MECHANISM AND ANN MODELING OF SULFATE-INDUCED HEAVE IN NON-SWELLING SOIL IN AN ACIDIC ENVIRONMENT

Abstract

Red earth (aluminosilicate) is mainly comprised of kaolinite mineral, which is of non-swelling in nature. The iron oxides present in the soil impart red colour to the earth, hence it is named red earth. In the case of natural 1:1 minerals, no significant volume changes are observed due to strong hydrogen bonding between their interlayers. Sulfate in an acidic environment can cause abnormal volume changes in natural red earth, owing to mineralogical alterations. Sulfuric acid solutions of varying concentrations are used to induce sulfate content in an acidic environment, which is a very common cause of acid sulfate soil contamination. When sulfuric acid starts interaction with red earth, the hydrated hydrogen ion starts attacking (due to smaller in size and/or ionic potential) on the red earth leading to a reduction in H-bonding between successive basic units of kaolinite and releasing the iron from iron oxides, results in the formation of new mineral rozenite (iron sulfate hydrate) in combination with sulfates. During this process, the heave reported in the red earth at nominal surcharge continued for a long period indicating that the formation of minerals was a rather slow process. The observed heave after considerable time lag, unlike the swell that occurs due to the adsorption of water by clay particles. The formation of new minerals is responsible for the observed non-hyperbolic nature of time swell relationships and the mineralogical and morphological changes are confirmed by scanning electron microscopy and energy dispersive analysis of X-ray studies. Besides, the sulfate-induced time-dependent swell percent in red earth in an acidic environment is studied by employing the Artificial Neural Network model with Levenberg–Marquardt (LM) algorithm. The network was programmed using MATLAB® code to process the information and predict the percent swell at any time, knowing the variable involved. Eight basic soil and fluid parameters were used as input feed to predict the percent swell to the proposed network. The study demonstrates that it is possible to develop a general ANN model that can predict time-dependent swell in sulfate-induced red earth with relatively high accuracy with observed data (R2 = 0.9986). The investigation results demonstrate reference to the engineering behaviour of acid sulfate soils, distributed worldwide under acidic environments.