Predictive Model for Flood – Induced Collapse Phenomenon in Residual Soils of Northern Edo, Nigeria

Abstract

Residual soils are in the category of questionable soils which have been experienced in the arid and semi-arid climatic zones of the world. The conditions in these zones favour the development of most unsafe collapsible soils. At their dry natural state, they possess awesome stiffness and high apparent shear strength, however upon flooding, may demonstrate a remarkable reduction in volume, consequently deteriorate in strength and collapse. In this research, the collapse phenomenon of residual soil collected from three locations in Auchi, Northern Edo, Nigeria has been investigated on undisturbed specimens by utilizing single Oedometer test. The results obtained from Oedometer tests were utilized to form the database to develop the Artificial Neural Network model for the prediction of collapse potential induced by flood. The influences of flood, flooding pressure, void ratio, dry density and porosity on soil collapse have been investigated. Six input parameters (i.e. Flooding Pressure, Initial void ratio, Initial water content, Initial dry density, Liquid limit and Initial porosity) are considered to have the most noteworthy influences on the degree of collapse and have been utilized as the model’s inputs while the model output will be the equivalent collapse potential. The proposed network was developed using Microsoft Visual Studio 2010 and the MS.NET Framework 4.0 and source codes were written in C-Sharp (C#). A supervised learning was utilized to train the Back Propagation feed forward multi-layer ANN algorithm with the momentum coefficient and learning rate as its parameters. The prediction performance of the Artificial Neural Network model was assessed by utilizing the primary statistical criterion proposed by Shahin, et al., [1] such as the coefficient of correlation, R2, and the root mean square error, RMSE. The model outcomes demonstrated that it has the aptitude to predict the collapse potential from single Oedometer test in residual soil samples with a good degree of precision with coefficient of correlation, R2 = 0.856 and root mean square error, RMSE = 166.199.