Abstract

Although the theory of water movement through an unsaturated, swelling soil has been developing for a number of years, the efficiency of the results has been limited, and the developed models have suffered from a lack of a mechanism to reliably define water flow and experimental techniques for measuring the unsaturated soil hydraulic properties. The intent of the present research is to elucidate and to investigate the theory of flow through unsaturated, swelling soil, to develop a three dimensional theoretical model based on a reliable soil swelling and unsaturated flow mechanism and, furthermore, to characterize the relationships between different soil hydraulic and swelling properties. In the first part, based on the water adsorption mechanism by the soil particles and the use of continuum theory principles, a theoretical model of unsaturated flow through swelling soil has been developed. Some new techniques were established to measure the different soil swelling and hydraulic properties and to analyze the effect of the soil's initial conditions. The reliability of the different existing empirical models, to be applied to the results obtained from hydraulics, shrinkage and swelling experiments, were investigated and modified where necessary. A static neural network model was developed to mathematically characterize the experimental results being used in the proposed numerical solution of the mathematical model. The neural network model works based on the back-propagation error method with adaptive learning rules by means of the neural network tool box of MATLAB software. A finite difference numerical model, based on a fully implicit method, was proposed to numerically solve the one dimensional governing equation (simplified form of 3D governing equation). For this part of the study, a computer program in C++ language was developed. Finally, the results of the numerical method were verified using the results obtained from infiltration tests. The tests were conducted for the semi-infinite soil columns having three different confinement conditions: completely confined, semi-confined and free swelling. The results show a clear agreement between the output of the developed numerical model and the experimental data. As one of the conclusions, the results of the numerical analysis of the developed mathematical model agreed reasonably well with those of the experiments. The study also indicated that the static neural network model, as one of the most promising models, can precisely characterize any experimental function related to the soil properties and cooperates with other numerical approaches such as finite difference method to solve a highly nonlinear partial differential equation such as unsaturated flow equation