate and precise treatment to care for the patient's health. However, it is not always possible for dentists to make an early diagnosis due to the limited number of symptoms or lack of experience. In this context, we propose developing a classification method based on deep convolutional neural network architectures that maximize caries classification performance on X-ray images. We build and explore three architectures named DCNN1, DCNN2, and DCNN3, which were trained and validated in an experimental data set formed by 1064 ROI samples with and without caries lesions. The DCNN3 model obtained the highest mean AUC score of 0.878 with 100 epochs during the validation in the training phase. However, the best-selected model was composed of the DCNN1 with 10 epochs. It reached a mean AUC score of 0.874, which did not represent a statistical difference compared to the DCNN3 model while being the simplest. Regarding the validation of the selected DCNN1 model in the test set formed by 106 samples, it obtained a mean AUC score of 0.830. The slight performance difference between the validation in training and test stages demonstrated the learning and generalizability of the proposed model, which is reasonable for the classification of caries lesions. However, any performance improvement depends on a larger data set, as the loss function behavior suggests.

ABSTRACT. In this article, we present the preliminary results of fine-tuning performed on a large language model (LLM) called Falcon, which comprises 7 billion parameters (7B) and is an open-source platform. Our primary focus lies in enabling the model to mimic university teaching, enabling it to respond to questions and emulate the instructional style of a teacher in the course “Circuit Theory 1”. To achieve this goal, we built a comprehensive database using all the online classes recorded during the pandemic for the aforementioned course. Leveraging the Whisper model, we accurately transcribed the audio content of these classes, thereby establishing the requisite corpus for finetuning the Falcon 7B model. To accomplish this task efficiently, we adopted effective techniques such as low-range adapters (LoRA) and data quantization (QLoRA), enabling us to conduct the fine-tuning process on a single GPU with a 15 GB capacity. Subsequently, we use GPT-4 compared the responses provided by the base model with those obtained after the fine-tuning process, using a set of 10 questions closely related to the course content. These comparisons enabled us to evaluate the effectiveness and precision of fine-tuning applied to the Falcon 7B model. The results obtained so far show a good result in fine tuning the Falcon 7B model to imitate university teaching and the quality of the responses about the course. This innovative approach has the potential for valuable applications within education, providing an efficient tool for interactive learning and answering online queries. Index

ABSTRACT. The exponential increase in devices connected to the Internet has rendered them vulnerable to various types of cyberattacks, disrupting their proper functioning. As a result, it is crucial to have reliable Intrusion Detection Systems (IDS) that can identify malicious activities on the network. However, the unpredictable nature of network behavior and the large volume of data to be audited pose significant challenges. This work aims to address these challenges by eliminating redundant data features, identifying a subset of relevant and representative features that enhance the anomaly detection performance of an IDS using many-objective optimization algorithms.

In this work, we have analyzed the recent TON_IoT dataset, which collects monitoring data on devices using Windows and Linux operating systems. The NSGA-II, NSGA-III, RVEA, and MOEA/D were used as many-objective feature selection algorithms. Our computational simulations identified good subsets of features that showed better classification performance and accuracy in comparison to the complete set of features. Notably, the results obtained using the NSGA-II stood out from the others.

ABSTRACT. As multi-agent systems transition from academia to industry, the integration between non-human agents and human operators is becoming the standard. Consequently, task allocation presents a complex challenge, where both identical and non-identical agents vie for assignments. When conventional task allocation algorithms are presented with human workers, they often fail to respond, resulting in unassigned tasks. The proposed algorithm facilitates human interaction within a multi-agent, heterogeneous robotic system, enabling them to compete for available tasks and report task completion using a Markov Decision Process (MDP) based approach. Through simulations, the algorithm demonstrates significantly improved performance when operating alongside non-human systems compared to classical algorithms. This advancement empowers humans to engage in both task generation and task completion. The algorithm showcases the interaction between human and non-human agents within collaborative environments, mirroring the dynamics anticipated in real-world industrial operations.

