Bayesian Approach to Quantifying Epistemic Uncertainty in a Processor Availability Model

摘要

It is often necessary to use a combination of testing and analysis to demonstrate compliance with quantitative performance requirements. Increasingly, the demonstration must establish that requirements are satisfied with a certain level of confidence. As an example, a processor intended for space applications may be required to have a certain reliability or availability (e.g., at least 99.9%) with high confidence (e.g., 95%). This paper examines the application of Bayesian statistics and Monte Carlo simulation to such situations, focusing on quantifying epistemic uncertainty associated with the quantity of available data, the validity of alternative models, probabilistically combining alternative models when appropriate, and estimating epistemic uncertainty associated with certain aspects of model completeness.