A friend and colleague brought a thesis paper by Louis J Hogge to my attention. Titled “Effective Measurement of Reliability of Repairable USAF Systems, the document proceeds to recommend the US Air Force to stop using MTBF and related measures and instead to use Mean Cumulative Functions (MCF) instead. Yeah! I certainly like this paper and the direction that I hope the Air Force and you will take going forward. Please give it a read and comment. And, if you are in the military – pass it along and continue the discussion around alternatives to using MTBF. Be sure to scroll down to see a nice long comment by Louis as part of a
Linkedin RIAC – Reliability Information Analysis Center discussion. You are encouraged to comment here, or directly to Louis via Linkedin (Louis’s Linkedin page).
Effective measurement of reliability of repairable usaf systems from FMS Reliability – written by Louis Hogge.
—
The following is Louis’s comment in a long running discussion on a Linkedin group.
| I’m pleased that my thesis has been read and is generating some discussion. That is a success in itself.Yes, as Leman B. said, my thesis is my view developed from my research and experience. I am still occasionally on the quixotic quest to remove MTBF as ‘the’ DoD/USAF measure of reliability. MTBF does have its place but I believe it is generally misapplied within the USAF. To frame any discussion of reliability analysis the context must be established. Are we discussing things that are not repairable or repairable? Are we discussing reliability prediction or measurement? Those are four different discussions with at least four different analysis processes. I’m going to make some general statements … for discussion. Of course there are always exceptions. And there is so much to say … someone could write a paper on the topic? Reliability analysis is concerned with time-to-event. Reliability analysis of things that ARE NOT repairable is concerned with time to end-of-life, usually referred to as Time To Failure (TTF). To make a prediction of the reliability of items that are not repairable a parametric failure distribution must be assumed. Controlled tests may be used to collect data to check and improve the failure distribution model. A predicted MTTF and other estimates such as mean expected life (not the same as MTTF) can be calculated from the failure distribution model. To measure the reliability of items that are not repairable a data set of TTF from a controlled population and environment is collected. The total usage life of the population, after all units in the population have failed, divided by the number of units in the total population is the measured Mean TTF (MTTF). There is no logical reason to expect predicted MTTF to represent the measured MTTF unless all of the assumptions and constraints used to develop the predictive model are valid through the life of the fielded population. Reliability analysis of things that ARE repairable is concerned with Time Between Failure (TBF). To make a prediction of the reliability of items that are repairable also requires application of parametric failure distributions but the models are far more complex than those for not-repairable items. Models may accommodate such things as effectiveness of repair, times to first failure and times to subsequent failures … . . Layers of failure distribution models are combined from the lowest level MTTF up through indentured repairable levels. At the desired level a predicted Mean TBF can be calculated. I suggest that the only value of the predicted MTBF is as a relative comparison to another design or implementation for the same application. The magnitude of the number in itself is not of value except in the most general terms. If the MTBF of one system is predicted to be 100 hours and another is predicted to be 200 hours it cannot be said the operational reliability of the second will be twice the first. If the predicted MTBF is 100 hours it cannot be said that the system could be expected to operate in the field for 100 hours without failure. To measure the reliability of items that are repairable the ordered sequence of Times Between Failure for each repairable item is collected. The total usage life of the repairable unit, after it has reached end of life, divided by the total number of failures that unit experienced in its life is the measured MTBF. There is no logical reason to expect predicted MTBF to represent the measured MTBF unless all of the assumptions and constraints used to develop the predictive model are valid through the life of the fielded unit. The measured MTBF is not interesting when you cannot know the number until the unit is retired from service. There are ways to measure the in-service performance of repairable units but it is not MTBF. |
| By Louis Hogge |
I use mean cumulative repair function calculations for our field data on repairable systems, some of which are for the USAF. Wayne Nelson wrote several good papers on the subject.
Hi Mark,
Yes Wayne has done good work on MCF and permitted me to post one of his papers. http://nomtbf.com/2012/02/graphical-analysis-of-repair-data/
cheers,
Fred