Augmenting system reliability analyses with observation priors
- Los Alamos National Laboratory
Occasionally, a system may fail a test without an obvious component being at fault. Instead, experts may know that at least one of a set of components has failed, but there is uncertainty about which members in the set were the actual failures. When no further information is available, this missing data may be imputed using standard data augmentation (DA). This process is already used in the current implementation of the JMP complex-system reliability modeling codes. In some cases when this situation arises, there may be some supplemental information about the nature of the failure that suggests which subset of components are more likely to have failed. the behavior of the system during the failure may make certain components more likely candidates, and lead the engineering experts to have certain prior beliefs about what occurred. In this case, it is still known that at least one of a set of components failed, but the experts have some idea that certain failure scenarios are more likely than others. This white paper addresses this situation by modifying the imputation process of data augmentation through the use of an observation prior. This prior is specific to particular observations, and a given outcome which is repeated several times could potentially have different observation priors associated with each occurrence.
- Research Organization:
- Los Alamos National Laboratory (LANL), Los Alamos, NM (United States)
- Sponsoring Organization:
- USDOE
- DOE Contract Number:
- AC52-06NA25396
- OSTI ID:
- 962294
- Report Number(s):
- LA-UR-09-01252; LA-UR-09-1252; TRN: US200919%%58
- Resource Relation:
- Conference: JMP TCG-XIV Meeting ; March 10, 2009 ; Albuquerque, NM
- Country of Publication:
- United States
- Language:
- English
Similar Records
Development and Validation of a Lifecycle-based Prognostics Architecture with Test Bed Validation
Summary of Glue Tests 1993