Accelerated Life Testing : Analysis and Optimization

This paper describes a methodology to conduct a sequential test defined by an optimal accelerated testing plan. This test plan is based on an economic approach defined in previous work, a prior knowledge on reliability parameters (choice of reliability function, scale and shape parameters …) and acceleration model (choice of model, model parameters …) to evaluate the proportions of failure at each accelerated level. When conducting a test, it is possible to verify the compatibility of results with prior knowledge from a consistency criterion that measures the compatibility between prior distribution and likelihood provided by the data. We can also reduce the censoring time in case of "Good results" while keeping the same level of risk. This possibility is authorized because the robust test is longer than basic optimal test plan. In the process of product development, we will use the qualification. Qualification is an application-specific process involving the evaluation of the product with respect to its quality and reliability. The purpose of qualification is to define the acceptable range of variability for all critical product parameters affected by design and manufacturing. The methodology will be illustrated by a numerical example. An Accelerated Life Test (ALT) is the process of determining the reliability of a product in a short period of time by accelerating the use environment. ALTs are also good for finding dominant failure mechanisms. Thus, an accelerated Life Test (ALT) is a test method which subjects test units to higher than use stress levels in order to compress the time to failure of the units. Conducting a Quantitative Accelerated Life Test (QALT) requires the determination or development of an appropriate life-stress relationship model. In the case of robust optimization of an accelerated testing plan, a plan is provided with a sample size, stress levels and censoring time. This plan provides a robust estimate of the probability of failure during the warranty period (reliability metric is selected to estimate the operation cost of the test plan), the robustness to guaranteeing the best cost with a given risk  integrating the uncertainties on input data (reliability, activation energy …). To define the test plan, in some approaches, we consider an objective function based on economic approach, Bayesian inference for optimizing the test plan. The prior knowledge is based on a feedback from expert's opinion, Field data analysis on old product, Reliability Standard... This information contains the uncertainty on the real reliability of the new product tested. During the test, the observation of data makes possible to verify the consistency of these points with the assumptions of prior information having fixed the test plan. The establishment of batch test to launch the tests needs to fix some parameters of the plan as the number of units tested, the level of each stress, and the number of units tested by stress level. However, other parameters can be modified. In this paper, an optimization test plan is proposed integrating the Bayesian inference and an objective function based on economical formulation. The proposed method consists of 3 subsequent steps: