Comparative Assessment of the Computational Fluid Dynamics and Artificial Intelligence Methods for the Prediction of 3D Flow Field around a Single Straight Groyne

Abstract

In this research, we investigated the three-dimensional time averaged flow pattern around a single straight groyne. To measure the three-dimensional velocity components in the laboratory, we utilized an Acoustic Doppler Velocimeter (ADV). We employed Computational Fluid Dynamics (CFD) and Artificial Neural Network (ANN) techniques to simulate the crucial flow characteristics. To validate these methods, we compared the simulation results with the measured data. The findings demonstrate that the ANN approach, with R2 values of 0.9152, 0.9150, and 0.9315, outperforms the CFD model, with R2 values of 0.8332, 0.8726, and 0.8051, in the prediction of the u and v velocity components as well as the velocity magnitude. The transverse velocity profiles indicate that the ANN method accurately predicts the velocity components and velocity magnitude, whereas the CFD method exhibits significant disparities from the measured data, particularly in the prediction of longitudinal and vertical velocity components, especially in the near-bed regions. The ANN method and the laboratory data display variations in their patterns across the shear layer and at the flow separation boundary, while the velocity profiles in the CFD method demonstrate a consistent increase from the right wall of the channel toward the main flow zone. Other flow features around the groyne, such as horseshoe vortex, secondary flow, clockwise and counterclockwise rotational flows around the groyne head and the length and precise center of the circulation zone are reasonably predicted by the ANN method.

Keywords

• groyne
• computational fluid dynamics
• artificial intelligence
• horseshoe vortex