101 / 2025-04-24 12:56:03

Prediction of pier and abutment scour depth under unsteady flow condition based on machine learning

With the continuous development of human society, hydraulic structures such as bridges play a crucial role in promoting regional connectivity. However, in mountainous regions, significant differences in water flow conditions between flood and dry seasons lead to intense riverbed evolution during floods, causing severe scour at the foundations of hydraulic structures and posing considerable safety risks. This paper analyzes the progress in unsteady flow scour research and explores its characteristics, introducing machine learning theory as a tool for further study of local scour under unsteady flow conditions. A backpropagation neural network (BPN) is utilized to establish a method for predicting the scour development process of bridge abutments and piers under both steady and unsteady flow conditions. The study develops an intelligent prediction model for the clear-water scour process of bridge abutments under steady flow, demonstrating that the BPN method effectively captures the nonlinear time-series characteristics of scour development. The model is also extended to predict scour for bridge piers in sediment-laden flows under steady conditions, with optimization techniques such as data augmentation and time windowing introduced to address challenges arising from limited and heterogeneous data. Furthermore, a time window-based model is developed for predicting abutment scour under unsteady flow, showing enhanced performance compared to existing empirical formulas. Additionally, a clear-water scour prediction model for bridge piers under unsteady flow with single-peak flow hydrographs is established using a superposition method, addressing issues of irregular fluctuations during flow recession phases. The results across all models demonstrate strong predictive accuracy, offering improved insights into the temporal evolution of scour.