ABSTRACT. In the context of the increasing global interest in sustainable transportation solutions, this research investigates South America's potential role within the electric vehicles (EVs) supply chain. The objective of this study is to conduct a comprehensive literature review to identify opportunities for South American involvement in various stages of the EV supply chain, with the aim of creating conductive scenarios for future research endeavors. The bibliometric analysis reveals limited scientific ouput in the region concerting the EV supply chain, with Brazil emerging as the most active contributor since 2010. However, recent developments show heightened interest from Chile, Colombia, Mexico and Peru. This research underscores the region's engagement in lithium extraction, crucial for EV batteries, but points out the lack of progress in other stages of the chain at the industrial level. Studies suggesting the feasibility of component production in Brazil through its economic relationship with Argentina, along with the proposed Bolivia-Paraguay integration to produce lithium-ion batteries and EVs, point out how regional collaboration can drive advances in other stages of the chain, also highlighting the importance of investment in research and the effective use of the region's resources. It raises the possibility that the region can play an important role in the EV industry if it seizes opportunities in a joint quest towards sustainable mobility.

ABSTRACT. The Internet of Things (IoT) based applications rapidly engulf human-driven systems. The inherent benefits, like process automation, have motivated researchers to propose novel IoT applications in multifaceted domains, contributing towards the economy of effort and conservation of precious resources. Existing literature emphasizes innovative water monitoring systems, but more attention is needed to efficiently use data from water sensors. The paper initially proposes a novel IoT-enabled water management system for data collection and analysis over an extended period. Based on the data collected after system deployment, we leverage the power of statistical techniques and Machine Learning to identify valuable trends from actual water consumption records, such as isolating the source of water wastage, forecasting future water demands, detecting anomalies in data and evaluating the impact of different environmental variables on water consumption. This paper also proposes a subsystem to control the usage of water.

ABSTRACT. —In the present century, abundant scientific evidence has been produced about the well-being generated by the bodies of water in cities, whether natural or artificial. Making an analogy with green areas, these bodies which can be lagoons, fountains, rivers, ponds and other outdoor water sources, have been denominated as "Blue Spaces", showing various positive impacts on mental health and on the incentive to physical activity of the population that access these assets. Today, the operational continuity of these spaces strongly depends on automation and in this context the visible light can contribute through the aesthetic function of Ornamentation, the scanner function of Measurement and the organic function of Communication. However, this automatic system, cause of entropy, it is exposed to the inexorable deterioration of its attitude, which must be recovered. Since it is unknown how these three functions are diminished and then recovered, in this article we propose a model inspired by biological phenomena. This model considers a working stage at night when the attitude of the system decreases as occurs in some animals while they work, and the next stage in the day when the attitude recovers due to the same animals resting. The aim of this bionic model is to contribute to improving the performance of Blue Spaces through the harmonic management of light as a beauty, monitoring and telecommunication agent.

ABSTRACT. In this work Data Mining techniques were applied to atmospheric contamination gases and meteorologic data collected during the pandemics of COVID-19 in order to extract relevant knowledge about its behaviour. These data were collected in the greater area of Asuncion, Paraguay during three time frames, before the start of pandemics, during the strict movement restriction and finally during the flexible movement restriction. Additionally, this data was correlated with daily new cases and deaths of COVID-19. Spearman correlation, association analysis and temporal statistics studies were used for obtaining significant knowledge. The conclusions showed that humid and cold conditions tend to offer better air quality and the highest contamination levels are between 10 AM and 10 PM. The contamination levels correlation with the pandemic data are mainly negative except the Ozone. Another conclusion is that atmospheric pollution by carbon and sulfur has an anthropogenic origin, mainly due to human mobility activities and particularly the fossil fuels based automotor traffic. This was observed as the Carbon Monoxide was reduced by 31% and the Sulfur Dioxide was reduced by 73% in areas with high automobile traffic. The data processing showed increase of up to 100% of Ozone levels and 32% in the particulate matter related to the normal prepandemic values.

