07.02.2013 Views

Optimization and Computational Fluid Dynamics - Department of ...

Optimization and Computational Fluid Dynamics - Department of ...

Optimization and Computational Fluid Dynamics - Department of ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

2 A Few Illustrative Examples <strong>of</strong> CFD-based <strong>Optimization</strong> 41<br />

In previous works [43, 44, 74], the optimization problem was simple (single<br />

objective) <strong>and</strong> the Simplex method [28, 63] has been used to speed-up the<br />

computational procedure. Here a multi-objective problem is considered <strong>and</strong><br />

an EA is employed to investigate the optimization domain for the retained<br />

input parameters. For the present problem, one evaluation is extremely costly<br />

in CPU time.<br />

One possibility to speed-up the evaluation is to perform every numerical<br />

simulation on parallel computers [42] or to use simplified methods to describe<br />

the chemical processes (e.g., the flamelet or tabulated chemistry approach<br />

[11, 25, 26, 27, 41, 52, 65]). In the presented application, the Navier-Stokes<br />

equations are solved for a two-dimensional flow. In the case <strong>of</strong> detailed chemistry,<br />

29 species conservation equations are solved additionally. Tabulated<br />

chemistry involves only three supplementary transport equations. In UGC + ,<br />

both the detailed reaction mechanism <strong>and</strong> the tabulated chemistry (FPI, for<br />

Flame-Prolongation <strong>of</strong> ILDM) are implemented <strong>and</strong> available for the computations<br />

[41]. In the first case, all chemical parameters are computed using<br />

the st<strong>and</strong>ard s<strong>of</strong>tware CHEMKIN [45] while FPI computations rely on tabulated<br />

values. The computation using detailed chemistry can be performed<br />

either with simple molecular transport using the unity Lewis number assumption<br />

or using detailed transport modeling [39]. The presented investigation is<br />

based on both detailed chemistry <strong>and</strong> detailed transport computation. This<br />

high level <strong>of</strong> physical accuracy is important to reduce the modeling uncertainty<br />

associated with the CFD solutions, especially for cases with minor<br />

differences in the objective functions. However, the stability <strong>and</strong> the speed<br />

<strong>of</strong> the evaluations are greatly enhanced by starting the simulation using the<br />

simplified method (FPI, see [41]) during 30 time-steps. The corresponding<br />

results are converted into an initial, detailed chemistry solution based on the<br />

tabulated values. The detailed chemistry computation is then continued for<br />

another 40 time-steps, leading to the “final” result <strong>of</strong> the evaluation, used to<br />

evaluate the objective functions after the necessary post-processing. Unfortunately,<br />

the detailed chemistry computations are extreme stiff. Therefore, in<br />

many cases, the residual values will still be very high after this fixed number<br />

<strong>of</strong> iterations, indicating convergence problems. Such simulations are marked<br />

as non-feasible results <strong>and</strong> dismissed from the optimization. “Non-feasible”<br />

means also that either no converged solution can be found or that the CPU<br />

time required until convergence would be unacceptably long.<br />

All CFD evaluations rely on adaptive time-steps in order to stabilize the<br />

solution procedure. A starting value <strong>of</strong> 0.1 s has been always used here for<br />

the restart <strong>of</strong> the detailed computation from the tabulated chemistry results.<br />

Further investigations are needed to check if other time-step values would<br />

improve the number <strong>of</strong> valid evaluations. Although the detailed computations<br />

considerably decrease the number <strong>of</strong> feasible solutions, they deliver also more<br />

realistic <strong>and</strong> accurate results.<br />

The smallest grid spacing employed in the present computations is 62.5 μm.<br />

This resolution is needed for very stiff intermediate radicals like HCO <strong>and</strong>

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!