Research Topic

Flow Optimization under Uncertainties

In simulations usually all parameters are assumed to be exact. But as a matter of fact, in physical problems often not all parameters are explicitly known precisely. Reasons for such uncertainties can be e.g. natural fluctuations, fabrication tolerances or the lack of knowledge.

In safety-relevant applications it may be necessary to take into account these uncertainties in order to minimize the chance of system failure.
For this purpose uncertain input parameter like geometric dimensions, material parameters, forces or boundary conditions, have to be modeled as random parameters.

The resulting advanced optimization problem requires significantly more computing power than the already computationally intensive flow optimization. The aim of this work is to research different possibilities for the reduction of the calculation time.

  • Multilevel methods
    Gains in importance, especially in the area of (flow)optimization; Combined with uncertainty quantification there even are two possible approaches to apply this method: The random parameter space and the flow region.
  • Parallelization
    Since,a collocation method for the uncertainty quantification is used an efficient adaptive scheme for a high-performance parallelization can be applied
  • Collection and usage of data
    Due to the huge number of function evaluations much information is collected that can be used for further calculations
  • Utilize similar start solutions
    By a skillful choice of the starting solution, the computation time can be significantly reduced

Key Research Area

Numerical optimization; Uncertainty quantification; Flow simulation; Parallelization


Christoph Thiem


Dolivostr. 15

D-64293 Darmstadt



+49 6151 16 - 24383


+49 6151 16 - 24404




thiem (at) gsc.tu...

 Print |  Impressum |  Sitemap |  Search |  Contact |  Privacy Policy
zum Seitenanfangzum Seitenanfang