RELATIVE IMPORTANCE OF UNCERTAIN STRUCTURAL PARAMETERS

摘要

A novel procedure for the estimating the responsesensitivity to input parameters of a complex FEmodel is presented. The method is specifically di-rected toward problems involving high-dimensionalinput parameter spaces, as they are encountered dur-ing uncertainty analysis with complex, refined FE-models. In these cases one is commonly faced withthousands of uncertain parameters and traditionaltechniques, e.g. finite difference, are unfeasible. Incontrast, the presented method quickly filters out themost influential variables. Hence, the topic is nothow to compute the sensitivity but how to identifyall those parameters for which probable variationshave the biggest influence on the response.. This isachieved by generating a set of samples with directMCS, which are closely scattered around the point atwhich the sensitivities are sought. From these sam-ples, estimators of the sensitivities are synthesizedand the most important ones are refined with a finitedifference calculation. In this paper, the underlyingtheory as well as the resulting algorithm is presented.Practical software issues will be also addressed.