MSA Study Type 1 not capable. We are at the limit. And manufacturing wants to continue producing.

[Monter]

Hello everyone,

I'm new to this website and have benefited from reviews many times.

I would like to describe a problem I encountered with an MSA Study Type 1.

The Cg / Cgk value of the MSA Study Type 1 is less than 1.33. But the production process must also be continued. Manufacturing engineers came to me with two suggestions.

1-Is it allowed to continue using the measuring process with a tolerance reduction (from 5 mu to 4 mu)?

Or

2-In this proposal, instead of individual measurements (for example 25 individual measurements), two or three measurements are taken each time and their mean value is evaluated as one measured value (for example, 25 measured values are obtained from 3 x 25 measurements).

I rejected the first proposal. Because I know that will increase the risks. On the other hand, instead of reducing the tolerance, I recommended an increase in tolerance.

I am also aware that the second suggestion will decrease the standard deviation and thereby improve the Cg / Cgk value.
Is the second suggestion allowed? And should we repeat MSA Study Type 1 according to the proposed solution (two or three measurements taken each time and their mean value evaluated as a measured value)?

I would be very happy if you can enlighten me.

 

[Jen Kirley]

Welcome to The Cove, Serra!

Is the tolerance from the drawing? If so, it should not be changed so as to make the MSA study look better.

Nor should we try to normalize inputs by averaging them. The MSA study's statistics are designed to present information in a way that helps drive decisions to improve the repeatability and/or reproducibility of measuring equipment and systems. Its purpose is not to check a box and get busy with the best numbers that manipulation can buy...

The risk of gaming MSA is to produce failing goods without adequate detection. Failure can get expensive. Production needs to be reminded of that.

 

[Ninja]

FWIW, it sounds to me like the test method is not (precise/accurate/repeatable/reproducible/reliable)...and that you aren't sure if there is an issue in the manufacturing line or not since you don't have good enough data.

The solution here is in the test method...either what method is used or how it is applied.
How do you design a better test, rather than gaming the data as Jen puts it.

Can you tell us a bit about what you are testing, and how the test is being performed?

Ideally, this would have been sorted out before production start. Since that isn't the case, time seems to be of the essence.

 

[Monter]

Hello Jen Kirley, hello Ninja,
First, thank you very much for the answers.
I had refused the tolerance reduction anyway.
But I find the answer to suggestion 2 valuable
At least I clearly understood one point. Measuring a component several times (instead of measuring once) and then evaluating its mean value is a manupilation. Although that looked like a solution in theory, it is not the right way to go in practice. That was a pure length measurement on a measuring device. I think I have to dive into the details. I'll be happy to give you more details later.
Best regards

 

[Ninja]

Trying to help solve the problem...

How long a length (0.0252mm, 47mm, 22m)?
What tolerance?
How are you currently measuring length (what tool)?

Many of the folks on here have measured all sorts of lengths (and most any other attribute).
I, personally, tend to work in micrometers (um, but I'm in America so they are 'microns')...but have had tight tolerances up to 6x6 inches.
Others work in feet or meters...you'll find good steering here for "don't bother with this tool" or "this worked great for me" here if you want it.

 

[Monter]

Hello Ninja,
This is a CMM and the tolerance is 1 micron. We measure a length (distance between to points) of 50 mm.
To avoid misunderstandings, I would like to repeat suggestion 2.
Here, the repeat measurement meant the measuring process (measuring a component several times, instead of measuring once, and then evaluating its mean value) during the manufacturing process. Not the MSA Study Type 1.
As far as I understand Jen Kirley correctly, a process should go the way it should.
Playing with the mean or standard deviations is just a manupilation.
Tomorrow i will look at the details with a measuring technician on site.
Excuse me. Maybe I haven't clearly described it.
Best regards

 

[Ninja]

What make/model of CMM?

I have used the MicroVu Vertex 120 and the OGP Flare... neither would pass an MSA study on 50mm with a 1um tolerance.
...but that was over 5yrs ago, I am not familiar with the current models.

Have you explored other tools, such as a laser micrometer?

 

[Welshwizard]

Hi Serra,

Repeating measurements is always good practice for obvious reasons, it can also be time consuming and even expensive. When formally trying to understand the impact of your measurement process on your part feature and spec you will need to compute the estimated standard deviation.

Of course, if it is enough for you to just quote good practice and carry on with production then that is your choice.

It would seem that you don't have much leeway here in terms of a specification width of 0.001 mm, you will need to be sure that all or a large part of that specification is not obscured by measurement variation. I have worked in national standards laboratories with very elaborate machines and tightly controlled environments that would struggle to hold a proportion of that specification.

Instead of the run chart of a type 1 study plot the results in time series on an individual moving range chart and check that the process is consistent. If it is compute the estimated standard deviation by: R bar/1.128 and compare what you get to the width of your spec.

Now, if the answers to the previous questions are good/encouraging you can formally estimate the effect of repeating the measurements on the standard deviation by computing: R bar/(1.128 x sqrt of the number of repeat measurements). If this is good/encouraging then go ahead and formally do the study with the number of repeat measurements you can then test out your point 2.

I have tried to plot a path for you taking into consideration what you have already written, computations can't replace common sense its true but I hope I have addressed that with use of the ImR Chart and the fact you cab use the data you have already gathered. What you actually do is based on your own quality needs and the time you have but I hope if it doesn't help now it may help in the future.