Simulation-Based Parametric Reliability Modeling Using Covariate Theory

Reliability analysis methods used for preliminary safety assessments of complex systems typically assume a predetermined and invariant set of input variable statistical properties. However, during the product development phase of the system and especially during in service operation, the characteristics of the random variables can themselves be subject to variation. Thus, the resulting failure probability distribution can vary greatly from early predictions. The objective of this paper is to explore a technique used to create a general parametric failure probability distribution as a function of key variables. This technique is constructed around covariate theory which is the basis of the familiar Accelerated Life Testing and Proportional Hazards Modeling approaches. Where these approaches have traditionally been used with physical experiments, they are applied within this study to Monte Carlo simulation data generated using an available component modeling and simulation environment of a gas turbine airfoil limited by a single failure mode. Necessary modifications to the traditional from of the covariate approach are identified for application to controlled Monte Carlo simulation data. Implications to potential safety improvements early on in the product development phase are discussed.