Vision System help

[Cam 123 0]

Hello Members;

 

   I require some assistance in building a vision system for assessing product defect. I have received a lot of biased & confusing recommendations so far from sale people.

I am considering a cost effective vision system for a bottling line that utilizes HDPE, PET bottles ranging in size from ( W X H) 1.5" x  2.5" to 4" x  7" ( 50cc to 400cc)  The line speed ranges form 30-80 bottles/min. The conveyor is 4&1/4"  wide with bottle positioned  approximately in the  middle of conveyor , thus could place the camera from 2" to 8" away from the bottle in order to capture the label in the field of vision.

 

 Would like to be able to 1) detect presence/absence  of label and lot/exp ( lot/exp are printed by a coder either on the label itself  in a predetermined space or on the shoulder of the bottle) , 2) label alignment ( crooked label)  3)  lid/cap presence on bottle, and lid/cap alignment.  I have been told that depending how accurate I want to detect a crooked label or  cross threaded cap would require a much higher resolution camera. I think 5 pixels is approx. 1.3mm. Is that to precise to detect. if so, could accept 2-3mm label misalignment.

 

3) If it could detect ripped, damaged or wrinkled labels that would an added bonus but don't want to add this feature if it greatly escalates the cost.

 

If  above defects are detected, would like to reject from conveyor via some sort of pneumatic blast of air. I think in some systems a PLC is not required and one can have the signal directly from the vision system of "NG" product trigger a 24v response sent to the pneumatic valve close to the camera, thus perhaps not requiring a timer or encoder?

 

Would like the system to be easy to set-up and change for different products. 

 

Is it feasible to set-up a in-expensive system to do this? There are many used systems and components on ebay at greatly reduced cost. Each Company ( Omron --ZFX c20/25, FLV, Keyence IV-H series,  Cognex Insight-2000 or checker 4G system) looks to have all-in-one models that have very easy user interfaces, are easy to install and set up as there is automation in the software for focusing and lighting etc.... Just not sure if they are capable of achieving the above  defect screening. I am open to more advanced systems if they are not to difficult to set-up and use. If you could give me some recommendations of model(s) controller, cam, light source that  would be capable of achieve the above would be greatly appreciate

 

Thank-you for your help

Cam 123

[kaiser_will 65]

Define "inexpensive".  Cost is relative.

If you want the vision camera's decision making to also control reject control, you may need a PLC.  Depending on the business constraints of "what if" an incorrect product makes it out the factory door, the camera's "Go or No Go" decision will most likely not be adequate for controlling your business constraints.

Is it feasible?  Yes.  You will need a PC plugged into the vision system, at least during debug phase, so plan accordingly.

Consider how the camera will be told to process an image.  Is your conveyor line continuous motion, or indexed?  If continuous motion, you may need a cam signal or a photoswitch to trigger the camera to take and process an image.

Will your process run (1) product or multiple products?  If multiple products are to be run, you will have to consider a vision systems that supports multiple setups and your line controller will have to instruct and interrogate the vision system (command and feedback).

Many PLCs or controllers, like Fanuc robot controllers, already have integration configured to network in a vision controller.  Cognex is the big dog in the market.  Mitsubishi has a PDF for integrating a Cognex vision system in with one of their PLCs.  https://www.cognex.com/support/downloads/ns/1/11/93/Vision%20System%20Connection%20Guide%20(COGNEX%20SLMP%20Connection).pdf

Overall, most vision systems look for pixels (i.e., Blob tool) or color.  Users configure identifying characteristics for the vision system to look for.

The 3 biggest factors that affect performance of a vision system:  (a) lighting [one may have to add lighting to insure the vision system sees a repeatable light level], (b) product presentation control [labels on rotating product may require 360-degree camera or multiple cameras or product escapement to control product placement for inspection], (c) speed [perform due diligence to understand the limitations of all parts of the process]

[HudsonT 0]

So kind of similar scenario, I am using a insight 2000 camera to scan pads and I need it to detect a very minor difference in two pads. We have it set up to set off an alarm if a pad goes un-punched under the camera. But this pad isn't supposed to get punched, yet gets picked up by the camera/program regardless. What would be the best way to try to differentiate them? I have tried to mess with the exposure and lighting, but nothing seems to work.