Non-replicable tests in MSA (Measurement Systems Analysis)


Andreas M?ller

Hi all!

This is my first time posting something on elsmar after a long time lurking! :)

I am having some troubles with MSA and what i believe to be non-replicable tests.

We are measuring differential pressure and opacity with a running car engine,
the exhausts move through a pipe to the object being tested and the pressure is measured at two points, before and after the object, to make sure that we're not experiencing a clogged object.

The opacity test is conducted after the object to see how much of the exhausts that are moving through the object and how much of the exhausts that are stopped. This is a percentage relative to no object existing (full exhaust).

Now there are a lot of variables here for us when we try to do a normal MSA, for example the car engine (is running idle) is not running at the exact same RPM and is not returning exact same thickness and "quantity" of exhaust gas. The gas could swirl more one time from another in the pipe giving us ups and downs in the readings of both opacity and diff.pressure.

So our problem is that we do not recieve the same presets for each part each time. Would a nested r&r be helpful here? And how would i go about to conduct the study?
(I am using Minitab 16 to interpret our results)

Thanks in advance!


Forum Moderator
In addition to Bev's excellent article, you may want to review the overview of Non-replicable MSA studies in my blog, as well as review the section in the AIAG MSA manual.

In some cases it may be difficult or impossible to control all of the variables. In that event, you will need to use the regression approach described in my blog and in the MSA manual. This approach requires that you know or can determine the relationship between the gas and engine RPM.

I recommend performing an additional study. While I can see that such a measurement may be highly variable in the short term, I would expect it to be very consistent over the long term once you have compensated for covariates such as engine RPM. If you have the freedom to adjust test parameters, you may want to consider longer term sampling and averaging techniques.

Andreas M?ller

Thank you both for the article and blog entry, i have a few ideas to discuss with my collegues now and we'll see where we end up.
Top Bottom