Perpustakaan Digital ITB

Data-driven methods has gained a rapid growth in recent years. It is driven by the rise of big data in various fields. Nowadays, Deep learning is the most famous data-driven method used in wide range of application. In Aerospace Engineering field, there are a lot of applications of deep learning such as in fluid mechanics. One of the biggest problem in fluid mechanics that has not been completely solved is turbulence. With massive success of deep learning in various real world problem, application of deep learning in turbulence modeling problem has became interesting research topic in recent years. Reynolds-average Navier-Stokes (RANS) turbulence model is the most used technique in modeling turbulent flow for real world problem. The main problem in RANS turbulence modeling is to obtain the closure relation that relates the Reynolds stress to the mean flow properties. In this work, deep learning is used to develop a model for turbulence closure modeling which gives the improvement from common turbulence model in computational fluid dynamics. There are two neural network architectures used to develop a neural network model for calculating Reynolds stresses in square duct flow problem. The first model is feed forward neural network (FFNN) and tensor basis neural network (TBNN). The performance of these two model is compared with RANS k ?? model as the linear-eddy viscosity model. From the results of this work, it is shown that both neural network models could give better performance on giving the closure prediction for turbulent in square duct flow and FFNN model proposed by the author could give the best closure prediction. Hence, the utilization of deep learning in turbulence closure modeling for square duct flow problem could give improvement from the common CFD model.