Wakes of bluff bodies that exhibit unsteady behavior are a topic of great interest in the study of fluid dynamics. Vortex formation in these... Show moreWakes of bluff bodies that exhibit unsteady behavior are a topic of great interest in the study of fluid dynamics. Vortex formation in these wakes depends significantly on the Reynolds number and the arrangement of the bluff bodies in the computation domain. To attain a comprehensive understanding of the unsteady wakes of adjacent bodies, we examine the emerged flow patterns in the wake of two bodies when subjected to different flow regimes and geometric configurations. This work aims to develop a reduced-order model that can capture the dynamics and predict the time evolution of specific parameters in the flowfield. Investigations including direct numerical simulations of two collinear plates normal to the flow were performed. Flowfield data and forces exerted on the plates were collected using a numerical code of an immersed boundary projection method (IBPM). The conducted numerical simulations pursued classifying the flow patterns by systematically varying the Reynolds number and the gap between the two plates. It was found that at small gap spacings, a typical von Karman vortex street is observed. Whereas at larger gap spacings, both a biased and a flip-flopping gap flow are detected. Prevalent coherent structures present in various flow regimes can be extracted via data-driven modeling techniques. The proper orthogonal decomposition (POD) method is used in this framework, from which projection-based reduced-order models are developed utilizing the governing equations of fluid flows. Single and broadband spectra are observed in the unsteady wake of the two-plate configuration. The amplitude and frequency of the time-evolution of the true POD modes and the predicted models are assessed using the spectral proper orthogonal decomposition (SPOD), an empirical method to extract coherent structures one frequency at a time from fluid flows. It was found that these reduced-order models are able to recover the frequency content from non-time resolved data. Show less