Incompressible transitional air flow over a wind turbine section

The theme of the 2019 challenge was distributed rendering, i.e. visualization of very large datasets that require parallel rendering on a cluster. Participants could use either their own data or the supplied dataset kindly provided for this competition by Joshua Brinkerhoff (UBC Okanagan). This dataset came from an OpenFOAM numerical simulation of incompressible transitional air flow over a wind turbine section.

First place

The first place was taken by Thierry Villeneuve, a Ph.D. student from the Department of Mechanical Engineering at Université Laval, for his time-dependent 3D volume rendering of the vorticity magnitude in a turbulent turbine wake. The dataset was taken from Thierry’s own CFD model using a DDES (Delayed Detached-Eddy simulation) approach with a domain composed of approximately 10⁸ cells. The simulation has been performed on Compute Canada’s Niagara cluster using 10 nodes (400 cores) and took about two months. The output with the three components of the vorticity vector was saved every 20 time steps, resulting in ~2200 frames. The blue / green semi-opaque regions represent low-vorticity zones, while the red regions show high-vorticity zones.

^{(Or click here to watch this video directly on Vimeo.)}

Second place

The second place went to Yohai Meiron, a Postdoctoral Fellow from the Department of Astronomy & Astrophysics at the University of Toronto, for his demonstration of VTK-m filters and rendering using the competition’s “default” dataset. VTK-m is a scientific visualization toolkit for multi-core processor architectures (both CPUs and GPUs). With an interface currently only in C++, VTK-m is a complete rewrite of the serial VTK algorithms and functions. Some functionality of VTK-m could be used through ParaView and VisIt.

While VTK-m supports MPI through a class called PartitionedDataSet, many if not most of its built-in algorithms do not know how to handle this class. The solution was to use another library called VTK-h that is basically a wrapper around VTK-m to enable hybrid parallelism.

The video below use four different rendering techniques. Of these the most interesting visually is the last part showing volumetric rendering of the z-component of the vorticity vector with nice coherent structures above and below the airfoil, and a gap between the lower structure and the airfoil itself.

^{(Or click here to watch this video directly on Vimeo.)}

Third place

The third place was taken by Joshua Ludwig, a Ph.D. Student in Plasma Physics from the University of Alberta. His video (below) was rendered using Joshua’s own parallel ray tracing rendering code on 5 nodes (160 cores) on Compute Canada’s Graham cluster.

The video shows plasma electron density (low in blue and high in red). The camera is moving at the speed of light to follow a short high intensity laser pulse (not shown). This laser pulse is so intense that it expels all the plasma electrons from its path and creates a bubble-like feature in the plasma. This bubble feature still contains positively charged ions due to their large mass and hence has strong electric fields that pull electrons into the bubble and accelerate them as they ride along with the laser pulse. This mechanism is called Relativistic Laser Wakefield Acceleration.

The simulation was performed on 32 nodes (1280 cores) of Compute Canada’s Niagara cluster using OSIRIS, a 3D fully relativistic Particle-in-Cell plasma modelling code. It solved Maxwell’s equations for the electric and magnetic fields on a 3600x1800x1800 grid covering the domain 160x120x120µm with ~23 billion electron particles. The output used specifically to make this video was about 20TB.

^{(Or click here to watch this video directly on Vimeo.)}