Author(s): Taymaz Esmaeili; Tetsuya Sumi; Nozomu Yoneyama; Abdolsattar Vakili
Keywords: Scour depth; Real scaled bridge piers; Neural network; Neuro-fuzzy
Abstract: Local scouring around hydraulic structures such as bridge piers is the consequence of the erosive process of stream flows on alluvial beds. Since, the local scouring around real scaled bridge piers endangers the safety and stability of bridges, it is essential to evaluate the maximum local scour depth. Wide variety of physical parameters (i. e. flow velocity, sediment characteristic and pier geometry) and non-physical parameters (i. e. flow regime, sediment transportation pattern and turbulent boundary layer characteristics) as an influential set of parameters involve in formation of local scouring around bridge piers. Therefore, it is necessary to assess the effects of influential set of parameters on local scour depth around real scaled bridge piers as far as possible. In this paper, artificial neural networks (ANNs) using a feed forward back propagation (FFBP) algorithm besides the adaptive neuro-fuzzy inference system (ANFIS) were used to estimate the local scour depth around real scaled bridge piers in non-cohesive bed materials. This study employed a field observed dataset consisting of 269 pier scour measurements. The field observed dataset in both original and non-dimensional forms introduced into the ANNs-FFBP and ANFIS models. The comparison of the results obtained by intelligence based models with the observed field data showed that ANNs-FFBP leads to a better prediction than ANFIS model for both original and non-dimensional dataset and also employing original dataset provides more accurate outcomes. Finally, sensitivity analysis shows that the mean flow velocity, flow depth, pier shape and debris effect are respectively the important influential parameters on the calculated local scour depth.