MSA For Solder Paste Inspection Machines

All,
This is my first post and I've done alot of reading through these forums and posts for something even remotely similar. Sorry if I missed anything.

Anyways, here is my situation. When trying to verify the capability of an inspection machine the customers typically run their own GRR test (using Minitab) on their own products.

Most inspection machines have a capability of running a pseudo-GRR test. In this case the machine would do the following for a 3x3 test:

Load Board
Align the system
Inspect the board
Align the system (without unloading)
Inspect the board
Align the system (without unloading)
Inspect the board
Unload the board
Repeat 2 more times.

The issue I have is that these boards typically have hundreds if not thousands of data points on them. I can usually suggest that the customer select only 10 points and use those for the test. However, this can still yield poor results because:
1. The machine is meaasuring the capability of another machine process (solder paste printing - which has significant variation)
2. The assumption is that there should be very little variation in the process or the items being measured - there is sometimes significant variation.

When reviewing the GRR results you will very often see that the % contribution to the variation from the inspection machine (measuring device) being <3% - thus the large part of the variation is due to the part to part variation.

However, when you look at the values that everyone is instructed to look at (% Study Variation) this can look really bad (3% variation contribution turns into a total system GRR of 25%).

I then dive deeper into the readings and run a normal capability test on each individual part and the Cp's are well over 12 (using the same tolerances)

All that being said (sorry for the dissertation).
My questions are:
1. Is this a proper test at all for measuring Machine Capability for this type of machine?
2. Are there more appropriate settings or measurement techniques for measuring the machine repeatability (which is really all I'm trying to measure).

If you've made it this far into this post - thanks! and I hope the answers help others.

-Joe

Can you attached some actual study results in Excel?

Sure...
Here is a simple data set the shows the above...

Can you explain what is in each column?

For example, are you measuring 34 different pads on component U7 that are numbered 659 through 692?

Are Width and Height nominal dimensions?

Are the remaining columns the actual measurements? What are the units and the specifications?

You are correct... I am indeed measuring 34 different pads on component U7 that are numbered 659 through 692.

The Width and Height are nominal dimensions in microns and these combined with a common thickness of 100 microns yield the target volume of each pad.

The volume column is the one of concern and is measured in microns3 (cubed)

The Specifications used are +/- 50% of the target. (Which is pretty wide, but it is the starting point)

Thanks
-Joe

I ran the R&R in Minitab. If you do not have Minitab, let me know , and I can copy it into Word and repost it.

Assuming that I correctly understood how the study was run and the data structured, your automated inspection machine is acceptable for inspection to the +/- 50% tolerances (GRR (%Tol) = 8.21%), but is not acceptable for use as a process control device (GRR (%SV)= 35.43% and ndc = 3).

This also means that the device is unsuitable for machine capability studies.

Thanks,
While those are similar results that I obtained using Minitab - that was also my dilemma.

Looking at Wheeler's "Honest Gage Study" and plugging in the same results comes up with a completely different story. This is where I am struggling. His theory essentially takes out the excessive "noise" captured in the measured results because it contributes too heavily on the machine and, based on his study, ranks the inspection much higher.

See attached.

Wheeler's approach is completely different from the standard AIAG approach. If your customer is willing to consider the Wheeler approach, I recommend that you go with it. It is based on sound theory and works in the real world.

The AIAG approach, though very widely used, has some major flaws in it (such as adding standard deviations). One of the drawbacks of this approach is that it tends to reject acceptable gauges.

Thanks,
That is what I figured.

Hi guys

I am looking at your file "for ELSMAR GRR VOLUME.xlsx"
Can you explain it? I need also to perform a gauge R and R to and SPI machine.
You’ve tested 1 board that contains components U7 numerous times? How many times did you inspect each pad? Can you explain to me the process?

thanks a lot!!!
Alon