Transferability of Reliability Data

In the strict statistical sense reliability data may only be used if the component to be assessed belongs to the population, whose observation served to determine the data. This is only true if plant-specific reliability data is used. Yet, this requirement is, as a rule, sufficiently satisfied if the reliability data does not stem from the plant under investigation, but 

In deciding on this matter it has to be borne in mind that it cannot be demonstrated that all features required to characterize the population have been recorded. 

If the above situations do not apply case-by-case decisions have to be taken. For example, if there are differences in the structure between the component to be assessed and the observed one often recourse can be had to the individual subcomponents (piece parts of components, e.g. the component pump consists of motor, power transmission, the pump itself, local controls and connected wiring). They enable one to compose reliability data for components made up of different numbers of sub-components or of sub-components with a different design and 

If there are differences in the working conditions between the component to be assessed and the one which has been observed it has to be examined to what extent these working conditions are encountered in any of the observed plants. An analysis of dependencies is helpful in this context. For example, generally reliability data of components exposed to aggressive media are worse than those to be expected in case of exposure to water. They then may serve as conservative estimates. The same is tme for cases of strong mechanical loads like, for instance, strong vibrations. Values obtained for such cases may also serve as conservative assessments for components under normal loads. 

In probabilistic analyses the influence of uncertainties of input data on the result is usually accounted for. In order to do this the reliability of components is represented by probabihty distributions instead of point values (usually expected values). Often log-normal distributions, which were presented in the previous Section, are used. The factor K95 then is a measure for the uncertainty. In general K95 lies between 2 and 10. 

---