Many of the “standard” reliability methods are intended for non-repairable systems. That is, when a component, sub-assembly or system fails, it is not repaired and returned to service. The Weibull distribution and other well-known distributions which effectively describe the time to failure assume the failures are “terminal”. That is, the whole system is replaced.
In contrast, repairable systems may fail multiple times during their lifetimes and this results in “recurrent events” in which system components may be repaired or replaced to bring the system back on line. In this case, a single system actually has multiple ages, i.e. components which have been repaired or replaced are “younger” than the rest of the system.
Reliability data comprised of recurrent events should be analyzed differently than time to failure data from non-repairable systems. In particular, it is important to recognize the sequence of the events for individual systems represented in the data. This is done by modeling the cumulative failures (repairs, or costs) versus the system age (time). This model can then be used to predict the total failures (repairs, or costs) at some future point in time.
As an example, warranty data are a collection of recurrent events on many products in the field. Events include repair, replacement and preventive maintenance. Warranty data can be analyzed to estimate the cost of extending the time on a standard factory warranty. The resulting model can be used to estimate such things as cost per unit or number of repairs per unit. This information can then be used to decide whether revenue would increase sufficiently to make a longer warranty period beneficial.
Training and consulting is available for repairable systems applications.
Greg Larsen, MS, CRE
Senior Reliability Consultant