High-Resolution PIV With Uncertainty Estimation With KNN

We introduce a novel approach to improving the resolution of Particle Image Velocimetry (PIV) measurements. The method merges information from different non-time resolved snapshots exploiting similarity of flow regions in different time instants. The main hypothesis is that the identification of similar flow structures at different time instants is feasible if a sufficiently large ensemble of statistically-independent snapshots is available. Merging individual vectors from different snapshots with similar flow organisation allows an artificial increase of the available information. This paves the way to refining of the interrogation region, i.e. increasing spatial resolution. The similarity can be enforced on a local scale, i.e. morphologically-similar regions are sought only among subdomains corresponding to the same flow region. The identification of locally-similar snapshots is implemented with an unsupervised K-nearest neighbours search in the space of significant flow features. Such features are identified with Proper Orthogonal Decomposition (POD) in subdomains of the original low resolution data. The refined bin size will depend on the number of “sufficiently close” snapshots: the more neighbours are identified, the higher is the “virtual” particle image density available, and consequently the smaller is the bin size. The statistical dispersion of the velocity vectors within the bin is then exploited in the estimation of the uncertainty. The optimal number of neighbours is the one corresponding to the minimum uncertainty. The method is tested and validated against datasets with a progressively increasing level of complexity: two virtual experiments based on direct simulations of the wake of a fluidic pinball and a turbulent channel flow; experimental data collected in a turbulent boundary layer.