Click the image to view the .avi movie
The direct numerical simulation above is for a temporally developing, hydrogen-oxygen shear layer flame at a pressure of 100 atm and a Reynolds number of 4,500. A detailed and pressure dependent reaction mechanism having 8 species and 19 steps was used, along with a real gas state equation, generalized heat and mass diffusion models, and real property models. The simulation was run on 3,840 processing cores on Clemson's Palmetto cluster. It required 976 x 1200 x 624 grid points (~ 3/4 billion). The video shows the flame temperature as a function of time along the centerline plane.
Click here to see a video of the temperature as a function of x3 at time t*=120
Click here to see a video of the hydrogen radical mass fraction as a function of x3 at time t*=120
Click here to see a video of the temperature as a function of x1 at time t*=120
Click here to see a video of the normalized vorticity magnitude as function of time
Click here to see a video of the normalized x3 vorticity component as a function of time
Click here to see a video of the water mass fraction as a function of time
Click here to see a video of the hydrogen mass fraction as a function of time
Click here to see a video of the hydrogen radical mass fraction as a function of time
Click here to see a video of the oxygen radical mass fraction as a function of time
Click here to see a video of the mixture kinematic viscosity as a function of time
Click here to see a video of the mixture thermal conductivity as function of time


Click the image to view the .avi movie
The direct numerical simulation above is for a temporally developing, hydrogen-oxygen shear layer flame at a pressure of 100 atm and a Reynolds number of 2,500. A detailed and pressure dependent reaction mechanism having 8 species and 19 steps was used, along with a real gas state equation, generalized heat and mass diffusion models, and real property models. The simulation was run on 2,016 processing cores on Clemson's Palmetto cluster. It required 560 x 720 x 336 grid points (~135 million) and took approximately 500,000 processor hours of run time. The video shows the flame temperature as a function of time along the centerline plane.
Click here to see a video of the temperature as a function of time in the streamwise y-z plane
Click here to see a video of the H2 mass fraction as a function of time
Click here to see a video of the O2 mass fraction as a function of time
Click here to see a video of the H2O mass fraction as a function of time
Click here to see a video of the H radical mass fraction as a function of time
Click here to see a video of the spanwise vorticity magnitude as a function of time


Click the image to view the .avi movie
The direct numerical simulation above is for a non-reacting, temporally developing, hydrogen-oxygen shear layer at a pressure of 100 atm and a Reynolds number of 2,000. The simulation was run on 512 processing cores on Clemson's Palmetto cluster using 384 x 384 x 232 grid points. The video shows the hydrogen mass fraction as a function of time along the centerline plane.
Click here to see a video of the O2 mass fraction
Click here to see a video of the mixture density
Click here to see a video of the spanwise vorticity component


Click the image to view the .avi movie
The direct numerical simulation above is for a reacting, temporally developing, hydrogen-air flame at a pressure of 35 atm and a Reynolds number of 2,500. The simulation was run on 2,016 processing cores on Clemson's Palmetto cluster using 560 x 700 x 336 grid points. The video shows the flame temperature as a function of time along the centerline plane.

RESEARCH INTERESTS:

Dr. Miller's research involves the large scale simulation and modeling of turbulent air-hydrocarbon mixing and reaction at both atmospheric pressure and supercritical pressures relevant to modern and forthcoming gas turbines and diesel engines. The research is directed at fundamental studies of single-phase and multiphase flows involving complex physics in relatively simplified geometries such as homogeneous turbulence, mixing layers and jets. The preferred research approach is computational fluid dynamics (CFD); in particular, the direct numerical simulation technique. Results obtained from the simulations are used both to study the flow physics and to aid in the development and testing of mathematical and stochastic models relevant to future engineering applications. Models used in large eddy simulations (LES) and probability density function (PDF) methods are of particular interest.

Direct numerical simulation, or `DNS', is an advanced computational technique in which all of the length and time scales of the governing equations are completely resolved, without resort to modeling. When conducted with high order accurate numerical algorithms, DNS therefore yields very nearly the exact solution to the problem. In this case, the results contain all of the information describing a flow at nearly all points in both time and space. This is incontrast to experiments in which information is known only where data gathering probes are installed in the flow. However, due to computational limitations DNS is a valid approach only for relatively moderate to low Reynolds number turbulence for which the range of length and time scales remains resolvable by available computer resources.

NUMERICAL APPROACH:

As mentioned above, DNS is performed using highly accurate computational algorithms in order to achieve desired levels of resolution and accuracy. The approach presently employed in Dr. Miller's research program is that of high accuracy finite difference schemes due to their relative ease of use on parallel computing architectures (as opposed to spectral methods or finite element algorithms). In particular, we employ several of the schemes detailed by Kennedy and Carpenter (1994) including both third and fourth order accurate Runge-Kutta time integration schemes, eighth order accurate explicit central finite differences, and fourth order accurate compact (tridiagonal) finite differences. For multiphase flow simulations a fourth order accurate Lagrange interpolation procedure is used to determine gas phase flow variables at droplet or particle locations.

COMPUTATIONAL RESOURCES:

All computational codes presently employed by Dr. Miller's research group are conducted on parallel processing architectures. All parellelization is based on the Message Passing Interface (MPI) subroutines based on full three dimensional domain decomposition. Partial computational support has been provided by the following:

"Kraken" at the National Institute for Computational Science at Oak Ridge National Laboratory and the University of Tennessee: >112,000 processing cores.

Clemson University's "Palmetto" cluster with 12,000+ processing cores.

The National Science Foundation through the National Computational Science Alliance (NCSA) under Grant CTS990040N utilizing the NCSA SGI/Origin 2000 (1024 processors).

The California Institute of Technology's Center for Advanced Computing Research (CACR) utilizing the Hewlett-Packard V2500 (128 processors).

Clemon University's Division of Computing and Information Technology utilizing a SUN Microsystems HPC 6000 (16 processors).

The Department of Mechanical Engineering's beowulf cluster of Pentium based PCs (48 processors).

The support of the National Science Foundation through the Faculty Early Career Development (CAREER) young investigator award, Grant CTS-9983762, and NSF Grant CBET-0965624 are gratefully acknowledged.

REFERENCES:

C.A. Kennedy and M.H. Carpenter, `Several New Numerical Methods for Compressible Shear-Layer Simulations,' Applied Numerical Mathematics, 14, 397-433, 1994.