Enhancing Grain Moisture Prediction with Artificial Neural Networks and Computational Fluid Dynamic

Abstract

Drying cereals is a vital process to reduce moisture content, enabling efficient storage. This study investigates the convective drying behaviour of cereals through numerical simulations and feedforward neural networks. Key parameters considered include temperature, ambient air speed, relative humidity, porosity, intrinsic permeability, density, thermal capacity, thermal conductivity, and initial relative humidity of stored grains. Training data were generated using numerical methods, solving heat and mass transfer equations based on Whitaker's model within COMSOL Multiphysics® 5.6. Simulation results reveal that moisture in cereals gradually equilibrates with the ambient environment, commencing from the exposed surfaces. The artificial neural network (ANN) demonstrates remarkable predictive accuracy using 100 data points, yielding an overall correlation coefficient of 0.99458 and a mean squared error (MSE) of 3.132 ×10-4. The combination of these two methods offers distinct advantages, with ANN saving computational time and numerical simulations not requiring initial samples. This combined approach proves promising for grain moisture prediction, though results must undergo rigorous validation to ensure reliability and accuracy.