Large Eddy Simulation

Jeremy Thornock & Ben Isaac

The images and movie in Figures 1 and 2 are taken from experiments run on a single, down-fired oxy-coal flame at different operating conditions. The flames are highly turbulent, with multiple scales and physics contributing the visible flame and ultimately the total heat flux including particle reactions, gas reactions, radiation, and turbulent fluid dynamics.

We observed in these experiments that the flame stand-off distance is strongly affected by the operating conditions of the boiler. Understanding how the flame behaves and the ultimate effect this has on the heat transfer within the boiler is critically important for obtaining maximum efficiency and material longevity.

Large eddy simulation (LES) offers a new paradigm for coal boiler simulation. LES captures a wide range of time and length scales directly on the CFD mesh, reducing the reliance on modeling. Surprisingly, many processes in the boiler are relatively slow, but still having an important effect on the heat flux. LES may capture directly many of these slow processes without relying on sub-grid modeling.

Reynolds Averaged Navier Stokes (RANS) methods as shown in Figure 3, have been only partially successful for coal simulations. The shortcoming of using RANS for boiler prediction is due to the lack of any explicit representation of important, time varying, scales seen in the experiments above.

A movie from Arches simulations of the Alstom Boiler Simulation Facility (BSF) is shown in Figure 4. The images show a volume rendering of the gas temperature. A slice through the domain shows the O₂ concentration.

Figure 1: Photos of a single, down-fired oxy-coal burner. (Exp. Credit: J. S. Zhang, U of Utah)	Figure 2: High-speed video closeup of a the coal flame. (Exp. Credit: T. Ring, U of Utah)
Figure 3: Fluent Simulation (industrial standard) 24 CPUH (1 Core) Eddy breakup Closure. Reynolds averaged simulation of a coal flame represents a typical industrial simulation.

LES to Exascale

The move to exascale computing will have an immediate impact on the physical accuracy of the simulation. Consider model spectra of the target application: a 350MW boiler. Figure 6 shows model spectra for the main flame envelope and the near nozzle regions. Two vertical lines show two proposed filter widths, or cut-off frequencies. The larger filter width (1cm) does not sufficiently capture enough of the spectrum in the near nozzle region, while the finer resolution cut-off captures roughly 80% of the energy in the nozzle region. A 1mm resolution would represent a resolution100 times finer than we have heretofore performed.

Figure 5: Turbulent flows can be characterized with the energy spectrum curve. Depending on the flow, the curve may deviate from the ideal shape, but still retain the basic concept of the energy cascade. With increased resolution afforded by the move towards exascale, our LES model resolves more of the energy cascade which increases the accuracy of the predictions.	Figure 6: Model spectra shown for the target application, a 350MW coal boiler. Cutoff frequencies for two resolutions are shown with the vertical dashed lines.

Proposed Work

• Developing and maintaining the LES algorithm and sub-grid closure
• Code translation into EDSL
• Orchestrate the incorporation of new physics (CQMOM, RMCRT, etc.)
• Ensure algorithmic scalability and computational efficiency
• Overall CFD verification