ABSTRACT. The global environmental crisis is one of the serious threats facing humanity, specifically water pollution, has become one of the main concerns of all societies, due to the harmful effects it has on health, as it endangers all living beings. This paper presents a state of the art in the advances of water quality monitoring systems that can help prevent these adverse effects, however, there is a gap between progress in this area in developed countries and Latin American countries. On the other hand, decision making to mitigate the negative impacts requires a programmable platform. IoT technologies, with real-time data collection, and the most used models based on Machine Learning in this field are reviewed

ABSTRACT. High energy demand has created management and organizational problems in many countries around the world. The world is experiencing a crisis that drives us to take action and contribute to new developments and research to influence the energy transition through a combination of control strategies to improve these processes. This promotes the use of non-conventional renewable energy to sustain life and promotes sustainable and efficient processes. The importance of developing new applications, using new technologies, and designing appropriate controls to optimize processes that require energy conversion. The focus of this work is the design, simulation, and implementation testing of an optimal control in the multilevel neutral point clamp (NPC) topology. The proposed control, by using an optimal state feedback based on the linearized model of the converter, allows to optimize the use of energy in the reference variations. Results are presented for current control, power control, and imbalance control between the coupling capacitances.

ABSTRACT. Fractional frequency transmission systems are capable of transmitting power through longer distances when compared with currently used transmission systems. However, research on the power electronics topologies required to transform the frequency of the system is still being carried out. In this paper, a control strategy based on model predictive control is proposed to regulate the cluster capacitor voltages of an AC-AC converter topology denominated as hexverter. The feasibility of the proposed control strategy is demonstrated using simulation results of a hexverter connected to two AC systems with different frequencies.

ABSTRACT. This study addresses the challenge of tuning a Finite Control Set Model Predictive Controller (FCS-MPC) in power electronics using Reinforcement Learning (RL). The focus is on a two-level voltage source converter (2L-VSC) equipped with an LC filter and resistive load, regulated by FCS-MPC. The primary strategy is to minimize a cost function that includes terms for reference tracking and current constraints, while an RL agent is used to adapt the cost function by also considering the Total Harmonic Distortion (THD). Two types of actor-critic RL agents, the Deep Deterministic Policy Gradient (DDPG) and Twin Delayed DDPG (TD3), are examined and compared. The study highlights the advantages of TD3, which exhibits more stable and rapid convergence in the training process.

ABSTRACT. Grid-forming inverters (GFI) have gained significant importance as they are used to supply alternating current (AC) in distributed generation systems and islanded microgrids, where inductive-capacitive-inductive (LCL) output filters are commonly employed to mitigate harmonics injected by voltage-source inverters (VSI). Advantages such as the simple operational principle, fast dynamic response, multivariable control, among others, have contributed to establishing Model Predictive Control (MPC) as a suitable alternative for controlling the inverter systems with LCL filters. However, it presents a significant disadvantage: it relies on the phenomenological model of the system, thus disturbances or random variations in parameters deteriorate control quality and performance. This work proposes the implementation of a model-free voltage predictive control for a grid-forming inverter with LCL filter, based on an ultralocal predictor using only input and output data from the system. The proposed method is validated through simulation analysis and experimental results to demonstrate its real-time applicability.

ABSTRACT. Grid-forming inverters (GFI) are important elements in modern AC microgrids, operating as high-bandwidth controllable voltage sources. Finite-Control-Set Model-Predictive Control (FCS-MPC) has demonstrated to be an effective alternative to conventional linear voltage controllers due to advantages such as fast dynamic response and intuitive design. However, the dependence on an accurate mathematical model of the plant is a drawback that limits the performance of FCS-MPC when facing model uncertainty or time-varying parameters in the controlled system. To overcome this problem, Model-Free Predictive Control (MF-PC) strategies have been developed using ARX time-series models for prediction, with parameters estimated online from available data mesurements. This paper proposes a variant of the MF-PC method with ARX predictor that fully eliminates the dependence on physical parameters and reduces the number of required measured variables and parameters to be estimated. The real-time feasibility of the proposed MF-PC is verified by Hardware-In-the-Loop (HIL) experiments.

ABSTRACT. Gemstones, such as diamonds, are used in various applications, from jewelry to technology, where they have recently been considered as semiconductor materials. However, the value of diamonds is difficult to measure due to their price being influenced by characteristics such as cut, color, clarity, and carat weight, making the estimation of diamond value a complex and sometimes subjective task. Currently, regression models are being developed to estimate the value of these precious stones. To support the estimation of diamond value and improve the training time of predictive models, this research proposes the multiclass classification of diamond values using standard ELM, regularized ELM, and weighted ELM. The classification was based on 4 value categories with respect to their prices: (a) less than US$500, (b) between US$500 and US$1000, (c) between US$1000 and US$1500, and (d) over US$1500. The results obtained are presented based on accuracy and model training time. Of the evaluated models, the regularized ELM presented the best results, with an accuracy of 0.8375 and a runtime of 109 seconds. The results demonstrate that ELMs can efficiently classify diamond prices, and the models are robust, showing the price trend, and the main classification errors of the models are generated in classes with prices very similar between diamonds.

