Author(s): Anubhav Baranwal; Bhabani Shankar Das
Linked Author(s):
Keywords: Local scour; Clear water scouring; Gamma test; MARS; M5Tree
Abstract: Local scouring occurs when water flow erodes river bed material around a bridge pier. It poses a significant threat to the safety and serviceability of bridges. The prime reason for the local scouring is the development of the horseshoe vortices upstream and wake vortices downstream of the bridge pier due to the flow obstruction. This study introduces a machine learning (ML) approach based on Multivariate Adaptive Regression Splines (MARS) and M5Tree to predict the local scour depth around bridge piers. Previous published data points from various literature, such as geometry parameters, hydraulic parameters, and bed roughness characteristics, were collected for clear water scouring (CWS). Five non-dimensional influencing parameters, i. e., ratio of pier width to flow depth (b/y), ratio of approach mean velocity to sediment incipient velocity (V/Vc), critical Froude number (Frc), ratio of pier width to mean sediment size (b/d50), and standard deviation of bed material (σg) were selected as input variables for the CWS depth model. A gamma test was performed to identify the optimal combinations of input parameters. It is found that the M5Tree model outperforms the MARS model and three other previous empirical models developed by other researchers, as evidenced by statistical indices, showing an improved scour depth ratio (SDR). The reliability of the models is indicated by coefficient of determination (R2) values above 0.90 for the CWS condition in both models. Therefore, due to its superior efficiency and reliability, the M5Tree model is recommended for estimating CWS depth around bridge piers. This study emphasizes the potential of the MARS and M5Tree models to enhance the accuracy of scour depth ratio predictions and optimize bridge design compared to traditional scour depth prediction methods.
Year: 2024