Identifying Mechanisms That Highly Accelerated Tests Miss

Extrapolating reliability from accelerated tests for technologies without field data always carries the risk that the accelerated tests do not show the mechanisms which dominate at operating conditions. In statistical terminology, such accelerated testing carries a risk of confounding. For linear models, there is theory which allows one to determine which models are confounded with others. This paper develops analogous theory for a simple kind of confounding model, evanescent processes, when kinetics is used as the basis of modeling accelerated testing. A heuristic for identifying simple evanescent processes that can give rise to disruptive alternatives (alternative models that reverse the decision which would be made based on modeling to date) is outlined. Then, we develop activity mapping, a tool for quantitatively identifying the parameter values of that evanescent process which can result in disruptive alternatives. Finally, we see how activity mapping can be used to identify experiments which help reveal such disruptive evanescent processes