Alternative Artificial Neural Network Structures for Turbulent Flow Velocity Field Prediction

Turbulence in fluids has been a popular research topic for many years due to its influence on wide range of applications. Computational Fluid Dynamics (CFD) tools are able provide plenty information about this phenomenon, but their computational cost often makes the use these unfeasible. For that reason, recent years, turbulence modelling using Artificial Neural Networks (ANNs) is becoming increasingly popular. These networks typically calculate directly desired magnitude, having input domain. In paper, Convolutional Network (CNN) predicting different magnitudes turbulent flows around geometries by approximating equations Reynolds-Averaged Navier-Stokes (RANS)-based realizable k-? two-layer model proposed. Using CNN, alternative network structures proposed predict velocity fields flow rectangular channel, with preliminary stage pressure and vorticity before calculating fields, obtained results compared ones basic structure. The demonstrate clearly outperform one, especially when becomes uncertain. addition, considering results, best configuration That tested domain multiple narrowing which domains conditions from training ones, showing fairly accurate predictions.