In this research, due to the ability of artificial neural network in modeling of highly nonlinear systems, intelligent turbulence model was proposed for prediction of channel flow. The flow analysis in the channel is done using the direct simulation method known as DNS.
A pseudo spectral method is used in Cartesian coordinates . The friction Reynolds number for channel flow is set to be Reτ= and the computational grids of are used in the x, y and z directions, respectively.
Also, neural network training using Leanberg-Marquardt algorithm (LM) is developed to predict the turbulent channel flow using information from DNS. This training was carried out using two novel approaches in determining the number of hidden layer neurons and multi-purpose training for simultaneous analysis of the vector of velocity field with the the achievement of performance of 0.5005 and correlation coefficient of 98%.
Due to the doubts in the accuracy of measuring the flow field of adjacent walls in the channel, theintelligent modeling of this section of the flow was conducted with 20480 samples, separately during the time steps. The results of network training and the analysis of velocity distribution of turbulent flow in different states indicate that the artificial neural network, like the usual experimental and numerical fluid dynamics methods, is capable of modeling turbulent flows well.
Rajabi, E., & Kavianpoor, M. (2018). Intelligent Modeling of Turbulent Channel Flow Based on Direct Numerical Simulation Data. Asas Journal, 20(50), 5-18.
MLA
Elham Rajabi; Mohammadreza Kavianpoor. "Intelligent Modeling of Turbulent Channel Flow Based on Direct Numerical Simulation Data". Asas Journal, 20, 50, 2018, 5-18.
HARVARD
Rajabi, E., Kavianpoor, M. (2018). 'Intelligent Modeling of Turbulent Channel Flow Based on Direct Numerical Simulation Data', Asas Journal, 20(50), pp. 5-18.
VANCOUVER
Rajabi, E., Kavianpoor, M. Intelligent Modeling of Turbulent Channel Flow Based on Direct Numerical Simulation Data. Asas Journal, 2018; 20(50): 5-18.
)