Life prediction - The need for a mechanistically based probability approach

Abstract

Concerns with life assessment and management for existing engineered systems and aging infrastructure worldwide have increased the emphasis on the development of methods for reliability and durability predictions. Increasingly it is being recognized that the traditional statistically and empirically based methods are inadequate. These methods are appropriate for interpolations, but their usefulness for extrapolation is limited. Effective predictors, i.e., those that provide precise estimates beyond the range of conditions employed in the development of supporting data and assessments of risk, must be based upon mechanistic models that capture the functional dependence on all the key external and internal variables. This type of modeling, to reflect typical industrial applications, requires multidisciplinary research that addresses the chemical and micromechanical processes that control damage evolution in materials and quantifies the stochastic aspects of these processes. The purpose of this paper is threefold. First, the differences between mechanistically based probability modeling and statistical modeling are demonstrated. Second, the use and applicability of the mechanistically based probability methodology is demonstrated. Third, a framework from which other applications can be addressed is provided. Specifically, the approach is illustrated through examples from creep crack growth in high strength steels and corrosion and corrosion fatigue of aluminum alloys. © 2001 Trans Tech Publications.