Estimating activation energies for multi-mode failures

• We model accelerated life testing data when we can have multiple failure modes.
• We develop a parametric model based on Bayesian Statistics.
• We compare various parametric models using Bayesian Information Criteria.
• We provides estimations for activation energy and observations membership probabilities.

In Accelerated Life Time Modeling, the goal is to estimate the activation energy and failure time distribution. Existing methods assume data sets come from just one mechanism of failure. However, in many applications, more failure modes can be involved and few data are available; hence, we have to develop a method to identify the number of failure modes and assign observations to the appropriate failure mode. We developed a methodology based on Finite Mixture models and Bayesian Model selection to identify multiple failure modes. The approach provides the probability for each observation being associated with a certain failure mode, and provides good estimates for the activation energy of each mode.