Quality Scan: Machine Vision Advances Improve Quality
In many industries, machine vision has taken hold at manufacturing plants, reliably ensuring product quality without the manufacturing process missing a beat. Better algorithms have made it more tolerant of environmental conditions, easier to program, and capable of communicating to the statistical process control systems that can use the quality data machine vision provides.
Thanks to improved pattern recognition algorithms, machine vision can reliably and quickly locate parts on the production line, despite the environmental changes that occur in the real world. Small changes—such as the color of the packaging—can have a significant impact on how the image is seen by a vision system. A shiny surface can change from a light-part-against-dark-background image to just the opposite when lighting conditions barely noticeable to the human eye occur. Even a replacing an old belt on the conveyor with a newer, brighter one can result in needing to tweak the pass/fail thresholds in a vision system. And moving the vision system relative to the production line or part can change the size that the part appears to be in the image. But newer, more robust, geometric pattern recognition algorithms improve the vision system's ability to reliably find parts even under challenging, changing conditions. Being able to quickly find the part is also key to adopting machine vision without sacrificing line speed. Better pattern recognition algorithms not only find parts reliably, but also exhibit less variability in the time it takes to provide an answer. Lesser algorithms keep searching for too long when there is nothing to find.
Another software technique making machine vision more practical and accurate on the factory floor is nonlinear calibration. Using this technique, when setting up the vision system, allows the software to correct for distortion in the image before locating or gaging the part. Benefits include measuring part tolerances more accurately and providing robots with more accurate positional data. Distortion is typically caused by one of two physical constraints. When physical space constraints require the vision system to be located very close to the part, but it still needs to see the entire part, a short-focal-length lens is typically used, resulting in a fisheye type of distortion. Perspective distortion is where the image looks like a trapezoid rather than a rectangle because the camera had to be mounted at an angle to avoid overhead equipment. Previously, machine vision was unusable for gaging parts accurately in these physical conditions. But now, with nonlinear calibration techniques, a distorted image can be "fixed" in software, allowing the vision system the flexibility to be mounted at up to a 45° angle without sacrificing accuracy.
Adopting machine vision is often the most cost-effective way to add 100% inspection to a production line and maximize line speed. Other choices require the part to stop in an inspection station. With more robust algorithms and improvements in imager technology to eliminate the blur of parts in motion, 100% part inspection can be achieved in a relatively modest amount of line space and without slowing the production process.
As manufacturers strive to make their production lines more flexible, using machine vision to find or inspect parts can also save money. Rather than using mechanical fixtures made especially for each part type the production line will encounter, machine vision saves capital expense by allowing the parts to be completely unfixtured on a conveyor or in a bin, without sacrificing the ability of a pick-and-place robot to quickly retrieve the part. Have the vision system inspect the part at the same time it is located, and a quality check has been added to the line with no additional equipment expense.
But perhaps most importantly, machine vision can provide a richer data set than simple pass/fail results. In conjunction with good statistical process control software, machine vision helps diagnose process failures with reason-for-reject data. To a machine operator trying to understand why parts are being rejected from the line, a picture can be worth a thousand words. And using the bandwidth provided by Ethernet combined with built-in OPC and ActiveX controls, vision systems can effectively inform the machine operator, using the same HMI interface that the rest of the control system is using. Additionally, the ability to communicate to a PLC using preconfigured industry-standard protocols such as Ethernet/IP, Profinet and Modbus TCP provides tighter integration between the vision system and the control system, while reducing setup time.
This article was first published in the July 2008 edition of Manufacturing Engineering magazine.
Published Date : 7/1/2008