Machine Vision Verification 1

[CurtainCall]

I'm not sure if this would be Inspection/Automation/Production so I just ended up here:

We have an Optical Vision System (Monochrome image based on previously programmed target) to find a specific part on a PCB and then perform a function (underfill).

This issue I have seems to be colour variability in my PCB stock, however, being new to the vision inspection circuit I may be missing some factors.

Since this is a Optical Vision System these are the factors I've controlled for:
Lighting (Brightness and location)
Surface Condition of PCB
Surface Condition of Mounting Jig
Cleanliness of Camera Lens

However, my system is still inconsistent and locates the target fine on some PCBs but not on others.

Some factors I haven't been able to control:
Sample Size - The product is high-margin, batch production, I have a limited number of samples to program the machine.
High Margin Product - Test samples are mostly "rejected" or "faulty" PCBs.
Batch Production - Test samples are often from Batches spanning years and possibly different suppliers.

Good news is that my success-rate with our latest batch seems to be >75% which is much better than the <50% of my original test samples.

I'll have another batch of boards to test in a couple days, but I wanted to tap some brains and see if there is another factor I've been missing, or if there is anything else I might have overlooked.

Thanks for the help!

 

[tygerdawg]

Many years of suffering through vision apps taught me lighting was my most variable criteria. Once I had a cyclic false positive that was traced to the light of a machine beacon (40 ft away) getting into my workcell. That 1 Hz flashing light hosed up my vision system. Control lighting, filter it, polarize it, transform background light, ...all types of things can be done.

If lighting is controlled, then you may want to think about a more robust inspection algorithm. Simple canned feature recognition routines may not be sufficient. Multiple features, checking background colors and making adjustments, redundant checks........

TygerDawg
Blue Technik LLC
Virtuoso Robotics Engineering

http://www.bluetechnik.com

 

[CurtainCall]

Thanks tygerdawg!

You seem pretty active on this forum and I appreciate both you assistance to the community and helping me personally.

I'll see if there is anything else I can do to control for the light issues, but I have a feeling you are right on the detection algorithm, I'll need something a bit more adaptive/complex/robust.

But in the mean time I'll have to try to control the light - maybe I'll just black-out the machine completely

If anyone else has any other ideas I'd be glad to hear them!

 

[PSE]

Agree with Tygerdawg. Filter or illumination selection can enhance contrast for your camera. A potential illumination trick if your equipment is up to it is to adjust illumination timing so that is asynchronous with ambient lighting (works with fluorescent type ambient illumination). It can allow you to capture images when the ambient light is effectively "off". As already indicated, control the lighting environment best you can and then begin honing the detection algorithms.

Regards,

Bruce Youngman

http://www.dynamicsolutions-mi.com