Article 8217

Title of the article



Mikhaylov Viktor Sergeevich, lead engineer, Central Research Institute of Chemistry and Mechanics named after D. I. Mendeleev (115487, 16а Nagatinskaya street, Moscow, Russia),

Index UDK





In the modern production of highly reliable, unique, complex products, the usual situation in which it is necessary to obtain an estimate of the probability of failure-free operation (PFO) based on tests that did not give failures has become. In this case, a test plan of type NBT is usually used, where N is the number of subjects of the same type; T – operating time (the same for each product); В – the characteristic of the plan, meaning that the working capacity of the product after each failure during the test period is restored. If the operating time is obeyed to the exponential distribution law with the parameter T0 (mean time to failure), the calculated value of the PFO for a given time (t) will be determined by the equality Р0(t,T0) = exp(–t/T0). The purpose of the article is to find a tool that will objectively judge (from the point of view of the effectiveness of the proposed evaluations) the proposed possibilities to estimate the PFO value different from zero and one by the results of tests that did not fail.
. To find an effective estimate, we used integral characteristics, namely, the total square of the deviation of the expected realization of a certain valuation variant from all possible values of P0(t,T0) for different failure flows of the set of products tested.
Results and conclusions
. For all variants of the NBT test plan, which do not allow the use of the P (t) estimate (due to the restriction t / NT> 0.99999999), an estimate of F(t) should be used to estimate PFO. When it becomes necessary to characterize PFO by a value different from zero and one by the results of non-failure tests, then the estimate F(t) in this case is an effective estimate. However, for highly reliable modern products that have been subjected to trouble-free testing, an PFO estimate of P (t) equal to one (1 = 0.999 ... 9 ...) is preferable to F(t), because Is closer to the true value than the result of the estimate of F (t).

Key words

mean operating time to failure, probability of failure-free operation, exponential distribution, test plan, point estimates

Download PDF


Дата создания: 08.06.2017 15:35
Дата обновления: 09.06.2017 09:57