Estimation of time dependent scour depth around circular bridge piers: Application of ensemble machine learning methods

Abstract

Scour is a major issue which impacts the life of a hydraulic structure. In this work, we have considered a bridge as an example of a hydraulic structure. Scour depth increases or decreases from time to time because of variations in flow scenarios. Scour phenomena have been extensively studied and many empirical as well as data-driven approaches have been proposed to estimate the scour depth. However, there are relatively few studies which can predict the time-dependent scour depth. In this study, we use a laboratory dataset compiled from several sources to predict time-dependent scour depth using ensemble and standalone machine learning methods. For scour depth estimation, various factors such as river bed properties (d50), sigmag, rhos, Ucr), flow properties around a pier (rhof, U, mu, g, y), bridge pier geometry (Al, Sh, Dp), and time (t, tR) are used. In this study, we apply two ensemble methods: Bagging Regressor (BR), AdaBoost Regressor (ABR), and one standalone machine learning method, Support Vector Regression (SVR), for the prediction of time-dependent scour depth. Out of these machine learning methods, both the ensemble methods provide superior predictions in comparison to the standalone variant and empirical equations, viz., Bagging Regressor (r2 = 0.913), AdaBoost Regressor (r2 = 0.887), followed by Support Vector Regressor (r2 = 0.814). We have also developed an open-source web-based tool to predict the time-dependent scour depth with the proposed machine learning methods. The web-based tool is generic enough to work with any dataset and allows the end user to select various input combinations, visualizations, and error metrics.