열화시험의 모형화 및 비용효율적인 실험계획

Abstract

In reliability analysis, degradation testing is an enchanting approach to obtain reliability information efficiently and effectively. The need for degradation testing is growing rapidly in recent years because degradation data can provide more reliability information and improve the product reliability inference greatly compare to regular failure-time data analysis.

But in order to come into its own as an "enchanting" approach, it has to satisfy three fundamental requirements: (1) degradation model which supports failure mechanism or suitable for degradation process. (2) appropriate statistical analysis method for degradation model. (3) degradation test plan having "good" estimation precision of the reliability measure under certain constraints.

Today, however, the newly developed products with complicated degradation mechanisms and fast-changing global marketplace make new issues about the three requirements, which require us to improve existing methods for degradation modeling, analysis, and test plan.

There has been extensive research on the analysis of degradation data and optimal degradation test plan. Most proposed approaches are based on simple degradation models, e.g., a (log)linear model. However, when complicated degradation models are required, existing studies may not be applied to the analysis and test plan for the degradation models. Furthermore, existing test plans generally considered the framework maximizing the precision of the reliability measure under limited total budget. Tight constraint of total budget may result in impractical results.

In this thesis, we introduce a nonlinear random-coefficients model for degradation testing data of newly developed devices, e.g., organic light-emitting diodes (OLEDs). The nonlinear model with random-coefficients can not only be a powerful tool to improve reliability estimation, but it can also support a framework to model the complicated degradation mechanisms. It can be shown that the framework minimizing the experimental cost under certain precision constraints is a reasonable assumption as long as some appropriate precision measures are available. We focus on the analysis of nonlinear degradation paths, along with cost-effective accelerated degradation test (ADT) design problem. The original contribution of this research as follows:

First, a bi-exponential model, which describes a two-compartment kinetic, with random-coefficients is introduced to characterize nonlinear degradation paths by complicated degradation mechanisms of the new devices. To estimate the parameters in the nonlinear random-coefficients model, several approximation methods are used. From practical examples, degradation mechanisms are investigated and validation of the nonlinear random-coefficients model is provided. Analysis result indicates that the reliability estimation can be significantly improved by using the nonlinear random-coefficients model.

Second, a new framework for degradation test plan is proposed. While most existing works maximize the precision of the reliability measure under the total budget constraints, the proposed framework minimizes the total experimental cost under the reasonable and intuitive precision constraints. The proposed framework explicitly considers the time-to-market requirement in the cost function, as well as the trade-off between termination time and sample size.

Finally, the established framework extended into an equivalent cost-effective accelerated degradation test plan to provide more flexibility to experimenters. Using a well-known constant-stress ADT problem as a target plan, the proposed equivalent degradation plan has a potential in reducing both the total cost and the termination time while maintaining the same precision of the reliability measure at use condition and the same number of total testing units.