|
|
 |
|
Ben Thacker with Southwest Research Institute |
|
Date: July 16, 2004
Time: 10:30am
Location: NIA, Rm 404
Speaker: Ben Thacker with Southwest Research Institute
Subject: "A Probabilistic Treatment of Expert Opinion"
Additional Information: Webcast
Many engineering applications are challenged by the lack of sufficient experimental data that are needed to produce accurate predictions. In an uncertainty analysis, this lack of available data confounds the selection of the probability density function (PDF) and introduces a level of arbitrariness in the ensuing risk analysis.
Quite often, the only available data are expert opinions, which are usually vague or conflicting. Several non-probabilistic non-deterministic methods have been proposed in the literature to perform a risk assessment under these conditions. In this presentation, we develop a methodology for incorporating expert opinion within a probabilistic framework using uncertain PDFs. This approach has the advantage of building from well-established and proven methods for performing probabilistic analysis.
The uncertainty on the PDF is characterized by treating the parameters in the PDF as random variables. Each expert opinion is considered to be a random sample from a parent distribution of expert opinions. The classical Bayesian updating scheme is extended to incorporate vague or imprecise interval (expert opinion) information. Conflict between experts is accounted for through the likelihood function and the updating process. The uncertain PDFs can be used in both simulation-based and MPP-based reliability methods.
The probabilistic framework presented establishes a direct link between the amount and quality of the available data and the computed reliability. This link allows the decision maker to weigh the expected value of additional data collection efforts against the expected improvement in reliability. This makes the proposed approach highly suitable for assessing alternatives early in the design process.
|
|
|
Home