Reference Frame Fields for Locally Adapted Vortex Detection and Visualization in Unsteady Flow
This thesis present three methods that advance the extraction and visualization of vortices in unsteady flow using reference frame fields to establish a cohesive framework for the analysis of unsteady vortices, advancing both their extraction through locally adapted reference frame fields and their visual verification.
Overview
The analysis of vortex structures plays a key role in flow visualization, where two main challenges must be addressed: feature extraction and visualization.
For vortex extraction, an established approach is to compute optimal reference frame fields relative to which the input vector field becomes as steady as possible. However, existing methods either optimize locally over a fixed neighborhood with uniform sample contributions, which might not match local features, or perform global optimization, which is slow. As the first method in this thesis, we present a reference frame optimization technique that automatically adapts to the flow field locally. Adaptivity is enabled by formulating reference frame optimization as a moving least squares (MLS) approximation that incorporates fluid features via a guidance field that determines sample contribution during local optimization. By using the finite-time Lyapunov exponent (FTLE) as a guidance field, local optimization is restricted to samples within the coherent region of the target position, improving the detection of small-scale vortices compared to prior methods.
Although locally adapted reference frame optimization improves vortex extraction, the previous pipeline limits its application to 2D, since it requires the precomputation of local sample contributions, and the computational cost and memory requirements increase several orders of magnitude in 3D. To enable locally adapted reference frames for 3D flow, for the second method in this thesis, we incorporate Monte Carlo Integration during local optimization with on-the-fly computation of sample contributions, avoiding precomputation and exposing progressively refining results. We use the Monte Carlo reference frame optimization to design an interactive visualization pipeline for observed line integral convolution (LIC). Interactivity is achieved by limiting the optimization domain to the active visualization slice and using instantaneous vorticity deviation (IVD) as a guidance field, which is significantly cheaper than FTLE.
In the last method, the focus shifts to vortex visualization, where visually verifying vortices in unsteady 3D flow remains difficult. A common approach is to compute streamlines around candidate corelines, but they depend heavily on the chosen reference frame and represent a temporal snapshot of the flow. Pathlines, by contrast, directly represent particle motion and are therefore a more natural choice. However, the local rotation of particles around a vortex is typically obscured by the bulk motion of the flow, making lab-frame pathlines difficult to interpret. To overcome these challenges, we present a visualization technique that associates pathlines with Lagrangian corelines. We compute a rotation-minimizing vortex-centric reference frame along Lagrangian corelines that depends exclusively on the deformation of the coreline. Using this reference frame, pathlines are transformed via a space deformation that avoids introducing a rotation component around the core's tangent and decouples the local swirl of particles from the bulk flow.
Together, these methods establish a cohesive framework for the analysis of unsteady vortices, advancing both their extraction through locally adapted reference frame fields and their visual verification through a pathline-based vortex-centric representation enabled by reference frame fields along Lagrangian corelines. Demonstrations on multiple unsteady flow fields confirm the effectiveness of each method, opening new avenues for flow visualization and analysis.
Presenters
Brief Biography
Julio Rey Ramirez is a Ph.D. student in Computer Science at King Abdullah University of Science and Technology (KAUST) advised by Prof. Markus Hadwiger. In 2018 he completed his B.S. in Electronics Engineering from the National University of Colombia (UNAL) in Bogotá, Colombia. He completed his M.S. degree in Computer Science at KAUST in 2022, working on reconstruction algorithms for cryogenic electron tomography under the supervision of Professor Wolfgang Heidrich.