Investigation of the drying kinetics of brewer's spent grain (BSG) using artificial neural networks and traditional empirical models

Brewer's spent grain (BSG) is the main by-product of the brewing industry. BSG can have diverse end-users and has a high moisture level. To guarantee good conditions for storage and trade, it is necessary to remove the moisture from the material and use the proper method for the drying process. The phenomenon of water content removal is represented by mathematical models. Empirical and phenomenological models, as well as artificial neural networks (ANN) can be employed for this purpose. Here, we compare the fitted curves between empirical models (such as Page, Midilli-Kucuk, and Newton models) and an artificial neural network. The fits were investigated quantitatively by the analysis of the mean squared error (MSE) and determination coefficient (R-2). The residues generated were analyzed qualitatively by the plots of histograms and qqplots (quantil vs. quantil). From the results that were obtained, it is possible to conclude that the ANN model had the best performance when compared to the empirical and semi-empirical models studied, with the lowest values of MSE and the highest values of the R-2 (0.999). The net presented adequate estimations for intermediate data as well, proving its use as a proper empirical model for the prediction of moisture data at several temperatures.