Principle Investigator: Dr. Alexei Poludnenko HPC System: Raptor, Garnet, Spirit Sponsor: Naval Research Laboratory, AFOSR (Air Force Office of Scientific Research), NASA (National Aeronautics and Space Administration)
Chemical Flame (Ambient Occlusion)
Methane-Air Mixture Pressure
Turbulent flames are pervasive both in our daily lives on Earth and in the Universe. They power modern energy generation and propulsion systems, such as gas turbines, internal combustion and jet engines. At the same time, they also have tremendous destructive potential being the primary driver of the majority of gaseous explosions. On astronomical scales, thermonuclear turbulent flames are at the core of some of the most powerful explosions in the Universe, knows as Type Ia supernovae. These are crucibles, in which most of the elements around us from oxygen to iron are synthesized, and in the last 15 years they have been used as cosmological distance probes to discover the existence of dark energy.
Despite this ubiquity in Nature, turbulent reacting flows remain poorly understood still posing a number of fundamental questions: What is the structure of turbulent flames at different turbulent intensities? What are the main mechanisms controlling the energy release rate (or the flame speed)? What is the stability of such reacting flows and are they susceptible to catastrophic transitions, e.g., formation of a detonation?
At the Naval Research Laboratory, a systematic investigation of the dynamics and properties of fast chemical and thermonuclear turbulent flames is underway. The focus of this study is on the model-free, first-principles modeling of the turbulence-flame interaction with the goal of understanding the fundamental physics of this process. A number of surprising phenomena have emerged in the course of this work. These include the ability of highly subsonic reacting turbulence to develop supersonic detonations spontaneously, pulsating instability and self-acceleration of turbulent flames, presence of the inverse energy cascade and strongly anisotropic turbulent transport, etc. These phenomena are unique to the reacting turbulence and are absent in its more traditional, nonreacting counterpart.
Here two examples are given of fully resolved, first-principles simulations of high-speed turbulent flames. One shows the process of spontaneous detonation formation by a chemical flame in a stoichiometric methane-air mixture under the upstream atmospheric conditions (first three images in the slide show and the bottom two movies on the right). The second example shows the structure of a thermonuclear flame in a degenerate, relativistic carbon plasma present in a stellar interior (last two images in the slide show and the top two movies on the right).
Both calculations were carried out using the code Athena-RFX on a uniform mesh. In the case of a chemical flame, the computational domain has size 1.3 cm x 1.3 cm x 42.5 cm and contains 537 million cells. In the case of a thermonuclear flame, domain size is 0.2 mm x 0.2 mm x 32 mm and it contains 4.3 billion cells. The latter simulation was performed on 32,768 cores on the Garnet platform (ERDC) and required 800,000 CPU hours. Substantial cost of the calculation associated with the large computational domain was compounded by the use of a detailed description of the nuclear reaction kinetics, equation of state, and thermal transport in degenerate plasma.
In the course of this study dozens of such calculations are required to fully sample the parameter space of turbulence-flame interactions for a typical reacting mixture. These simulations allow one to study in great detail the fundamental processes controlling the dynamics of burning in such complex, turbulent reacting flows.
The DAAC team has produced several images and movies for Dr. Poludnenko. These images and movies have appeared in the Walk-In/Walk-Out video at SuperComputing 2013, a poster submission for IEEE Vis 2012, and in the HPC Insights magazine.
The making of these images begins with the data created by the simulation. The data has ranged in size from around 10TB to over 20TB. Managing data this size in any capacity is very challenging. We chose to use VisIt, running in a client-server mode. The server would run on whichever HPC system the data resides on. Inside of VisIt, an isosurface was extracted that represented the surface of the flame. This geometry was saved into a Stanford PLY format.
After the geometry is created, To get the values needed to create a volume rendering of the pressure waves, we created a custom program that combed through the data and picked out values if they were above a minimum threshold. These chosen values were then translated into a PRT format which is used by Krakatoa , a plugin for 3D Studio Max. Krakatoa is a particle renderer, which is perfect for volume rendering projects.
To learn more about the details on how these images were rendered, including the custom flame shader, ambient occlusion shader, and Krakatoa, click here.