Statistical Framework for Sensitivity Analysis of Structural Dynamic Characteristics

The uncertainty involved in the structural parameters inevitably leads to uncertainty in predicting the resulting structural dynamic characteristics, and this relationship is important to quantify. There is a large volume of work dealing with quantifying the overall uncertainty propagated from parameters to the structural dynamic responses, whereas little work has been done in measuring the contributions of the individual parameters and groups of parameters to the overall uncertainty, which is corresponding to the variance-based global sensitivity analysis (GSA). The variance-based GSA allows for providing a robust assessment of the relative influences of individual parameters on the structural dynamic characteristics. Although it is a powerful tool, the variance-based GSA suffers from the limitation of high computational cost, especially when applied to the expensive-to-run complex systems such as the long-span cable-stayed bridge in this study. To alleviate the computational burden, a fast-running nonparametric Gaussian process model (GPM), a fully specified statistical model, is used as a surrogate model. The highlights of the developed metamodel-based approach for the variance-based GSA are: (1) adopting the full GPM, rather than the mean of GPM, where the latter drops the valuable uncertainty information of the prediction variance; (2) enabling the full GPM-based method not to be restricted to the calculation of the first-order sensitivity index, and to be applicable for the computation of sensitivity indices of groups of parameters; (3) generalizing this approach to be suitable for the cases with arbitrarily distributed parameter uncertainty. Then, this full GPM-based approach is applied for sensitivity analysis of structural dynamic characteristics of a long-span cable-stayed bridge.

Language

  • English

Media Info

Subject/Index Terms

Filing Info

  • Accession Number: 01639537
  • Record Type: Publication
  • Files: TRIS, ASCE
  • Created Date: Jun 21 2017 3:01PM