Gauging method comparison - Air Gage vs. CMM Measurement

[achorste]

Afternoon all (or morning to those of you far away from Britain!)

Looking for some advice:

We machine tapers on hip joints (amongst many other things). These are currently CMM inspected for spigot angle. However we have had some issues with rejections from the customer which measure fine on the CMM when we get them back. It turns out they are using air gauging to measure the angle & diameter.

Anyways along with our process improvements to reduce variation, we are equipping ourselves with identical equipment as the customers goods inwards inspection.

My plan was to perform a GR&R on each method using the same parts - however this would not give me a direct comparison.

The question is - how do I directly compare each piece of equipment? Some mention has been made of a "Two tailed T Test" that will give me the answer. However I neither know what one of these is or what to do to perform one effectively. (We do have Minitab by the way).

Any advice is greatly appreciated.

[Bev D]

The statistical techniques you need are: Youden plots to visualize the difference including any biases, paired t test (NOT a two tailed t test) to determine if a real difference exists after removing the systematic bias and the Bland-Altman analysis of the importance of the difference between the two methods.

I would add that this type of analysis is not well served by the traditional gage R&R (the one that calculate %EV, etc. using the tolerances with rther small numbers of parts =10) You are performing a true "method comparison" and the techniques listed above are for this type of analysis.

One thing to understand is how each guage sytem actually perfroms the measurement. Air gages are notorious for being misapplied - they may not be accurately measuring the real characteristic of interest.

My recommendation would be to then measure approximately 30 or so parts twice with the CMM and twice with the air gage. Plot the two sets of results for each guage on a Youden Plot. This is a scatter plot with a 45 degree 1:1 line drawn on it. No measurement error will be exhibited by the points falling on the 45 degree line. (DO NOT PERFROM REGRESSION OR DRAW A BEST FIT LINE. MEASUREMENT ERROR DOESN'T FIT STANDARD REGRESSION TECHNIQUES.) At this point you could calculate the number of distinct categories or the Discrimination ratio. But really, looking at the data will tell you whether or not each method is repeatable given the range of total variation. If one or more of the techniques is nto repeatable you need to start there. any other analysis after this step is meaningless unless both methods are repeatable.

If both methods are repeatable, you can proceed to plot the first measurement result for each guage on a Youden Plot. Then you can follow the procedures as listed in the referenced articles. OR post your data at this point and we can step you thru it from there.

References:
"STATISTICAL METHODS FOR ASSESSING AGREEMENT BETWEEN TWO METHODS OF CLINICAL MEASUREMENT" by Martin Bland and Douglas Altman, published in The Lancet, February 8, 1986.

You may also search for Bland Altman, Youden and Paired t tests...

[achorste]

Absolutely fantastic - just the kind of guidance I was looking for.

Muchos gracias, Danke Schon, etc etc

[bobdoering]

Note - be sure to mark and measure the same location on each part. You want to identify measurement variation - not physical variation.

[Miner]

If you don't find Youden Plot, it is also known as an Iso-plot.

[mboucher]

Your CMM software is no doubt defaulting to LSQ (Least Squares Algorithm) this is typical with every CMM software.

Try switching the calculation method to match air gaging. You may need to increase the amount of points taken and switch from LSQ to a maximum inscribed circle. This will change the calculation to the maximum size pin that will fit in the diameter. Since the air gaging is a two-point measuring device you may have to do a point to point measuring check.

Are they sweeping the entire circle and giving you an average? Just remember a CMM is not a air gage and it will average the points taken to 'fit' a circle.

Mark Boucher

[Bev D]

Bingo!

the measurement method must match the intent of the thing it's measuring.

generically speaking the choice between an air gage and a CMM (or hwo the CMM works anyway) is whether you are essentially passing a flexible (liquid or gas) object thru a path or an inflexible object. The air gage will give total volume while the CMM typically provides an effective circle as you described.

as in most cases the physics of function must be understood first. then comes physics of failure and statistical analysis...

[Bev D]

In Reply to Parent Post by Miner

If you don't find Youden Plot, it is also known as an Iso-plot.

There should be much more out there regarding the Youden plot.
but basically it is as I've described. a scatter diagram with a 1:1 (45degree) line. The standard deviation of measurement error (the SD of the difference between the two repeated readings) is used adn not hte standard error in the Y direction about some best fit line as in regression. the SD of measurement error has a vector perpindicular to the 1:1 line....