A Comparison of Uncertainty Quantification Methods on Benchmark Problems for for Space Deployable Structures

摘要

The efficacy of several uncertainty quantification methods are compared on two deploy-able structures benchmark problems. These benchmarks include typical pathologies found in large deployable structures, including contact with flexible components under low rate conditions and geometrically nonlinear displacements. The performance of the Sierra Solid Mechanics computational platform on these benchmarks was previously investigated. This paper continues this exploration by assessing competing methods for sensitivity analysis and uncertainty quantification on these benchmarks. A particular objective was to consider the mixture of both epistemic variabilities (such as machine tolerances or design uncertainties) and aleatory variabilities (such as calibrated material constants) using the DAKOTA uncertainty quantification platform. The results demonstrate that these benchmarks can exhibit very high off-nominal reaction forces due to the parameter variations. Moreover, it was found that the treatment of the epistemic variables as uniform random variables, as is typically done in Monte Carlo analysis, may lead to significantly different margin estimates than a hybrid epistemic-aleatory analysis.