The multivariate analysis of failure data in a component event data bank using randomization and linear logistic regressionFunded under: JRC-REACTSAFE 5C
Component failure-events collected from NPPs in various countries are stored in the Component Event Data Bank (CEDB) of the European Reliability Data System (ERDS) developed by the JRC. The databank stores information on operational history and failure related data on various component families, identified by their engineering and operation characteristics. Databank data are sparse and "intelligent" selections of data sets for model building for reliability have to be made. This paper presents an approach different from "Cox regression". It fits linear models to the logit of the probability of survival given the time the component operated and other concomitant variables. Before logistic regression is fitted, the observed times to failure or suspension are randomly assigned to components and are treated as if they were prespecified and fixed. Responses (component alive or failed) are determined as they would have been observed at the times assigned by randomisation and are then regressed upon the randomised times and selected concomitant variables. Estimation and testing for the significance of variables in the model are done by simulation. The proposed method is applied on CEDB data on 42 extraction or feed booster pumps of the condensate and feedwater system. Application proves to be significant for the dependence of the logit on time. Reliability of the components in time is estimated for the two pump types.
Bibliographic Reference: Paper presented: 7th International Conference on Reliability and Maintainability, Brest (FR), June 18-22, 1990
Availability: Available from (1) as Paper EN 35313 ORA
Record Number: 199011566 / Last updated on: 1994-12-02
Original language: en
Available languages: en