Modelagem do tratamento de efluentes fenólicos sintéticos via processos foto-Fenton aplicando a técnica das redes neurais artificiais
Data
2022-07-27
Autores
Teixeira, Flora Morena de Conceição 
Orientadores
Moraes, José Ermírio Ferreira de
Tipo
Dissertação de mestrado
Título da Revista
ISSN da Revista
Título de Volume
Resumo
Nos últimos anos, tem-se observado um interesse crescente no desenvolvimento de trabalhos relacionados à modernização dos processos de manufatura através da aplicação de novos conhecimentos nos campos da automação, controle e tecnologia da informação. Nesse contexto, uma das principais ferramentas tem sido as redes neurais artificiais (RNA), consistindo em uma importante técnica empírica de modelagem, que é inspirada no funcionamento dos neurônios do cérebro humano, com potencial significativo de uso para predição de variáveis críticas de processos industriais. Adicionalmente, observa-se também a preocupação cada vez maior de indústrias e órgãos governamentais no impacto dos processos produtivos no meio ambiente. Por consequência, processos de tratamento de efluentes mais eficazes e limpos tem sido cada vez mais estudados. Nesse contexto, o presente trabalho apresenta, como objetivo principal, o desenvolvimento de um modelo, via técnica das redes neurais artificiais, capaz de representar, de forma satisfatória, o processo de tratamento de um efluente sintético fenólico, aplicando-se o sistema foto-Fenton. Para esta modelagem, foi utilizada a linguagem R, com o uso do software RStudio, operando-se com apenas uma camada oculta. Neste trabalho foram avaliadas as seguintes variáveis de entrada: tempo de reação, área irradiada, potência das lâmpadas UV, concentração de peróxido de hidrogênio e a concentração de íons ferrosos (Fe2+). Já, como variável de saída, adotou-se o teor de carbono orgânico total (COT). Com o desenvolvimento do trabalho, foram obtidos 3 modelos, a partir de três conjuntos de dados experimentais fornecidos, sendo analisado, para cada modelo, o número ótimo de neurônios da camada oculta. Como resultado, verificou-se uma significativa capacidade de predição da variável de saída COT e capacidade generalização principalmente para o 2° modelo, obtendo-se valores de R2 iguais a 0,985 e 0,957, entre os valores preditos e experimentais, considerando os conjuntos de dados de treinamento e de teste da rede neural, respectivamente. Desta forma, foi possível verificar a eficácia da utilização da RNA para a obtenção de um modelo matemático eficiente para representar processos foto-oxidativos que, frequentemente, apresentam um comportamento complexo, envolvendo mecanismos de reações químicas via radicais livres, podendo assim, de forma rápida e contínua, inferir a qualidade do efluente na saída do seu processo de tratamento.
In recent years, there has been a growing interest in the development of works related to the modernization of manufacturing processes through the application of new knowledge in the fields of automation, control, and information technology. In this context, one of the main tools used has been artificial neural networks (ANN), consisting of an important empirical modeling technique, which are inspired by the functioning of neurons in the human brain. with significant potential of use for the prediction of critical variables of industrial processes, especially in processes where the phenomenological models are not completely known. In addition, there is also a rising concern of industries and government agencies regarding the impact of production processes on the environment. Consequently, more effective and cleaner effluent treatment processes have been increasingly studied. In this context, the present work presents, as main objective, the development of a model, through the technique of artificial neural networks, capable of representing, in a satisfactory way, the process of treatment of a phenolic synthetic effluent, applying the photo- Fenton. For this modeling, the R language was used, applying the RStudio software, operating with only one hidden layer. In this work, the following input variables were evaluated: reaction time, irradiated area, UV lamp power, hydrogen peroxide concentration and concentration of ferrous ions (Fe 2+). As an output variable, the total organic carbon content (TOC) was adopted. With the development of the work, 3 models were obtained from three sets of experimental data provided, being calculated, for each model, the optimal number of neurons in the hidden layer. As a result, there was a significant ability to predict the output variable TOC and generalization ability mainly for the 2nd model, with R2 values equal to 0.985 and 0.957, between the predicted and experimental values, considering the data sets training and testing of the neural network, respectively. In this way, it was possible to verify the effectiveness of the use of RNA to obtain an efficient mathematical model to represent photo-oxidative processes that, frequently, present a complex behavior, involving mechanisms of chemical reactions via free radicals, thus being able to, quickly and continuous, infer the quality of the effluent at the exit of its treatment process.
In recent years, there has been a growing interest in the development of works related to the modernization of manufacturing processes through the application of new knowledge in the fields of automation, control, and information technology. In this context, one of the main tools used has been artificial neural networks (ANN), consisting of an important empirical modeling technique, which are inspired by the functioning of neurons in the human brain. with significant potential of use for the prediction of critical variables of industrial processes, especially in processes where the phenomenological models are not completely known. In addition, there is also a rising concern of industries and government agencies regarding the impact of production processes on the environment. Consequently, more effective and cleaner effluent treatment processes have been increasingly studied. In this context, the present work presents, as main objective, the development of a model, through the technique of artificial neural networks, capable of representing, in a satisfactory way, the process of treatment of a phenolic synthetic effluent, applying the photo- Fenton. For this modeling, the R language was used, applying the RStudio software, operating with only one hidden layer. In this work, the following input variables were evaluated: reaction time, irradiated area, UV lamp power, hydrogen peroxide concentration and concentration of ferrous ions (Fe 2+). As an output variable, the total organic carbon content (TOC) was adopted. With the development of the work, 3 models were obtained from three sets of experimental data provided, being calculated, for each model, the optimal number of neurons in the hidden layer. As a result, there was a significant ability to predict the output variable TOC and generalization ability mainly for the 2nd model, with R2 values equal to 0.985 and 0.957, between the predicted and experimental values, considering the data sets training and testing of the neural network, respectively. In this way, it was possible to verify the effectiveness of the use of RNA to obtain an efficient mathematical model to represent photo-oxidative processes that, frequently, present a complex behavior, involving mechanisms of chemical reactions via free radicals, thus being able to, quickly and continuous, infer the quality of the effluent at the exit of its treatment process.