Compressed representation of separation bubbles from a vast database

The present work describes a wide experimental database of laminar separation bubbles formed on a flat plate for different Reynolds numbers, adverse pressure gradients and free-stream turbulence intensity levels. The database accounts for 72 different combinations of the aforementioned parameters, for which both short and long bubble types have been observed. For each flow case, a set of 6000 snapshots has been acquired using a fast response Particle Image Velocimetry system. In this work all the 72 × 6000 images have been used simultaneously to construct a large snapshot matrix containing both the statistical and the dynamic response of the bubble to the flow parameters variation. To handle such a large snapshot matrix, a parallelized version of the Proper Orthogonal Decomposition routine is presented to be adopted in High-Performance Computing environment. The reduction of the high dimensional database into a low-order model retaining few POD modes and coefficients has been obtained. Particularly, two subsets of modes capturing the time-averaged and dynamic response of the bubble have been detected based on the frequency spectra of the related coefficients. It will be shown that dimensionality can be reduced up to about 1%, while retaining the generalized response of a laminar separation bubble. The present work therefore shows the existence of a reduced state space over which the response of a LSB evolves for a wide range of the leading influencing parameters. • LSB to changes in Reynolds number, turbulence intensity and pressure gradient. • POD on large datasets is a promising tool to extract key LSB features. • 15% of POD modes capture the statistical and dynamic response of LSBs. • Time-mean and dynamic response of database LSBs evolves in a low-dimensional space. • The low-order state space is able to reconstruct statistics and dynamics of LSBs.