Estimation of Vortex-Induced Vibration Based on Observed Wakes Using Computational Fluid Dynamics-Trained Deep Neural Network

Abstract

The vibration of a circular cylinder due to fluid forces is of interest in various engineering fields. In this study, we investigate an approach to estimate the fluid forces acting on a circular cylinder in a flow field based on experimental flow visualizations using a deep neural network (DNN). Specifically, the wake patterns and fluid forces are correlated in a computational fluid dynamics (CFD) simulation, and the forces in the experiment are estimated by comparing experimental and computational wake patterns using a DNN. The approach is tested via dye-ink visualization around a circular cylinder at a Reynolds number of 560, referring to Seyed-Aghazadeh et al. (2015, “An Experimental Investigation of Vortex-Induced Vibration of a Rotating Circular Cylinder in the Crossflow Direction,” Phys. Fluids, 27(6), p. 067101). First, the CFD simulation of a circular cylinder with forced vibration in the crossflow direction is conducted with various vibration frequencies. Subsequently, the visualized wake images of the resulting flow fields and corresponding fluid forces are used as training data for the DNN. In the estimation, the images from the experiment are detected by the CFD-trained DNN. Thus, we can recall the correlated fluid forces using CFD simulation. The average drag coefficient and peak value of the lift coefficient estimated using streaming experimental images, have standard deviations of 2.1–13.7% and 6.6–18.6%, respectively, depending on the number of training images. The root-mean-square value of the lift coefficient obtained from the estimation is 0.82, which is comparable to the experimental value of 0.8, under the same flow and oscillation conditions.