Ensemble-correlation algorithm for accelerated, adaptive PIV processing

Kurzfassung

Without doubt the PIV technique can now be considered a standard flow measurement technique in the field of experimental fluid mechanics, both for fundamental as well as applied research. At this point it is commonplace to acquire ever increasing amounts of image data with inflationary trends very similar to Moore’s law for computer hardware. Considerable effort has been devoted to the development of adequate processing algorithms that push spatial resolution and accuracy close to the theoretical limits. Nonetheless numerous tuning parameters are involved to optimize acquisition and subsequent evaluation, such as seeding density, pulse separation, sample size, grid resolution, peak fit, choice of correlation method, validation parameters, etc. The proper choice of many of these parameters can usually only be gained through experience. This makes the quality of a PIV measurement dependent on subjective judgment of the individual user. Thus any tools that are capable of objectively assessing the quality of a PIV measurement are of great value. Ideally these tools should be applicable at the time of image acquisition (e.g., for near real-time adjustment of pulse delay and seeding density) and should provide initial parameters for subsequent processing and validation. Another shortcoming of common algorithms is that most parameters are applied in a global sense, that is, they are constant throughout the field of view while the dynamics of the examined flow field would call for spatial (or temporal) variation of PIV processing parameters. The presentation explores the possibilities of utilizing the statistical information provided through the analysis of PIV image series (not necessarily time-resolved) on the basis of the so-called ensemble-correlation. Originally introduced for the evaluation of micro-PIV images to overcome effects due to Brownian motion and to increase the spatial resolution in spite of sparse seeding, it has received little attention for its possible use in classical macroscopic PIV applications. The main concept behind ensemble correlation is to average locally sampled cross-correlation planes over many images rather than averaging the extracted displacement vectors. The mean displacement is then obtained from the average correlation planes whose correlation peaks typically have a considerably larger signal-to-noise ratio than the correlation peaks in the correlation data obtained from the individual image pairs. One great advantage of ensemble-correlation is that the mean displacement field can be obtained considerably faster than by using conventional image-by-image analysis. This can be exploited to rapidly assess PIV image quality as well as provide initial estimates for evaluation of the individual image pairs. The class of proposed algorithms suggested here uses ensemble correlation on a given PIV image series to recover the mean displacement and in addition evaluates the averaged correlation peak shape to extract detection bounds for processing the individual PIV image pairs. The algorithms inherently offer the capability of spatially varying detection criteria. The presentation illustrates the performance and potential of the algorithm on PIV data sets both from experiments as well as numerical simulations.