�ݺ�ߣ

Six Sigma
Six Sigma: DMAIC; Y=f(x)
Is it a Goal, a Measure, a Process, a
Tool or an expletive deleted?

 D efine
M easure
A nalyze
I mprove
C ontrol

I’m not conducting a typical gage R&R, but rather comparison to a standard by two different
systems – test machines. A brief summary if I may to help you understand my problem if you
will please: I work for an ice cream company and I our raw ingredients are milk, skim
milk, cream and other substitutes like butter milk and whey. Unlike a discrete manufacturing
industry this is a commodity driven process industry and we inventory our raw dairy fluid
ingredients based on their butter fat and total solids content. The higher the BF and TS the
more money it’s worth impounds of weight. We have two systems, actually test
machines, one is made be accompany call Foss and the other is made by a company called
CEM, both do the same thing to measure butter fat and total solids in a given sample of a
dairy fluid product like milk or cream. However, they use completely different technologies to
do the same thing and of course they each claim that their system is the most accurate to
laboratory test standards.

It is algal requirement in the state of Iowa that milk be tested by a
reputable, independent, unbiased third party tester. This is to ensure that the dairy farmer
doesn’t get short-changed on his tanker load of milk. We pay the farmer based on the third
party’s test results. We still create inventory on our side when we receive a tanker of milky
our own BF and TS test results. Once a week our procurement department compares a
report on our results with the third party’s test results and we've notice that over the course
of a year our test results were testing higher than the third party’s results, so we were paying
for less than we took in. In our unique business .5% shrink is an acceptable loss, but we
recreating a financial paper loss of 2.5% this equated to over a million dollars loss.

I was assigned to see what was going on. My first investigation showed that
despite the Foss testers had better repeatability the CEM testers; the CEM
testers came closer in agreement with the third party’s test results. We took
the Foss testers offline and are using the CEM testers for all raw dairy
product testing. This has improved our financial reporting. The thing is the
Contesters takes 7 minutes to test a sample, the Foss testers only takes
45seconds. With our high season approaching the CEMs will not be able to
keep up with testing demands. So, to the end I’ve been investigating why
the Foss testers went out of calibration. It didn’t take long for me to find out
that the Foss testers had not been re-calibrated on over three years;
someone dropped the ball on that one. I’ve been working hard with the
Foss people to get the Foss testers back to their previous performance
levels, plus a new feature they have is that these Foss testers can now be
hooked up to the internet and Foss can condition monitor them remotely
and warn us of any anomalies.

So, I now have the Foss testers re-calibrated. The re-calibration process has been extremely
time and labor consuming as I have to have specially made to order from a dairy laboratory
all our raw dairy product standard test sets across thebe and TS range
3.25%, 3.50%, 3.75% and so on. Run duplicates in both the calibration and temperature
compensations at 39, 69 and 104 degree F –what a job this has been… Ultimately, I need to
demonstrate that the Foss testers are accurate to the third party test results. I see this a
boxing match with CEM in one corner and Foss in the other with DQCI (the third party
tester)as the referee. The third party tester uses a very accurate wet chemistry method
called Moonie for BF and TS testing, it’s an industry best practice.

I'm planning tomorrow to get the samples from the six dairy products and run them through
both testers twenty four times and send the other same test sample vials to the third party for
them to test twenty four times and send me their results. Typically we are looking for a
standard deviation of not more than0.06.

To this end Dana, what would be a to the point statistical tool for me to use to compare these
two systems (the Foss and CEM testers) BF and TS results with the third party’s using the
same raw dairy test samples to see who is closer in agreement with the third party. I have
attached my data collection sheet I plant use and have populated the cream test results from
a previous investigation.

To try and keep this simple I will summarize. You can
visualize what I am saying in the graphics
T2 vs. Master
• The mean of the Data is not significantly Different
• The variation between T2 and M is significantly different
T1 vs. Master
• The mean of the Data is significantly different
• The variation between T1 and M is not significantly different

I am going out on a limb using the word Significantly

With the little data I have to go on I cant provide much more than
the data provided says that T1 and T2 are significantly different
than M.

Questions
Was the MSE Run 10 samples x 3 measurements by 3 Techs?
Was the study randomized?
Was it crossed or nested?
If the Study was run again with the Master would it be similar?
Can you provide the MSE Data

If you are wanting to assure your measurement system is accurate, repeatable, stable, and precise. The below will not answer
that question. I guess I did not understand when I sent you the initial response. But let me try and clarify what I did and what an
alternative is to that first even though I do not feel this will be the correct path for you to take now that I have more information.

I will have to stick by what I provided originally with just one question that you will have to take into account. To begin with if the
samples you are taking are independent i.e. samples drawn from 2 different populations you would use a two-sample t-test as I
did. What I did not know was if the samples were dependent. If they are dependent then a paired t-test would be appropriate.
Both of these tests will tell you if there is a difference.

Here is the catch. These tests will only tell you if there is a difference between the samples. They will not give you any indication
of where the sources of variation are coming from in your samples. It can come from within or between samples, your
measurement system or other sources I am not aware of not being knowledgeable of your processes. This subject of variation
within and between samples takes into account 4 questions that are based on time..

1 and 2 below is considered short term variation.

1 - What is changing within your samples?
2 - What is not changing within your samples?
3 and 4 below is considered long term variation?
3 - What is changing between samples?
4 - What is not changing between samples?

Now that I think I understand you want to evaluate your measurement system I would recommend you perform this study first.
You can do this 2 ways. A crossed or nested study. Minitab explains this very well and has great examples as well to walk you
through the process. It will also provide you with an interpretation of the examples. In summary a crossed study is used when
your 3 appraisers measure the same 10 parts 3x each. For a crossed study your 3 appraisers never measure the same part more
than once. This is normally done during a study that is destructive in nature or renders the sample no longer useable. There is no
other way to move forward with comparing samples without knowing your measurement system is capable first. Once that is
established then you can evaluate differences between your samples with confidence.
I wish I was in a better position to get involved with the study but with the position I am currently in I have very little time to help
except after hours. I really would like to give you the best I can and have to say you must do a true Measurement System
evaluation first. If you cannot measure then your data from the testing will not give you an accurate account of the variation in the
product you are trying to assess not matter how you analyze it.

�ݺ�ߣ

Mse June 24 2011

Recommended

More Related Content

Viewers also liked (16)

Similar to Mse June 24 2011 (20)

More from ExerciseLeanLLC (13)

Mse June 24 2011