Determinação das propriedades mecânicas de géis de caseinato de sódio via redes neurais artificiais
Data
2019-11-22
Autores
Carvalho, Beatriz Castro Ferreira 
Orientadores
Martins, Tiago Dias
Tipo
Trabalho de conclusão de curso
Título da Revista
ISSN da Revista
Título de Volume
Resumo
As proteínas e os polissacarídeos são dois tipos de biopolímeros responsáveis pela estrutura, textura e estabilidade dos alimentos. Existem poucos trabalhos que desenvolveram modelos matemáticos que correlacionem o comportamento mecânico-estrutural sobre a gelificação de proteínas com ou sem a presença de polissacarídeos, e os modelos encontrados para esse tipo de sistema complexo ainda são muito limitados. Pensando nisso, o presente trabalho teve como objetivo estudar a previsibilidade das propriedades mecânicas dos géis de caseinato de sódio com ou sem a adição de polissacarídeos por meio da aplicação de redes neurais artificiais. Assim, para os dados de entrada foram inseridos valores numéricos de concentração da proteína caseinato de sódio e polissacarídeos, tipo de polissacarídeo (inulina, carragena e goma jataí), temperatura de gelificação e razão glucona-δ-lactona (GDL)/caseinato. Para os dados de saída, foram inseridos valores de tensão na ruptura, deformação na ruptura e módulo de Young. Dessa forma, variou-se e analisou-se o número de camadas intermediárias, bem como os diferentes métodos matemáticos utilizados para treinar as redes neurais artificiais (RNA). Para o problema proposto se utilizou como critério os menores valores de Função Objetivo e os maiores de R2, encontrando a rede 5-16-12-3(5 neurônios na camada de entrada, 16 na primeira camada intermediária, 12 na segunda camada intermediária e 3 na camada de saída) como a mais adequada para o estudo. Porém a mesma não atingiu resultados satisfatórios para a deformação de ruptura, logo foi feito uma nova RNA apenas para essa variável encontrando uma configuração ótima de 5-4-18-1. Os resultados mostraram como redes neurais podem ser ferramentas versáteis para muitos estudos inclusive para o estudo proposto.
Proteins and polysaccharides are two types of biopolymers responsible for food structure, texture and stability. There are few studies that have developed mathematical models that correlate mechanical-structural behavior on protein gelation with or without the presence of polysaccharides, and the models found for this type of complex system are still very limited. The objective of this study was to study the predictability of the mechanical properties of sodium caseinate gels with or without the addition of polysaccharides by the application of artificial neural networks. Thus, for the input data, numerical values of the concentration of sodium caseinate protein and polysaccharides, polysaccharide type (inulin, carrageenan and locust bean gum), gelation temperature and glucone-δ-lactone ratio (GDL) / caseinate were entered. For the output data, tensile values at rupture, strain at rupture and Young's modulus were entered. Thus, by varying and analyzing the number of intermediate layers, as well as the different mathematical methods used to train ANNs (Artificial Neural Network). For the proposed problem we used as criteria the lowest values of Objective Function and the highest of R2, finding the network 5-16-12-3 (5 neurons in the input layer, 16 in the first intermediate layer, 12 in the second intermediate layer and 3 in the output layer), as the most appropriate for the study. However it did not reach satisfactory results for the rupture deformation, so a new ANN was made just for this variable, finding an optimal configuration of 5-4-18-1. The results showed how neural networks can be a versatile tool for many studies including the proposed one.
Proteins and polysaccharides are two types of biopolymers responsible for food structure, texture and stability. There are few studies that have developed mathematical models that correlate mechanical-structural behavior on protein gelation with or without the presence of polysaccharides, and the models found for this type of complex system are still very limited. The objective of this study was to study the predictability of the mechanical properties of sodium caseinate gels with or without the addition of polysaccharides by the application of artificial neural networks. Thus, for the input data, numerical values of the concentration of sodium caseinate protein and polysaccharides, polysaccharide type (inulin, carrageenan and locust bean gum), gelation temperature and glucone-δ-lactone ratio (GDL) / caseinate were entered. For the output data, tensile values at rupture, strain at rupture and Young's modulus were entered. Thus, by varying and analyzing the number of intermediate layers, as well as the different mathematical methods used to train ANNs (Artificial Neural Network). For the proposed problem we used as criteria the lowest values of Objective Function and the highest of R2, finding the network 5-16-12-3 (5 neurons in the input layer, 16 in the first intermediate layer, 12 in the second intermediate layer and 3 in the output layer), as the most appropriate for the study. However it did not reach satisfactory results for the rupture deformation, so a new ANN was made just for this variable, finding an optimal configuration of 5-4-18-1. The results showed how neural networks can be a versatile tool for many studies including the proposed one.