ABSTRACT. In datasets, the preponderance of imbalanced classes impedes accurate cyberattack categorization. While high aggregate accuracy is sought, it's paramount to adeptly classify all attack types, especially the under-represented ones. Existing methodologies, such as Ensemble techniques and the Synthetic Minority Oversampling Technique (SMOTE), address these disparities, yet the dynamic nature of underrepresented cyberattacks in cybersecurity remains a concern.

To address this, we introduce a nested cascade model tailored for diverse cyberattacks within imbalanced datasets. This model leverages binary classifiers across tiers, each targeting a specific attack type. Before initializing the cascade, SMOTE is applied to counterbalance class disparities.

The cascade's classification sequence employs a dual strategy: an initial one-vs-all binary classifier approach for pending classes, followed by prioritization based on model performance.

We assessed our approach using the UNSW-NB15 dataset. Preliminary results indicate approximately 80% efficiency across metrics like accuracy, recall, and F1-score. Notably, SMOTE's integration yielded significant improvements for underrepresented classes.

ABSTRACT. Sokoban-type puzzles and their variants involve complex planning and decision-making challenges in wide and deep search spaces. Such scenarios are interesting for testing path-planning algorithms. This document uses A* and IDA* (Iterative Deepening A*) algorithms to solve Zen Puzzle Garden game boards; a Sokoban-type puzzle where you need rake the sandy area completely. The algorithms are evaluated in terms of speed and node expansion in moderately complex boards. Additionally, we implement three improvements: transposition tables, move ordering, and deadlock tables. The results show that IDA* with the three proposed improvements is on average faster than A*. Although its strength is its low memory usage, it can be seen that it is very competitive versus A* in terms of time usage. Transposition tables and deadlocks tables reduce by 500 times the number of nodes expanded by the IDA* without improvements. On the other hand, the ordering of nodes is not very effective for some boards, which causes a worse performance of the IDA* algorithm, giving rise to greater exploration in boards that do not lead to the solution. Studying these algorithms can be beneficial in the field of path-finding for Sokoban-type game domains or solver-based procedural content generators.

ABSTRACT. Currently there is no standard metric for difficulty in games that can be applied in all domains, regardless of the game. This is not a surprise, because of the subjective factors involved in defining difficulty. However, studies that address the design and application of metrics that can be applied to multiple games are still important, because they can still be used to increase the quality of content generated by automated agents. We take a look at applying information entropy concepts in the game Asteroid Escape and compare the results with other metrics to assess difficulty used in these types of games. This approach is largely based on future work proposal from a previous study done on The Witness puzzles that yielded positive results on using an automatic method to assess the difficulty of user rated puzzles. This study also includes a modification of BFS and A* capable of finding every shortest solution. From the implementation, it is possible to look at contradicting results with the literature, where basic metrics had a better result correlating with an official definition of difficulty from the game designers. The impact expected from this study is to propose an update to the the corpus of puzzle games where the uncertainty metric for difficulty has been applied. As having an accurate assessment of difficulty can open the doors to other studies, such as improving the quality of the output from procedural content generation methods in games or propose an automated method of difficulty adjustment based on this metric.

ABSTRACT. The use of biometric systems for individual identification is an alternative that provides great data security since this technology uses unique and distinctive information about the person. Palm vein recognition has been raised as an efficient technique for detecting people without compromising data security because vein structures are under the skin and only can be captured in a living body. Besides, palm vein patterns are related to soft biometric traits such as gender and age, allowing the development of multi-task learning models. This paper introduces a multi-task extreme learning machine (ELM) model for palm vein multiclassification to simultaneously process identity and soft biometrics data. The proposed methodology evaluates a single ELM model that shares the hidden layer weights and has three outputs for identification, as well as gender and age classification of individuals. The evaluation was performed on the VERA palm vein database, which includes these soft biometric labeling metadata. Although no improvements were reached in terms of training time compared to a standard ELM, it is possible to simplify the number of neurons to optimize the multi-task ELM. The proposed method contrasts with the standard ELM, which requires generating a specific model for each task and adjusting the number of neurons for each one. Experimental results on multi-task classification show an average accuracy of 95% with only 700 hidden neurons. These results indicate the feasibility of simultaneous identification and classification using a multi-task ELM on palm vein patterns.

