FRSF - INVESTIGACIÓN Y PRODUCCIÓN CIENTÍFICA

Permanent URI for this communityhttp://48.217.138.120/handle/20.500.12272/534

Encontrará los objetos que son resultados de las investigaciones

Browse

Author: search.filters.author.Vega, Jorge RubénSubject: search.filters.subject.Soft sensor

Search Results

Now showing 1 - 3 of 3
  • Thumbnail Image
    Item
    Estimation of quality variables in a continuous train of reactors using recurrent neural networks-based soft sensors
    (Chemometrics and Intelligent Laboratory Systems, 2024) Perdomo, Mariano Miguel; Clementi, Luis Alberto; Vega, Jorge Rubén
    The first stage in the industrial production of Styrene-Butadiene Rubber (SBR) typically consists in obtaining a latex from a train of continuous stirred tank reactors. Accurate real-time estimation of some key process variables is of paramount importance to ensure the production of high-quality rubber. Monitoring the mass conversion of monomers in the last reactor of the train is particularly important. To this effect, various soft sensors (SS) have been proposed, however they have not addressed the underlying complex dynamic relationships existing among the process variables. In this work, a SS based on recurrent neural networks (RNN) is developed to estimate the mass conversion in the last reactor of the train. The main challenge is to obtain an adequate estimate of the conversion both in its usual steady-state operation and during its frequent transient operating phases. Three architectures of RNN: Elman, GRU (Gated Recurrent Unit), and LSTM (Long Short-Term Memory) are compared to critically evaluate their performances. Moreover, a comprehensive analysis is conducted to assess the ability of these models to represent different operational modes of the train. The results reveal that the GRU network exhibits the best performance for estimating the mass conversion of monomers. Then, the performance of the proposed model is compared with a previously-developed SS, which was based on a linear estimation model with a Bayesian bias adaptation mechanism and the use of Control Charts for decision-making. The model proposed here proved to be more efficient for estimating the mass conversion of monomers, particularly during transient operating phases. Finally, to evaluate the methodology utilized for designing the SS, the same RNN architectures were trained to online estimate another quality variable: the mass fraction of Styrene bound to the copolymer. The obtained results were also acceptable
  • Thumbnail Image
    Item
    Classifier algorithms for tuning multi-model soft sensors : application to the estimation of quality variables in a continuous industrial process
    (WCCE11, 2023) Perdomo, Mariano Miguel; Clementi, Luis Alberto; Sanseverinatti, Carlos Ignacio; Vega, Jorge Rubén
    In this work, a multi-model soft sensor (SS) is proposed to estimate non-measurable variables in continuous processes. The proposed approach involves a first stage of clustering, using Gaussian mixture models, to identify the clusters that represent the multiple working conditions of the process. Then, for each identified cluster, multivariate linear regression sub-models are calibrated. Finally, the required non-measurable variable is estimated through a linear combination of the estimations from each sub-model. The weight coefficients for each sub-model are calculated using a classification algorithm. The performance of four different classification algorithms is evaluated in terms of the capability of their resulting multi-model soft sensor to estimate the mass conversion in a numerical simulation of a continuous emulsion polymerization for industrial production of Styrene-Butadiene Rubber. The results showed that the classifier model plays an important role in the multi-model soft sensor performance. Furthermore, a multi-model soft sensor that assigns the weights through Gaussian mixture models performs better than cases where a multi-layer perceptron, a linear discriminant analysis, or a K-nearest neighbors are used.
  • Thumbnail Image
    Item
    Estimación de variable de calidad en un tren de reactores continuo utilizando soft-sensors basados en redes neuronales recurrentes
    (XX RPIC, 2023) Perdomo, Mariano Miguel; Clementi, Luis Alberto; Vega, Jorge Rubén
    La producción industrial de caucho SBR (Styrene-Butadiene Rubber) utiliza un tren de reactores tanque agitados continuos para la obtención de un látex. En esa etapa, es importante contar con estimaciones en tiempo real de algunas variables en el último reactor del tren, a efectos de lograr productos de calidad con buen nivel de producción. Los soft sensors (SS) son una alternativa adecuada para cumplir con esta tarea, en comparación con las estimaciones típicamente brindadas por laboratorios analíticos y analizadores en línea. El proceso incluye dinámicas complejas y múltiples modos de operación, resultando necesario el uso de SS capaces de capturar esas características. En este trabajo, se propone el desarrollo de un SS basado en redes neuronales recurrentes para estimar la conversión másica del látex a la salida del tren de reactores. La selección de la ventana temporal necesaria para efectuar las estimaciones se basa en un análisis de correlaciones cruzadas y una búsqueda por grilla. Por otra parte, el modelo que conformará al SS se selecciona a través de una exploración de distintas arquitecturas y sus espacios de hiperparámetros. Los resultados de simulación muestran que la adopción de una red neuronal recurrente basada en GRU (Gated Recurrent Unit) y una ventana temporal de 800 minutos exhibe el mejor desempeño para estimar la variable de interés.