Optimization and Computational Fluid Dynamics - Department of ...
Optimization and Computational Fluid Dynamics - Department of ...
Optimization and Computational Fluid Dynamics - Department of ...
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
12 Dominique Thévenin<br />
three-dimensional flow geometry discretized using millions <strong>of</strong> elements <strong>and</strong><br />
involving complex physics, typical computing times will be days or weeks.<br />
Should we ab<strong>and</strong>on the hope <strong>of</strong> any optimization for such interesting problems?<br />
Not necessarily! Since the duration <strong>of</strong> the complete optimization process<br />
reads in principle “(number <strong>of</strong> evaluations)×(duration <strong>of</strong> a single CFD evaluation)”,<br />
there are at least two clear possibilities to still use CFD-O for such<br />
problems:<br />
1. the first one consists in speeding-up as much as possible the evaluation<br />
procedure;<br />
2. the second one consists in reducing as much as possible the number <strong>of</strong><br />
required evaluations.<br />
To speed-up each CFD-based evaluation, different solutions might again<br />
be employed <strong>and</strong> intelligently combined. It is in principle possible to use,<br />
separately or simultaneously, a physical, a mathematical or an algorithmic<br />
point <strong>of</strong> view:<br />
• From the physical point <strong>of</strong> view, speeding-up the evaluation means reducing<br />
the model complexity while keeping all (or most) <strong>of</strong> the needed<br />
coupling processes describing the important physics. This will for example<br />
be demonstrated in the last chapter <strong>of</strong> this book, where reduced equations<br />
will be partly used instead <strong>of</strong> the full Navier-Stokes equations, leading<br />
to a tremendous reduction in the needed computing time. It is also employed<br />
in the next chapter when considering optimization <strong>of</strong> turbulence<br />
model parameters: instead <strong>of</strong> solving the full multi-dimensional Reynolds-<br />
Averaged Navier-Stokes (RANS) equations to describe turbulent channel<br />
flows, a reduced model will be used instead. Model reduction is clearly<br />
a very efficient procedure <strong>and</strong> should be used every time this is possible.<br />
But it requires <strong>of</strong> course a clear underst<strong>and</strong>ing <strong>of</strong> all physical processes<br />
controlling the application considered!<br />
• Applied mathematics should also be considered to reduce as much as possible<br />
the needed computing time <strong>and</strong> computer memory needed by CFD.<br />
This means that the most efficient solution procedures should be employed.<br />
This can lead for example to multigrid acceleration, perhaps to solve the<br />
pressure/velocity coupling equation; or to the implementation <strong>of</strong> a highly<br />
complex, three-dimensional adaptive unstructured grid or Adaptive Mesh<br />
Refinement, in order to drastically reduce the needed number <strong>of</strong> discretization<br />
elements; or to an efficient pre-conditioning <strong>of</strong> the system. None <strong>of</strong><br />
these are specific to CFD-O. But clearly, an experienced CFD specialist<br />
has a better chance to succeed also in CFD-O!<br />
• Finally, the practical coding <strong>of</strong> the CFD evaluation should also be improved<br />
as much as possible. From an algorithmic point <strong>of</strong> view, this will<br />
mean in particular the adaptation <strong>of</strong> the code to the properties <strong>of</strong> the<br />
employed computer: perhaps a vector, more <strong>of</strong>ten today a parallel system.