Correlation of Field Data with Reliability Prediction Models
Abstract
This report considers the factors influencing the goodness of fit of MIL-HDBK-217C prediction models. Although it is not possible to statistically separate casual factors in every case, areas in which the models are deficient are identified and quantified. Possible causes are reviewed and the most likely casual factors identified. Where positive interference are possible, a range of statistical methods are used to give an unbiased assessment. The underlying distribution of time to failure is investigated since MIL-HDBK-217C assumes a constant failure rate model. Results suggest that no great error will accrue from such an assumption although strictly it is not always valid. The statistical models developed for this study may be used for future model evaluation whenever an unbiased assessment is required. The correlation matrix/ ratio plot method may be used interactively to construct an optimal model but least squares regression analysis is preferred. The ratio plot method allows empirical confidence intervals on predicted failure rates to be readily evaluated.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 01, 1981
- Accession Number
- ADA111258
Entities
People
- K. A. Dey