EFFICIENT BAYESIAN PARAMETER IDENTIFICATION

Abstract eng:
Many important parameters influencing structural behaviour involve unacceptable uncertainties. An extensive development of efficient methods for stochastic modelling enabled reducing these uncertainties in input parameters. According to Bayes’ rule, we obtain a more accurate description of the uncertain parameter involving an expert knowledge as well as experimental data. The aim of this contribution is to demonstrate two techniques for making the identification process more efficient and less time consuming. The first technique consists in replacement of the full numerical model by its polynomial approximation in order to reduce the computational effort. The particular approximation is based on polynomial chaos expansion constructed by linear regression based on Latin Hypercube Sampling. The obtained surrogate model is then used within Markov chain Monte Carlo sampling so as to update the uncertainty in the model inputs based on the experimental data. The second technique concerns a guided choice of the most informative experimental observation. Particularly, we apply sensitivity analysis to determine the most sensitive component of the structural response to the identified parameter. The advantages of the presented approach are demonstrated on a simple illustrative example of a frame structure.

• Export as: BibTeX | MARC | MARCXML | DC | EndNote | NLM | RefWorks
• Add to your basket