Vision System Part Rejects

[Bob O]

I have a question for those that use or integrate vision systems. What is your acceptable % of a good part being rejected? I am normally in the range of 0.0% - 0.2% and always looking to make it better but I have had people tell me that if it is <= 1% they do not even worry about it. Thanks, Bob Edited April 20, 2006 by Bob O

[IO_Rack]

Sorry to elaborate so much, I didn't read that correctly the first time.

[Smoke]

I would say that this % number must be decided by the process and plant standard. We have integrated many vision systems e.g. Cognex, DVT, Keyence, Acuity. There are many lessons learned. As IO_Rack has mentioned. Also fixturing. If the test object is closer or further from the lens the size of the test object shrinks or grows. Different visions systems handle light changes better than others. We have gone as far as to rotate or slide the test object into an enclosure where the ambient light is constant. I don't know if your application is close to an operator, but if the operator is wearing a white shirt one day and a black the next it can change everything. To get back to your question, in our facility if there are 4 good parts scraped in a shift the process is considered down until fixed. This is at a production rate of 300 to 1000 part per hour depending on the process. I vision is very capable of these numbers.

[ianbuckley]

Each application will vary slightly on what constitutes acceptable reject levels (not rejected bad, rejected - good). Never heard of anyone complaining about inspection system for rejecting bad parts - manufacturing process maybe, but not the inspection doing its job. It all depends on what the consequences are for bad parts getting through to the customer. If it is a matter of personal safety (could someone get hurt or die?), then obviously the system will be designed to never possibly let a bad part through - even if it means a lot of rejected good parts. You do what you can from lighting and presentation to minimize rejected good parts, but anything marginal is tossed. If the bad part is easily replaceable with no real harm done, and each part is expensive to manufacture, then you aim to avoid rejecting good parts.

[Bob O]

Sorry I have not replied sooner. It is nice to get confirmation and hear what others are striving for reject wise. I just check a line from home, I am at .01%, total fail, and this is on a problematic label from the factory. Here is a link to what we are manufacturing on this line. http://www.bluebunny.com/WhatsNewInnovations.aspx I am inspecting for proper job number on the lid and container. The lid travels on a flat conveyor and does rotate in the horizontal plane a bit so I have to track that which isn't to big of an issue. The biggest problem with these is if the print registration is off. I have to do certain things in the product file to account for the worst case and then all is well. The container travels upside down and rotates causing the job number to shrink and grow depending on its position when inspected. This really is not too big of an issue any more since I believe I figure out the tricks for it. Now I need to investigate inspecting the whole container for imperfections. $$$$$$$$ Edit..Sorry about the link. My links never seem to work when I insert them. Edited April 22, 2006 by Bob O

[BobLfoot]

You might want to check out a company I did my co-op senior thesis for in 1985. They were doing glass bottle whole container inspections then and still are now. Their technology might have a cross-over application.

[rookie]

I am a sales engineer for a Canadian vision systems integrator. We inspect everything from automotive castings to pharmaceuticals, and have been doing this for 11 years. A vision system if properly configured should be capable of storing your images for offline reviewing. Typically, we will collect images (good and bad parts) in seperate folders, and run the images through an offline inspection mode. Since you are attempting to read a printed code, you should not be expecting to adjust your thresholds for pass or fail. Since your part rotation is uncertain, you could find an anchor point for correcting the orientation offsets, and then perform the code inspection. Mechanically you could try to crowd the part to minimize the skew, and then perform your inspection. I assume that you only need to correct for x, and y, and that the z value is constant. Proper illumination is key to code reading and label verification. What type of vision sensor are you using? What type of lighting? Is the lighting strobed or constantly on? Do you have capabilities to store images? (for this type of inspection, about 1,000 sample images would be a good baseline to fine tune your sensor.) The question now becomes..."What is the acceptable level of false rejects vs false passes."