Skip to main content
The ARC Training Centre for
Transforming Maintenance through Data Science
Presentations

The Effect of Informative Priors on Mean Lifetime Estimation from Censored Reliability Data

ryan-leadbetter

Virtual - Researchers Catch-up

Friday 13 August 2021

https://ctmtds.atlassian.net/wiki/x/DYJJAQ

For large populations of similar components, replacement policies are often designed based on the mean lifetime of the components and whether or not the risk of failure increases with time. A data-driven approach to identifying this information is to fit a Weibull distribution to the lifetime data of the components and estimate the parameters of the distribution.

However, lifetime data for components of this type are often heavily censored due to limited historical data, long component lifetimes, and past maintenance strategies. This heavy censoring results in biased parameter estimates when using traditional methods to estimate the Weibull parameters, such as Maximum Likelihood. One solution to mitigate the effects of censoring is to carry out the survival analysis in a Bayesian framework using an informative prior. However, it is essential to carefully select the prior distribution when taking this approach to consider the sample size and information contained in the data.

Ryan presented a simulation study that looks at the impact that different prior distributions have on the behaviour of the Weibull analysis and summarise how a Bayesian approach can be leveraged to obtain better-behaved inference about the Weibull parameters in the case of highly censored data.