ANALYSIS AND SIMULATION OF COLLOID DISPERSION WITH NEURAL NETWORKS

Colloid dispersion is an important earth surface process that causes the disaggregation of colloids, such as clays, by altering the fluid and surface chemistry to create a stable suspension. Previous workers have shown that dispersed colloids cause decreases in hydraulic conductivity, piping erosion, slope failure, degradation of soil productivity, increased erosion, and increased potential for transport of adsorbed contaminants on colloids. It is important to be able to predict what natural or anthropogenic conditions cause sediment to disperse. No single mathematical model describes colloid dispersion in the natural environment because the controlling factors are complex. Several forms of linear statistical analysis based on common geochemical parameters have been used to describe dispersion behavior. These attempts are fundamentally flawed because the problem is non-linear and the data is often limited. Multilayer neural networks are universal approximators that are capable of dealing with non-linear problems and incomplete or noisy data. In addition, with this artificial intelligence technique knowledge the physics or mathematics of the problem at hand is not necessary. Thus, it represents an ideal tool for analyzing and simulating colloidal dispersion. The analysis of 19 soil samples for the affects of sodium adsorption ratio (SAR), electrical conductivity (EC), pH and mineralogy on dispersion using a neural network resulted in a significantly improved mathematical model when compared to traditional statistical methods. The network quantitatively described the interactions and dynamics of dispersion in accordance with chemical theory. Further investigation of colloid dispersion processes combined with neural network analysis procedures could lead to a global solution for the simulation of dispersion for any geochemical, mineralogical and environmental condition.