Redes neurais artificiais e modelos de regressão na predição de propriedades reológicas de méis brasileiro

• Autor(a) principal: Silva, Vanelle Maria da
• Data de Publicação: 2016
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Repositório Institucional da UFLA
• Texto Completo: http://repositorio.ufla.br/jspui/handle/1/11567
• topic: Apicultura - Modelos matemáticos; Mel - Propriedades físico-químicas; Mel - Propriedades reológicas; Mel - Viscosidade; Redes neurais (Neurobiologia); Regressão não linear; Bee culture - Mathematical models; Honey - Physical and chemical properties; Honey - Rheological properties; Honey - Viscosity; Neural networks (Neurobiology); Non-linear regression; Química, Física, Físico-Química e Bioquímica dos Alim. e das Mat.-Primas Alimentares

The rheological properties of honey are of practical importance for beekeepers and industries, since their determination allows for processing and quality control of honey. The relationships between physico-chemical and rheological properties are considered complex nonlinear systems. Therefore, artificial neural networks (ANNs) and linear and non-linear regression models were used to predict the rheological properties of 40 Brazilian honeys from different floral sources based on easily obtainable measurements. A rheological characterization of honeys was performed by means of shear tests at steady state for determining the viscosity (η) at different temperatures (10°C to 60°C) and small amplitude oscillatory shear (SAOS) testing for determining the parameters storage modulus (G’), loss modulus (G’’) and complex viscosity (η*) in temperature scans (0°C -75°C-0°C) and frequency scans (0.1 Hz to 10 Hz) at different temperatures (10°C to 60°C). All honeys showed liquid behavior at the evaluated temperatures and mechanical spectra. The Arrhenius model was the most appropriate for estimation of η for all honeys and η* for some of them, where the Williams-Landel-Ferry (WLF) model was the most appropriate for predicting η* of the orange blosson, multi-southest and multi-southern honeys. Simplified models were proposed to determine η and η* from the combined effect of temperature and concentration, which showed coefficient of determination (R²) equal to 0.9540 and 0.9334, and root mean square error (RMSE) equal to 8.00 and 10.44 for η and η*, respectively. In estimating the viscosity (η) from shear measurements at steady state, an ANN (model 1) with architecture of 2-12-1 neurons in its layers showed good performance in the test phase, with RMSE and correlation coefficient (r) values equal to 0.0430 and 0.9681, respectively. In prediction of the parameters G’, G’’ and η* from the temperature scans during heating and cooling, ANNs with architectures of 2-9-3 (model 2) and 2-3-3 (model 3) presented RMSE values equal to 0.0261 and 0.0387 in the test phase, respectively. For all the determined parameters, nonlinear exponential models showed similar results to models 1, 2 and 3. An ANN with 3-9-3 architecture (model 4) presented RMSE and r values for G’ equal to 0.0158 and 0.7301, for G’’ equal to 0.0176 and 0.9581 and for η* equal to 0.0407 and 0.9647, respectively, in the test phase for the frequency scan data. These results were far superior to those obtained using second order multiple linear models. The acquisition of all models is an important application for the processing of honey and honey-based products, since these properties are essential in engineering, quality control and product shelf life calculations.