Abstract:

A coupled artificial neural network model with a genetic algorithm optimization model is developed for a practical case of a single cutoff. The proposed cutoff is of a soil-embedded vertical part with an inclined extension. The model successfully found the optimum dimensions of the vertical and inclined parts, the optimum angle of inclination, and the optimum length of protection downstream of the cutoff for a factor of safety of 3 against piping. Two thousand one hundred cases are modeled first using Geo-studio software to find the required length of downstream protection against piping for different lengths of the vertical, inclined lengths of the cutoff, its angle of inclination, soil layer depth, and degree of anisotropy. Then the created data set was used to develop an Artificial Neural Network (ANN) model for finding the length of protection required. The ANN model showed high performance with a determination coefficient of (0.922). The genetic algorithm model needs a minimum number of randomly generated populations of 100000 and three crossover iterations to produce a stable optimum solution. Running the model for different practical cases showed that the optimum angle variation was low and fluctuated around 30o. This low angle variation was due to its lower effect on the downstream soil protection length compared to the other decision variables. At the same time, the other dimensions varied with input variables, such as the depth of the soil layer, the seepage driving head, and the degree of isotropy. For a degree of anisotropy (ratio of vertical to horizontal hydraulic gradient) less than 0.5, the results showed no need for protection against piping; hence it is recommended to use minimum dimensions for such a case. The coupled model can easily obtain the optimum dimensions for any given input. Importance analysis showed that the optimum length of the downstream protection was highly affected by the vertical and inclined length of the cutoff, while it was less affected by the angle of inclination. Correlation analysis supported the importance analysis.  

Keywords: