Author(s): L. Brandimarte; E. Toth
Linked Author(s):
Keywords: No Keywords
Abstract: In the most recent years, the use of Artificial Neural Networks (ANN) models has been proposed in the literature for the prediction of scour depth around bridge piers, taking advantage of their capability to flexibly reproduce the highly non-linear nature of the relationship between input and output variables, also when such relationship is not explicitly known a priori. Nonetheless, so far, no attempt has been made to implement an ANN model that takes specifically into account the fact that profoundly different phenomena govern the formation of pier holes under clear-water and live-bed conditions. This study aims at further investigating the potentiality of the ANN approach: a wide set of both field and laboratory data is used to test different architectures of ANN for predicting the local scour depth as a function of the variables characterizing the flow, the sediments and the pier. In particular, special focus is given to the analysis of the scour depth under clear water and live bed conditions. The work analyses the impact of the use of different training data sets on the performances of the model over an external, validation set, whose data are not used in any way in the calibration of the models. As a standard of reference, the scour depths estimates predicted by the ANN models are compared to the estimates obtained by empirical formulae conventionally used in the literature and in the current engineering practice.
Year: 2010