Partial Image Error (PIE) in Digital Particle ImageVelocimetry (DPIV)

摘要

This paper examines and quantifies Partial Imaging Error (PIE) that is caused when an image is divided up into smaller interrogation regions, cutting particle images at the edge of a region in two. Original investigations have assumed that this source of error is negligible, yet this paper will prove that PIE is in fact a major contributor to measurement error and is more significant than the non-uniform weighting of the correlation function usually associated with measurement discrepancies. Results show that this error is significant for typical seeding densities and commonly used interrogation region sizes. If the correlation of regions that are 16 x 16 pixels or less is attempted PIE can prohibit a meaningful result from being obtained despite there being a valid correlation peak. It will also be highlighted that processes usually associated with improving the accuracy of measurements are unable to account for the effects of PIE and hence this inherent error remains. In order to reduce the effects of PIE and to increase measurement accuracy, it is necessary to normalise the correlation field before using a curve fit estimator on the correlation peak. However, it is shown that normalising by overlapped area, which is typically used as a normalisation function, is able to reduce mean bias error by correcting for the non-uniform weighting of the correlation function yet has no effect on RMS error. To correct for PIE, normalisation of the correlation field by signal strength (NSS) is presented here as an effective means of reducing both the mean bias and RMS error.