Application of Machine Learning Algorithms for Estimation of Filter Dimensions for an Earthen Embankment Dam

Seepage through an embankment must be controlled to prevent concealed internal erosion and migration of fine materials. It is extremely important to control the seepage flow and inhibit removal of the soil particles comprising the dam body. Modern design practice incorporates this control into the dam design through the use of internal filters and adequate drainage provisions. The seepage analysis theories proposed by researchers like Casagrande, Schaffernak, Dupuit, and Pavlovsky for homogenous earthen dam resting on impervious base finds their application in estimation of filter dimension. This paper reports the utilization of machine learning in this regard. A large dataset is generated by using Schaffernak’s theory for phreatic surface assessment and by varying the governing parameters in all possible range that affects the filter dimension. The dataset is used for training in suitable algorithms related to Multilayer Perceptron (MLP), Random Forest (RF), Support Vector Regression (SVR), Ridge Regression (RR), and Xtreme Gradient Boosting (XGBoost) algorithms. The results illustrated that XGBoost algorithm could potentially be used to estimate filter dimension and exit discharge. These trained models can be used as a ready reference solution by the practising engineers, which provides them a preliminary idea for designing toe filters.