An Evaluation of Neural Networks Capability for Scour Depth Prediction around the Abutments

Scour a very complicated process, is one of the main factors in abutment destruction. Complication of stream pattern around abutments, and the variation of effective factors on scour,have lead to the develop of innumerous empirical equations, which have many restrictions due to the limited experimental conditions. Applicability of multilayer pereceptron(MLD)networks to prediction of the maximum scour depth was evaluated for abutments having vertical, winged and semi-circural walls. The ANN results were compared with the calculated values by an empirical equation suggested by Barbhuya and Dey (2004). Eight scenarios were defined using effective parameters and several networks with different important parameters to predict scour depth. Comparison of the results of different scenarios showed that the one, which used only two parameters and to predict scour depth around abutments was the most efficient indoing so. Results of sensitivity analysis indicated that the two parameters, and, had the most effect on the scour depth prediction around abutments. A comparison of the obtained results using the ANN model, and the empirical equation with the experimental data revealed that the ANN model presented more precise results than the empirical equation for scour depth prediction around abutments. Moreover, the ANN model is more applicable to the scour depth prediction around the obutments wilk vertical walls than the other 2 types of walls.