Non-Linear Models With Autoregressive Error Structure for Studying Bean Seed Drying Kinetics

Abstract

Objective: To use the non-linear regression models (Lewis, Overhults, Page, Midilli, and Three-parameter simple Exponential) to describe the drying kinetics of bean seeds as a function of time (hours).   Theoretical Structure: The research project shows the steps taken to conduct and analyze data.   Method: The model parameters were estimated using the least squares method and the Gauss-Newton convergence algorithm. The assumptions of normality, homoscedasticity, and independence of residuals were tested using the Shapiro-Wilk, Breuch-Pagan, and Durbin-Watson tests, respectively. If the assumption of independence of residuals was violated, this dependence was modeled with an autoregressive error structure AR(1). The adjusted coefficient of determination (Raj2), Akaike information criterion (AIC), residual standard deviation (RSD), and Bates and Watts curvature measure were used to assess the goodness of fit of the models.   Results and conclusion: The results showed that the Midilli model presented a good quality fit to the data, and is the most suitable for describing the drying kinetics of bean seeds, with the drying rate averaging 0.4681 g of water/hour.   Research Implications: The research contributes to the literature with practical information about the drying process.   Originality/value: Highlights the importance of adjusting non-linear regression models to the drying kinetics of biological products. These models are used to represent the decrease in the amount of water in a given food over time.