Evolutionary streamlines for automatic flow feature extraction

Figure: Visualization of blood flow simulation data in an Aneurysm: (a) Exhaustive exploration of the volume by seeding 5000 streamlines within the domain. Features of interests are hidden in the dense set of flow lines. (b) Filtering the lines from (a) reveals the central vortex in the aneurysm showing the 97 selected lines. (c) A direct optimization of 100 individuals using an evolutionary algorithm finds the central vortex after 20 iterations reducing the computational effort by one order of magnitude.

In this work we explore the potential of evolutionary algorithms to generate expressive visualizations. Evolutionary algorithms find close-to-optimal solutions for tasks by imitating biologically mechanisms like selection, mutation and recombination. In this work we consider the task of finding a set streamlines expressing the overall behavior of a flow field, while highlighting features of interest. Our approach directly optimizes the solution candidates with respect to a user selected fitness function. As such it provides a powerful alternative to filtering methods commonly used in visualization where the space of possible candidates is densely sampled in a pre-processing step from which the best candidates are selected and visualized. At the same time the problem of possible under-sampling is solved since we are not tied to a preset resolution. We demonstrate our approach on the well-known flow around an obstacle as case with a two-dimensional search space. The blood flow in an aneurysm serves as an example with a three-dimensional search space, see Figure 1.