PIV data clustering of a buoyant jet in a stratified environment

Three spatial clustering approaches of a high-Reynolds number transient buoyant jet in a linearly stratified environment are applied along with proper orthogonal decomposition to identify similar/consistent regions in the domain of interest. The velocity fields analyzed are obtained from an experimental test with large scale, time-resolved, particle image velocimetry (PIV) measurements. Clustering is performed by the k-means method considering: (a) crosssection velocity profiles, (b) point-wise energy spectra, and (c) point-wise Reynolds stress tensor components. Three metrics are used for the assessment of clustering approaches, namely: (a) within-cluster sum of squares, (b) average silhouette, and (c) within-cluster number of POD modes required to resolve prescribed levels of total variance/energy. Results are promising and lay the foundation for an in depth analysis of local features of complex flows as well as the formulation of efficient reduced order models.