Flow Variables Prediction Using Experimental, Computational Fluid Dynamic and Artificial Neural Network Models in a Sharp Bend

Abstract

Bend existence induces changes in the flow pattern, velocity profiles and water surface. In the present study, based on experimental data, first three-dimensional computational fluid dynamic (CFD) model is simulated by using Fluent two-phase (water + air) as the free surface and the volume of fluid method, to predict the two significant variables (velocity and channel bed pressure) in 90º sharp bend. The CFD results are compared with experimental data, and CFD model is verified with average RMSE, 0.02 and 0.13 and MAE, 0.018 and 0.1 respectively for the velocity and the pressure. Then, two multi-layer perceptron artificial neural network (MLP-ANN) model is trained by observed datas. The results show that the value of R2, 0.984 and 0.99 respectively to predict the velocity of flow and pressure by ANN models are acceptable accuracy. ANN model acts more accurately with average erro value of MAE, 0.048 than the CFD model with average MAE, 0.06 to predict the velocity and pressure. The velocity and pressure pattern of flow is predictable through both numerical models, CFD and ANN models in every part of the channel.