Data-Driven Identification of Robust Low-Order Models for Dominant Dynamics in Turbulent Flows

This work presents an automated process minimising input parameters for the study of turbulent flows. The goal is to gain insight into the flow dynamics by deriving a data-driven reduced-order model (ROM). Spectral proper orthogonal decomposition (SPOD) is used to efficiently separate the flow dynamics and project the flow field onto a low-dimensional subspace to represent the dominating dynamics with a reduced set of modes. A polynomial combinations of the temporal modal coefficients defines a function library to describe the dynamics by a linear system of ordinary differential equations. In a two-stages cross-validation procedure (conservative and restrictive sparsification), the most important functions are identified and combined in a final ROM. The process is demonstrated for PIV data of a circular cylinder undergoing vortex induced vibration (VIV) Re = 4000.