Studying the seepage phenomena under a concrete dam using SEEP/W and Artificial Neural Network models.

Abstract

Seepage under hydraulic structures is considered to be a dangerous phenomenon which may cause the collapse of the structure over time if neglected. In this research, a SEEP/W model was developed to find the seepage rate and exit gradient under a concrete dam provided with two sheet piles. The independent variables were head difference; coefficient of soil permeability; and the spacing, lengths, and inclined angles of the sheet piles. The model was run for three different values of each independent variable. The results obtained from SEEP/W model were then used to create two neural artificial network (ANN) models (A and B) in which the output variables were the seepage rate (model A) and exit gradient (model B). The most appropriate structure, which gave minimum relative errors, was (7 3 1) nodes for both models. The results of the ANN models indicated that the variable with the most effect on seepage rate was the coefficient of soil permeability, with an importance ratio of about 76%, followed by the difference in the head (8%), the distance between piles (5.5%), length of downstream pile (5%), length of upstream pile (4%), and downstream and upstream inclined angles of the sheet piles, with ratios of about 1% and 0.5%. In terms of exit gradient, the most influential factor was the distance between piles at 35%, followed by the downstream inclination angle, length of downstream pile, head difference, length of upstream pile, inclined angle of upstream pile, and soil permeability with importance of about 23%, 19%, 14%, 7.5%, 1% and 0.5%, respectively. These results are in agreement with an analysis of the SEEP/W model.