In spite of its fundamental technological importance, our knowledge of basic combustion processes is surprisingly incomplete. Theoretical combustion science is unable to address the complexity of realistic flames, and laboratory measurements are difficult to interpret and often limited in the types of applicable flames or levels of detail they can provide.

The Center for Computational Sciences and Engineering (CCSE) in Berkeley Lab's Computational Research Division has teamed with the Environmental Energy Technologies Division (EETD) to build a high-performance computing solution to flame simulation and analysis, with a unique potential to make dramatic progress in combustion science research.

The CCSE group has created the first detailed simulations of laboratory-scale turbulent premixed flame experiments, using 3-D, time-dependent software with chemical fidelity and fluid transport at an unprecedented level of detail. Researchers including Marc Day and John Bell of CCSE and Robert Cheng of EETD are working together to validate simulations with experimental data, then probe the computed results for information not easily obtainable from experiments in any other way.

The investigations are focused in two primary areas: how turbulence in the fuel stream affects the local combustion chemistry, and how emissions are formed and released in the product stream. The work has applications for devices including power generators, heating systems, water heaters, stove, ovens, and even clothes driers.

Simulation at such a level of detail was impossible just a few years ago. However, over the past five years algorithmic improvements by applied mathematics groups such as CCSE, funded by the Department of Energy, have slashed computational costs for these types of flow studies by a factor of 10,000. These savings enable key improvements in the fidelity of the chemical and fluid dynamical descriptions of the flows, to the point that real experiments may now be simulated without ad hoc engineering models for under-resolved physical processes.

Nevertheless, simulation of practical-scale combustion devices remains an immense undertaking. CCSE has implemented its advanced simulation algorithms on state-of-the-art parallel computing hardware to increase the number of variables available for describing the system, from hundreds of thousands of variable five years ago to more than a billion today.

The research approach taken by CCSE has explicitly targeted both the temporal and spatial multiscale aspects of combustion modeling. The group takes advantage of key mathematical characteristics of low-speed flows, common to most combustion applications, to eliminate components of the model relevant only to high-speed scenarios. For low-speed flows, these components have little effect on the system dynamics, yet they drive down the simulation efficiency by unnecessarily limiting the maximum numerical time step size.

The integration algorithms are implemented in a set of software tools based on adaptive mesh refinement (AMR), a dynamic grid-based system that automatically allocates computational resources to regions that contain the most interesting detail. The AMR methodology allows one to simultaneously incorporate large-scale effects that stabilize the flame, as well as the very fine-scale features of the combustion reaction zone itself.

The detailed solutions computed by CCSE are being validated with experimental data provided by the EETD Combustion Lab. Comparisons include global observables like mean flame locations and geometries as well as statistics of instantaneous flame-surface structures. In addition to simply validating the computed solutions, however, the research groups probe massive amounts of data generated by the computation in order to learn more about flame details, such as the localized effects of large and small eddies on the structure of the combustion reaction zone.

For example, the distribution of hydrogen atoms in the thermal field is tightly coupled to key chain-branching reactions required to sustain the combustion process itself. The detailed models accurately represent the transport of hydrogen with respect to the other chemical species in the context of this turbulent flow. CCSE is presently using detailed chemistry and transport models, containing 20 to 65 chemical species and hundreds of reactions.

Examples of investigations by the group include the simulation and analysis of a three-dimensional V-flame experiment in the EETD Combustion Laboratory. In this experiment a thin rod is placed across the exit of a circular nozzle issuing premixed methane-air fuel vertically into an unconfined region open to the lab. The rod stabilizes a robust V-shaped flame that is highly corrugated and time-dependent.

Rod-stabilized V-flames: in the laboratory (left), photographed at a relatively slow shutter speed, and from an animated simulation at a higher frame-rate (right)

The figure above shows a photograph of a laboratory V-flame and a representative calculation under identical conditions. Due to the finite shutter speed of the camera, the surface of the turbulent flame appears smeared out over a "flame brush thickness." Simulation results are depicted on the right as a volume visualization of the temperature gradient field, which provides a convenient marker for the flame surface. The simulation was based on 20 chemical species and 84 fundamental reactions. The computed surface exhibits large-scale wrinkling of the instantaneous flame surface, and when averaged over time, shows remarkable agreement with the laboratory photos, in terms of flame brush thickness, spreading and growth rates.

In the experiment, the instantaneous location of the flame may be visualized using particle image velocimetry (PIV), in which inert particles are distributed in the unburned gas with a uniform density. The upward-moving gas expands as it passes through the flame; the density of tracer particles shows a corresponding decrease.

The location of the abrupt change in particle density, as captured in the photograph below (left) indicates the instantaneous flame position in a vertical plane through the center of the flame-stabilizing rod. The photo may be compared to a representative planar slice of the simulated fuel concentration (right), since fuel is consumed at the flame front. The still images demonstrate exceptional agreement, both in terms of overall flame shape and brush growth characteristics, but also in terms of fine-scale wrinkling of the flame surface.

The flame's location can be modeled through particle image velocimetry (PIV). The photograph on the left closely matches the simulated density of methane particles on the right, created using an adaptive low-Mach model.

CCSE is now working with EETD researchers to develop additional statistical measures of both the simulation and experimental data so that they can obtain more detailed quantitative comparisons. Beyond experimental validation, the solutions will be used to track the fate of fuel particles as they wash through the flame and undergo a range of chemical reactions to contribute to pollutant or unburned hydrocarbon emissions, for this and other interesting experimental configurations.

Also, since processing and understanding the wealth of data generated by these simulations continues to raise open questions, CCSE is constantly building and improving innovative diagnostic techniques that will help shed new light on the complex behavior of turbulent reacting flows.

Additional information

• More about the Center for Computational Sciences and Engineering, with animated models of flames
• For more about the CCSE's adaptive methodology for low Mach-number combustion modeling applications, contact Marc Day or John Bell.