Information Technology Reference
In-Depth Information
Analysis of Bulky Crash Simulation Results:
Deterministic and Stochastic Aspects
Tanja Clees, Igor Nikitin, Lialia Nikitina, and Clemens-August Thole
Fraunhofer Institute for Algorithms and Scientific Computing, Schloss Birlinghoven
53754 Sankt Augustin, Germany
{Tanja.Clees,Igor.Nikitin,Lialia.Nikitina,
Clemens-August.Thole}@scai.fraunhofer.de
Abstract.
Crash simulation results show both deterministic and stochastic
behavior. For optimization in automotive design it is very important to
distinguish between effects caused by variation of simulation parameters and
effects triggered, for example, by buckling phenomena. We propose novel
methods for the exploration of a simulation database featuring non-linear
multidimensional interpolation, tolerance prediction, sensitivity analysis, robust
multiobjective optimization as well as reliability and causal analysis. The
methods are highly optimized for handling bulky data produced by modern
crash simulators. The efficiency of these methods is demonstrated for
industrially relevant benchmark cases.
Keywords:
Surrogate Models, Bulky Data, Multiobjective Optimization,
Stochastic Analysis.
1
Introduction
Simulation is an integral component of virtual product development today. The task
of simulation consists mainly in solution of physical models in the form of ordinary or
partial differential equations. From the viewpoint of product development the real
purpose is product optimization, and the simulation is "only" means for the purpose.
Optimization means searching for the best possible product with respect to multiple
objectives (multiobjective optimization), e.g. total weight, fuel consumption and
production costs, while the simulation provides an evaluation of objectives for a
particular sample of a virtual product.
The optimization process usually requires a number of simulation runs, the results
form a simulation dataset. To keep simulation time as short as possible, "Design of
Experiments" (DoE, [1]) is applied, where a space of design variables is sampled by a
limited number of simulations. On the basis of these samples, a surrogate model is
constructed, e.g. a response surface [2], which describes the dependence between
design variables and design objectives. Modern surrogate models [3, 4, 12-15]
describe not only the value of a design objective but also its tolerance limits, which
allow to control precision of the result. Moreover, not only scalar design objectives
but whole simulation results, even highly resolved in space/time, (”bulky” dataset)
can be modeled [12-15].
Search WWH ::
Custom Search