Clustering-based threshold estimation for vortex extraction and visualization

Research efforts have been devoted to extraction and visualization of vortices in an unsteady (turbulent) flow. Characterizing the behaviors of the flow, vortices are identifiable as regions using a vortex detector known as the lambda2-criterion. Isosurface visualization renders vortex regions based on a chosen isovalue. However, it is highly challenging to choose one isovalue suitable for visualizing vortex regions of the entire flow field. A solution is the approach of maxima score that localizes vortex regions identified by the lambda2-criterion based on similarity scores relative to local extrema. The approach is however sensitive to noise or floating-point errors in the flow, leading to clutter in vortex visualization. As a feasibility study, this paper presents a threshold estimation to overcome this sensitivity. The estimation involves clustering on local minimum differences in lambda2 scalar values derived from the gradient tensor of the velocity field, and yields multiple values of the threshold without user intervention. Tested on several flows in various size and Reynolds number, the results of the threshold estimation confirmed overcoming the sensitivity of the maxima score approach. This indicates a potential of the threshold estimation to improve the robustness of the approach for vortex extraction and visualization.