Some recent developments on failure processes with masked failure data

The starting point for the optimisation of a maintenance policy for a physical system is usually in possession of the failure process for this system. Such a failure process is normally a stochastic process that models time between failures and that should be estimated based on failure data collected either from the lab or from the field. Field data are often noisy. For example, for a multi-component series system, times between failures are recorded but which components causes the system to fail may be missing. Namely, only masked failure data are available. In this case, we need to estimate a failure process from the aggregate, masked failure data. This paper reviews three recently developed methods on modelling the failure process with masked failure data.