Sensitivity of value of information to model and measurement errors

Mohammad Shihabuddin Khan, Siddhartha Ghosh, Colin Caprani, Jayadipta Ghosh

Research output: Contribution to journalArticleResearchpeer-review

Abstract

The value of information (VoI) framework, based on Bayesian preposterior analysis, can be used to estimate the most likely benefit associated with a particular structural health monitoring (SHM) strategy. The errors within the VoI framework can be traced to the underlying predictive models and the inspection instruments. Conventional VoI analysis assumes a nonerroneous predictive model. Also, it considers only the (unbiased) random errors associated with inspection instruments. In this paper, the authors propose a VoI framework that explicitly considers the different uncertain errors within the predictive models and inspection instruments. Global sensitivity analysis and parametric investigations are performed to study the sensitivity of the VoI framework to various error parameters by estimating Sobol' indices through Monte Carlo simulations and polynomial chaos expansions. It is found that the VoI framework is highly sensitive to the errors within the predictive model. This study recommends that any VoI analysis should be preceded with a thorough quantification of the errors within the predictive models lest an inaccurate estimate of the VoI is obtained.

Original languageEnglish
Article number04020038
Number of pages13
JournalASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, Part A: Civil Engineering
Volume6
Issue number4
DOIs
Publication statusPublished - Dec 2020

Keywords

  • Measurement error
  • Model error
  • Sensitivity analysis
  • Sobol' indices
  • Structural health monitoring
  • Surrogate modeling
  • Value of information

Cite this