Analysis of degradation process with measurement errors

Degradation tests are often applied on highly reliable products when the product performance can be repeatedly measured. In this paper, we compare two common types of degradation models - a nonlinear regression model and a stochastic process model. Particularly, we discuss the effects of measurement error on model parameter estimation and model selection. Using an example of photovoltaic product degradation, we (1) demonstrate the use of linear models for estimating model parameters, and (2) provide a hypothesis test for the statistical significance of product performance degradation. It shows that some forms of measurement errors, such as the drifting error of tester, can be easily incorporated into the analysis of stochastic degradation models.