Conditional space-time POD extensions for stability and prediction analysis
The correlation and extraction of coherent structures from a turbulent flow is a principle objective of data-driven modal decomposition techniques. The Conditional space-time Proper Orthogonal Decomposition (CPOD) offers insight into transient dynamics, revealing the causation of specific flow phenomenon - or events, in a customizable manner. This work exploits the temporal evolution of CPOD modes in a reduced subspace, resulting in new extensions and adaptations that meet or exceed the capabilities of other decomposition methods. Chiefly, it is demonstrated that the subsequent application of dynamic mode decomposition (DMD) to CPOD modes, provides a flexible tool to investigate targeted flow instabilities, both tonal and convective in nature. By extending the CPOD time-horizon to educe the former type, it is shown that CPOD-DMD can exactly reproduce Spectral POD modes. Regarding the latter, a multi-resolution framework (CPOD-mrDMD) yields a refined "cause and effect" stability analysis, capable of diagnosing the natural forcing mechanisms within the flow, and the resulting unstable modes. In a separate application, CPOD properties are appreciated in the context of reduced order models, with an example of real-time flow prediction of extreme events derived from an active sensor correlated to a CPOD mode. The various CPOD functions and perspectives in this work are demonstrated on: the nonlinear chaotic Lorenz system, 3D intermittent turbulent spots in supersonic boundary layer transition, Schlieren video processing of unstarted inlet buzz, the aeroacoustic feedback forcing of a resonating impinging jet, and prediction of intermittent bluff-body wake structures impinging on a channel wall.