My first Gage R&R - Paper/Poly/Foil Material Bond Test, need help

[Ninja]

Since the process never stops, once you're at steady state for the job there shouldn't be much variation within a roll. There might be normal variation from rolls at beginning, middle and end of a job, and even within rolls.

Also, you cant take a sample from the middle of a roll. You would have to unwind thousands of feet to get to it, and we must maintain certain roll sizes per specification.

Hi Zac,
I work in the "web world" too, though not the same as yours.
Since your run load will generate a number of in-spec roll sizes, and it is the "run" you are trying to quantify variation on...take a sample immediately following a discharge roll change. This is the middle(ish) of your "run" which is what you want anyway.

You don't have to break a roll to get your sample...just grab one on the roll change.

Another great question. Conditionioning of the samples is very important. I will age them all under controlled conditions for 5 days, that way the time difference between tests (each on a different shift) will be small compared to the overall time since manufacture.

Keep one thing firmly in mind as you proceed. In order for the GRR to be useful to you, your measurements for GRR should replicate the measurement method you'll use later.

In other words, do your GRR sample prep THE SAME WAY you'll use the test later. Doing GRR with a 5day conditioning period only works if you'll test your in-line production samples after conditioning for 5-days too...which isn't quite "in-line" anymore.

If your GRR is run on a special case (5 days) and your normal testing is done with 30min aging, your data will essentially conclude "This method that we don't use works very well". I'm sure that isn't what you want.

 

[zac2944]

Beauty is in the eye of the beholder.

What is the end use to which you will put your gauge?

Thanks again Miner for your insight. I'm using this gauge to determine conformance of finished product to spec (scrap or good?).

The %Tolerance of 17.47 means that the gauge is adequate for use as an inspection device. Your earlier concern about rejecting good product and accepting bad product is laid to rest.

Awesome. Makes me feel more comfortable about the data we see from this gauge, and now I can back the assertion that this is a valid test for determining goo from bad product. That's really all I wanted to get out of this GRR.

On the other hand, the %Study Variation of 74.86 and an ndc of 1 means that this gauge is not suitable for SPC and will inflate the process variation observed and make your process capability look worse than it is in reality.

We don't our processes by SPC (yet), but we do make adjustments (tamper?) if the values from this test are too low. We know there are process parameters that can affect bond levels measured by this test and we will adjust them if need be. That said, we don't use the output of this gauge to fine tune our process.

One other thing to keep in mind is that since this is a nested study, we are assuming homogeneity of product. If this is not true, the gauge will appear worse than if it were true. Unfortunately, I do not see a way around this issue.

If I understand the statistics correctly, this might explain why my Repeatability is 74.39(%SV), since it would contain any of the within batch variation that violates the homogenous assumption. Is this correct to say?

Also, I believe that my data also show that operator 1 may add more variation than operators 2&3. I didn't notice any gross differences during the testing, perhaps an audit of their methods is in order?

 

[zac2944]

Hi Zac,
I work in the "web world" too, though not the same as yours.
Since your run load will generate a number of in-spec roll sizes, and it is the "run" you are trying to quantify variation on...take a sample immediately following a discharge roll change. This is the middle(ish) of your "run" which is what you want anyway.You don't have to break a roll to get your sample...just grab one on the roll change.

Thanks! That's exactly how we do it. We take a sample from the top of every roll as it comes off the line for QC testing. We did it the same way for this GR&R.

Keep one thing firmly in mind as you proceed. In order for the GRR to be useful to you, your measurements for GRR should replicate the measurement method you'll use later.

In other words, do your GRR sample prep THE SAME WAY you'll use the test later. Doing GRR with a 5day conditioning period only works if you'll test your in-line production samples after conditioning for 5-days too...which isn't quite "in-line" anymore.

If your GRR is run on a special case (5 days) and your normal testing is done with 30min aging, your data will essentially conclude "This method that we don't use works very well". I'm sure that isn't what you want.

We thought about this for some time before conducting the test, but aging for 5 days seemed to be the best option. From what we can tell, this particular product stabilizes to ~99% before it gets to the lab for regular testing. So we don't believe the extra time makes any impact in this case. We chose this process for the GR&R for just that reason. Also, since my experiment had to run over 3 shifts, if I pulled and test samples same day, the %difference in time between operators would have been very large. Operator #1 would get them right off the line (~30min), #2 would get them 8hs later, and #3 at 16hrs later. Waiting 5-day, the %difference in time between shifts becomes much smaller.

This stuff was a lot easier in school with the text-book "perfect case" scenarios. You just don't get those opportunities in a fast paced manufacturing environment. There are just so many more variables.

 

[Ninja]

This stuff was a lot easier in school with the text-book "perfect case" scenarios. You just don't get those opportunities in a fast paced manufacturing environment. There are just so many more variables.

TOTALLY Agreed...{chuckles}

The folks above are giving great advice. The only reason I chimed in was the whole "Value of your work" aspect.

Fast paced manufacturing environment...wait 5 days for testing...Something sounds risky here.

If you validate your testing via GRR based on 5-days, then your test is only validated after waiting 5 days.

Do you do a totally different test for in-line process control? How do you know the bond is good BEFORE you run 45,000 feet?
Or is this me just missing the point and derailing the topic?

 

[Miner]

If I understand the statistics correctly, this might explain why my Repeatability is 74.39(%SV), since it would contain any of the within batch variation that violates the homogenous assumption. Is this correct to say?

You are correct.

Also, I believe that my data also show that operator 1 may add more variation than operators 2&3. I didn't notice any gross differences during the testing, perhaps an audit of their methods is in order?

The effect of Operator in the ANOVA table was not statistically significant, in this test. If you are concerned about the possibility, you could design a study to investigate that.

 

[zac2944]

TOTALLY Agreed...{chuckles}Fast paced manufacturing environment...wait 5 days for testing...Something sounds risky here.

If you validate your testing via GRR based on 5-days, then your test is only validated after waiting 5 days.

Do you do a totally different test for in-line process control? How do you know the bond is good BEFORE you run 45,000 feet?
Or is this me just missing the point and derailing the topic?

Not a derail. Good questions.

We would normally get the machine running to the correct process conditions, then splice in a new roll at the master rewind and kick out the roll of "make-ready" material. We test the top of this make ready roll while the first good (assumed) roll is rewinding. If the top of the make-ready roll is good, and process conditions haven't changed, then roll #1 is considered good.

Testing usually happens 20-30min after the roll is produced, and sometime with a forced cure (oven) and somethine without depending on the product structure. In the case of the process used for this GRR, the material is at >99% of its finished properties before we can even get it into the lab. Therefore my assumption stands that 5 days or 5 minutes will not have an affect on the test data.