In-depth simulation of netted collars on scour depth control using machine-learning models

The present research aims to forecast the safeguarding efficacy of a mesh collar, of hole diameter d, in developing of scour depth around a cylindrical bridge pier of diameter D under the steady and clean water conditions utilizing three machine learning models (MLMs), namely Support Vector Machine (SVM), Gene Expression Programming (GEP), and Multilayer Perceptron (MLP). A total of 240 laboratory measured scour depth data were employed in this study. The experimental setup involved the installation of four distinct mesh collars, configured in the shapes of a circle, square, rectangle, an d triangle by shape factor (SF) of 1.78, 1, 2.3, and 1.69, respectively. The mean size of non-cohesive sand particles was selected with a particle size of 1.3 mm. Employing dimensional analysis, three dimensionless parameters, namely SF, d/D, and U-c/U were identified as independent variables adopting for the input variables for the MLMs. The performance assessment metrics involved Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), coefficient of determination (R2), and the Developed Discrepancy Ratio (DDR). The simulation results demonstrated that MLMs exhibit a high degree of accuracy in predicting relative scour depth (RSD) influenced by the presence of mesh collars. Among the aforementioned three models, the GEP model demonstrated its superiority with corresponding values for (RMSE, MAE, R-2, DDRmax) indices of (0.11342, 0.08642, 0.85058, 2.54) and (0.0787, 0.0624, 0.8959, 3.66) during the training and testing phases, respectively. Finally, an equation was extracted to predict RSD using the GEP model.