Data validation

In particle image velocimetry the measurements contain a number of spurious vectors. These vectors deviate unphysically in magnitude and direction from the nearby vectors that are, in turn, physically meaningful. They originate from those interrogation spots that contains insuffcient number of particle images, or whose signal to noise ratio is very low. In post processing process, the first step is to identify these vectors and subsequently discard them to form a valid data set. The detection of either a valid or spurious displacement depends on the number and spatial distribution of particle image pairs inside the interrogation spot. In practice, there should be at least four particle image pairs to obtain an unambiguous measurement of the displacement (Westerweel, 2000). The number of particle images inside an interrogation spot is a stochastic variable with a Poisson probability distribution. Hence an average of 10 particle images per interrogation spot at an average in-plane displacement of will give a probability of 95% of finding at least four particle image pairs. Here, is the size of the interrogation spot. The valid data yield can be improved by increasing the seeding density. But by increasing the seeding density we increase the influence of the seeding on the flow.

There are various way to detect spurious vector in a velocity field. Three mainly used tests are the global mean test, local mean test and local-median test. The global mean and the local mean are both linear estimators of valid vectors. The local median test is a nonlinear estimator that is often used in outliers identification. The outliers in turn, are identified by the median of the sample data. Out of the above three, Westerweel [176] has shown that the local median test has the highest effciency. In these techniques, the value at a grid point is compared with the neighboring grid points; if it exceeds a certain threshold, the value is discarded.