Intelligent Vision Stops Bypass of Quality Control
|Other articles in the May 2009 Control Engineering North American print edition supplement:
– 415 Parts Seen
– Motion and Vision Combine to Detect Flaws
– The New Look of Facial Recognition
Operators sometimes employ clever and often low-tech strategies for bypassing a vision-based label quality-control system. Sometimes they simply shut off the system and let products go through. Other times, they place an image of a good product in front of the camera to allow the lesser quality labels to go by. Makers of a global brand of personal care products recently found some of its own operators were breaking the rules this way.
The company asked CI Vision (Aurora, IL), for a new vision inspection system with a mechanism to prevent plant staff from bypassing it. CI Vision’s “Anti-Vision Circumvention Technology” turned out to be a useful feature of its Pro Series inspection system.
The base model of the Pro Series system has two cameras in steel housings connected via the GigE Vision camera interface standard. Up to four cameras can be installed, and the vision system is controlled either by a custom-built PC or Matrox 4Sight X industrial PC. The CIVCore software that controls the image acquisition and inspection tasks is built from the Matrox Imaging Library (MIL).
|CI Vision’s label image acquisition and inspection software is built from the Matrox Imaging Library|
To inspect a product batch, an operator sets up the inspection type on a touchscreen. When an image is acquired, normalized grayscale correlation is used to determine the presence or absence of the label itself, as well as the location of the label’s region of interest (a bar or matrix code, or a lot number). Code 128 bar and two-dimensional data matrix codes are read by MIL’s code module for correctness. The CIVCore’s OCV module, which is built from MIL’s pattern matching and image processing modules, reads text-based lot numbers and label control numbers.
The Pro Series system features an optional tool that detects if the same label appears on the front and back of the product. The system uses the search model from the presence/absence feature to determine which label is in front of Camera 1. With that result, the system automatically instructs Camera 2 to look for the other label. If the second camera fails to find it, the system concludes that the product has the same label on both sides.
If a product fails any inspection, it gets flagged and diverted off the line. But, as this manufacturer had discovered, sometimes operators let non-conforming labels go through. Rick Koval, lead software engineer at CI Vision, implemented a three-part strategy to ensure the system inspects 100% of the products:
Step one . Make sure that the vision system operated with other automation equipment at all times. When the labeler is running, the vision system must also be running.
Step two . During operation, use image acquisition trigger inputs as a mechanism for determining the product’s presence so it can be inspected. If a certain number of seconds pass without a trigger while the labeler’s ‘run’ signal is detected, the vision system faults and outputs a signal to shut down the line. Similarly, when there is no ‘run’ signal and the vision system receives triggers, the labeler also signals to shutdown the line.
Step three . To detect fake images, check for movement between acquired images by comparing translation results. Locations are compared from image to image over the sequence of acquired images. If a certain number of images indicate no movement between them, the current part and all future parts will fail.