OPTIMAL DESIGN OF ACCELERATED DESTRUCTIVE DEGRADATION TESTS IN PRESENCE OF COMPETING RISKS

Accurate prediction and evaluation of products reliability is one of the concerns of the companies for technical and managerial decisions. In this regard, failure data of products are used for reliability estimation. However, for products with high reliability, these data require a very long time and a large number of test units that is usually costly and impossible. Therefore, optimal usage of resources such as test units, time and facilities is necessary. Accelerated reliability tests are usually used for gathering failure data of products. However, for products with high reliability no failure or few failures may happen even by accelerated methods. So, evaluation of the products reliability just with using traditional life tests is difficult because it only records the time of failures. Therefore, measuring the degradation process have been proposed as an alternative to failure time data to obtain fast and convenient data in manufacturing industries. Accelerated Degradation Tests (ADTs) are useful tools in evaluating lifetime of high reliability products. Many products degrade due to multiple failure mechanisms. In other words, multiple failure mechanisms (competing risks) compete to put an end to the life of a product. In some applications, the process of degradation measurement destructs physical and chemical characteristics of test units. This kind of accelerated test is called Accelerated Destructive Degradation Test (ADDT). In applying these tests resource and estimation issues should be taken into consideration. Sample size, sequence of destructive measurements, number of measurements in each stress level, and other decision variables are important in effectively designing and implementing an ADDT. Therefore, in this paper, the optimal design of an ADDT in presence of competing risks is studied. In this regard, asymptotic variance of the failure time quantile is minimized. A sensitivity analysis on the results is reported to evaluate the robustness of the results to model parameters.