Neural Network Based Modelling of Soil Particle Diameters Under Varying Quantity of Sodium Hexametaphosphate

Abstract

In this study, the particle diameter of the soil was simulated and modelled by using artificial neural network method. In order to determine the particle diameter of the soil based on passing time, hydrometer reading, temperature of the solution and the quantity of the sodium hexametaphosphate, the quantity of the sodium hexametaphosphate 0, 10, 20, 30, 40, 50 and 60 were respectively selected. As pointed out in the Turkish Standard 1900, the soil particle diameters in the solution prepared with 40 g sodium hexametaphosphate was taken as reference. It was found that the average soil grain diameter for 0 g sodium hexametaphosphate was about 4.5 times bigger than the reference grain diameter, for 10 g was 3.9 times, for 20 g 3.46 times, for 30 g 2.12 times bigger. However, the hydrometer reading could be done only up to the 260th min for 50 g sodium hexametaphosphate and for 60 g sodium hexametaphosphate the hydrometer couldn't be read. The relationships between experimental results and artificial neural network (ANN) model exhibited good correlation. The correlation coefficients square were found as R² = 0.99 for training set and R² = 0.94 for testing set with ANN. Based on the results of the study, it could be said that the ANN method could be used for modelling of the particle diameter of the soil according to the passing time, hydrometer reading, temperature of the solution and the quantity of the sodium hexametaphosphate.