Neural network estimation of the scour cone geometry around outlet in the pressure flushing

When flushing carry out as pressure condition, a scour cone is performed around the outlet. As the
flow around the outlet in the pressure flushing is three dimensional, therefore that it is difficult to
establish a general empirical model to provide accurate estimation for scour cone volume and
length. In this study artificial neural network (ANN) with multi-layer perception which using backpropagation
algorithm (MLP/BP) was used. The scour cone volume (Vf) and length (Lf) were
modeled as a function of three variables; water depth (HW), mean flow velocity through outlet (uf)
and mean grain diameter (D50). For training and testing model, experimental data in two forms of
original and non-dimensional are selected. The results of this research indicate that MLP/BP model
can predict the scour cone volume and length. Finally, sensitivity analysis with original and nondimensional
data set show that mean flow velocity through outlet (uf) and uf / √(g (Gs-1)D50) have a
greater influence on scour cone volume and length rather than other parameters.