Reliability prediction through degradation data modeling using a quasi-likelihood approach

Reliability estimation in new product development is essential, particularly when the design concept itself is new. The common way in which these estimations are carried out is through the use of life tests. These tests may not yield failure observations among the products being tested. In particular, due to the strong pressure on "Time to Market" in many design processes, time to perform these tests is simply not available. Therefore a recent alternative technique has been to look at degradation data of the performance characteristic in question. In such an approach, the degradation data is modeled and this model is used to predict the reliability of the design or its statistics. Very often the performance characteristic distribution, at any given time instant, is non-normal and the data across time are correlated. This deters the use of the usual likelihood ideas in modeling the data. However, the method of Generalized Estimating Equations (GEE) based on the quasi-likelihood approach becomes a useful technique for this purpose. This paper attempts to demonstrate an application of the GEE modeling approach on a typical set of degradation data whose marginal distribution is Poisson. The model is used to predict the characteristics of the Poisson distribution, at any time interval, from which the reliability is estimated. The estimates of the confidence bands for the reliability are obtained through the use of multivariate Monte Carlo Simulation.