Variable Data Test Never Fails

Howdy folks!
Pull test for a glue bond with a 5lb minimum spec is returning 60+lb values, and the glue bond is not actually failing.

I suggested just calling the non-failure a failure and running stats on variable data.

Someone else suggested converting to attribute data (P/F) and testing more samples.

What would you do? Thanks!


Forum Moderator
Staff member
Just for clarification, are you seeing a cohesive failure of the substrate, not an adhesive failure of the glue to the substrate nor a cohesive failure of the glue itself?
occasionally the substrate fails, but more often than not the tester is (manually) pulling up to 65-70 lbs of force and then 'giving up' or terminating the test. My argument is we're so far above the spec, just call it a failure and call it a day. Someone else argued to use P/F criteria, which would require testing an additional 29 samples. Not sure which method is more sound (but my option saves time and samples)


Forum Moderator
Staff member
interesting problem. I always lean in favor of variable data, but you have two problems with your data. The first is that the mean will be lower than it would be had they tested to failure. Given the leeway between 60 and 5, this probably isn't much of an issue. The bigger problem is that the variation that you have is not the variation in strength, but is the variation in when the tester decided to stop testing. If you have historical data on the variation, you could use that as a proxy.

There is a more complicated, but legitimate approach that you can use called probit analysis. This is a technique used in reliability analysis of stress to failure testing where you have censored data (i.e., no failures at a given stress level). If you have Minitab, probit analysis may be found under the Reliability menu.


Stop X-bar/R Madness!!
Physical testing, especially tensile testing, has repeatability issues due to the variety of failures you can generate. Can you share a typical test curve - or are you just collecting max value off of a digital scale?


Looking for Reality
Just a stray thought more than anything else...

If the test "never" fails...and is really testing either the machine shutoff setting, the substrate inherent (cohesive) strength or the pull tab tensile strength...

All of these presume that you've been doing this test with the same results for many batches..."many" being subjective:
1. consider why you run this test at all.
2. If you still need to (customer requirement), consider radically reducing your sample size

Tests are important.
But tests cost money.
Lack of attributable result (not testing the glue beyond P/F) lowers the value of the test.
If you aren't getting your money's worth...why are you spending it?
If you get the same value from less money...spend less.

I see that you're trying to analyze the data...not a bad thing at all...but it then costs more money for the analysis (your paycheck).

Step back from the nitty gritty and re-evaluate what's important. If it's still important after that perspective...dig in.



Stop X-bar/R Madness!!
When you exceed your requirement by 100 times, you really have all the data you and your customer need. Knowing the exact catastrophic failure data beyond 60 is an interesting academic question, but of absolutely no practical value. Report ">60" as your variable data (so as not to be trapped into increased testing for "attribute" data), and walk away from it. There has to be something more important to work on than a process that is 100X better than the spec.

John Predmore

Involved In Discussions
To satisfy yourself that your inspection apparatus is working properly (and able to distinguish defective glue joints), prepare test specimens where you compromise the joint integrity, for example by applying less glue, or remove substrate material to reduce surface area. (As a safeguard, clearly label your defective test samples so you don’t lose sight of them and they get shipped as valid product.)

If you want to understand better what is occurring within the joint, instead of a force gage use a tensile tester. A tensile tester measures both force and displacement. Force x Distance = Work. The area under the Force/Displacement curve is Energy. A tensile tester can also apply force at different load rates or at different angles, which might provide valuable information.

Top Bottom