Parametric method of observation results processing with regard to missed data

The matters of ensuring dependable and safe operation of NPP facilities is of significant relevance. That is due to the fact that the proportion of equipment at the end of assigned service life in the nuclear power industry is very high, thus dependability analysis of NPP elements and systems is required. In the process of dependability characteristics analysis a number of problems occur, i.e. evaluation of residual life of equipment, justification of life extension decisions. Also, it is required to provide spare parts for elements and systems, select maintenance strategies, etc. That increases the value of activities aimed at analyzing the dependability of nuclear power facilities and, subsequently, the requirement to develop the methods of analysis of statistical information on the operation of NPP elements, subsystems and systems for the purpose of identifying their performance parameters. At nuclear power plants, activities are organized to collect information on the operation of various facilities, i.e. failures and defects of system components, maintenance procedures, operating modes, storage conditions, etc. The information provided by the NPPs has a number of distinctive features. That is due to the following factors: presence of censorship of failure data, absence of sufficient service hours within the given observation interval and the limited volume of available data. All those factors cause an uncertainty in the resulting evaluations and, subsequently, lower that optimal accuracy on dependability characteristics calculation. In the process of evaluating the dependability of facilities in operation a certain part of facilities and systems often does not fail over the period of observation. In such situations statistical analysis of dependability is required that is based on the so-called right censored samples of which the distinctive feature consists in the fact that the inspected product does not fail within the period of observation. In some cases the operation times of specific facilities are unknown. For instance, at the initial stage of facility operation information on its performance was not collected, and the decision to collect data was taken later. In this case the required method must take into consideration the missing information that was not collected at the initial stage. The limited volume of information is due to the fact that the nuclear energy facilities fall into the category of highly dependable equipment. Failures are rare events. Therefore in order to increase the reliability of dependability indicators estimation all the available information must be used. Thus, taking into account all the available information enables more accurate results that can be used to calculate NPP facility service life. The purpose of this article is to show the application of the method of repeated sample and examine its efficiency. The main focus is on missed data that are to be recovered. The authors provide the results of evaluation of the exponential distribution law parameter subject to right censored and missed data. The suggested method of repeated sample is compared with the bootstrap method and mean substitution method. For evaluation of exponential distribution law parameter the authors suggest using the maximum likelihood method. Statistical characteristics calculation is provided. All the calculations and results are based on test cases.

1. Antonov АV. Sistemny analiz [System analysis]. Moscow: Vyshaya shkola; 2004. Russian.

2. Antonov AV, Nikulin MS. Statisticheskie modeli v teorii nadiozhnosti: Ouchebnoie posobie [Statistical models in the dependability theory: A study guide]. Moscow: Abris; 2012. Russian.

3. Zangiyeva IK. Reshenie problem nepolnykh dannykh massovykh oprosov [Solving the problem of incomplete data of mass survey]. Rossiyskaya sotsiologia zavtrashnego dnia [Russian social science tomorrow]. 2008: 84 – 95. Russian.

4. Zloba E, Iatskiv I. Statisticheskie metody vosstanovleniya propushhennykh dannykh [Statistical methods for recovery of missed data]. Computer Modeling & New Technologies. 2004; 6: 55 – 56. Russian.

5. Cox DR, Oakes D. Analiz dannykh tipa vremeni zhizni [Analysis of survival data]. Moscow: Financy i statistika; 1988. Russian.

6. Little RA, Rubin DB. Statisticheski analiz dannykh s propuskami [Statistical analysis with missed data]. Moscow: Financy i statistika; 1991. Russian.

7. Efron B. Netraditsionnie metody mnogomernogo statisticheskogo analiza [Unconventional methods of multivariate statistical analysis]. Moscow: Financy i statistika; 1988. Russian.

8. Bischl B, Mersmann O, Trautmann H. Resampling methods in model validation. Algorithm Engineering Report. 2010 Aug; 9: 14 – 31.

9. Meeker WQ, Escobar A. Statistical Methods for Reliability Data. New York: John Wiley & Sons, Inc.; 1998.