GR&R for an Automatic Optical Inspection Machine (AOI)

[Karl2109]

Hello Everybody,

I'm new to this forum so I ask for forgiveness in advance for asking a question which may already have been answered and I just did not see it !

My question(s) is in regard to performing a GR&R for an optical inspection machine during an on-site evaluation at a customer site. The customer evaluates different AOI systems from different vendors simultaniously.

(1) Am I correct in saying that there cannot be a GR&R for a fully automatic (in-line) optical inspection machine and that there can be just a GR result ?

(2) I read in some of the previous feeds that Cp/Cpk must not be used for the evaluation of an AOI system. Is that true and if so why ?

(3) If I use the simplied formular to determin the GR (stability test) - what determines if the factor 5.15 or 6.00 should be used ?

(4) And here comes the big one - what determines the spec limits/tolerance in those applicable formulas/GR&R calculators ? I think it should be the tolerance (spec limit) of the AOI system (x/y + optical resolution) while some of my colleagues believe it should be the spec-limit of the sample/part the AOI system is inspecting (e.g. tolerance to an area covered with a fluid).

Any feedback would be sincerely appreciated.

Thanks and regards, Karl

 

[caversluis]

Hej Karl,

For some time ago I did a six-sigma project on a AOI machine. We use orbotech machines in electronics manufacturing. The machines are used for detectings shorts etc. If you do that, a regular GR&R can not be used as you work with attribute data. Thefore I performed a Attribute gage R&R where the machines is treated as an operator and the results are compared to the findings of an expert.

In fact you will have to do this for every single things you look for on the board. In my setup a used 15 boards with 88 opportunities for defects (ofd). Every ofd was inspected twice both machine and expert. The result was analyzed in minitab.

Hope this helps,

Cornelis

 

[Karl2109]

You are right on the money Cornelis - Our equipment and the Orbotech equipment address the same market segment (2D/3D PCB inspection).

What differenciates a regular GR&R from the 'Attribute GR&R' you have mentioned ?

Any thoughts concerning my other (original) questions -- particuarly concerning Cpk data and why this may not be the way to go for an AOI machine ?

Thanks, Karl

 

[caversluis]

Well let us say you produce shafts. You will then be able to measure the diameter on the shaft. The measurements will be "real" like 60,345. Due to this, you can calculate mean and deviation. As you will have tolerances, you can calulate Cp and Cpk values.

Now you are not producing shafts but PCB's. This complicates things a bit since you can not measure f.ex. shorts. You can see whether you have a short or not. So your "measurement" will be good / bad. This is called attribute data. The problem is that when you have attribute data, you can not calculate mean and deviation and therefore you can not calculate Cp /Cpk.

The same goes for GR&R. The statistics behinds this rely on "real" data. When you do not have "real" data, you have to use another method.

For attribute data, you can use Attribute Gage R&R. This was in fact invented for operator inpsection. You would have a number og boards inspected by an expert and then you would have several operators inspect the same boards. When the operators find the same mistakes, they are good operators.

This method can be uses for your AOI machine. Just consider it an operator! Wou will find a pretty good explanation on Attribute Gage R&R here

europe.isixsigma.com/library/content/c070516b.asp

I also can recommend the book "six sigma memory jogger" which can be bought from amazon. It explains the most important six sigma tools - it includes an explanation of GR&R and attribute GR&R.

With kind regards,

Cornelis