Predicting the scouring depth around a rectangular pier with a round nose by artificial intelligence methods

Abstract

Local scour around bridge piers constitutes a significant phenomenon in river engineering. The development of artificial intelligence models and their capabilities has resulted in the widespread adoption of these models. In the experiments, various factors were considered, encompassing three different pier angles (0, 5, and 10°), six submerged vanes with two angles (20 and 30°) aligned with the flow direction, three vane heights (0, 1.25, and 2.5 cm) on the bed, and a collar. The study evaluated the performance of distinct machine learning models, namely, the multi-linear regression (MLR) model, radial basis function neural network (RBF), multilayer perceptron neural network (MLP), support vector machine (SVM), and adaptive neuro-fuzzy inference system (ANFIS) in predicting the scouring depth. Application of these models (RBF, MLP, ANFIS, SVM, and MLR) yielded positive outcomes, with corresponding R2 values of 0.99, 0.98, 0.97, 0.96, and 0.90 for scour depth assessment. The results demonstrated that the RBF artificial neural network model exhibited excellent generalizability in comparison to the other models. It consistently provided accurate predictions with minimal errors across varying training set sizes. Conversely, the ANFIS model exhibited limited generalizability compared to the other models, as reducing the training set size resulted in a significant increase in prediction errors during testing. Furthermore, although the MLR model demonstrated good generalizability, its prediction error was relatively high compared to the alternative models.