Infant Failure Patterns Are Random!

... and RANDOM failures aren't always so common.

He sat pensive in the bright green electric cart, cob-webs in the rear basket, tinted safety glasses concealing his eyes. The road from the maintenance shop to the chemical plant sloped ever slightly down to the river; the draft from the facilities multiple cooling towers showing clearly in the cool morning. As I emerged from the maintenance shop to meet our pump reliability engineer, he greeted me with a smile. I sat in the cart. “Well,” he said, “here we go again.” With that, we were off to see a pump that had seized late the night before.

Equipment failures are part of the risk of operating any facility. One way to reduce or minimize risk from equipment failure is by understanding failure patterns. There have been numerous studies related to failure patterns; perhaps most notable is that by Nowlan & Heap in Reliability Centered Maintenance. Much like the story of the engineers going to physically look at a pump failure to better understand the failure, when we try to understand failure patterns in our facility, we need to be sure we are physically looking at the failure data from our facility. We all understand that the way one pump fails today may very well not be the same as a similar (or different) pump may fail at an entirely different location. We need to apply this same understanding to the prevalence of the different failure patterns at different facilities.

• What if I told you that you can expect 1 in 4 facilities to have less than 10% of their Assets exhibit a RANDOM failure pattern? Similarly, that you could expect roughly 1/2 of all facilities to experience less than 25% INFANT MORTALITY failure patterns for their Assets?

By understanding the failure patterns in our own facilities we can objectively answer these questions. We can only do that by integrating life data analysis into our maintenance & reliability toolbox.

Several studies have been published as they relate to failure patterns in different industries. The table below shows some of these. 

From these results, we can begin to understand the prevalence of the different failure patterns. Further, we can perform statistical analysis to paint a picture of what we can expect when looking at multiples of facilities. By fitting the results of previous studies to a beta distribution, we have generated the following plot (a histogram of a hypothetical study of failure patterns across 100 facilities).                         

There are many “take-aways” from this plot. One such take-away is regarding INFANT MORTALITY.

• The occurrence of infant mortality may be random. The oft-repeated statistic from Nowlan & Heap of 68% Infant Failure pattern (for United Airlines in 1968) is not a rule of thumb or what you may likely expect in any particular facility / industry / study. We should not presume that this statistic applies to our own facilities or industries.  As the graph shows, the % of Assets exhibiting this failure mode is expected to be fairly flat across multiple facilities (i.e. we can expect to find a similar amount of facilities (4-6 facilities) who have 75-80% of their Assets exhibit Infant Mortality failure patterns as we would for 15-20% (4-6 facilities out of 100)).

Once we understand how much variation there exists in these results, we begin to appreciate the need to objectively determine what our own facilities look like (in particular from a life data analysis perspective).

Is life data analysis a key part of your maintenance & reliability tool box? 

Do you track the failure patterns in your facility? If so, how do your results compare?