Estimation of oscillation parameters of a circular cylinder from its downstream vorticity fields

Life around us is full of fluid flows with oscillating objects such as swimming fish and hovering birds. Even if flow phenomena involving such oscillating bodies are simply modeled as an oscillating cylinder in a uniform flow, slight deviations in the oscillation parameters can drastically change the flow in the wake. In this study, we consider a flow around an oscillating circular cylinder under different conditions (i.e., amplitude, frequency, and angle) and attempt to estimate these oscillation parameters from the vorticity fields in the wake using machine learning. In this presentation, we will report the results of machine learning with convolutional neural networks (CNN) and a multilayer perceptron (MLP) applied to the classification and regression problems of amplitude, frequency, and angle. The results show that in both cases, the classification problem is simpler than the regression problem, and classification can be performed with a higher accuracy. Furthermore, for the regression problem, there was a noticeable difference in the estimation accuracy depending on the frequency and amplitude. In order to identify the cause of this difference, we investigate the similarity of the flow field and the number of POD modes, and discuss the significant influence of the complexity of the flow field.