Vision Measurement System - Software Program Qualification


Involved In Discussions
Hi there!
I read posts in Vision Inspection System Validation
and I have a question, if I may.

I am in a similar situation whereas, I have a Vertex 251 vision system (already validated IQ-OQ-PQ), and I am writing programs for measuring parts (medical device components).

Agreed - programs need to be validated. Can anyone recommend the type of validation/verification/qualification for each program?
We have hundreds of parts (different colors, shapes and sizes) so I dread a full GRR or even a t-test comparing with a manual MicroVu for each... any suggestion? Is a t-test be acceptable?, since the human factor is very much reduced in these tests.
Can I use a generic qualification for groups (parts that are of similar features)? or generic qualification of a feature like "taper length" etc.?
(The appropriate will be to qualify each program due to light differences, part-color differences, - but too time consuming)



Looking for Reality
Trusted Information Resource
Hi Sue,

thanks for posting in public.

by PM, i mentioned a "every feature we ever measure is on this part" GRR, and a program validation per program. ..but i dont know if that flies in medical.

you reference "manual MicroVu" not sure what that it the operator using the tools manually outside of running a program?
If so, i would group your products (measurements, really) into od ranges, length ranges, id ranges, etc. And run grr on each of these....shouldn't be more than about 16....way better than hundreds...


Involved In Discussions
Hi Ninja,
By "manual microvu" I mean a different microvu machine which is manual type (not programmable). With this machine we have years of experience, and it is a "gold standard" to us, so if I compare measurements with both machines and they "agree" (t-test or alike), I can qualify the program on the "programmable machine".

Does it make sense?


Looking for Reality
Trusted Information Resource
makes sense to me....

the big question is whether it makes sense to your customers or other regulatory bodies...

The only "gold standard" generally accepted that I am aware of is NIST.

What was the validation you used on your manual machine that customers signed off on?
Perhaps build your verification around that.
I can't believe you did 100's of GRR's on the manual machine.


Involved In Discussions
:notme: GRR on the manual machine...
I am not talking about hardware-software validation. This was already done with all bells and whistles.
I'm referring to specific parts or features like taper length, radius etc that I have to program a test for. I am not aware of NIST references to use...
Can you clarify?




I also work for a Medical company and we have several Micro Vu of the series 311HC and 410, would you like to share information, I need some thoughts on how to program a rutine and I would love to have your help on it.

Also, currently Im running a Gage R&R for our production lines.

PLease drop me a message and I will also see how to assist you with recommedations.


Involved In Discussions
Hi Elbujo,
I think my plan works. Although no auditor really looked at it, but it makes statistical sense to me.
First, I validated the software and hardware (IQ, OQ, PQ), I included orthogonal regression model (
Once that validation was done, I had to validate "programs" (routines)...that is the hard part. I did a GRR using a part that can represent a group of products (similar sizes and similar features, using same program steps).
Then I wrote programs to all parts that are in the group - I ECOed each program and attached the GRR study to the ECO as justification and explained that this "program uses the same routines and the parts are similar, so the GRR of another part is applicable" . Still I had to do a few GRRs to verify different geometrical features, but that methodology saved me a lot of time. To be more safe, I also compared one part's measurements using the program with a results coming from a manual vision system, just to make sure I didnt do a gross mistake.

Hope it helps,
Top Bottom