nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

Gauge R&R For Volume Cutter

mmeewes

Hi all,

I'm hoping to get some insight on best practices for conducting a gauge R&R study for one of my critical processes at work. For the sake of background, I work for a food manufacturer that makes . The point of interest is at the front of the line right after a large slab (Approx. 60" Wide) is formed. Operators at this location use a circular cutter to extract samples from the slab and then weigh them. The cutter itself is the same volume as a finished product, so the intent is that by cutting out the same volume and weighing the core sample we can better predict the finished weight of 1 unit. I've done several correlation studies between the core sampler and the finished unit downstream and have found little to no evidence of a correlation, other than the fact their variances are equal meaning they are coming from the same population.

These results have lead me to the following hypotheseis', 1) The product itself suffers from extreme variation within the window of tolerance that we are attempting to control. 2) The cutter tool used is contributing an excessive amount of variation to the test due to testing methods, product being tested (various SKUs), and conditions of the tool.

I've conducted Gauge R&R studies in the past with say 3 operators and 10 samples, with each operator measuring each sample twice in random order. All of these studies however have been on devices such as balances or calipers. I've never tested on volume before, so question 1 is how should I collect the responses? Question 2 is how should I design a study on a product that is inherently variable? The slab does not stop as testing is being conducted, it is moving at roughly 10' per minute. Sample 1 could be 8 to 10 grams different from sample 2 just due to natrual variation alone.

A target unit is 40 grams with a "tolerance" 6 grams. Our control charts tell us however that we are only able to commonly control to an 8 to 10 gram window. Now, the data used for these control charts is collected using this cutter so the variation may be skewed if my hypothesis is correct.

Any help that anyone could provide would be extremely helpful and very appreciated. Thank you in advance!

Matt

Accepted answers

Peter_Bartell

From the way you've described how you'd like to set up the experiment, each sample is indeed nested within an operator since they are creating the samples and the samples are NOT crossing the operators. You should be OK to go through the classic Gage R & R if you specify the correct model. I also suggest you take a look at Don Wheeler's approach called EMP (also supported in JMP) for analyzing the experiment. I like Wheeler's approach because it's much more visual than the classic tabular Gage R & R and their nonsense percentages...that's another debate for another time. You may find my Mastering JMP On Demand Webinar informative:

https://www.jmp.com/en_us/events/ondemand/mastering-jmp/evaluating-and-monitoring-your-process-using-msa-and-spc.html

One thing I strongly encourage though is to take a look at your documented work instructions for the test method and process...there should be some in the company's quality management system if your company is ISO9001 registered...and maybe even if not registerd but at least has a QMS. Then go to GEMBA and watch, for yourself, how the operators are actually creating, handling, and measuring the samples during the conduct of your experiment AND actual production...and don't be surprised if you see people doing all manner of things not included in the documented work instructions. Again, my past experience found all sorts of 'improvement' opportunities just by watching.

Good luck on your certification...I'd tell your certifying body this looks like a Black Belt project to me

All comments

Peter_Bartell

As I see it you've articulated two separate but related problems.

First problem is the inherent variation of the product itself is far beyond the specification tolerance.

Second problem is an unknown measurement system variation.

The first problem is one where you'll have to put your root cause analysis hat on and work through all the classic Define, Measure, Analyze, Improve and Control phases of a classic Six Sigma project...at least that what it sounds like to me. During a prior lifetime I led inumerable projects just like you've described. A process that is just too variable for commercial viability and a team forms to reduce variation. The toughest type of Six Sigma project I've encountered. If you aren't familiar with the Six Sigma model for process improvement maybe you'll find watching this video helpful:

https://www.jmp.com/en_us/events/ondemand/non-series/jmp-and-visual-six-sigma.html

A more end to end treatment of the Six Sigma approach can be found in this great book:

https://www.sas.com/store/prodBK_67155_en.html?storeCode=SAS_US

A piece of that project should also be problem #2. From everything you've described it sounds like a good old fashioned Gage R & R approach is viable. So I must be missing something? Why can't you just take the samples that are for use in the study and execute the designed experiment behind the Gage R & R? The only thing I can think that might get in your way is if somehow measuring the samples is destructive? Doesn't sound like it...but maybe?

mmeewes

Thanks for the reply! So I apologize for not saying so up front but you hit the nail right on the head, in that this issue that I posted about is part of a Green Belt project I've taken on for certification. As for the Gauge R&R portion of your question, the act of taking the sample itself is destructive. If you think about it, there are really two measuring systems in play here. The 1st is the cutter itself that is used to extract the sample from the slab. The second is the balance used to weigh the sample. I've done studies with these operators where I focus on the balance itself, and I've been able to statistically conclude that the balance is an acceptable form of measurement for our process.

With those facts, you can see that simply weighing the same core sample twice for repeatability would be another test for the balance, not the cutter. As I see it, I am looking at a nested destructive gauge R&R study to measure the cutter itself. If I were to set out right now, I think I would design the study as follows:

1) Have an operator cut two samples from similar locations on the slab.

2) Weigh the 1st sample and record the response.

3) Weigh the 2nd sample and record the response.

4) Repeat steps 1 - 3, 9 more times.

5) Repeat steps 1 - 4 twice more with 2 different operators.

From my research it sounds like a nested study could be acceptable so long as you can state that the two separate samples are mostly identical. It is the last part of that sentence that give me pause. I'm not convinced that the two samples required for repeatability will be similar enough to be considered mostly identical. Will JMP be able to distinguish between the Part-to-Part variability and R&R variability accurately? Will I just need an asterisk next to the results that explain that the results might be bias due to the variation of the product itself?

Peter_Bartell

https://www.jmp.com/en_us/events/ondemand/mastering-jmp/evaluating-and-monitoring-your-process-using-msa-and-spc.html

Good luck on your certification...I'd tell your certifying body this looks like a Black Belt project to me

mmeewes

Thanks for all you're help!

It feels like a black belt project!