In failure-analysis work, we often want to know why the damage happened. If it’s a fracture, a true answer is always a variation of “the stress exceeded the strength.” This answer, however, is not terribly useful. What we really mean by “Why did it break?” is something along the lines of “Why did this one break?” or “Why did this one break now?” or “Why did this one break after such a short life?”
Asking a more specific question provides context for planning the investigation. Follow-up questions might address whether the failure has safety implications or significant financial implications. Once this is known, we can make a reasonable business decision as to whether it is worth doing multiple tests and testing multiple additional parts (exemplars) in order to gain more confidence in the findings.
It isn’t usually enough to determine whether the part “met the spec.” The part does not know what the spec is. The part rarely fails solely because of lack of conformance to a spec.
In order to address this issue, we need to test “exemplars.” Most people are more familiar with the term “control part.” This term can be misleading, however. What exactly makes a good control part? If we have been producing component X for 20 years, and it has always performed well in the field, and then we suddenly have a rash of failures from parts made two years ago, the one possible reasonable control part would be samples that are older than the two-year trouble spot. But how many people have old parts lying around?
Another potential good source of control parts is durability-tested parts that passed the durability test. Yet again, how many companies scrap these out as soon as the test is complete?
If you can’t find known good parts as control samples to compare to the failure, then exemplar is a more precise term than control. It’s better than nothing as a comparison, but it could be misleading if people assume it is a good part.
The fact is that most medium-technology and low-technology parts have engineering prints and material specifications that are very wide open. The parts work because a supplier’s employee understood the use and made sure that the part would function. Now that employee leaves, and the replacement person decides to cave in to cost-reduction pressure. The part still meets the spec. Maybe the strength is even greater than the original parts. But maybe there is some other characteristic that causes a problem. Maybe the part is a bit thinner. This slight difference may kick the part into a finite cycle life from the nominally infinite life it had before. A part on the edge of the good durability envelope may not need much variation to become an occasional failure.
I’ve often thought that there’s a business opportunity out there to store successful durability-test parts. When we have them to cut up and compare to a field- or durability-test failure, we get a much greater chance of having data that is straightforward to interpret.