ABSTRACT. In the field of geosciences research, due to the large amount of complex data, clustering methods are a necessary as a first approach to provide insight into the data without previous analysis. Uniform Manifold Approximation and Projection (UMAP) has recently gained popularity as an effective clustering algorithm in several fields. We explore the application of UMAP as a pre-processing technique to enhance clustering results on 89 geochemical samples from the sediments of the Rapel River fluvial system with 25 variables. The clustering method was performed using Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDSBSCAN), and the results were evaluated using the Silhouette Score. The generated clusters were then compared to the data origin results and interpretation. We demonstrate that UMAP can construct a favorable representation for clustering purposes, distinguishing different rivers denoting their own characteristics based on their geochemical data.

ABSTRACT. In the global mining industry, periodic monitoring of large Mining Waste Deposits (MWDs) is essential. This paper presents an innovative approach leveraging advanced artificial intelligence techniques combined with high-resolution Sentinel satellite imagery to accurately geolocate MWDs in Chile. Our methodology involves segmenting satellite images, expert labeling, and training state-of-the-art models for precise detection. Experimental results, as summarized in the table below, highlight the superior performance of the YOLOv7 model with an AP Box of 0.75 and an AP Mask of 0.72, outperforming other tested configurations like Mask RCNN with FPN 3X. The results demonstrate the potential of this approach in optimizing the management and monitoring of MWDs, contributing significantly to safety and sustainability in the mining sector. This research not only provides insights into the capabilities of AI in geospatial analysis but also sets a benchmark for future studies in the realm of mining and environmental management.

ABSTRACT. Currently, chloride leaching is the most efficient technique available for copper recovery in the low-grade mining segment (below 0.4% CuT). No better biological or hybrid alternative processes have been found to date. However, there is still a lack of knowledge regarding the optimal operational and design parameters for leaching heaps to ensure sustainability. Identifying optimal operational values involves determining the optimal dosages of water, sodium chloride, and/or calcium chloride, as well as the optimal temperatures for the process at various stages, aeration requirements for the heaps, input and output humidity, among other factors. This work proposes a system to diagnose the health status of a chloride leaching heap with the support of machine learning and to forecast copper recovery levels.

ABSTRACT. In this paper, we employ an experimentally designed Model Predictive Control (MPC) framework coupled with a Kalman filter to regulate a point absorber Wave Energy Converter (WEC). This methodology entails the utilization of a mathematical model representing the point absorber, enabling the controller to anticipate the energy converter's future behavior. Executing the MPC necessitates real-time data from sensors, which can be subject to white noise. Through the application of the Kalman filter, these inherent errors are effectively mitigated, enabling accurate estimation of adjusted state variables. This, in turn, leads to the generation of control signals and system outputs that correspond to position and velocity. The outcomes of the study highlight that this control strategy exerts a positive influence on power extraction, as it ensures alignment between velocity and excitation force, ultimately enhancing the efficiency of the process.

ABSTRACT. The objective of this work is to study how to minimise wave reflection in the wave channel of the Universidad del BioBio Wave Lab by means of different porous media. The experimental study was made defining six initial wave states, in which all experiments were carried out. Four situations were compared: first, without any wave absorber; second with a sponge wave absorber already present in the channel; third with a porous plate absorber in series; and finally with a triangular mesh absorber with sponges. Wave train-free surface elevation data were collected using three carefully spaced ultrasound sensors and used to calculate the reflection coefficient associated with each wave structure and parameter using Matlab code. The results revealed that the first scenario showed the highest reflection, while the second scenario showed the second highest reflection coefficient. On the other hand, the third situation presented the lowest reflection coefficient, and in the last situation, a slightly higher coefficient than the third case was obtained. These results highlight the efficiency of the two proposed absorber structures and their performance in reflection reduction. In conclusion, the porous plate absorber showed higher mitigation of the reflection, followed by the triangular mesh absorber and sponges. This study could contribute to future research conducted in the wave-generating channel, providing valuable insight for practical applications in the field of coastal research and engineering

