Is X-Ray Data Variable or Attribute? How to do a capability study?

[Caster]

Hello again. I need some help with a new concern.

We x-ray parts.


The test results can range from Level 0 (perfect) up to Level 7 (pretty awful) in increments of 1.


Trained technicians make a judgment based on hard copy x-ray film master defect pictures.

So for example we might have a Level 2 porosity defect. The customer specification might be Level 4 MAX.

There are no in between values, for example there is no such thing as 2.2 or 2.345, only 2 or 3.

Sooooo…..the customer wants (demands) a capability study. I can plug the data into Minitab and it will do the math, but I have a feeling this calculation is bogus since this is not really variable data. Then again maybe it variable data with very limited discrimination?

I seem to recall that this may be called classified attribute data – and that there are methods to determine capability for data in this format….but I can’t find the references.

Any ideas how to treat this data?

 

[Bill Ryan - 2007]

We have X-rayed in the past but now use a fluoroscope. My experience is that it is still very subjective as to whether an area is a "2" or a "3" and I have always submitted as an attribute capability study (and have always tried to get by without needing to submit a Gage R&R ). Having said that, please keep in mind our products are, pretty much, thin walled die cast housings for automotive Tier 1 customers. While internal integrity is important, it is not usually a "study" requirement. We tend to use the fluoroscope more for noting anything changing in our casting process - not as an inspection/sorting tool.

 

[Steve Prevette]

Any ideas how to treat this data?

You could treat it like categorical survey data, using https://www.hanford.gov/safety/vpp/survey.htm as an analysis methodology.

 

[Statistical Steven]

Hello again. I need some help with a new concern.

We x-ray parts.


The test results can range from Level 0 (perfect) up to Level 7 (pretty awful) in increments of 1.


Trained technicians make a judgment based on hard copy x-ray film master defect pictures.

So for example we might have a Level 2 porosity defect. The customer specification might be Level 4 MAX.

There are no in between values, for example there is no such thing as 2.2 or 2.345, only 2 or 3.

Sooooo…..the customer wants (demands) a capability study. I can plug the data into Minitab and it will do the math, but I have a feeling this calculation is bogus since this is not really variable data. Then again maybe it variable data with very limited discrimination?

I seem to recall that this may be called classified attribute data – and that there are methods to determine capability for data in this format….but I can’t find the references.

Any ideas how to treat this data?

You can always use a Chi-Squared test to determine if the "trained" inspectors are repeatable and reproducible by taking known films (have an independent person grade them) and see how well they are able to grade the film versus the expected value. If you use more than 30 films, you can do a t-test for testing if an inspectors mean is significantly different than zero.