Cement take estimation using neural networks and statistical analysis in Bakhtiari and Karun 4 dam sites, in south west of Iran

Water seepage from dam foundations causes reservoir water loss and raises the risk of dam instability. One method of remediation for controlling instability and leakage of these rock foundations is grouting. Since a considerable portion of the costs for dam construction is allocated to grouting, as a result, study of the influencing factors of cement take in grouting jobs is of paramount importance for each site. The most dominant parameters which play a decisive role in the efficiency of grouting are rock mass strength and permeability, grouting pressure, water-to-cement ratio, and properties of jointed rock mass such as joint aperture, roughness, and spacing. In this paper, the relationship between cement take and Q-system, aperture and spacing of joints, Lugeon number, depth of grouting, and grouting parameters such as grouting pressure and water-to-cement ratio is investigated via statistical analysis and artificial neural networks for two large concrete dam sites, Bakhtiari and Karun 4, located in southwest Iran. Karun 4 has been constructed while Bakhtiari is still under construction with respective heights of 230 and 325 m. The mentioned parameters, the relationships of which are found in relation to cement take, are determined based on engineering geology reports for all the 5-m segments of the trial grouting boreholes. Bivariate statistical analyses showed that the highest correlation (R = 0.64) exists between cement take and Q-system by a logarithmic relationship. In addition, statistical investigations based on multivariate analyses between cement take and all the mentioned variables show a poor correlation (R = 0.48) which encouraged the authors to use neural networks for finding a relationship between cement take and the influencing variables. This resulted in a higher correlation (R = 0.77, RMSE = 9.2) with respect to the statistical method.