Recurrent Event Data Analysis

In life data analysis, it is assumed that the components being analyzed are non-repairable; that is, they are either discarded or replaced upon failure. When analyzing the failure behavior of non-repairable components, the data points are typically either times-to-failure or times-to-suspension. For a group of non-repairable units coming from a single population, the time-to-failure of one unit in the sample does not affect the time-to-failure of other units in the sample. Therefore, the lifetimes of non-repairable systems are considered to be independent and identically distributed (i.i.d).

On the other hand, for complex systems such as automobiles, computers, aircraft, etc., it is likely that the system will be repaired (not discarded) upon failure. Failures are recurring events in the life of a repairable system, and data from such a system are obtained by recording the age of the system at the time when each failure occurred. This type of data is known as recurrent event data.

The failure behavior of a repairable system is dependent on that system’s history of repairs; therefore, traditional life data analysis methods, such as the Weibull distribution, are not appropriate because those methods treat every failure event as identical and independent from the previous one. In order to analyze recurrent event data, Weibull++ includes a choice of two methods: non-parametric and parametric analysis.

The non-parametric RDA folio is based on the mean cumulative function (MCF). This analysis provides a plot of the MCF to illustrate the average number of recurring failures of a system, or a group of systems, over a given period of time (or distance, cycles, etc.). The MCF can be used to evaluate whether the number of failures increases or decreases over time, to predict the future number of failures or to compare different data sets from different designs, operating conditions, production periods, etc.

The parametric RDA folio uses the General Renewal Process (GRP) model to analyze the failure behavior of a repairable system. The GRP analysis method takes into account the effectiveness of repairs on the condition of the system. For example, a repair may bring the system to an as-good-as-new condition, to an as-bad-as-old condition or to some stage in between. If the system is only partially rejuvenated after the repair, then this may affect how the system fails in the future. Over time, the recurrence rate of failures may remain constant, increase or decrease. The GRP model allows you to analyze the failure behavior of a partially restored system over time so you can obtain estimates such as the cumulative number of failures, mean time between failures (MTBF) and failure intensity.