ABSTRACT. The Modular Multilevel Matrix Converter (M3C) has been indicated as a suitable option for Low-Frequency Alternating Current (LFAC) transmission due to some characteristics such as simple power and voltage extensibility, controllability and high efficiency. The control of the M3C in LFAC is simplified because the floating capacitor voltage oscillations are relatively small. However, when connecting to AC ports of different frequencies, AC transformers at fractional or nominal frequencies may be required. This can lead to an increase in the built-in ratio of such systems. To address this issue, this paper proposes a simple control strategy for a Back-to-Back M3C equipped with a medium-frequency port. Simulation results for validating the topology and control system of a high-power LFAC application are presented for a 396 MW case study.

ABSTRACT. This paper presents a comparative study of three control methods to control the output voltage of a nine-level Flying Capacitor Converter (FCC) in DC/DC buck applications. The methods considered are: Linear controller with Phase Shifted Pulse Width Modulation (PS-PWM), Finite Control Set - Model Predictive Control (FCS-MPC), and Sequential Phase Shifted - Model Predictive Control (SPS-MPC). MPC methods require a detailed model of the system to obtain good results, so a refines loss model is incorporated into the standard FCC equations to minimize output errors, particularly at lower voltage levels and high switching frequencies. This model includes the output inductor and GaN semiconductor's behavior. Simulation results are used to compare the dynamic and steady-state behavior of these three methods showing the excellent performance of SPS-MPC in both cases.

ABSTRACT. In this work, we present the speed control of a six-phase induction machine, employing an inner loop predictive current control based on the model, powered by a matrix converter. This system aims to harness the advantages offered by multi-phase systems and matrix converters, utilizing their multimodular topology in three-phase modules. We showcase the design of the proposed control strategy and the analysis of the results obtained in the simulation environment in terms of current and machine speed tracking.

ABSTRACT. Radial flux permanent magnet machines are of widespread use in several applications, particularly in high- performance applications since permanent magnets offer high- efficiency in a wider range of power and working speed. Electrical machines suffer from several mechanical imperfections such as manufacturing tolerance and eccentricity, the latest become more critical in high-speed operation. Eccentricity is a phenomenon affecting particularly radial flux machines introducing mechanical vibrations, reducing the machine efficiency, and even lead to motor failure. This article provides with a mathematical approach to estimate the airgap magnetic flux density of radial flux permanent magnet electrical machines.

ABSTRACT. Metagenomics studies the genetic information of microbial communities in different contexts. As metagenomic DNA is often fragmented and then sequenced into small reads, these reads can be assembled into longer sequences called contigs. An important step in the metagenomic analysis pipeline is Binning, which corresponds to the classification (supervised) or clustering (unsupervised) of reads or contigs. In the case of unsupervised Binning, several Machine Learning algorithms that use DNA sequence descriptors, such as k-mers Frequency and GC Content to perform clustering, have been employed. This paper proposes the use of Unsupervised Extreme Learning Machines (US-ELM) for Metagenomic Binning. The experiments use three datasets with different numbers of species present, and compare the results obtained by US-ELM with respect to the k-means and Maximization Expectation (ME) algorithms. The performance comparison employed metrics widely used in the problem, such as Accuracy, Rand's index, and Clustering Computation Time. From the experiments, we can see that US-ELM outperforms the other two clustering methods both in quality and in computational time, showing that it has an interesting potential in the Metagenomic Binning problem.

ABSTRACT. This paper presents as a proof of concept the methodology for the detection of movement of individual fingers (thumb, index, and middle finger) using EMG sensors placed on two muscles on the forearm of one person. Two MyoWare Muscle sensors are used together with an Arduino UNO microcontroller board that works as a data acquisition system to collect electromyography signals from muscles. These signals are then transformed using short-time Fourier transform with the aim of having the signals in the time-frequency domain. After filtering, the highest peaks in both signals are identified in real-time to detect the movement of an individual finger. Finally, the signals for the thumb, index, and middle fingers were identified as an initial step towards the development of a prosthesis in which each finger will have its own movement in future work.

