Assessment of Data-Mining and Some Empirical Methods in Scour Depth Estimation at Briclge Piers

Abstract

Local scour at . bridge pieds is one of the numeroussafty hazaraels that the eaten their stability. An abundance of such facfors and their complexities, along with a mulfiplicity ofempirical relationships, make the development of an integrated approach for estimation of scour depth very difficult. However, the presence of novel data-mining approaches such as the artificial nevral networks (ANN) and the M5 Tree Model has facilitated the solution of complicated engineering problems. In this study by using laboratory data and identifying 10 scenarios including different combinations of effective parameters in scour depth, the performance of ANN and M5 tree models have been investigated and results compared with 3 empirical relationships (Melville, Mississippi and HEC-18). The results indicated that the M5 Tree Model via presenting 2 simple if-then rules and may CC=0.95 in comparison with the other ANN and empirical approaches may estimate scour depth with high accuracy. The results ahso indicated that between the 3 used empirical relations, HEC-18, Mississippi and Melville relations presents high accuracy, respecfively.

Keywords

• local scour
• data mining
• m5 model tree
• artificial neural networks
• empirical equations