Gage R&R Destructive Test With Breakaway Fixture To Simulate A Part

[formerrocker]

We perform a destructive test using a digital force gage to measure the strength of a container joint. The problem is this: the process is out of control to the point that sequential parts have a lot of variation. With a typical Gage R&R, one would take parts from one batch and assume that they are nearly identical. In our case, we know that the range of measurements is too large to assume this. In fact, I did a Gage R&R and the result was almost 200%! I made a breakaway fixture with latches that seems very repeatable (std. dev. of a little more than a pound with a mean of approximately 40 pounds). However, if I use multiple operators to take measurements using this breakaway fixture, and plug this into the traditional formula that assumes I am retesting the same part, I get a very large (over 30%) unacceptable R&R result. I believe on or more of my assumptions may be incorrect. If I can get a reasonable number, we can start our effort to improve the manufacturing process itself. I have no reason to believe that the force gauge or method is not repeatable. I've used two different gauges with similar results. Any ideas?

 

[Miner]

Can you attach your data and analysis? We can help more with this information. You didn't say whether you were assessing to the tolerance or to the process variation and this makes a huge difference.

 

[formerrocker]

As requested, here are the numbers which are the test results of using the breakaway fixture, not real parts. 3 columns for operator number 1, 3 columns for operator 2, and 3 columns for operator 3. I'm primarily interested in determining the tolerance for the gauge. Thanks for your help!

 

[Miner]

I have a few more questions:
1. Are the three operators actually testing 10 different latches?
2. If 10 different latches, does the variation in the breakaway force of the latches represent the variation in the actual process that they are meant to represent?

Your first post seems to imply that the variation in the latches is much less than the actual process. The measurement variation in this study is +/- 3.6 units. How does this compare to the actual process variation that you are seeing?

The reason that your %GRR is so high is that the variation in the latches is only +/- 1.4. The measurement error is so large that the gage cannot see the part to part variation in this study. If your actual process is this tight, you have problems. However, if your actual process is +/- 14, the gage is fine.

The variation between latches should reflect the actual process variation. Please remember that as you improve a process, a gage that was once acceptable, may become unacceptable as the process variation decreases.

 

[formerrocker]

The breakaway fixture has a number of latches that overcome spring loaded ball bearings. The fixture simply snaps together and a motorized test stand mounted force gauge breaks loose the latches. The fixture gets snapped back together and this is repeated for a total of 30 times for each operator. There are 3 operators, so I posted 90 measurements.

The actual manufacturing process is sometimes fairly stable, but has shown sample reading ranges of around 10 pounds to over 30 pounds at times. Examining the destroyed bonds, there is physical evidence to show that some bonds were indeed stronger than others when comparing items that measured 15 or 20 pounds apart in the samples exhibiting a large range.

We have identified some root causes underlying some of this range variation, and are in the process of making improvements. We would like to of course identify the measurement system's contribution. My gut feeling is that the measurement process is good enough for our purposes, but I need to be able to prove it.

Again, thank you for your time and expertise!