Artificial neural network model for the kinetics of canola oil extraction for different seed samples and pretreatments

Abstract

In this work, a multi-layer feedforward artificial neural network (ANN) was used for modeling and predicting the oil extraction yields of three canola samples with three pretreatments (unpretreatment, hydrothermal, and microwave pretreatment), considering extraction time and temperature as variables. Based on the results of the training, validation, and testing of the network, a neural network with eleven neurons in one hidden layer was selected as the best architecture for predicting the oil extraction yield response. Results obtained by the ANN model were compared with models from the literature (modified Fick's diffusion models), generally obtaining a more accurate fit with the ANN model. Practical applications: Existing models of canola oil extraction kinetics have some limitations since they are not able to describe various conditions, such as variability among samples and pretreatments. Artificial neural networks (ANN) are powerful and high-precision computational statistical modeling techniques that can address different problems. The aim of this work was to model the kinetics of canola oil extraction under different conditions (varying temperature, samples of canola, pretreatments applied) with an ANN, which presents several advantages over other reported models, allowing to describe a process that depends on many variables even when the data are incomplete or contain errors, thus facilitating its industrial application.