Accuracy of three-dimensional, thermomechanical model computations in rock salt (reliability of predicted data). Final report

Based on the ''response surface approach'', an effective stochastic method has been developed minimizing the computation effort with the ANSALT software system by applying methods of optimal test planning. The following results can be reported: the sensitivity of the predicted results to changes in input parameters can be expressed by the computable quantity of ''sensitivity''. The sensitivity coefficient for characteristic event quantities proves to be a suitable measure for discriminating relevant input parameters from less relevant parameters. If the input parameters are stochastic variables with a given density function, the probability density of the system reponse can also be computed. Further, the regression function allows an assessment to be given of the system behaviour in response to any input parameter combination. The accuracy of the reported method mainly depends on the degree of accuracy achieved by the regression analytical model in reflecting the system response, with the modelling of interactions and of non-linearities being of essential significance. (orig./HP)