| Summary: | <p>This paper concerns the acceleration factor estimation at strenuous tests of products in case of unstable production process (different batches of the same products may be with different index of reliability).</p><p>Kartashov has developed techniques for determining the invariant functional from batch to batch, which convert accelerated test results to the normal mode. Its highlight is to conduct socalled preliminary tests of one sample of products, including pre-tests in the variable mode. The standard procedure for the preliminary tests (examination) is as follows: testing the n groups of products, with m elements in each group, begins in the normal mode. As soon as one of the products in the group fails, tests of remaining products start in the accelerated mode. In addition to tests in the variable mode, there are also tests conducted in the constantly normal mode. As a result of such tests of products from one batch, an acceleration factor of strenuous tests is determined for this type of products for any batch.</p><p>The described procedure has the following shortcomings:</p><p>• tests duration in the normal mode is long and, as a result, is very much time-consuming and cost-demandable;</p><p>• tests conducted in the variable mode to the failure of all the products have the same drawback.</p><p>This paper proposes a method for conducting the preliminary studies. It does not require testing in the constant mode and, additionally, allows tests duration in the variable mode to be restricted by the moment of the first failure in the r group. To analyze the results of such tests the Renyi type criterion of homogeneity of two samples is suggested. A method for calculating the distribution of its statistics for the finite sample sizes is developed and implemented in the software complex. The asymptotic distribution of the statistics is given. The estimate of the acceleration factor is provided by its minimization. It is shown that for real sample sizes of products only exact distribution should be used, as the asymptotic distribution approximates satisfactorily the exact distributions starting with n 500.</p>
|
|---|
