Attribute MSA study

One of the trainers told that you can select multiple defects for Attribute MSA study. I was not convinced. My question is how can we do MSA study using multiple defect samples?

Can you elaborate on your question? Why do you think it can’t be done? Can you give an example?
In general your trainer is correct. There are many ways to perform an attribute MSA where parts have multiple defects - bot multiple defects on a single part and multiple defects across many parts.

SREEDHARA HN said:

One of the trainers told that you can select multiple defects for Attribute MSA study. I was not convinced.

Click to expand...

Is the source of your skepticism rooted in one of the following concepts of attribute testing?

1. My test is to identify "good parts".
2. My test is to identify specific defects which would disqualify parts.

Tidge said:

Is the source of your skepticism rooted in one of the following concepts of attribute testing?

1. My test is to identify "good parts".
2. My test is to identify specific defects which would disqualify parts.

Click to expand...

Dear Tidge,

Thanks.
My test is to identify "Good parts"and disqualify the parts with any visual defects (Not restricted to some specific defects)

Bev D said:

Can you elaborate on your question? Why do you think it can’t be done? Can you give an example?
In general your trainer is correct. There are many ways to perform an attribute MSA where parts have multiple defects - bot multiple defects on a single part and multiple defects across many parts.

Click to expand...

Dear Bev D.
Thanks for the response ...
My doubt is if you select multiple defects across many parts, then it will become "Poisson test". Where as if you select same defect across many parts then it will let you know whether the appraiser is capable of effectively identifying the defect parts. (Marginally not OK towards both USL & LSL).

SREEDHARA HN said:

Dear Bev D.
Thanks for the response ...
My doubt is if you select multiple defects across many parts, then it will become "Poisson test". Where as if you select same defect across many parts then it will let you know whether the appraiser is capable of effectively identifying the defect parts. (Marginally not OK towards both USL & LSL).

Click to expand...

What do you mean by “Poisson”? Are you referring to the statistical distribution (poisson) or a a poisoned test? Not sure what the latter would mean.
In reality the appraiser is better tested by being presented with parts that have multiple defects and parts that include defects that span the range of what would normally be seen. I have successfully performed many of these types of tests. When the appraiser knows there will only be one type of defect present they are more alert to that defect. But when the real world inspection population has defects that reflect reality you gain better insight into the appraiser’s over all ability.

An example: I have set up sample sets for worm detection in canine feces. We had to collect samples that contained round worm, hook worm, whip worms and heartworm eggs. Some samples also had to contain pollen which can look like parasite eggs. Some samples had to have pollen and a worm egg and a few sample had to contain two types of eggs. We also had to have samples with grass and other stuff in it as that contamination was supposed to be removed during sample prep - without removing the eggs. Any remaining contaminant would cause the slide cover to not lay flat obscuring the focal depth of the miscroscope used to see the eggs. Then we had a lot of ‘clean’ samples. We analyzed the results in several ways. First we analyzed how well the appraisers detected samples that were known to have eggs no matter what type of egg it was. Then we analyzed the results for each separate type of egg. Simple analysis but tedious….I’ve also done this with complicated diagnostic instruments that can have several different types of failures in a single instrument, semiconductor chips with numerous types of defects, automobiles with various types of fit and finish defects…there is no reason to not have multiple types of defects. If the part can have multiple types of defects then that is what you use in your sample set.

So we would need to know your specific concern here: is it the math? Is it the difficulty in selecting the appraisal set that represents reality? Or is it something else?

SREEDHARA HN said:

My test is to identify "Good parts"and disqualify the parts with any visual defects (Not restricted to some specific defects)

Click to expand...

This is likely where your confusion is coming from. It is impossible to "prove" that a part is "good". The only things a test can do are:

1. verify if a part meets a specified requirement
2. identify specified defects

As @Bev D points out: any given part can have more than one defect and not necessarily complicate an Attribute study. Some critical thinking has to be applied. For example, if the slide for canine feces was completely devoid of a sample, it would be useless to try to use that slide for an assessment of the ability to detect a certain type worm egg. I'm not sure what the appropriate mathematical jargon would be for excluding it, but it would probably be something like it "not being a representative member of the ensemble of potential test candidates."

we called it a null sample. It didn’t get counted in either the numerator or the denominator of any analysis. It did have to be documented as a null. It wasn’t a zero. Zero is a value…it’s one of the first things I have to teach people who analyze data and who enter data - don’t confuse the two.

Sincere Thanks to both Bev D and Tidge for the support. I have got the answer from your clarification.