Applying Bayesian updating methods to a new combined lifecycle failure distribution

Reliability based modeling methods are the cornerstone for optimizing large scale system-of-systems supply chain and prognostics and health management (PHM) models. A new reliability distribution, the CoMBined Lifecycle (CMBL) distribution, arising in practice from systems lifecycle work at Sandia National Laboratories, is presented and evaluated. This bathtub shaped distribution has easily interpretable parameters useful in modeling both older components where significant failure data is readily available or new components where only expert opinion is available. Three methods to update the CMBL distribution when new data becomes available are presented. The first method attempts by-section sequential Bayesian updating, the second uses a Poisson process and employs multiple changepoint modeling, and the third attempts to use the underlying probability functions with multiple changepoint modeling. Since the changepoint models generally require 50 to 100 failure times to provide consistent and reasonably accurate CMBL distribution updates, a data supplementation scheme is explored.