Pattern recognition analysis of the turbulent flow past a backward facing step

A pattern recognition technique for the investigation of large-scale coherent structures, is applied to analyze the turbulent separated flow over a backward facing step (BFS) at a Reynolds number Re-h = 5.0 x 10(3). The instantaneous two-dimensional velocity distribution is obtained by means of digital particle image velocimetry (D-PIV) measurements. High spatial resolution (Delta r/h = 1/25) is achieved with the application of an iterative window refinement image processing algorithm. The measurement plane is oriented in order to investigate spanwise aligned vortices footprints. The detection algorithm is based on velocity pattern spatial cross correlation. An additional isotropy condition is imposed to improve the detection of vortices and shear layer. The structure of the shear layer emanating from the step edge is examined emphasizing the role of coherent fluctuations with a length scale d ranging from 0.12 h to 0.44 h. A characteristic statistical spatial occurrence is found for the educed spanwise-aligned rollers: a quasi-linear spreading region extends from x/h = 0.8 up to x/h = 3.5. Within the same region the production of turbulent kinetic energy exhibits a maximum. At smaller scale, the vortices show a significant presence of counter-rotating structures inside the free shear layer suggesting that the spanwise rollers undergo early three dimensional instability and breakdown within a few step units. Conditional data averaging is also applied to the results and structural properties (coherent velocity, vorticity and turbulence production) are highlighted: close to the step edge the coherent vorticity distribution is strongly distorted showing an intense interaction between the rollers and the shear layer. A roughly circular pattern is recovered downstream x/h = 4. (C) 1999 American Institute of Physics. [S1070-6631(99)00512-7].