Data-driven prediction of unsteady flow over a circular cylinder using deep learning

Abstract

Unsteady flow fields over a circular cylinder are trained and predicted using four different deep learning networks: generative adversarial networks with and without consideration of conservation laws and convolutional neural networks with and without consideration of conservation laws. Flow fields at future occasions are predicted based on information of flow fields at previous occasions. Predictions of deep learning networks are conducted on flow fields at Reynolds numbers that were not informed during training. Physical loss functions are proposed to explicitly impose information of conservation of mass and momentum to deep learning networks. An adversarial training is applied to extract features of flow dynamics in an unsupervised manner. Effects of the proposed physical loss functions and adversarial training on predicted results are analyzed. Captured and missed flow physics from predictions are also analyzed. Predicted flow fields using deep learning networks are in favorable agreement with flow fields computed by numerical simulations.

Prediction of a typhoon track using a generative adversarial network and satellite images

“…During training a mini batch size of mbs = 8 and learning rates of 0.00004 and 0.02 have been chosen for the generator and discriminator network. These values have proven themselves to be valuable in the work of Mathieu et al 24 , as well as in the study of Lee and You 25 .…”

Deep unsupervised learning of turbulence for inflow generation at various Reynolds numbers

“…This method can be utilized for the inflow turbulence generation problem. In general, convolution neural networks (CNN) are used to better represent the spatial correlation of flow [13,14,15,16,17,18]. However, the typical CNN structure [16] generates blurred fields over time when the recursive prediction, which uses the predicted information as the input, is carried out.…”

Prediction of aerodynamic flow fields using convolutional neural networks

“…where U , and V are the x-component and y-component of the velocity field respectively, and P is the scalar pressure field. m is the batch size, n x is the number of grid points along the x-direction, n y is the number of grid points along the y-direction, and L is the number of layers with trainable weights, and n l represents number of trainable weights in layer l. MSE is the mean squared error, and GS is gradient sharpening or gradient difference loss (GDL) (Mathieu et al, 2015;Lee and You, 2018). In this paper, we use gradient sharpening based on a central difference operator.…”

“…Quantitative results are presented in Tables 3 and 4. 3.2.3 Shape, angle of attack, and Reynolds number variation with gradient sharpening To penalize the difference of the gradient in the loss function, and to address the lack of sharpness in predictions, we use gradient sharpening (GS) (Mathieu et al, 2015;Lee and You, 2018) in the loss functions combination and present the cost function over the training set as,…”