Artificial neural network for prediction of local scour depth around bridge piers using MATLAB

Scour is the consequence of the moving of bed material through the waterway, besides, scour undermines the stability of bridges, leading to a huge loss in the economic and human lives. In this study, several feed-forward artificial neural network (ANN) models with backpropagation algorithms have been established using deep learning toolbox in MATLAB software, besides, trial and error processes have been used to determine the number of nodes in the hidden layer. Moreover, 400 laboratories dataset from different resources have been used in this study, which included pier width, flow velocity, flow depth, sediment critical mean velocity, particle size distribution as input parameters, and equilibrium scour depth as a target parameter. The regression and mean squared error values have been employed for the evaluation of models. The results showed that the Levenberg-Marquardt training algorithm gives the best performance of prediction compared to other algorithms, moreover, sensitivity analysis showed that the elimination of pier width from input parameters decreases the efficiency of prediction more than other parameters. Also, ANN models with dimensional parameters provide better results than non-dimensional parameters. It has been found that ANN could be used as a reliable technique for prediction of equilibrium scour depth.