Vision systems/ Visual ID

[Madbrace]

I just started a new project and need some advise-Ideas. I work for an automotive parts remanufacture. The project I am working on consists of a system that will identify each part-core that comes in by part #. We have over 1600 part #s. I am currently working on a trial-Demo vision system that I just found out is not capable of processing 1600 unique parts.

The Idea is to basically have this part-core pass by a set of cameras and be identified by part # in a matter of seconds. Anyone have any experience with vision systems? Any good companies too look at?

Thanks

 

[tygerdawg]

I have a lot of painful experience with vision systems. It all depends on LENSING & LIGHTING. Modern vision systems are powerful and user-friendly.

To positively identify 1600+ separate items, you'll have to have a powerful system that can store 1600+ separate templates upon which to compare. This probably is not realistic because of the cycle time involved.

Another approach is to develop a decision algorithm based upon features of the parts or families of features.

For that many parts, you should try to put on some sort of identifier. A series of dots or holes, barcodes, something that will make the job of implementing a vision system easier.

TygerDawg

 

[Madbrace]

Tops one minute. These are cores so we do not have any info on the parts coming in. We currently have guys id the parts by tribal knowledge that have years of experience. Even then they can take up to a couple of minutes on the parts they cannot id, they process about 2000 parts each for a total of 20,000.

 

[tygerdawg]

"Tops one minute." Does that mean you are dsyelxic? Sorry, couldn't resist after my day today.

Camera recognition via a template means teaching a known good part and storing that info. The vision system would be presented a part, process the image, and compare to your other 1600 stored parts. This would take a lot of memory and juice. Probably not practical, at least in this century.

Hence my idea about creating a algorithmic decision tree...vision system searches through a series of features to identify a part. If the parts have enough good discriminatory features, then it's conceivable this is possible. Just a lot of programming. But if you could do this and set up parallel identification lines, it may work within an acceptable cycle time.

I said "lensing & lighting", I should have added "money" too. I'm sure if you call some vision system integrators, they would start drooling like some of ol' Pavlov's dogs over your project description.

TygerDawg

 

[charlie2go2ubet]

You asked for company names on visual inspecton. Cognex, Keyence, Omron and DVT are ones I have seen most often as canned solutions, but their mark up over cost is significant. They also would have a problem on the 1600 number unless you did it using several cameras/set-ups in a process of ellimination, as mentioned earlier.

http://www.cognex.com/

It is also possible to get visual recognition software and have a programmer develop a custom solution that processes digital images from a digital camera. This requires more time and expertise. A smart programmer might be able to handle the 1600+ features you spoke of earlier given enough resources and time.

VMM's...Visual Measurement Machines are also available. Versions of these machines have edge and image recognision, along with a servo controlled light table. They can sometimes automate the inspection procedure to a degree, but were originally intended as high tech optical comparitors.

http://www.pemstar.com/viewer/services_test-dev_aoi.asp

Best,

Charlie