ABSTRACT. This study proposes a methodology for iris-based diabetes detection in 130 subjects, in which geometric transformations and changes in brightness and contrast were applied to increase to 1300 images, and a selection of 10% of the pixels were selected, and 13 principal components were used to feed an Extreme Learning Machine with the Adam optimization algorithm, a learning rate of 0.01, 256 neurons in the hidden layer, and a batch size of 128. After performing five-fold crossvalidation, the results demonstrated balanced performance, with a mean accuracy of 0.9992, mean F1-score of 0.9988, and mean AUC of 0.9999 for diabetes detection.

ABSTRACT. This work proposes exploring two active contour models, the Geodesic and Chan-Vese, to maximize the mass segmentation performance on mammography images. Both models were optimized in terms of initialization radius and number of used iterations and validated on an experimental data set containing 115 images with mass lesions. The best-selected Chan-Vese model, with a radius of 50 pixels and 436 iterations, outperformed the best Geodesic model, attaining a mean Dice score of 0.812 versus 0.558. This result highlighted the successful performance of the Chan-Vese model in segmenting mass lesions from different images. It also demonstrated the Geodesic model’s tendency to get stuck in local minimums. The median and CLAHE filters were essential to improve the mass lesion boundary quality before the segmentation step.

ABSTRACT. The European hazelnut is one of the most important nut fruit trees worldwide. However, there is little information on the interaction between different productive variables that influence fruit yield per unit area in different regions of the world. In this regard, the present research aims to identify the most important auxiliary variables of high spatial resolution that allow describing the yield of the European hazelnut. The proposed methodology allowed the identification of seven productive management variables, which would be closely related to the yield of this crop. These variables would be the multispectral indices calculated using satellite images Sentinel-2; Advanced Vegetation Indices (AVI), Enhanced Vegetation Indices (EVI) and Soil Adjusted Vegetation Indices (SAVI), together with two-dimensional and three-dimensional leaf area, xylem water potential and diameter and total number of axles per plant. The selected variables showed a high spatial dependence in the experimental semivariogram with values > 0.8. Additionally, a high correlation with performance was identified (R > 0.6). These variables allow us to understand that the yield achieved by the European hazelnut tree depends on the vegetative expression of the plant achieved within the season. On the other hand, these variables allow us to characterize the spatial variability of the European hazelnut and understand the high spatial dependence that exists between the different sampling sites at the intra-farm scale of the orchard.

ABSTRACT. Physiological processes in plants exhibit dynamism, necessitating the utilization of innovative mathematical tools for comprehension. Electrical signals play a critical role in both intercellular and intracellular communication. The biomathematical modeling of these electrical signals requires a comprehensive understanding of models that involve periodic signals. This article provides a concise overview of the fundamental principles of oscillatory signals, focusing on a specific arithmetic approach known as proportional arithmetic. Furthermore, the introduction of q-periodic functions, proportional models for heat and wave phenomena, and the Fourier and Laplace proportional transform is presented. These mathematical tools possess unique characteristics and hold significant potential for modeling electrical signals in plants. The application of this biomathematical modeling approach will advance the understanding of plant electrophysiology and the mechanisms that underlie it.

ABSTRACT. Dormancy, a plant physiological phenomenon, plays a crucial role in the adaptation of crops to adverse environmental conditions. This article presents a thermocyclic activation temperature method for geospatial endodormancy release. In particular, it provides an accurate estimate of the date of dormancy interruption in deciduous crops in the Ñuble Region, Chile and would have a direct impact on agricultural decision-making.

ABSTRACT. The digitalization of biomathematical models offers new opportunities for understanding and predicting the spread of Gray mold in grapevines. By collecting phenological, meteorological and epidemiological data, computational models can be formulated so they simulate the progress of infection under different environmental conditions and grape growth stages. Implementing these digitalized models allows for the analysis of various scenarios and evaluation of the impact of key variables such as temperature, humidity, and the presence of \textit{Botrytis} spores on disease incidence and severity. This provides valuable information for vineyard growers and winemakers, enabling them to make informed decisions regarding Botrytis management and control practices. The digitalization of biomathematical models also facilitates the integration of real-time data, such as automated weather stations and field sensors, improving prediction accuracy and responsiveness to environmental changes. Additionally, these models can be used as training and educational tools, helping growers and adviser the wine industry gain a better understanding of the biological and epidemiological processes related to gray mold.