New methods for pre- and post-processing of stochastic simulations

Abstract

The aim of numerical simulation is a reliable prediction of real system's behaviour, which is influenced by numerous uncertainties. In this work, a modelling process is developed for spatially localised modelling of uncertainties by random fields. The developed differential scale-space representation allows the derivation of a stochastic model that can be used to generate synthetic realisations based on a given sample. The post-processing of such stochastic simulation results is usually performed by means of statistic methods. However, this way only values of selected points in space and time can be evaluated and the associativity to the geometric model is lost. To overcome these limitations, a new concept to visualise geometric scatter and dependent statistic measures from stochastic simulation results is developed. The visualisation of the expected value geometry with statistic measures of dependent variables as well as surrounding probability envelopes by superimposed transparent geometry constitutes a holistic approach to evaluate stochastic simulation results.

Advances in Validation of Computational Mechanics Models

“…Researchers, again from the innovative aerospace field, present a holistic approach to evaluate such a family of models resulting from stochastic simulations. 4 Stochastic simulations may be seen as a step in the process of model tuning or model updating with a view of increasing the matching quality of the model and the experiment. The implicit question of whether a model tuning by adapting free parameter values to get a better fit with the experimental data is a recommendable procedure is tackled in a paper by Charlotte Werndl, 5 a researcher with a background in philosophy.…”

“…Researchers, again from the innovative aerospace field, present a holistic approach to evaluate such a family of models resulting from stochastic simulations. 4…”