Developing artificial neural network models to predict the bearing capacity of cohesionless soil for shallow foundations.

Abstract

The bearing capacity of soil is considerably very important for geotechnics and engineering geology, since it is a significant parameter for the foundations design. This research proposes the use of artificial neural network to predict the bearing capacity of cohesionless soil for shallow foundation. In this paper 145 datasets were used to train and validate the model. Four parameters (friction angel of the soil (), unit weight (), footing width (B) and footing length (L)) were used as the model inputs. Relating to these input parameters, in the ANN model is forecasted the ultimate bearing capacity. The last parameter been set as the required target (i.e. output) in the ANN model. Performance comparison of the developed models (interms of coefficient of correlation (R) and Mean Squared Error (MSE) ) revealed that the developed artificial neural network models could be effectively used at the preliminary stage of estimating the bearing capacity of cohesionless soil for shallow foundations, instead of the coventional methods. Coefficient of correlation (R) equals to 0.989 and Mean Squared Error (MSE) equals to 0.000297, strongly implies that the ANN model shows a high level of reliability in forecasting the bearing capacity of cohesionless soil for shallow foundations.