Engineers who have not been involved in machine vision recently may be surprised at how far the field has progressed. There have been several significant developments in the technology over the last 15 years. Originally, video analysis took place through custom-designed electronic hardware. These early vision systems were fast but expensive and lacked flexibility and sophistication. It was typical for vision hardware on a production line to perform one, and only one, task. Alterations in vision algorithms came only through designing new circuits.
Software and more sophisticated electronics have, of course, dramatically changed matters. Thanks partly to ever more-powerful computing hardware, pattern-recognition algorithms have progressed beyond simple tasks such as determining the presence or absence of a cap on a juice container, to more complicated analysis such as finding a fine crack in a glass beer bottle.
Vision systems are increasingly deployed to guide robots in performing assembly tasks. In addition, they facilitate necessary processes such as SPC and material planning by reading date and lot codes, and bar codes. Cosmetic defects, such as torn labels on packages, are fair game for detection by vision even if they are difficult for the human eye to see. This sensitive level of inspection can be performed at production rates up to 3,000 items/min.
Growth of technology
One thing that hasn't changed is the importance of illumination. Proper lighting on the object to be inspected continues to be a key factor in making a vision system work properly.
The object of lighting for vision is to make defective products stand out somehow from good products. Early lighting schemes accomplished this goal but had some well-known drawbacks. High-voltage xenon gas strobes, for example, had a reputation for inconsistent light output. They also needed to be relamped periodically and were a safety concern. Also used were fiber-optic lighting with quartz halogen sources. The source lamps degraded over time, required frequent relamping and the fiber-optic device that projected the light on the part being inspected was expensive.
Advances in solid-state light-emitting diodes (LEDs) now let them serve as illumination for machine-vision cameras. This new light source has a consistent light output, operates from low voltage, and can be controlled relatively easily. It also is relatively inexpensive and lasts a long time compared to the older technologies.
The use of LEDs in machine vision has opened up many new applications. Several of these were previously unworkable or prohibitively expensive when xenon strobes or fiber optics were the only options. LED lighting can be configured to project light in an almost infinite number of ways. Vision manufacturers can now design LED lighting systems specific to the product being inspected and can control the light automatically through software.
Manufacturers of vision systems combine lighting schemes with electronics and cameras to produce systems that are turnkey. It is no secret that the ever-increasing power of the desktop PC has dramatically changed machine vision, as it has other industries. These now powerful computers, and lower prices, have been instrumental in making vision inspection useful and economical. PCs can now run sophisticated and robust algorithms for video-image analysis. For new inspection capabilities, one merely loads new software.
Costs have also come down in camera technology. A recent advance in this area is something called a smart camera. Smart cameras have a small CPU chip built in. Typically a user develops an inspection task with the camera attached to a desktop PC. Then the inspection gets downloaded to the processor in the smart camera. The camera can then operate independently of the PC. Multiple cameras can be set up this way.
However, this low-cost approach has limits. Inspection tasks must not be too fast or complex. Typically these low-end systems are used to read bar codes and other simple tasks.
The combination of more powerful algorithms and lower pricing has let vision systems replace humans for redundant tasks such as inspecting pharmaceutical caps on a conveyor. Further, vision systems combined with robots have taken over repetitive assembly tasks such as building up automobile axle bearing components. These heavy metal parts are moved into place with vision-guided robots rather than by human operators who manipulate them into place by hand.
Another relatively recent development in vision is the introduction of colorful graphic touchscreen controls. These take care of certain setup details so a high-schooleducated production operator can key in vision-inspection parameters that include lighting intensity, camera and lens settings. It generally takes operators only about 5 min to tap in these settings for a typical vision task.
Once the inspection is running, the operator can use the terminal to make sensitivity changes on the fly. The best vision systems let the operator test proposed sensitivity changes against a "challenge set" of images saved in system memory. This is a collection of video images of good parts and defective parts.
Only if the vision system can discern good parts from bad with the new settings will the sensitivity change be accepted. All this activity takes place without interrupting the current inspection taking place on actual production. This is true multitasking and takes all the risk and uncertainty out of making a change to the vision inspection.
Advances in machine-vision inspection are now being brought to bear on some of the more challenging inspection tasks. Several of these are in consumer packaging that often proceeds at very-high production rates. For example, cap-making machinery for various beverage containers can turn out product at close to 3,000 caps/min.
Real-time operating systems coupled with inspection algorithms optimized for high speed and specialized high-framerate video cameras can handle such jobs. At these high speeds, part tracking becomes a critical element; the vision system must be able to successfully reject the faulty product as it enters the reject station. Unfortunately no off-the-shelf hardware has the capability to track parts at these high speeds. So manufacturers such as Applied Vision have developed custom hardware that can track parts at rates of 3,000 parts/min or higher.
The inspection of metal packaging constitutes another challenge. Aluminum beverage containers delight the eye with their highly reflective surfaces. For inspection purposes the task becomes one of inspecting irregularly shaped mirrored surfaces at rates close to 2,700/min. Here proper scene lighting is the key. The illumination system must show up tiny flaws without generating glare or overly dark areas.
In vision applications, as in other engineering endeavors, complexity becomes evident as one examines tasks in detail. This is why plans for vision installations should have as much detail as possible, particularly with regard to the objects the system will view. Though it may seem like an obvious mistake, customers often seek quotes for such systems before they've fully defined the task to be accomplished. Finger pointing and frustration are a frequent result.
Take as an example a case where a food producer wants to inspect the inside of a jelly jar for foreign contaminants before fill. The components used in the vision system will depend on factors that include whether or not all the jars are the same size, have the same shape, or will be clear glass. The size of the contaminants to be rejected will also be important. With advance planning, vision systems inspecting for contaminants can also find glass-manufacturing defects as well. Engineers defining the system will need to decide how quickly job changes must be carried out and whether the system will be washed down with water.
It would be easy to think up a dozen more questions for this particular application. A point to note is that the answers to some of these questions, such as size variations of jelly jars to be inspected and whether the system will be in a water wash-down area, can change the price of the system by a factor of two or more.
Finally, it is interesting to examine the cost range of vision systems now on the market. A simple verification of a bar code with a smart camera and offthe-shelf lighting can sell for $6,000.
This system may mount easily to a conveyor with hardware included in the kit. A multicamera system with specialized material handling for glass containers in a water washdown environment can exceed $200,000 and will take production offline for several days during installation.
Experience is a great asset when fielding such systems. A knowledgeable vision-system salesperson can save time and money. But whether you buy vision components and develop a solution, or bring in a turnkey system that includes material handling, don't skimp on training for the factory-floor personnel who will be operators.
Those operating a complex vision system might need a one-week class offsite. Taking production people off the line and sending them away to training sounds expensive. But the alternative is to have an automatic device mounted to your production line which is incorrectly rejecting good product or unable to reject true defects. And none of your crew will know how to fix it.