STRUCTURAL RELIABILITY. THE THEORY AND PRACTICE
A higher economic efficiency of equipment dependability testing can be achieved by means of reduced testing time or smaller quantity of test samples. The reduction of testing time increases sample censoring, while a lower quantity of test samples decreases the volume of equipment operating times sample. Test parameters may be reduced only if information processing methods ensure the validity of the calculated dependability indicators.
As the result of the tests, small censored samples of equipment mean times to failure are generated. Dependability calculation using such samples is performed through the maximum likelihood method. The article presents the findings of experimental studies of precision of maximum likelihood parameter estimation of exponential law over small singly right censored samples. The studies were performed by means of computer simulation of censored samples similar to those generated as the result of equipment dependability testing. These experimental data show that most maximum likelihood estimates obtained over small singly right censored samples have significant deviations from true values.
This paper features regression models establishing dependence between deviation of maximum likelihood estimates from true values and parameters defining the sample structure. They allow calculating and introducing corrections to maximum likelihood estimates. Experimental studies of their application efficiency were conducted. The accuracy of maximum likelihood estimates significantly increased upon application of the developed models and correction of maximum likelihood estimates. Software for application of regression models was developed.