Abstract
This study attempted to develop a method to predict the degree of soil compactness in terms of bulk density, which varies as soil is subjected to vehicular loads. In situ experiments were carried out by measuring the stresses in soil underneath operating tractors using stress state transducers. The soil at the test site was sampled and investigated in terms of its pertinent properties. A neural network model was then developed to memorize the functional relationships that govern the changes in bulk density as a function of soil properties and stress parameters. The results demonstrated that the model could predict bulk density by a root mean squared error of 4.28% with a correlation coefficient of 0.838. Furthermore, the sensitivity analysis was performed to prioritize the significance of each input parameter presented to the network by adopting a method called the Relative Strength of Effect (RSE). The results indicated that both normal and shear stresses were the dominant factors influencing soil compaction. Besides, susceptibility to compaction was found to be dependent mainly on the initial bulk density and the moisture content of soil.
