Scour can have the effect of subsidence of the piers in bridges, which can ultimately lead to the total collapse of these systems. Effective bridge design needs appropriate information on the equilibrium depth of local scour. The flow field around bridge piers is complex so that deriving a theoretical model for predicting the exact equilibrium depth of local scour seems to be near impossible. On the other hand, the assessment of empirical models highly depends on local conditions, which is usually too conservative.
In the present study, artificial neural networks are used to estimate the equilibrium depth of the local scour around bridge piers. Assuming such equilibrium depth is a function of five variables, and using experimental data, a neural network model is trained to predict this equilibrium depth. Multilayer neural networks with backpropagation algorithm with different learning rules are investigated and implemented. Different methods of data normalization besides the effect of initial weightings and overtraining phenomenon are addressed. The results show well adoption of the neural network predictions against experimental data in comparison with the estimation of empirical models.