Adaptive PIV processing based on ensemble correlation

Abstract

The ensemble correlation approach to PIV processing relies on averaging the correlation planes of many images before performing a peak search and subsequent estimation of the average displacement. Generally this processing method has found wide-spread use for the analysis of sparsely seeded micro-PIV image data and can achieve resolution at the single pixel level if the number of images is sufficiently large. The article investigates the potential of the ensemble correlation approach for macroscopic applications driven by two primary motivations: (1) to increase processing speed for the retrieval of mean and fluctuating components, and (2) to provide an estimator to aid in the processing of the individual frames at a later stage. The ensemble correlation algorithm, implemented as a coarse-to-fine grid refinement approach with adaptive sampling window offset, provides mean velocity data at least one order of magnitude faster than the conventional frame-by-frame PIV interrogation schemes and shows rapid convergence and minimal deviation from averaged data obtained by conventional PIV. Estimates of the fluctuation terms (RMS-values) are derived using a modified peak detection and fitting algorithm based on least squares fitting an elliptically shaped Gaussian distribution. The method is demonstrated on PIV recordings obtained from a subsonic free air jet and compared to results obtained by conventional (pair-by-pair) PIV processing. Mean displacements are matched within the expected uncertainty of PIV processing except in regions of increased turbulence which exhibit a systematic bias. At this point fluctuating components can only be roughly estimated but can be used the limit the peak search area while the processing of the individual frames at a later stage. A second example demonstrates the method on PIV recordings obtained from the flow of a transonic compressor. Aside from recovering the mean displacement an order of magnitude faster than conventional PIV processing the sampling resolution could be improved roughly 8 times (32x32 vs. 12x12 pixel) allowing much a higher spatial resolution of compression shocks