Browsing by Author "Salas, Juan Carlos"

Now showing 1 - 6 of 6
  • Thumbnail Image
    Item
    An Overview of Microgrids Challenges in the Mining Industry
    (2020) Gómez, Juan S.; Rodriguez, Jose; Garcia, Cristian; Tarisciotti, Luca; Flores-Bahamonde, Freddy; Pereda Torres, Javier Eduardo; Nuñez Retamal, Felipe Eduardo; Cipriano, Aldo; Salas, Juan Carlos
    The transition from fossil fuels to renewable energies as power sources in the heavy industries is one of the main climate change mitigation strategies. The carbon footprint in mining is related to its inherent extraction process, its high demand of electric power and water, and the use of diesel. However, considering its particular power requirements, the integration of microgrids throughout the whole control hierarchy of mining industry is an emergent topic. This paper provides an overview of the opportunities and challenges derived from the synergy between microgrids and the mining industry. Bidirectional and optimal power flow, as well as the integration of power quality have been identified as microgrid features that could potentially enhance mining processes. Recommendations pertaining to the technological transition and the improvement of energy issues in mining environments are also highlighted in this work.
  • Thumbnail Image
    Item
    Forecasting copper electrorefining cathode rejection by means of recurrent neural networks with attention mechanism
    (2021) Correa Hucke, Pedro Pablo; Cipriano, Aldo; Nuñez Retamal, Felipe Eduardo; Salas, Juan Carlos; Löbel Díaz, Hans-Albert
    Electrolytic refining is the last step of pyrometallurgical copper production. Here, smelted copper is converted into high-quality cathodes through electrolysis. Cathodes that do not meet the physical quality standards are rejected and further reprocessed or sold at a minimum profit. Prediction of cathodic rejection is therefore of utmost importance to accurately forecast the electrorefining cycle economic production. Several attempts have been made to estimate this process outcomes, mostly based on physical models of the underlying electrochemical reactions. However, they do not stand the complexity of real operations. Data-driven methods, such as deep learning, allow modeling complex non-linear processes by learning representations directly from the data.We study the use of several recurrent neural network models to estimate the cathodic rejection of a cathodic cycle, using a series of operational measurements throughout the process. We provide an ARMAX model as a benchmark. Basic recurrent neural network models are analyzed first: a vanilla RNN and an LSTM model provide an initial approach. These are further composed into an Encoder-Decoder model, that uses an attention mechanism to selectively weight the input steps that provide most information upon inference. This model obtains 5.45% relative error, improving by 81.4% the proposed benchmark. Finally, we study the attention mechanism’s output to distinguish the most relevant electrorefining process steps. We identify the initial state as a critical state in predicting cathodic rejection. This information can be used as an input for decision support systems or control strategies to reduce cathodic rejection and improve electrolytic refining’s profitability.
  • Thumbnail Image
    Item
    Identificación de estados fisiológicos orientada a la prevención de accidentes laborales en base a datos fisiológicos no invasivos y técnicas de inteligencia artificial
    (2021) Fouere Andrade, Francisco Antonio; Salas, Juan Carlos; Núñez Retamal, Felipe Eduardo; Pontificia Universidad Católica de Chile. Escuela de Ingeniería
    Los estados fisiológicos alterados, como el estrés y la fatiga, muchas veces asociados al contexto laboral, afectan negativamente a las personas, promoviendo la aparición de enfermedades cardiovasculares, asma, trastornos del sueño, ansiedad, depresión, entre otros. Adicionalmente, los estados fisiológicos alterados afectan el rendimiento laboral de los trabajadores, disminuyendo los indicadores de productividad de las empresas, y transformándose en una causa de accidentes laborales. En este trabajo, se desarrolló un sistema que apunta a identificar los estados fisiológicos en los cuales se encuentra cada trabajador minuto a minuto, mediante una recolección de datos cardio-respiratorios no invasiva, que representa una gran oportunidad para mitigar los perjuicios ya expuestos. Para ello se tomaron datos de electrocardiograma y frecuencia respiratoria de seis personas durante un periodo de una semana. Con estos datos se entrenaron una serie de modelos de aprendizaje profundo para clasificar en qué estado fisiológico se encuentra cada persona minuto a minuto. El modelo con mejores resultados según una función de pérdida de entrenamiento que considera la divergencia de Kullback-Leibler y el error cuadrático medio, fue el modelo llamado “What Color”, que se compone de un autoencoder variacional con redes convolucionales al codificar y decodificar, arquitectura de la cual se toma el espacio latente para generan tres clústeres que corresponden a tres niveles de estados fisiológicos para cada persona, mediante la técnica de mezcla de gaussianas. La significancia de cada clúster se justifica a partir de cinco análisis, la observación visual de las señales de intervalos RR que vive en cada uno de los clústeres, la relación entre el estado fisiológico, la postura de la persona y los periodos de sueño, los valores promedios y desviaciones estándar existentes de las variables relevantes de cada clúster, análisis del estado del arte sobre estados fisiológicos que surgen de forma espontánea en los resultados de la investigación, y finalmente, la aplicación del modelo a un repositorio externo de datos fisiológicos obteniendo resultados acordes a los esperados.
  • No Thumbnail Available
    Item
    Neural Network-Based Model Predictive Control of a Paste Thickener Over an Industrial Internet Platform
    (2020) Núñez Retamal, Felipe Eduardo; Langarica Chavira, Saúl Alberto; Díaz Titelman, Pablo; Torres, Mario; Salas, Juan Carlos
    This article presents a real implementation of a neural network-based model predictive control scheme (NNMPC) to control an industrial paste thickener. The implementation is done over an Industrial Internet of Things (IIoT) platform designed using the seven layer reference model for IIoT systems. Modeling is achieved using an encoder-decoder with attention recurrent neural network, while MPC search is done using particle swarm optimization. An industrial evaluation is presented, which highlights the set-point tracking and disturbance rejection capabilities of the proposed NNMPC technique.
  • No Thumbnail Available
    Item
    OSTIA: A Low Cost Alternative for Short Summative Assessments in Massive Programming Courses
    (IEEE, 2020) Salas, Juan Carlos; Munoz Gama, Jorge
    The following topics are dealt with: computer aided instruction; educational courses; educational institutions; teaching; computer science education; Internet; further education; learning (artificial intelligence); engineering education; pattern classification.
  • Thumbnail Image
    Item
    Process-Oriented Feedback through Process Mining for Surgical Procedures in Medical Training: The Ultrasound-Guided Central Venous Catheter Placement Case
    (2019) Lira, Ricardo; Salas, Juan Carlos; Leiva Ruiz, Luis Fernando; Fuente Sanhueza, René Francisco de la; Fuentes Henríquez, Ricardo Sergio; Delfino, Alejandro; Hurtado Nazal, Claudia; Sepúlveda Fernández, Marcos Ernesto; Arias, M.; Herskovic, Valeria; Muñoz Gama, J.