Prediction of three-dimensional flood-flow past bridge piers in a large-scale meandering river using convolutional neural networks

TL;DR

This paper develops convolutional neural networks to efficiently predict turbulent flow statistics in large-scale meandering rivers with bridge piers, reducing reliance on computationally intensive simulations.

Contribution

The study introduces CNN models trained on LES data to accurately predict turbulent flow statistics in large rivers, offering a faster alternative to traditional simulations.

Findings

CNNs accurately predict turbulent statistics in validation scenarios

The approach reduces computational time compared to LES

CNN predictions closely match LES results

Abstract

The prediction of statistical properties of turbulent flow in large-scale rivers is important for river flow analysis. Large-eddy simulations (LESs) provide a powerful tool for such predictions, however, they require a very long sampling time and significant computing power to calculate the turbulence statistics of riverine flows. In this study, we developed encoder-decoder convolutional neural networks (CNNs) to predict the first- and second-order turbulent statistics of the turbulent flow of large-scale meandering rivers. We trained the CNNs using a dataset obtained from the LES of the flood flow in a large-scale river with three bridge piers, which formed the training testbed. Subsequently, we employed the trained CNNs to predict the turbulent statistics of the flood flow in a river with different bridge pier arrangements, which formed the validation testbed. The CNN predictions for the validation testbed river flow were compared with the simulation results of a separately performed LES to evaluate the performance of the developed CNNs. The results showed that the trained CNNs can successfully produce turbulent statistics of the flood flow in large-scale rivers, such as that chosen for the validation testbed.